A pragmatic randomised controlled trial to compare antidepressants with a community-based psychosocial intervention for the treatment of women with postnatal depression: the RESPOND trial

Classification and aetiology of postnatal depression

Postnatal depression is defined as a non-psychotic depressive episode meeting standardised diagnostic criteria for a minor or major depressive disorder, beginning in or extending into the postnatal period. 7 In the two major classification systems, International Classification of Diseases version 10 (ICD-10) and the Diagnostic and Statistical Manual of Mental Disorders, fourth edition (DSM-IV), ICD-10 categorises mental disorders that occur postpartum as ‘puerperal’, but only if they cannot otherwise be classified, whereas DSM-IV allows ‘postpartum onset’ to be specified for mood disorders starting within 4 weeks of delivery. The WHO Guide to Mental and Neurological Health in Primary Care (also known as ICD-10 PHC) does classify postnatal disorders separately (F53), and offers guidance concerning diagnosis and management to primary care health professionals in the UK. 8 There is some evidence for the specificity of the diagnostic concept – one study has found that women who experienced PND with no previous episodes of depression were at increased risk of further episodes of PND but not depressive episodes outside the perinatal time frame, whereas women who experienced PND as a recurrence of a depressive illness were at increased risk of repeated PND as well as the development of depression outside the perinatal time frame. 9

The aetiology of postpartum depression remains unclear with little evidence to support a biological basis,10,11 though a small experimental study reported that stimulating hormone changes after delivery led to a significant change in mood in five of eight women with a previous history of PND, compared with none of eight control women, suggesting differential sensitivity to hormone changes. 12 Premenstrual dysphoric disorder has also been reported as a risk factor for PND13 and is suggestive of a risk of hormonal sensitivity in some women. However, despite considerable research, no single causative factor has been isolated and a multifactorial aetiology has been suggested. Consistent findings suggest the importance of psychosocial variables. 11,14 In particular, stressful life events,15,16 marital conflict,15–18 a past history of depression, younger maternal age19 and the lack of social support10,15,17,20,21 have been found to significantly increase the risk of postpartum depression.

There are still unanswered questions as to whether there is an increased relative risk of depression within a given time after the birth of a child, what these time limits might be, whether there are any characteristic clinical features that distinguish PND from any other kind of depression, and whether aside from the coincidence with childbirth, there are any other pathognomonic aetiological factors associated with it. Women with PND characteristically exhibit symptoms of uneasiness, irritability, confusion and forgetfulness, anhedonia, fatigue, insomnia, anxiety, guilt, inability to cope and thoughts of suicide. Frequently exacerbating these symptoms are low self-esteem, lack of confidence and unrealistic expectations of motherhood. It can be seen to be very similar to depression occurring at other times in a woman’s life. The most recent controversy surrounding PND concerns the timing of its onset and the question as to whether it is an extension of antenatal depression or a totally different entity. 22

Although the range of mental-health problems occurring in the postnatal period is not dissimilar to those affecting all adults, the nature, treatment and impact are different in several ways. When considering the diagnosis and treatment of postnatal mental-health problems, there is more than one patient to consider. A severe depressive illness after the birth may place the infant as well as the mother at risk. Lengthy PND may have long-term adverse effects on the child’s development and the marital relationship. Follow-up studies have found that approximately one-third of cases of PND become chronic or recurrent. 23 Women who have suffered from postpartum depression are twice as likely to experience future episodes of depression over a 5-year period. 9 The last Confidential Enquiry into Maternal and Child Health (CEMACH) report24 identified one of the leading causes of maternal death as suicide, with more than half of the women who died having an underlying history of mental illness not all of which was depression. Many of these women had regular contact with health-care professionals. There is a tendency in the literature and in practice to use the term ‘postnatal depression’ in a generic sense for all postnatal mental illness. This is not entirely helpful because it leads to other types of mental illness being missed or treated inappropriately. The need to prevent, identify as early as possible and treat such potentially life-threatening conditions optimally is clear.

Antidepressants

The prescribing of antidepressants in the UK over the last few years has increased considerably. 49 There is now good evidence regarding the efficacy of antidepressants compared with placebo for patients with major depression and/or dysthymia in primary care. Their efficacy in minor depression has been poorly studied and results are conflicting. 50 It does however appear that for acute-onset depression, the size of the antidepressant–placebo difference increases with severity of depression and that there is uncertainty about the clinical importance of any benefit in milder depression. The recent HTA-funded THREAD (THREshold for AntiDepressant response) trial51 does provide some evidence in favour of selective serotonin reuptake inhibitors (SSRIs) when combined with supportive care over supportive care alone for mild to moderate depression although this study did not have a placebo so the estimate of benefit includes any placebo response.

The SSRIs have become the first-line choice of antidepressant52 with strong evidence for continuing therapy for at least 6 months to prevent early relapse. 53 In trials comparing antidepressants with a variety of different psychological treatments or in combination with such treatments [e.g. cognitive behavioural therapy (CBT), interpersonal psychotherapy, problem solving53], it appears that there is little difference in efficacy between antidepressants and psychological therapies in mild and moderate depression, with little extra benefit from a combination of the two except possibly in more severe depression. 52 The literature with regard to non-directive counselling shows modest benefit compared with the usual treatment provided by the GP. 53

Although PND is in essence no different from depression at other times in respect of its phenomenology, and therefore in its likely response to antidepressants, there has so far only been one placebo-controlled trial of an SSRI for PND. 54 This found that fluoxetine was significantly more effective than placebo, and after an initial session of counselling, as effective as a full course of CBT. There have been two other trials in this area – one comparing nortriptyline and sertraline, which showed no differences between the tricyclic antidepressant and SSRI,55 and a second comparing paroxetine and CBT. 56

The only special issue to consider is that of breastfeeding. Mothers are often reluctant to take antidepressants because of concerns about transmission in breastmilk or potential side effects. 57 A recent review58 concluded that ‘with the exception of a few cases, no serious adverse events have been reported in infants exposed to antidepressant medications through breastmilk.’ The available evidence suggests that antidepressants are relatively safe for breastfed infants. A pooled analysis of studies found that plasma levels of sertraline are usually undetectable in breastfed infants, with paroxetine levels somewhat higher, but citalopram and fluoxetine produced infant plasma levels above 10% of the maternal plasma level (in 22% and 17%, respectively). 59 There is therefore no prima facie case for excluding breastfeeding women from a treatment trial of SSRIs, even though the British National Formulary60 advises caution in their use. However, fluoxetine should probably not be used in breastfeeding women because of its long half-life and citalopram should be used with caution.

Psychosocial interventions

Many patients, and perhaps especially women, state a preference for non-medical treatments or talking therapies for depression in primary care. A Cochrane review considering the effectiveness and cost-effectiveness of counselling in primary care including eight trials, concluded that counselling offers significantly greater improvement than usual care from a GP in the short term but not in the long term and might not be associated with increased costs. 61 There has been substantially more research on the effectiveness of psychological and psychosocial interventions for PND than the use of antidepressants. The role of psychosocial factors in the aetiology of PND has been clearly demonstrated, and research findings support the view that ‘therapeutic listening’ and extra support may be particularly helpful in depression occurring at this time. The treatments that have been evaluated include various types of psychotherapy including CBT and interpersonal psychotherapy, psychoeducational strategies, psychodynamic therapy, counselling and other less well-defined psychosocial interventions such as social support. 62 A variety of different counselling interventions has been studied. For instance, six, weekly 1-hour counselling visits from Child Health Care clinic nurses were found to be more effective in relieving PND than the usual primary care. 63 However, this was a very small trial with only 20 women randomised to receive counselling, commencing about 4 months postpartum and the outcome was measured 1 week after the final counselling session. A psychoeducational group intervention was more effective than routine primary care in reducing depressive symptoms but had no impact on psychosocial outcomes. 64 However, again, the sample size was small, 23 women randomised to the psychoeducational groups, and CBT only comprised one part of the intervention. A Cochrane review on caregiver support for PND included two trials (137 women) and concluded that professional and/or social support may help in the treatment of PND. 65 The provision of counselling by the voluntary sector may also be effective in alleviating symptoms of PND. 66

Health visitors (HVs) are the health-care professionals best placed to offer psychological support for women with PND in the UK because they are in regular contact with women throughout the first postnatal year. Corney found that HVs can help alleviate depression by visiting clients frequently and encouraging the expression of feelings. 67 Hennessy,68 however, reported that a HV had recognised only 27% of mothers she identified as depressed, and Briscoe and Williams observed that HVs should be given opportunities to develop their counselling skills. 69 The first controlled trial of counselling by HVs for women with PND in general practice took place in the early 1990s. 70 It is this non-directive counselling intervention that has been widely taken up by HVs as the model most likely to be implementable in the service setting. However, with the recent reduction in the number of HVs, their associated increased workload, particularly in terms of child protection, and the reduced use of routine screening with EPDS for PND following the report from the National Screening Committee,41 the availability of health visitor non-directive counselling is not as widespread as might be anticipated.

A recent Cochrane review on psychological and psychosocial interventions for treating PND62 included four trials that evaluated non-directive counselling,63,70,71,72 three in the UK and one in Sweden. The intervention in all four trials was provided by trained HVs/nurses. The earliest of these studies was the first to report the potential effectiveness of non-directive counselling for PND, although the trial was rather small, 26 women randomised to counselling, and the final outcome was measured very soon after the intervention ended. 70 Overall, these interventions were effective for the depressive symptomatology in postpartum women, suggesting that this treatment modality may be an important option for mothers with mild to moderate postpartum depression but the methodological quality of these trials was rather poor, particularly in terms of concealment of allocation to groups. These trials also demonstrated the feasibility of population-based screening and the application of home visiting using trained health professionals.

Non-directive client-centred counselling is a form of counselling that is based on Rogerian principles. 73 The understanding is that in many situations, people can resolve their own problems without being provided with a solution by the counsellor. In particular, the counsellor’s role is to listen to people, be empathic with them, encouraging the person to express their feelings but not suggesting what decision the person should make. By listening and reflecting back what the person reveals to them, the counsellor helps them to explore and understand their feelings. With this understanding, the person is able to make the decision that is best for them. It is the listening quality that has resulted in the synonym ‘listening visits’ for non-directive counselling when offered by health professionals. A qualitative study of women who had undergone listening visits revealed that they felt positive about the intervention if they had a good prior relationship with their HV but were less positive if that relationship was poor. Women perceived the intervention as supportive rather than therapeutic and ascribed their recovery to other factors. 74

Study population

All eligible women were recruited between January 2005 and August 2007, from 77 collaborating practices in the UK located in Bristol (21 practices), London (Croydon and Bromley) (21 practices) and Manchester (35 practices). These practices served a wide range of neighbourhoods including both affluent and socioeconomically deprived urban areas. All 18-week follow-ups were completed by March 2008.

Eligibility criteria for screening phase

Inclusion Recently-delivered women aged 18 years or over; who had a live birth; and were living with their baby.

Exclusion Women who had a stillbirth or neonatal death; whose baby was more than 26 weeks old, whose baby had been fostered or adopted; those with psychosis; alcohol or drug abuse; or those already receiving treatment for depression.

Initial postal screening

General practices receive two notifications when a baby is born to a woman registered at the practice: from the hospital where the baby was born and from the local PCT. During the recruitment phase of the study, these notifications were collated by a clerical assistant at each collaborating practice, and reviewed by the PHV to exclude women who did not fulfil the eligibility criteria.

Eligible women were then sent an invitation to participate in RESPOND when their baby was 5–6 weeks old. This invitation pack was prepared by the research team but completed and sent from the general practice by practice staff, and included information about the RESPOND study (see Appendix 2), a consent form (see Appendix 3), a screening EPDS75 questionnaire, and a prepaid envelope for return to the study team.

Women were sent a reminder letter and further screening EPDS if they had not replied after 2 weeks. If a woman returned a signed consent form and scored 11 or more on this first EPDS, her GP was asked to review the exclusion criteria and exclude her if she was not eligible to participate further. All eligible, consenting women who had scored 11 or more on the screening EPDS were invited to participate in a home visit to further assess their eligibility for the trial. If a woman was ineligible, her GP and PHV were informed of her initial EPDS score. Trial recruitment procedures are summarised in Figure 1.

FIGURE 1.

Summary of RESPOND recruitment procedures. Note: If suicide ideation was reported at any stage in this process the self-harm protocol was implemented (see Appendix 4).

Other recruitment methods: referral by health professionals

Although postal screening was intended as the primary route for entry into the study, collaborating GPs and PHVs were also able to refer women who became depressed between 6 and 26 weeks postnatally. Women who agreed to be referred were invited to participate in a home visit with the research associate.

Home visit (baseline)

Once a potentially eligible woman was identified, a research associate would arrange a visit, usually in the woman’s home. This visit was usually conducted at about 8 weeks postpartum (2 weeks after the screening EPDS had been completed); however, women could receive a home visit up to 26 weeks postpartum. Women were given further information about the study (see Appendix 5), given the opportunity to ask questions, and completed a second written consent form (see Appendix 6). To confirm eligibility, women completed a baseline EPDS and the revised computerised Clinical Interview Schedule (CIS-R). 76 However, at the start of the trial, women who scored below 13 on the EPDS were not always asked to complete the CIS-R so these data are not available for all women who received a home visit.

Women were eligible for entry to the intervention phase of the trial if they met the inclusion criteria (see Participants) and:

• scored ≥ 13 on the baseline EPDS

• received an ICD-10 primary diagnosis of depression on the CIS-R

• were proficient in English at a level to complete all research assessments (two women whose first language was not English had some language assistance in completing the assessments and/or listening visit intervention)

• their recently delivered baby was less than 26 weeks old.

Women were excluded if they met any of the exclusion criteria or:

• were already taking psychoactive medication or receiving psychological therapy

• were actively suicidal.

Demographic data and psychiatric history were collected by self-report questionnaire. This included age, employment and income, ethnicity, marital status, parity, baby’s age and method of feeding the baby, and history of depression and its treatment. The questionnaire also included three adapted items on social support77 (see Appendix 7). If the woman had a partner, she was asked to provide details of his employment, income and ethnicity. These data were collected for descriptive purposes because these factors may contribute to the depressive episode and affect outcomes.

Women who were eligible for randomisation were asked to complete further questionnaires to assess quality of life [Short Form Health Survey (SF-1278), EuroQol Self-report Questionnaire (EQ-5D79)], Maternal Adjustment and Maternal Attitudes Questionnaire (MAMA80) and the quality of their relationship with their partner if applicable [Golombuk–Rust Inventory of Marital State (GRIMS81)]. For details of these instruments, see Measures. Furthermore, if an eligible woman had a partner, she was asked if a short battery of questionnaires comprising the General Health Questionnaire (GHQ-12),82 SF-12, PAPA83 (the paternal version of the MAMA; i.e. Paternal Adjustment and Paternal Attitudes Questionnaire) and GRIMS, and prepaid envelope, could be left for her partner to complete and return to the study centre. Trial follow-up procedures are summarised in Figure 2.

FIGURE 2.

Summary of intervention and follow-up procedures. Note: If suicide ideation was reported at any stage in this process the self-harm protocol was implemented (see Appendix 4).

If a woman was ineligible, her GP and PHV were informed of her EPDS score and CIS-R diagnosis. Eligible women were randomised to receive either antidepressants or listening visits using the procedure outlined below.

Randomisation

Before the baseline home visit, the woman’s trial identification number, date of birth and trial centre were entered into a web-based randomisation program. After eligibility had been determined and consent had been obtained at the home visit, the researcher telephoned the remote computerised randomisation service and responded to a series of questions by keying numbers (e.g. patient identification number, baseline EPDS score) on the telephone keypad. The researcher then immediately informed the woman of her treatment allocation and provided her with further information about the allocated treatment (see Appendix 8).

Randomisation was stratified according to the baseline EPDS score (13–15, or > 15) and trial centre. The randomisation sequence was generated using a computer program with block sizes of six, eight and ten, varied randomly. The methods of sequence generation were concealed from the researchers involved in enrolling and randomising the women into the trial. Participants, researchers and those delivering the interventions were not blinded to the treatment allocation.

Follow-up schedule

Follow-ups were scheduled at 4 weeks, 18 weeks and 44 weeks for randomised women. (It should be noted that in the original RESPOND protocol, a 44-week follow-up was planned; however, funding to extend the trial only provided for data collection up to 18 weeks.) At these three time points, postal questionnaires and prepaid envelopes were sent to collect outcome measurements (EPDS, SF-12, MAMA, GRIMS, EQ-5D, adherence with antidepressant medication if applicable) and to obtain information on employment and income. Women who had agreed for their partners to receive a questionnaire at baseline were provided with a further short battery of questionnaires (GHQ-12, SF-12, PAPA and GRIMS) for their partner to complete at each follow-up. If the woman reported thoughts of self-harm at a follow-up assessment, and her GP had not previously been notified of this, the research associate asked the woman for permission to contact her GP (see Appendix 4).

The target time frame for follow-up was 3–6 weeks for the 4-week follow-up; 16–20 weeks for the 18-week follow-up; and 42–48 weeks for the 44-week follow-up. Women who had not returned their questionnaire after 1 week were contacted by phone or post and reminder questionnaires and prepaid envelopes were posted when necessary. If the woman had not replied after these reminders, the research associate attempted to telephone her to collect the EPDS outcome measure over the telephone.

Over the course of the trial, it became evident that follow-up rates were lower than anticipated. To obtain more complete data and reduce attrition, women were offered home visits by the research associate at 4 weeks and 18 weeks to collect the questionnaire data. The questionnaires were completed by the woman but the research associate was able to answer any queries if required. Research health visitors (RHVs) were also able to help with collection of outcome questionnaires at 4 and 18 weeks for women receiving listening visits by requesting women to complete their self-report questionnaires before commencing or continuing with the listening visits.

Strategies to monitor and improve recruitment rates

To monitor recruitment rates, practice staff were asked to record the number of women aged 18 years and over who had given birth, the number excluded by their GP or PHV, and the number of invitations sent out, on a monthly basis. To corroborate these figures, the relevant PCTs were also approached periodically and asked to provide statistics for the number of women aged 18 years and over who had given birth in each collaborating practice.

To improve the trial profile and recruitment rates, members of the research team and local Mental Health Research Network hub clinical studies officers visited collaborating GPs and PHVs to promote the trial and address any concerns or queries. In addition, the protocol was amended to include a number of additional strategies. Research associates and RHVs visited antenatal and postnatal clinics to publicise the trial and to distribute EPDS screening questionnaires to postnatal women. Posters advertising the RESPOND trial were distributed to collaborating practices and HV clinics. Three newsletters were sent to GPs and PHVs, locally adapted for each centre, to promote the trial and keep health professionals informed of trial progress. PHVs offered to remind women to complete the screening EPDS, and were provided with RESPOND leaflets and screening invitation packs to distribute to women at postnatal home visits and clinics. In addition, the patient invitation and information sheets were condensed and adapted at each centre to make them more accessible and reduce the burden on newly delivered mothers. Finally, women who had previously completed an EPDS screening questionnaire and had been ineligible to participate had the opportunity to be referred back to the trial by their GP or PHV if they were thought to be depressed; and the time frame for all health professionals’ referrals was increased from 3 months to 6 months after the birth.

Antidepressant medication

Women randomised to antidepressants were asked to make an appointment with their own GP as soon as possible, to discuss the prescription of an appropriate antidepressant. Women were contacted approximately 1 week after randomisation by the research associate to ascertain what progress they had made with this.

The woman’s GP and PHV were informed of the antidepressant group allocation by fax. The PHV was asked not to start any form of counselling or other psychological intervention specifically aimed at alleviating the symptoms of depression. The GP was told to expect the woman to make an appointment at the practice, and was asked to prescribe an antidepressant according to a simple clinical practice guideline developed by the research team (see Appendix 9) based on the North of England Guidelines84 and the British Association for Psychopharmacology 2000 Guidelines. 85 Although an SSRI was recommended as a first-line treatment, a pragmatic approach was employed whereby the GP and the patient agreed which antidepressant medication should be prescribed. The guidelines recommended the prescription of fluoxetine (20 mg) if not breastfeeding, sertraline (50 mg), paroxetine (20 mg), citalopram (20 mg) or escitalopram (10 mg) as first-line treatments. (It should be noted that since the study was designed it has been reported that citalopram treatment results in significant infant plasma levels. 52) Lofepramine (70 mg twice daily) or reboxetine (4 mg) were recommended as second-line treatments. However, it was noted that there is a lack of safety data on the use of escitalopram and reboxetine by breastfeeding mothers. The guidelines also advised noting the past response to an SSRI, previous adverse effects of any SSRIs, any concurrent medication and potential interactions, and the profile of the preferred SSRI regarding breastfeeding. The guidelines suggested that women be monitored after 2 weeks to assess side effects, and at 4 weeks to review treatment efficacy, and then every 4 weeks until 28 weeks. GPs were also guided on increasing the dose, changing the antidepressant medication or stopping pharmacotherapy altogether.

Information on the antidepressant prescribed and treatment adherence was obtained through women’s self-report at all follow-up points, and by recording prescribing information from women’s medical notes.

Health visitor non-directive counselling (listening visits)

Listening visits are a psychotherapeutic intervention that uses a form of non-directive counselling, often referred to as ‘active listening’ when delivered by HVs to women with PND, and originates from client-centred psychotherapy. 70 Non-directive counselling employs a person-centred approach and focuses on the feelings of the client, and as such the content of listening visits is determined by the client themselves. Using listening visits, HVs aim to facilitate the expression of the woman’s feelings without interfering, in a way that is respectful, empathic and reflective. HV-delivered non-directive counselling is referred to throughout this report as ‘listening visits’. During the visits, HVs discourage discussion of the practicalities of baby care and encourage expression of the mother’s feelings.

If a woman was randomised to listening visits, her GP and PHV were notified that this allocation had been made and that visits would commence after a 4-week waiting period. The protocol specified the 4-week waiting period to (1) test the effectiveness of antidepressants against usual care and (2) replicate the likely wait a woman might have for referral to a counsellor.

To avoid contamination between treatment groups, GPs were asked to provide GSC to these women, and not to prescribe antidepressants or any extra psychological care unless absolutely necessary. Furthermore, PHVs were asked not to provide any psychological intervention aimed to alleviate the women’s symptoms of depression. Although certain PCTs offer training in non-directive counselling to HVs to use with women with PND, we worked closely with HV leads in the participating PCTs to ensure that this was not the case for the practices recruited to this trial.

Change to alternative intervention

Women allocated listening visits were able to visit their GP at any time during the study period as part of their GSC. Although GPs could prescribe antidepressants for these women at any time during the study period, they were asked not to offer a prescription before 4 weeks unless absolutely necessary. RHVs also discussed the use of antidepressants with women during listening visits if this was thought to be necessary. This discussion most commonly occurred at the fourth visit as part of a review of treatment progress, but the use of antidepressants may have been suggested earlier if the woman was particularly unwell.

When the 4-week follow-up assessment had been completed, women allocated antidepressants were reminded by the research associate that they could also receive listening visits. In addition, a letter was sent to the GPs of women allocated antidepressants 4 weeks after randomisation (see Appendix 10) to remind them of the treatment options available to women allocated antidepressants. This included offering listening visits instead of, or in addition to, antidepressants. Women who requested these were then offered a series of up to eight listening visits with the RHV which were scheduled to commence as soon as possible after the crossover was requested.

Diagnosis of depression

The EPDS,75 a 10-item self-report questionnaire, was used to screen for PND. Several validation studies have been carried out in English-speaking samples as well as in other languages. 75,87 The case definition for probable depression is a score of ≥ 1388 and this threshold was used to determine eligibility to the trial. At the screening stage to reduce the chance of missing false negatives we used a threshold of ≥ 11. The EPDS was completed at the screening stage, the baseline home visit and at all follow-ups.

The self-administered computerised version of the CIS-R was used at baseline to obtain a more accurate measure of the woman’s clinical state, to confirm a diagnosis of depression and to ensure eligibility. Questions were presented to the woman on the computer screen and she responded independently using the keyboard. The CIS-R is a fully structured psychiatric assessment for 14 common symptoms of depression and anxiety in the week before interview. 76 It has been used widely, including in the UK Psychiatric Morbidity Surveys. 89 The questions in the CIS-R enable the ICD-10 criteria for depressive episode, generalised anxiety disorder, phobias, panic disorder and obsessive compulsive disorder to be defined. The CIS-R also generates a total score that ranges from 0 to 57. There is excellent agreement between the interview-administered version and the self-administered computerised version. 90 In particular, there is no evidence of any consistent bias between the two methods of administration. The original CIS was used in the first trial of listening visits for PND,70 and the CIS-R was used in a more recent trial comparing an SSRI with CBT. 54

Self-harm risk

Thoughts of self-harm in the past 7 days were assessed using a question on the EPDS at all time points and with appropriate questions from the CIS-R at baseline. Risk of self-harm was defined as:

• an answer of ‘Sometimes’ or ‘Yes, quite often’ to question 10 of the EPDS ‘The thought of harming myself has occurred to me’

• reporting that ‘life isn’t worth living’, suicidal thoughts or suicidal plans on the CIS-R.

Primary outcomes

The primary outcome was depression assessed using the EPDS at 4 and 18 weeks after randomisation to address research objectives 1 and 2, respectively. The EPDS was selected as the primary outcome (rather than a longer, more thorough assessment of psychopathology such as the CIS-R76) because the EPDS is shorter, has been used in previous studies more often than any other measure, allowing comparability with previous research, and is easier to complete, hence facilitating more successful follow-up rates. The primary outcome was the binary variable of whether or not the woman had ‘improved’ her EPDS score such that she no longer satisfied the entry criterion of an EPDS ≥ 13. The comparisons between the randomisation groups at the two follow-up time points therefore compared the proportions of women whose scores were < 13 as opposed to ≥ 13.

Secondary outcomes

All secondary outcomes were collected at baseline and at both follow-ups. The main secondary outcome was the continuous score on the EPDS at 4 and 18 weeks, analysed both separately and together.

Quality of life was measured using the standard SF-12 version 2 questionnaire. This 12-item measure is a widely used and well validated78,91 generic measure of functional quality of life. The questionnaire was modified (see Appendix 7) to include three additional items from the SF-3692 to incorporate the five-item Mental Health Index (5-MHI), a measure gaining in popularity as a short self-report questionnaire for emotional well-being,93 but not used in our analysis. Mental and physical component scores were calculated using standard algorithms, with higher scores indicating better health. If a woman returned an incomplete SF-12 questionnaire, it was not scored and was regarded as missing.

Health-related quality of life was measured using the EQ-5D79 questionnaire, with the intention of using the data to estimate Quality Adjusted Life-Years (QALYs). This widely used instrument includes questions about mobility, self-care, usual activities, pain/discomfort and anxiety/depression to describe health status. Utility scores for each health state have been determined using valuations from the general population. 94 We also recorded personal health state valuation using the EQ-5D visual analogue scale (VAS), a ‘thermometer’ ranging from 0 to 100, where 100 represents the best state imaginable and 0 represents the worst state. Total scores were not calculated for women with incomplete responses to this scale, and were regarded as missing.

Maternal attitudes were measured using the relevant 12-item subscale (attitudes towards pregnancy and the baby) of the MAMA questionnaire (postpartum version). 80 This questionnaire included 12 items (see Appendix 7) and is acceptable to women. 80 Higher scores on this measure indicate a more positive attitude. Incomplete responses to this scale were not scored and were regarded as missing.

The quality of the marital relationship was assessed using the short form of the GRIMS. 81 This questionnaire (see Appendix 7) has been used in previous studies with acceptable validity (Golombok, personal communication 2008). This measure could only be completed by women in a current relationship. Total scores were calculated with higher scores indicating a poorer relationship. Incomplete responses to this scale were not scored and were regarded as missing.

The woman’s partner was invited to complete a small number of questionnaires. These included the GHQ-1282 to assess mental health, their role as a parent using the PAPA,83 the GRIMS81 and the modified SF-12 as detailed above. The data obtained from partners will be presented in a separate publication.

Process measures

Attendance for listening visits was as recorded by the RHV. Adherence with antidepressant medication was assessed at 4 and 18 weeks using a modified version of the Morisky Adherence Scale95 and four items adapted from a scale reported by Schroeder et al. (see Appendix 7). 96 Provided the woman had given consent, GP medical notes were also examined to extract information on consultations and medication prescribed for depression (date prescribed, drug name, strength, dose, amount issued), during the period between randomisation and the 18-week follow-up.

Resource use

As requested in the HTA commissioning brief, the original protocol described a full cost-effectiveness analysis at 44 weeks, i.e. around the time of the child’s first birthday. However, when the recruitment extension request was curtailed to allow data collection only as far as the 18-week assessment, this part of the work plan ceased.

Original sample size justification

In the original protocol, the primary outcome was the binary EPDS score at 4 weeks, comparing women randomised to antidepressants with those receiving GSC, and repeated at 18 weeks to compare randomisation to antidepressants plus or minus listening visits with listening visits alone. A positive outcome was defined as an EPDS that had ‘improved’ to the extent that the woman no longer satisfied the entry criterion for EPDS – namely, an EPDS < 13. For the original sample size calculation, it was anticipated that about 40% would improve by 4 weeks among women receiving GSC compared with 55% receiving antidepressants, and that in the same group of women this would rise to 50% at 18 weeks given the addition of listening visits compared with 65% in those receiving antidepressants. A difference of 15 percentage points in the proportions scoring < 13 on the EPDS was considered by clinical judgement to be substantial and worth detecting at each follow-up time point.

For a two-sided 5% significance level, 211 women were required in each group to have an 85% power of detecting a difference of 40% versus 55%, requiring a total of 468 women to be recruited to allow for 10% attrition at the 4-week follow-up. Assuming an 85% follow-up rate at 18 weeks, this sample size would yield 83% power to detect a difference of 50% versus 65% as significant with a two-sided 5% significance level. Attrition would have to exceed 22% for either power to fall below 80% for these specifications, and any sample size such as that derived here would be expected to have considerably higher power to detect clinically important differences in mean EPDS when the outcome is considered as a continuous score.

Assuming a prevalence of PND of 10% and an acceptance rate of 50%, recruiting 468 women was calculated to require a total of 9360 births to be covered by the recruitment period for the participating practices. Assuming that a practice with a list size of 10,000 has about 100 births per year, then 63 practices over an 18-month period was thought sufficient for the trial. It was intended that each centre would therefore recruit 21 practices for the trial.

Revised sample size calculation

Due to the difficulties in recruiting sufficient numbers of participants, and a higher than anticipated attrition rate, a revised sample size calculation was derived in late 2006/early 2007, in conjunction with the Data Monitoring Committee. For these calculations, an attrition rate of between 10% and 20% was assumed for the primary end point of 4 weeks, and (conservatively) up to 30% for the 18-week follow-up. Moreover, a total sample size of 250 women was projected, which would therefore yield 100–112 women available for analysis in each group at 4 weeks, and (for a range of 20–30%) 88–100 at 18 weeks. For the original target difference of 15 percentage points (40% versus 55%), power was very low, at between 45% and 56% for the two follow-up times depending on the attrition observed (again all with a two-sided 5% significance level). As indicated in Table 1, however, power was just over 80% for a target difference of about 20 percentage points across the same range of assumed attrition rates. In the event, with 254 women randomised and attrition rates of about 15% and 20% observed at 4 and 18 weeks respectively, the study had 80% power to detect a difference of 20–20.5 percentage points at the two follow-up points.

Although it was still felt preferable to base the sample size calculation (conservatively) on the binary version of the EPDS score at follow-up, this approach is in general highly conservative. As an indication of this, with a two-sided 5% significance level a sample size of 250 would yield 80–90% power to detect differences on the continuous EPDS scale of 0.4–0.5 standard deviations (SDs). Assuming an SD of about 5 (as observed at initial screening and at baseline in this trial) we would have adequate power to detect differences of 2.0–2.5 points, all of which are smaller than differences that would be considered clinically significant. 97

Data handling and software

All data were entered into Microsoft^® access 2000 databases at all three study centres and then merged onto the central databases (currently in access 2003). Data validation was conducted in both Microsoft^® access 2003 and stata version 9.2, including checks for missing and inconsistent values, plus logic and range checks. The original paper questionnaires were checked where necessary and values were amended in the database. Statistical analysis was conducted according to a predefined analysis plan, using stata version 9.2.

Preliminary analyses

Descriptive statistics were obtained to compare: the number of births recorded and screening invitations sent out and returned; characteristics of women who screened positive and negative at the initial screening; and characteristics of women randomised to the two treatment groups. The purposes of these analyses were to assess the epidemiology of PND, consider the generalisability of the group of women randomised, and to assess the baseline comparability of the two randomised groups.

Primary comparative analyses

The primary outcomes of the trial were the EPDS scores (as a binary variable, ‘improved’ < 13 versus ‘not improved’ ≥ 13) at the 4-week and 18-week follow-up time points. At 4 weeks and 18 weeks the primary analysis was an intention-to-treat (ITT) logistic regression of the EPDS score at follow-up, adjusting for baseline EPDS as a continuous variable and centre as the other stratification variable.

Secondary analyses

Using ordinary linear regression models at each follow-up point separately, the ITT analysis was then applied to all the secondary outcomes: the continuous version of the EPDS score; SF-12 (mental and physical components); EQ-5D (five-item utility score and VAS valuation); MAMA and GRIMS. All the following remaining secondary analyses were conducted on the full set of primary and secondary outcomes.

The primary analyses were repeated after additional adjustment for the precise time that had elapsed between randomisation and the date of completion of the relevant follow-up questionnaire. This was conducted by first adjusting for this time as an additional covariate in the primary (logistic) regression models and second by restricting the analysis to those who completed the questionnaire between 3 and 6 weeks and 16 and 20 weeks for the 4-week and 18-week follow-ups, respectively. The next secondary analysis adjusted for any variables that exhibited potentially influential imbalance at baseline from the 23 variables considered.

The next set of secondary analyses comprised explanatory investigations of the effects of the interventions, comparing women based on treatment(s) received. Although the crude (biased) treatment(s) received analyses are noted for comparative purposes where it is deemed helpful, the tabulated results derive from Complier-Average Causal Effect (CACE) analyses employing instrumental variables regression to estimate unbiased treatment effects. 98,99 These were performed for all primary and secondary outcomes at each follow-up, using ordinary linear regression models for continuous outcomes. In the absence of well-established methods of this kind for logistic models, the following approach was adopted for the (primary) binary outcome: ordinary instrumental variables regression for outcome percentages, plus probit models as sensitivity analyses to check that model assumptions were not seriously violated. In all cases the probit models gave very similar p-values to those from the ordinary regression models, and hence only the latter are presented here given their advantages in terms of interpretability of the regression coefficients (as differences in percentages).

All CACE analyses included the same women available for the corresponding ITT analysis. At 4 weeks these analyses compared those who had received antidepressants with those who had received GSC, with the CACE method accounting for selection effects operating after randomisation. At 18 weeks, some women in both randomisation groups had received all four possible combinations of the two interventions, that is: neither (GSC only); antidepressants only; listening visits only; both interventions. Whereas at 4 weeks a conventional instrumental variables regression is possible, at 18 weeks there is a larger number of observed combinations than randomisation groups and a combined instrumental variables regression is not possible. At 18 weeks, therefore, each intervention was considered separately in two parallel CACE analyses.

In respect of the listening visits, the explanatory analyses always used the RHV records. For antidepressants, two sets of data were used: first, the self-report data from the follow-up questionnaires and second the prescription data from the primary care records (specifically, whether or not there was evidence of a prescription for antidepressants by the relevant time point). Before the regression analyses, the levels of agreement between the two sources of antidepressant data were compared at the two time points using cross-tabulations and kappa statistics. At 18 weeks it is emphasised that this comparison has a number of limitations. First, the self-report data at 18 weeks only refer to the previous 4 weeks. Second, the prescribing data were only collected up to the target time of the 18-week follow-up rather than when this follow-up actually took place – hence the time period for the ‘18-week’ record-based analyses was up to the actual 18-week follow-up or 18 weeks after randomisation, whichever was the shortest. For both these reasons a certain degree of discrepancy between the sources at 18 weeks would be expected. It remains that at 4 weeks the timing of the two sources was identical – namely, the date at which the 4-week EPDS was completed.

The remaining secondary analyses involved various sensitivity analyses to reduce the impact of missing data. First, a repeated measures (logistic) regression model was fitted to the 4-week and 18-week (binary) primary outcome data combined, using generalised estimating equations. This analysis initially investigated the interaction between intervention group and time (to assess whether there was evidence of a change in magnitude of effect between the two follow-up times), and then the main effect of intervention, in both cases adjusted for baseline EPDS and centre. Not only does interpretation of the latter effect depend on the presence or absence of an interaction with time, but in this study particular care is needed in the emphasis to be placed on the results of this analysis given that the interventions being compared changed considerably between the two follow-up times. The repeated measures analysis was also conducted for the continuous EPDS score, again adjusting for baseline EPDS and centre.

The second (missing data) sensitivity analysis involved the application of multiple imputation by chained equation (MICE) methods100 in stata (ice procedure: 25 April 2008 version 1.4.6). A total of 25 data sets were generated by this procedure and 10 switching procedures were undertaken. The imputation model included any potential confounders where there was any suggestion of a relationship with missing EPDS scores at either 4 or 18 weeks (ascertained by simple cross-tabulations between relevant factors and inclusion or attrition by 4 weeks and 18 weeks, separately). Initially it was the continuous EPDS scores that were imputed, from which imputed binary outcome variables of < 13 versus ≥ 13 were calculated. The primary ITT analyses were then repeated for four EPDS outcomes (the binary and continuous versions at each of the 4-week and 18-week follow-ups), with in each case the results compared with the original complete-case primary analyses.

Subgroup analyses

The preplanned subgroup analyses were all performed for the primary ITT regression models, adjusting for centre. The primary subgroup analysis investigated the interaction between the CIS-R score at baseline and randomisation group in regression models for the (binary) 4-week and 18-week EPDS outcomes. Further (more pragmatic) subgroup analyses repeated these analyses replacing the CIS-R with the continuous baseline EPDS and its binary version in the form of the stratification groups (baseline EPDS < 16 and ≥ 16). To potentially increase the power for these essentially exploratory analyses, the subgroup analyses were repeated for the continuous version of the EPDS score.

Recruitment and sampling of women and health professionals

In order to contact women who had completed the trial, i.e. had their final outcome measures for the trial taken, the research associates posted a letter explaining that interviews were being held with trial participants to all women taking part. The research associates then telephoned women to ask if they would be willing to be interviewed and, if they were, whether or not they could pass their contact details on to the qualitative research associate (RA) responsible for conducting the interviews. If the woman agreed to this, the research associate completed a ‘release of personal details form’ and gave this to the RA, along with information about which trial site the woman had been recruited in and to which intervention arm she had been randomised.

To contact individuals who had declined to take part in the trial, either before or at the stage of randomisation, at the time that the individual declined, the research associate present informed the individual about the qualitative study. The research associate then asked the individual if her contact details could be passed to the RA and followed the same procedure as described above.

Using the information provided by the research associates, a purposive sampling approach was used to ensure that interviews were held with both ‘completers’ and ‘decliners’ and in the case of the former, with women who had been randomised to different treatment arms and recruited in different centres. Within this sampling approach, maximum variation was sought in relation to age, socioeconomic background, ethnicity and women who had and had not complied with their allocated treatment.

Thirty-seven women were sampled and contacted by telephone to discuss the study further and, if they were still willing to be interviewed, to arrange an interview time and place. On being contacted, 28 women (27 completers and one decliner) agreed to be interviewed. Before interview, they were posted a letter confirming the interview time and an information sheet with details about the qualitative study. Two versions of the information sheet were available: one for completers and one for decliners. Written consent to take part was secured at the time of interview.

Sampling was also purposive for the health professionals’ interviews. Maximum variation was achieved in relation to GPs’ age, sex, length of time in general practice, practice size, and level of deprivation in the area. An attempt was made to interview GPs who had and who had not referred patients to the trial. For the HVs, variation was attained in relation to time since completion of training, length of service in that area and level of engagement with the trial. At the start of the trial, most HVs were practice-based, but corporate working was introduced to all areas during the first year of RESPOND. An attempt was made to interview HVs who were still practice-based, as well as those who had been moved to a corporate way of working.

Thirty-seven GPs were approached by letter and telephone and 19 agreed to be interviewed. Twenty HVs employed within participating PCTs were invited to participate and 14 agreed to be interviewed. For both sets of interviews, data collection and analysis proceeded in parallel, and recruitment ended when data saturation had been reached.

Interviews

The interviews with the women and with the health professionals were semi-structured in nature. Flexible interview guides were used to ensure that key areas were covered, while allowing participants to raise issues that were salient to them. The interview guide used with the women who had completed the trial (see Appendix 11) explored a number of topics: experiences of PND; treatment preference at the time of randomisation; expectations and experiences of antidepressants, listening visits and GSC; and views of the trial. The guide used with women who had declined to take part (see Appendix 12) also focused on their experiences of PND and expectations of the treatments delivered in the trial, as well as asking about their reasons for not taking part. The interview guides for health professionals (see Appendices 13 and 14) covered the following areas: understanding of PND; diagnosis and management of PND; and the patient–professional relationship, professional–professional relationships, and experience of the RESPOND study, in particular, views on the interventions.

Interviews with trial participants were held between November 2006 and June 2007. Two of the 28 women interviewed were interviewed over the telephone. The remaining women were interviewed face-to-face in their own homes. The interviews lasted between 40 minutes and 2 hours. The interviews with the health professionals were conducted between January 2006 and February 2007. Participants were interviewed at their place of work, and interviews lasted between 25 and 67 minutes. With participant consent, all the interviews were audio-recorded and transcribed verbatim.

Analysis of the interview data

For both the women’s and the health professionals’ interviews, analysis proceeded in parallel with data collection, allowing for modification of the interview guides in the light of emerging themes and for thematic categories identified in the initial interviews to be tested or explored in subsequent interviews where disconfirmatory evidence was sought. 102 For both data sets, analysis was inductive and a thematic approach was taken. 103 Hence, each transcript was read and re-read to gain an overall understanding of the participants’ views and experiences. Emerging themes were then noted and coding frames (one for the women’s data and one for the health professionals’ data) developed. Transcripts were read and discussed by researchers from different professional backgrounds (primary care and psychology), so that the analysis and the coding frames could be debated and refined through discussion.

To allow electronic coding and retrieval of data, transcripts from the women’s interviews were imported into the software package QSR nvivo (version 2). Several transcripts were independently coded by K.M.T. and the RA, who then met to discuss areas of consensus and discrepancy. This led to further codes being developed and to existing codes being defined more clearly. Once all the transcripts had been electronically coded, data were analysed using a framework approach. 104 Using this method, K.M.T. summarised in tables what participants had said in relation to specific issues (e.g. concerns about medication, effect of antidepressants), and then made comparisons both within and across interviews to identify thematic patterns and deviant cases. As a result of the pragmatic nature of the trial, with women being able to have both treatments (either simultaneously or sequentially) at some point, the comparisons made were between participants who varied in terms of treatment preference and use, rather than their treatment allocation.

The health professional transcripts were independently coded as data were collected by C.A.C-G. and a research associate (E.C.), supported by a medical student (L.C.) and another research associate (C.R.). This led to further codes being developed and to existing codes being defined more clearly as data collection progressed. Regular meetings were held at all stages to discuss discrepancy and to achieve consensus. Once all the data had been collected and transcripts had been coded, data were analysed using a framework approach. C.A.C-G. and E.C. summarised in tables the themes agreed and made comparisons across transcripts, both within and across the two groups (i.e. GP and HV).

Supervision of trial

The trial was independently supervised by a Trial Steering Committee, which comprised an independent academic GP as chair, statistician/trialist, psychiatrist, representative from the Community Practitioners and Health Visitors Association, a consumer representative, the lead investigators and the trial co-ordinator. The trial was also supervised by a Data Monitoring and Ethics Committee, which comprised an academic health services researcher (chair), statistician and psychiatrist. Members of these committees are named in the Acknowledgements.

Practices

Figures obtained from the practice monitoring visits and the PCTs are shown in Table 2, and relate to the 21 practices from each of Bristol and London, and 35 in Manchester. This table shows that the total number of invitations sent to women by practices was less than the number of births recorded by the practices, even after the exclusions by GPs were considered. Moreover, the PCTs in each centre reported more births than the practices recorded although some of the PCT figures were estimated. The PCTs were requested to provide monthly figures. However, because of other priorities, some PCTs were only able to provide annual totals for each practice and as practices were recruited at different times and for different periods, annual figures were calculated pro rata where necessary, for practices that had been active for only part of the year.

Patients

Practices sent a total of 10,666 invitations to recently delivered women and 4409 replies were received, including 15 who were referred to the study directly by the GP or HV. However, 62 of the invitations had been sent to women who had given birth twice during the recruitment period and had responded to the invitation to participate on both occasions. These second responses were therefore discarded giving a total of 4347 responses. After scrutinising the data, it was found that a further 19 replies were from women who had responded to both the initial invitation and to a reminder sent shortly afterwards and these duplicate responses were also removed leaving a total of 4328 responses. The overall response rate was therefore 41%. The numbers of responses/referrals from each centre were 2251 from Bristol, 951 from Manchester, and 1126 from London, with response rates of 44%, 29% and 53%, respectively. Of the 15 women referred directly by practitioners, two declined a home visit, four did not have an ICD-10 diagnosis of depression, four were randomised to antidepressants and five to listening visits.

The ‘screening’ Consolidated Standards of Reporting Trials (CONSORT) diagram (Figure 3) shows the outcome of the 10,604 (10,666 minus 62 second births) invitations to participate in the RESPOND study across the three centres. Of the 4328 replies received (after the removal of the 19 duplicate responses described above), 155 (3.6%) women did not complete the initial (‘screening’) EPDS questionnaire. A further 113 women who returned the questionnaire scoring < 11 (and hence included in the 3184 within Figure 3) indicated on the consent form that they wished no further contact with the study, and 181 women who scored ≥ 11 on the EPDS questionnaire also refused further participation or could not be contacted after returning the initial questionnaire and consent form. Therefore, the total number of women to refuse the invitation to participate at the initial screening stage was 449 (10.4%). Women who declined to participate were entered onto the database with all identifiable information removed including the woman’s date of birth even if this was available.

FIGURE 3.

Screening CONSORT diagram for all sites.

The total number of women who returned a completed screening EPDS (n = 4158) or who were referred to the study by their GP/HV (n = 15) was therefore 4173 (termed from now on as the ‘questionnaire responders’). Of these 4173 responders, 3184 (76%) women scored 10 or less on the initial EPDS questionnaire and a total of 989 (24%) women scored ≥ 11, or were referred directly by their GP/HV (Table 3). These 989 (23%) women were offered a home visit assessment to assess eligibility for the trial. A diagnosis of PND was confirmed if the women scored ≥ 13 on a further EPDS at the home visit and had a diagnosis of depression according to ICD-10 criteria as assessed by the CIS-R.

The mean (SD) age and EPDS score of the 4173 responders were 31.4 (5.5) years and 7.3 (5.5), respectively. The mean (SD) age of their babies was 7.5 (3.0) weeks. Corresponding figures for each centre are shown in Table 4, which also shows distributions of these characteristics and the proportion of women answering 2 or 3 on the self-harm item on the EPDS questionnaire – that is, ‘the thought of harming myself has occurred to me’ either ‘Sometimes’ or ‘Quite often’. Women from Manchester tended to be younger, have slightly older babies, have higher EPDS scores and be more likely to admit to thoughts of self-harm.

A comparison of age and EPDS data is shown in Table 5 for those women with an EPDS score < 11 and for those with an EPDS score ≥ 11 or referred. Women who had high EPDS scores early in the postnatal period tended to be younger and were much more likely to admit to thoughts of harming themselves (15.2% compared with 0.2%).

Eligibility screen

The purpose of the eligibility screen was to ascertain whether women (n = 989) who scored ≥ 11 on the initial screening EPDS (n = 974), or who were referred (n = 15), met the criteria for a diagnosis of PND. For the purposes of Figure 3, the following hierarchy of exclusions was applied:

• refused/did not reply

• already receiving treatment

• excluded by GP/HV

• improved (interim telephone EPDS < 8 or home visit EPDS < 13)

• no ICD-10 diagnosis of depression

• other exclusion criteria.

The screening CONSORT diagram (Figure 3) shows the outcome for each of the 989 women who were invited to an eligibility home visit. A total of 361 women who scored ≥ 11 on the screening EPDS did not proceed to the home visit for the reasons given in the (screening) CONSORT diagram (Figure 3). One hundred and eighty-one women declined or did not reply. Their reasons for declining the home visit interview included ‘feeling better/not wanting to take antidepressants’ and ‘lack of time’. However, this information was not systematically obtained or recorded. A further 180 women were found to meet one or more exclusion criteria before the interview took place, including 88 women who were already receiving treatment for depression and the 55 women who scored < 8 on the interim telephone EPDS (see Chapter 2, Trial monitoring). Reasons for exclusions either by GPs/HVs or researchers were not always given but reasons that were recorded are as follows: poor English (n = 8); infant older than 6 months (n = 3); infant-related reasons such as foster care/died/child protection (n = 4); GP concern (n = 4).

A total of 628 home visit assessments were conducted during which a further six women were found to be receiving treatment for depression. Of the remaining 622 women, 298 women scored < 13 on the EPDS so were ineligible and 54 scored ≥ 13 on the EPDS but did not have an ICD-10 diagnosis of depression and so were also ineligible as was one woman whose baby was more than 26 weeks old. Of the 269 eligible women (those who scored above 12 on the home visit EPDS and had a diagnosis of depression according to the CIS-R), seven were excluded by their GP or HV after the home visit had taken place and eight declined randomisation. Therefore, including the 94 (88 of the 989 and six of the 628) women who were already being treated for PND, the overall prevalence of PND in the 4173 questionnaire responders was 8.7% (95% CI 7.9 to 9.6%).

Table 6 (and Figures 6–8 in Appendix 15) shows the breakdown of these exclusions by centre. This table incorporates the initial screening and the eligibility assessment. At the time of the eligibility EPDS, women in Bristol and London were almost twice as likely as those in Manchester to have ‘improved’ – that is, to have an EPDS score below the cut-off point and/or fail to get a diagnosis of depression as assessed by the CIS-R. As well as being more likely to be eligible for the study, women in Manchester were also more likely to have already been diagnosed with depression and be receiving treatment. In addition 73 (45%) of the women in Manchester were in the more severe stratum (EPDS at home visit ≥ 16) compared with 74 (25%) in London and 52 (31%) in Bristol.

Comparison of randomised and not randomised women

Table 7 shows the sociodemographic characteristics of the women who were eventually randomised compared with all other women who completed the home visit interview. Although all sociodemographic variables were collected by the researcher at the time of the home visit interview for all randomised women, this was not initially the case for women who were found to be ineligible. After the study had been recruiting for several months, it was decided that these variables were necessary for epidemiological purposes and so these data were then collected retrospectively by telephone by researchers at each centre.

Women who were randomised were less likely to be married or living with a partner and tended to have less social support. They were also more likely to report previous antidepressant treatment, to have fewer qualifications, and to be in routine occupations.

Randomisation

A total of 254 eligible women were randomised and allocated to either the antidepressant arm (n = 129) or to the listening visits arm (n = 125). A further eight women were randomised incorrectly: four with no ICD-10 diagnosis of depression, one who scored < 13 on the home visit EPDS, one who was later found to be receiving counselling; and two who had a second birth during the recruitment period and were randomised on both occasions. Data for these women have been removed from the analysis of the randomised controlled trial, and in Figure 3 they appear in their ‘correct’ boxes, albeit on a post hoc basis.

The sociodemographic variables collected at the eligibility home visit were compared for the two treatment groups and are shown in Table 8. These variables were collected using a self-report questionnaire and as a result data were incomplete, particularly for questions relating to partner data.

Table 8 indicates a good balance between the two randomisation groups, although there are some differences that may need to be adjusted for in secondary analyses – in particular, diagnosis, number of children, previous antidepressant treatment, breastfeeding and employment status. The small imbalance in respect of the GRIMS is of less general concern because this will reflect their current partner status (i.e. can only be completed if there is a current partner). Hence this baseline variable was only adjusted for in analyses involving the GRIMS outcome itself.

Follow-up

The ‘intervention’ CONSORT diagram (Figure 4) shows the progress of the randomised women up to the 18-week follow-up. Individual CONSORT diagrams for each centre appear in Appendix 16 (Figures 9–11). Outcome measures were scheduled to be collected at 4 and 18 weeks after randomisation within a target time frame of, for example, 1 week before to 2 weeks after the due 4-week follow-up date. These time frames were not always achieved and some women withdrew or were lost to follow-up. Table 9 shows the actual time of questionnaire completion at the 4-week and 18-week follow-ups.

FIGURE 4.

Intervention CONSORT diagram for all sites.

The numbers of questionnaires returned at 4 and 18 weeks were 218 and 206, respectively. The overall follow-up rate was therefore 86% at 4 weeks (Bristol 90%, Manchester 88%, London 77%) and 81% at 18 weeks (Bristol 84%, Manchester 87%, London 70%). More women in the antidepressant group withdrew or were lost to follow-up [4 weeks: antidepressants 23 (18%), listening visits 13 (10%) p = 0.090; 18 weeks: antidepressants 32 (25%), listening visits 16 (13%) p = 0.015].

In terms of the interventions actually received by those who were lost to follow-up, of the 36 women not followed up at 4 weeks, none received listening visits as part of the study. In addition, information on prescribing was available from the primary care records for all but three of these women. From this source, eight (24%) of 33 received a prescription for antidepressants; the remainder apparently received GSC only. Of the 48 women not followed up at 18 weeks, information about treatment actually received was known from RHV and primary care records for all but seven women. Of the remaining 41 women: 11 (27%) received GSC only; 13 (32%) received antidepressants, nine (22%) received listening visits and eight (20%) received both interventions. At both time points, the women who were lost to follow-up received substantially less in the way of intervention than those who were included.

Table 10 and Table 11 present the relationships with attrition at 4 and 18 weeks for a selection of variables. This selection was driven by the requirements of the multiple imputation analyses (see Chapter 2, Statistical methods) so as to identify the potential confounding variables to be included in the (imputation) models for missing EPDS outcome data. From these results it was considered sufficient to include the following variables in the imputation models for (missing) EPDS at follow-up: intervention group; centre; EPDS itself at all time points including baseline; diagnosis; previous antidepressants; mother’s age; number of children; living with partner and employment status. Overall, the primary EPDS outcomes were known for 218 and 204 women at the 4-week and 18-week follow-up points, respectively. Moreover, this outcome was only missing at both time points for 22 women, since 14 of the 36 scores missing at 4 weeks had known values at 18 weeks, and 26 of the 48 missing at 18 weeks were known at 4 weeks.

The mean (SD) time interval between randomisation and completion of the questionnaires was 5.4 (2.1) weeks at the 4-week time point and 160 (73%) were completed between 3 and 6 weeks after randomisation (Table 9). At 18 weeks, the mean (SD) time interval was 20.9 (5.0) weeks and 134 (65%) were completed between 16 and 20 weeks.

Adherence to protocol

A total of 129 women were randomised to receive antidepressants and 125 to receive listening visits. Inevitably, not all adhered to the protocol. Furthermore, the protocol stated that women randomised to the listening visits arm and who did not respond (EPDS score ≥ 13 at the 18-week follow-up) should be offered antidepressant treatment after the 18-week follow-up. However, practically and ethically it was not possible to prevent women accessing treatment options recommended by NICE. Consequently after the 4-week follow-up, both randomised groups had access to both treatments and so by 18 weeks some women had received either antidepressants or listening visits, some both, while others had refused both treatment options and opted for GSC from their PHV and/or GP. The diversity of treatment received by 18 weeks in each treatment arm is shown in Table 12.

A total of 125 women were randomised to listening visits, which commenced approximately 4 weeks after randomisation, and 117 women had at least one visit. In addition, 68 women in the antidepressant arm requested listening visits after the 4-week follow-up. Of these, 64 had at least one visit. During the course of the study, a total of 181 women therefore received some listening visits (Table 13). Among the 206 women providing primary outcome data at 18 weeks, from Table 12 a total of 162 (79%) received some listening visits according to RHV records, 57 (59% of the 97 followed up) and 105 (96% of 109 followed up) in the antidepressants and listening visits groups, respectively.

Adherence to medication was ascertained by self-report at 4 and 18 weeks with the intention of verifying these reports with prescription data collected from women’s medical records. At the 4- and 18-week follow-ups, women were asked to complete an adherence questionnaire relating to the previous 4 weeks if they had been prescribed antidepressants. Results are shown in Table 14. At 4 weeks only 59 (56%) of the 106 women followed up among those randomised to the antidepressants and who completed the questionnaire reported taking any antidepressants. In the listening visits group seven (6%) of the 112 women followed up also reported taking antidepressants. This makes a total of 66 (30% of 218) women who reported taking antidepressants at 4 weeks, all of whom responded to the adherence questionnaire depicted in Table 14. At the 18-week time point, the numbers in each group who reported taking antidepressants during the previous 4 weeks were 62 (64% of the 97 followed up) and 37 (34% of 109 followed up) in the antidepressants and listening visits groups, respectively. This makes a total of 99 (48% of 206) women who reported taking antidepressants at the 18-week follow-up, of whom all but 10 provided data for Table 14 (nine and one missing from the two groups, respectively).

The antidepressant prescriptions data collected directly from practices were not available for 20 women: 12 who had refused consent and eight who had signed a version of the consent form that was not acceptable to practices for this purpose. Prescribing data were available for 201 of the 218 women followed up at 4 weeks; of the 97 and 104 in the antidepressant and listening visits groups, respectively, the numbers (percentages) being treated as intended were according to the notes, 52 (54%) and 99 (95%). These are very similar to the corresponding figures of 56% and 94% according to the self-report data, and agreement between these two sources was reasonably high. In particular, among the 97 in the antidepressant group, crude agreement was 78% with a kappa of 0.56. Of the 55 women who self-reported taking an antidepressant, 43 (78%) had been prescribed such a drug; conversely, of the 42 who claimed not to have taken an antidepressant, 9 (21%) had such a prescription noted in their primary care records. Overall, by 4 weeks, 57 women had according to their records been prescribed antidepressants (52 in the antidepressant group, five in the listening visits group); 144 women had not (45 and 99, respectively).

At 18 weeks, prescribing data were available for 193 of the 206 women followed up (90 and 103 in the antidepressant and listening visits groups, respectively). According to the notes the number (percentage) being treated with antidepressants in the antidepressants group was 58 (64%); likewise, in the listening visits group the number (percentage) receiving at least one visit was 95 (92%). These are nearly identical to the corresponding figures of 64% and 90% according to the self-report data, and agreement between these two sources was very high. In particular, among the 90 in the antidepressant group, crude agreement was 84% with a kappa of 0.66. Of the 58 women who self-reported taking an antidepressant, 51 (88%) had been prescribed such a drug; conversely, of the 32 who claimed not to have taken an antidepressant, seven (22%) had such a prescription noted in their primary care records. Overall, by 18 weeks 92 women had according to their records been prescribed antidepressants (58 in the antidepressant group, 34 in the listening visits group); 101 had not (32 and 69, respectively).

Descriptive statistics

Of the 106 women allocated to receive antidepressants with EPDS ≥ 13 at 4 weeks, 48 (45%) had improved as defined by having an EPDS score < 13 at follow-up. This compares with 22 (20%) out of 112 women allocated to receive listening visits. At 18 weeks the corresponding figures were 60/97 (62%) and 56/109 (51%), respectively.

Proportion improved at 4 weeks

Differences in proportions improved (that is, EPDS < 13) at 4 weeks between those randomised to antidepressants and those randomised to listening visits were analysed using a logistic regression model adjusting for baseline EPDS score and centre and analysed on an ITT basis. Results are shown in Table 15, row a. Women in the antidepressant group were more than twice as likely to have improved 4 weeks after randomisation as those randomised to the listening visit group.

As there was considerable variation in the elapsed time of follow-up, the effect of this was examined first by adding the elapsed time to the regression model and then by restricting the analysis to those women who completed the questionnaire within the target time frame (3–6 weeks). The inclusion of the elapsed time in the model had no effect either on the adjusted odds ratio or its confidence interval (Table 15, row b). Only 161 women (73 antidepressants, 87 listening visits) completed the questionnaire in the target time frame. Restricting the analysis to these women barely altered the results (Table 15, row c).

The model was then adjusted for the baseline imbalances apparent in Table 8 (diagnosis, number of children, breastfeeding, previous antidepressant treatment, employment status). This had very little effect, if anything increasing the odds ratio slightly (Table 15, row d).

The efficacy of drug therapy during the first 4 weeks of the trial was explored using explanatory analyses to derive CACE estimates: first involving self-report data and second prescription data. Both analyses gave very similar results.

As detailed in Table 14, during the first 4 weeks of the trial, 66 (59 antidepressant, seven listening visits) of the women who returned the questionnaire reported taking antidepressants whereas 152 (47 antidepressants, 105 listening visits) were receiving only GSC. In this case, the difference in the percentages improving by the 4-week follow-up was reduced (41% and 28% for the antidepressant and listening visits groups, respectively), but from the corresponding instrumental variables regression model there was still strong evidence in favour of antidepressants – namely, an adjusted difference in percentages of 48% albeit with a wide 95% CI of 23 to 73% (Table 15, row e). This difference is much larger than that from both a corresponding (ITT) difference according to allocated group (24%; 95% CI 12 to 36%) and a crude, biased comparison of the 66 women who reported taking antidepressants versus the 152 who did not (15%; 95% CI 2 to 28%). The latter comparison is based on the two observed proportions (Table 15, row e), albeit after adjustment for baseline EPDS and centre; the CACE estimates, on the other hand, are very different from the crude difference, reflecting the considerable selection effects operating after randomisation. Very similar results were obtained comparing the 57 women followed up who were prescribed an antidepressant according to practice data compared with the 144 who were not (Table 15, row f).

Proportion improved at 18 weeks

The above analyses were repeated to examine the differences in proportions improved 18 weeks after randomisation (Table 16).

In the ITT analysis, the likelihood of improvement was 10 percentage points higher in the antidepressant group than in the listening visits group, although from the logistic regression analysis there was no clear evidence of benefit for one group compared with the other (Table 16, row a). Adjustment for timing of questionnaire completion and baseline imbalance had very little impact on these results (Table 16, rows b to d).

Self-report and practice data were again used to examine treatment effects for antidepressants in CACE analyses (Table 16, rows e and f), and RHV records for listening visits (Table 16, row g). For both interventions the observed differences were small [up to 6 percentage points with a standard error of 7–8 percentage points in crude treatment(s) received analyses]. The differences from the CACE analyses were larger (in one direction or the other) than the crude differences, and were plausible in light of the selection effects apparent from the ITT and crude treatment(s) received results. In addition to reservations about the assumption of independent effects of the two interventions in these (separate) CACE analyses, all the estimates in these models had wide confidence intervals and were within chance variation (Table 16).

The lack of evidence for differences at 18 weeks is likely to be the result of a combination of reduced power consequent on the original sample size not being achieved and a genuinely reducing effect over time, exacerbated by the considerable degree of switching across the two interventions by the later follow-up. Indeed, the precision of the CACE analyses will very likely be reduced by the latter, especially for listening visits which such a large proportion had received by the later follow-up.

A repeated measures logistic regression analysis was also performed to compare the improvement in the EPDS over time in both randomisation groups, adjusted for baseline score and centre. An interaction test led to evidence of a reduction in the odds ratio between the randomisation groups between 4 and 18 weeks (p = 0.032); ignoring this differential effect led to an overall odds ratio between the groups of 2.0 (95% CI 1.3 to 3.0), p = 0.001.

EPDS score as a continuous variable – descriptive statistics

The mean (SD) EPDS scores of both randomisation groups from the initial screen to the 18-week follow-up are shown in Table 17 and Figure 5. Both groups had similar mean scores at the initial EPDS screen and at the home visit eligibility assessment but the improvement was greater for the antidepressant group at the 4-week follow-up. However, by 18 weeks there appeared to be convergence. The distribution of EPDS was investigated by considering the histograms of the change in score from baseline to the 4-week follow-up within each intervention group separately for various candidate transformations (using the gladder command in stata). In no case was there any indication that the distributional assumption of the regression models would be improved by transforming the data, and in any case the sample sizes are generally large enough for such methods to be robust. These investigations were repeated for all the other secondary outcomes, with the same conclusion drawn in each case.

FIGURE 5.

Comparison of the mean (± 2SE) EPDS scores for the women randomised to antidepressants or listening visits, from initial screening to the 18-week follow-up (with lower scores reflecting better mental health).

EPDS score at 4 weeks as a continuous variable

Differences in EPDS scores from baseline to 4 weeks between those randomised to antidepressants and those randomised to listening visits were analysed using multiple linear regression models for the 4-week EPDS score adjusting for baseline EPDS and centre on an ITT basis. The mean EPDS score was about 2 points lower in the antidepressant group than in the listening visit group, with a margin of error of about 1 point (Table 18, row a). The point estimate corresponds to a standardised difference of about 0.4 SDs. Neither adjustments for time elapsed to follow-up, nor baseline imbalance made any substantial difference to these results (Table 18, rows b to d).

From the CACE analyses, an instrumental variables regression for the continuous EPDS score at 4 weeks (adjusting for baseline EPDS and centre) led to an estimated difference in means between those who reported taking an antidepressant and those reporting not doing so of – 4.2 (95% CI – 6.8 to – 1.6), p = 0.002 (Table 18, row e). This difference is larger than that from both the corresponding ITT analysis (Table 18, row a) and a crude (biased) comparison of the 66 women who reported taking antidepressants versus the 152 who did not (– 1.1, 95% CI – 2.4 to 0.3). The latter comparison is based on the two observed means (Table 18, row e), albeit after adjustment for baseline EPDS and centre; the CACE estimates, on the other hand, are very different from the crude difference, reflecting the considerable selection effects operating after randomisation. Very similar results were obtained comparing the 57 women followed up who were prescribed an antidepressant according to practice data compared with the 144 who were not (Table 18, row f).

EPDS score at 18 weeks as a continuous variable

The number of questionnaires returned at the 18-week follow-up point was 206 (97 antidepressants, 109 listening visits). The analyses for the 18-week EPDS scores are shown in Table 19. Although the mean EPDS score at 18 weeks was still lower for those in the antidepressant group than the listening visits group in all analyses performed, the difference between the groups had reduced considerably and the scores for both groups appeared to be converging (Figure 5). For the primary ITT comparison and the initial secondary analyses (Table 19, rows a to d), the differences between the means of the randomisation groups reduced to less than 1 point (about 0.2 SD) by 18 weeks. Moreover, the p-values all indicate lack of evidence of differences beyond chance and the 95% confidence intervals spanned zero.

Self-report and practice data were again used to examine treatment effects for antidepressants in CACE analyses (Table 19, rows e and f), and RHV records for listening visits (Table 19, row g). As with the primary outcome, for all CACE analyses the differences were larger (in one direction or the other) than the crude differences, and were plausible in light of the selection effects apparent from the ITT and crude treatment(s) received results. In addition to reservations about the assumption of independent effects of the two interventions in these (separate) CACE analyses, again all the estimates in these models had wide confidence intervals and were within chance variation (Table 19).

As with the primary outcome, the lack of evidence for differences at 18 weeks is likely to be the result of a combination of reduced power and the considerable degree of switching across the two interventions by the later follow-up, especially for the CACE analyses.

A repeated measures analysis was also performed to compare the improvement in the EPDS score over time in both randomisation groups depicted in Figure 5. The outcome EPDS scores at 4 and 18 weeks were again adjusted for baseline score and centre. An interaction test led to marginal evidence of a reduction in the difference between the randomisation groups between 4 and 18 weeks (p = 0.070); ignoring this possible differential effect led to an overall difference between the groups of – 1.4 (95% CI – 2.4 to – 0.3), p = 0.013.

Other secondary outcomes

There were no adverse events or serious side effects of treatment in the trial. For the continuous secondary outcomes, results for the mental and physical component z-scores for the SF-12, the EQ-5D score and thermometer, the MAMA and GRIMS scores are presented in Tables 20–31.

From these tables (in which the sample sizes vary because of missing scores; see Chapter 2), the results for the SF-12 mental component score (Tables 20, 21) are very similar to those for the EPDS score, especially in its continuous version (Tables 18, 19). Specifically: the comparative benefit in scores for the antidepressants group at 4 weeks is apparent for the primary ITT analysis and all of the initial secondary analyses; likewise for the CACE analyses at 4 weeks: at 18 weeks the differences were reduced and not beyond chance variation albeit with wide confidence intervals especially for the CACE analyses. There was no evidence in any of the analyses for differences between the various groups in terms of the SF-12 physical component score (Tables 22, 23).

For the EQ-5D utility score (Tables 24, 25) all analyses indicated marginal evidence in favour of the antidepressant group at 4 weeks, with no differences at 18 weeks. For the EQ-5D VAS valuation (Tables 26, 27) the pattern was similar to the score although in this case the evidence was even weaker for the 4-week analyses.

For the MAMA attitude to baby subscale (Tables 28, 29) there was weak evidence of benefit to the antidepressant group at 4 weeks from the ITT analyses, and secondary analyses including the CACE models. At 18 weeks there were mostly similar differences but the evidence was even weaker.

For the GRIMS relationship scale (Tables 30, 31) there was no evidence of differences at 4 weeks in any of the analyses, and while the differences were slightly greater at 18 weeks, there remained no convincing evidence of any differences.

Sporadic low p-values among these secondary outcomes should be interpreted with caution given the number of such outcomes considered. Nonetheless, overall the results for the secondary outcomes support an effect specific to mental-health benefits of antidepressants at the 4-week follow-up, with equivocal findings at the 18-week follow-up (albeit in the context of the above-stated caveats about power at 18 weeks).

Quality-adjusted life-years

We estimated QALYs using data from the EQ-5D administered at baseline, 4 weeks and 18 weeks. We had complete data for 192 (76%) women. Those in the antidepressants group achieved 0.263 QALYs over the 18 weeks, an annual equivalent of 0.76, and those in the listening visits group achieved 0.253 (annual equivalent 0.73). These results provide no evidence of a difference between the two groups.

The 4-week wait

Some women described how they had found the 4-week wait difficult but bearable because they knew they would be receiving help and the treatment they wanted. Most of the women, however, detailed how they had been desperate to start. One individual had even worried about harming her child:

‘I felt at first when she said you had a wait a month or whatever, anything could happen in that month, you know, what if I lost my temper, lost my rag with this baby?’ (Participant 20, listening visits then antidepressants)

Another individual (participant 10, listening visits only) recalled wondering why she was not receiving treatment when she had just been told that she needed it.

Although women detailed how they had been desperate to start the visits, only two of them mentioned receiving any support during this period: one individual had been visited by her own PHV and another participant by her aunt. The only evidence of women proactively seeking help during the 4-week wait came from one woman who had called the RESPOND research associate, asking if the visits could start early.

The listening visits

Only one individual, who had listening visits, had received just four visits. All the other women who had received this treatment had the maximum number of visits available in the trial, i.e. eight. All the women who had received listening visits reported that they had found the visits helpful. Women appeared to have benefited because the visits gave them an opportunity to talk, because they had found the process of talking therapeutic, and because they had found the support and advice given by RHVs helpful.

Ending the visits

Fourteen out of the 22 women who had received listening visits stated that they felt eight visits had not been enough. Having completed the visits, 10 of these women went on to see their GP. Six of them were then prescribed antidepressants, two of them received counselling and the remaining two women were referred to a psychiatrist. The other four women had gone on to see a private counsellor, had stayed on medication having started antidepressants before the visits, or had not sought further help because they were nervous about going to their GP or talking to their PHV.

Most of the women who had visited their GP reported that their concerns about antidepressants had been addressed, indicating that the GP consultation could be a factor in changing women’s views of antidepressants. GPs had discussed women’s fears and explained how antidepressants might help. In addition, they had prescribed antidepressants in a way that had been reassuring:

‘She [GP] said, “Well just try it and see how you get on…I’ll give you something very light…it’s not that addictive…this is a low dosage one.” And I think that probably helped as well…You’re thinking, okay, I can cope with that.’ (Participant 19, listening visits then antidepressants)

However, a few of the women described how, despite talking to their GP, their concerns about taking medication had remained and it was apparent that they had gone on to take medication because of a lack of treatment choice. Antidepressants had been used to tide women over while they waited for counselling arranged through their general practice, or were turned to when the individual was unhappy with the counsellor available through her practice. It was also apparent that the responsibilities of parenthood, and the symptoms of depression, could mean that visiting a counsellor was not feasible:

‘I did say was there any counselling that was available that I could access, and they said “not really…[and] they don’t come for you at home…” It was very difficult because I have two children to look after, in my present state of mind as well, like just driving a car and catching a bus is something that would be a nightmare for me. And they said the other option is antidepressants, and they started me on antidepressants.’ (Participant 22, listening visits then antidepressants)

When comparing the accounts of women who commented that they felt eight visits had not been sufficient with those of women who reported that they had received enough visits, it appeared that women who had wanted further support were individuals whose PND did not relate primarily to their experiences of motherhood or to their circumstances during the postnatal period, which appeared mainly to be the case for the other women. Women who had wanted more listening visits described how they had felt depressed before their pregnancy or detailed how they had struggled with depression for years, or mentioned that the birth of their child had raised memories of past negative events, e.g. terminating a pregnancy, death of a father. Some of these women also described how they had overdosed in the past, had suicidal thoughts and/or self harmed. In contrast, the women who felt that four or eight visits had been sufficient, related their PND to a hormonal imbalance, needing to adjust to parenthood, having a particularly difficult time following the birth of their child, or to problems that had now been resolved, e.g. relationship difficulties, unrealistic expectations of self. It was also apparent that most of these women, while receiving the visits, had put other sources of support in place, e.g. they had started Sure Start, made new friends, attended parenting classes, started taking antidepressants which they felt were now working.

During the interviews held with women who had felt a need for further treatment, women talked about how they had felt angry that the visits had ended, and individuals described how they felt that they had been ‘left hanging’ and ‘completely exposed’. In addition, it was pointed out that that if no further support had been available through the GP, the individual would have felt that the listening visits had been pointless:

‘Just me thinking about it [the idea of no treatment after the visits] now makes me feel quite panicky. I am feeling like, oh my god, oh that’s awful…what would have been the point of ripping off the plaster and starting to abrade the wound, only to then just say, oh well.’ (Participant 10, listening visits only)

Coming off antidepressants

At the time of interview, seven women were still on antidepressants. In some cases this appeared to be because of a fear of stopping:

‘You persevered with them [antidepressants]?’

‘I still am (laughs)…I’m too scared to come off them.’ (Participant 26, antidepressants then listening visits)

‘I sometimes feel what’s the point in taking these but I know they’re in my system, so I’d like to just keep it like that.’ (Participant 22, listening visits then antidepressants)

Women who had stopped taking their antidepressants had done so because they had felt better. In some cases, individuals had stopped without consulting their GP.

While 11 out of the 21 women whose first treatment was listening visits had gone on to take antidepressants, only two of the five women whose first treatment was medication went on to have listening visits. Although one individual could not remember why she had started the visits, the other individual explained that she had always intended to have them as this had been her treatment preference at the time of randomisation. The reason so few women taking antidepressants went on to receive listening visits might have been because three of the five women who started on antidepressants as their first treatment, were individuals who had wanted this treatment and had found antidepressants sufficient for them to cope with their PND.

The diagnosis of postnatal depression

Disclosure

Some GPs described strategies used to facilitate disclosure and offer women ongoing support:

‘Once you kind of know they’re in distress you don’t just give them one session, you ask them to come back always…you get them to come back 2 weeks later to see how they’re doing.’ (GP, L1)

In addition, a minority of HVs described exploring depressive symptoms as a routine part of their interactions with women:

‘…but I think most people have actually heard about it now and don’t find it unusual that you ask them about their mood and how they are feeling.’ (HV, B2)

Both GPs and HVs, however, described the current systems of care as hindering disclosure of symptoms of PND:

‘You know, we’re not user-friendly in the health services…say someone is de-motivated because of low mood, then they ring for an appointment and they can’t get through and then we ask patients “do you really need to be seen today?”…they have to jump through hurdles.’ (GP, B6)

‘They know it’s a hurried environment, and they know it’s hardly the environment for them to give you clues or confide in you that they’re depressed.’ (HV, M5)

Health visitors, however, suggested that there was limited value in identifying women with PND as all they could do was to refer a woman to the GP and they viewed the GP’s role as limited to the prescribing of antidepressants:

‘So, there’s almost an ethical dilemma of, well, is there any point in identifying them if you can’t do anything with them other than send them to the GP for antidepressants, which isn’t good, you know.’ (HV, M5)

This view pervaded the HV transcripts and was suggested to be a major reason why HVs did not encourage women to disclose their feelings.

Team work

Whose responsibility is postnatal depression?

Although a few GPs talked about the pivotal role they perceived HVs to play in the detection and management of women with PND, most alluded to the recent reorganisation and change in role with a perception that women with PND go undetected. Other GPs describe how HVs have actively declined to support women with what was dismissed as a ‘mental-health problem’, although agreed that HV teams still had a role in offering practical support to women with PND:

‘Well, our health visitors tend to say that it’s a mental-health problem ‘nothing really to do with me’, which is disappointing really. They do go in and offer support but it’s very vague what that support is. They cover some practical things around sort of nursery services for the children, stuff like that and it depends which health visitor team it is because not all will take an interest.’ (GP, M7)

Many HVs suggested that there was little point in identifying women with PND, as they felt they had nothing to offer women themselves and no resources to refer women to:

‘In an ideal world we’d want to pick them up and then offer them more support, but we can’t do that. So there’s almost this ethical dilemma of well is there any point in identifying them if you can’t do anything with them other than send them to the GP for antidepressants, which isn’t good, you know?’ (HV, M5)

The HV view that there is limited value of referring a woman back to the GP is seen again in these data. Other HVs, however, were quite clear that it was the GP who was the responsible clinician. HVs agreed that they no longer saw the detection and management of women with PND as a priority for them:

‘…erm, at the moment we’re short-staffed, so we’re really on priority cases at the moment.’ (HV, M1)

And a few HVs described negative aspects of providing support to women with PND with a risk of dependence on them:

‘You cannot emotionally and mentally prop somebody up for years and years, it’s got to end…sounds awful, doesn’t it?’ (HV, B4)

So both GPs and HVs reported that the current systems within which they work constrain the care that can be provided to women at risk of PND and HVs no longer see the management of women with PND as an integral part of their work.

Views on the RESPOND trial

Despite exploration with all respondents about their experiences with the RESPOND trial and the interventions that women in their practice or HV team had experienced, few GPs or HVs were able to discuss experiences or outcomes of women. There were few data in the interviews with GPs about the HV-delivered intervention in RESPOND, but they did talk positively about the trial itself:

‘Well, I’m not quite sure what’s going on now. I realise I should know because I was there at the meetings, and I know [receptionist’s name] has a pile of those things that she sends out to people when she sends the postnatal thing and I presume they fill them in. I don’t, actually, I’ve forgotten what happens after that.’ (GP, M1)

They were particularly positive about the organisation of trial and information given about prescribing:

‘I’ve been enormously impressed, the communication side is excellent, the information that you supply, that a patient has been randomised to antidepressant treatment and they’ll be coming in, and an appointment’s arranged and the information sheet about antidepressants and advice about breastfeeding is there…so it was all good to see.’ (GP, M2)

‘It highlighted a few patients who we saw earlier and who we treated, which we probably wouldn’t have otherwise. I think that it was worth doing just for that if nothing else.’ (GP, L1)

Some GPs suggested that the intervention being delivered by a separate HV might be beneficial:

‘I think the benefit [of being involved in RESPOND] is the patients always feel that there’s someone else interested in them, so they have benefited in that respect, otherwise no.’ (GP, B2)

On the whole, the GPs valued the trial as providing a service for women who might otherwise not receive one:

‘I think probably the two women diagnosed [within RESPOND] were not by our GPs, I think we feel confident that people are being looked after with their postnatal depression.’ (GP, B5)

This view resonated with some HV accounts, where respondents were unaware of the intervention delivered by the RHV, possibly because they had insufficient women in the trial:

‘I’m afraid on my caseload there’s only one that has sort of completed the process, so I feel unable to comment on the value to individuals.’ (HV, B2)

Others described how having women in the trial provided support to their own service:

‘I think postnatal depression is very important, I think it’s an unmet need so I carry a big guilt complex there, and this [being involved in RESPOND] will ease the guilt complex very, very slightly.’ (HV, M5)

Health visitors suggested reasons why the trial might be useful for women, in particular, that it gave women permission to disclose how they felt:

‘I think sometimes they don’t like to disclose it because it’s seen as, they think they’re seen as not coping, you know, that’s why I think this study has been useful because it’s specifically for them, you know, they’ve done it in their own time, had it at home, read it through and decided that they do want help.’ (HV, M1)

‘…I think with the study [RESPOND] somebody new goes in and says “it’s all about you, I’m here to talk about you.” Whereas they probably do get a bit of conflict thinking you’re looking at the children all the time.’ (HV, M1)

There were, however, some negative comments from HVs who felt the trial had encroached on their role:

‘I’m finding it quite difficult, really, taking a back seat, I had a kind of rhythm going and that’s been stopped.’ (HV, M3)

‘…going back to the RESPOND trial, I mean, it’s sort of taken away some of what we’re doing if you see what I mean, but then we haven’t got time to do it.’ (HV, L3)

And some HVs expressed concerns about the EPDS being sent to women at home:

‘It’s not brilliant, unfortunately, and I’m sure they don’t fill them in.’ (HV, L5)

In summary, the interviews yielded insufficient data to allow a full exploration of the acceptability of the trial intervention by the primary care professionals, who had only experienced one or two of their patients being part of the trial. Although having mostly positive views of being involved in the study, they did not express any views on the HV-delivered intervention.

Strengths

This chapter reports a qualitative study embedded in a randomised controlled trial. The use of qualitative methods allows practitioners to raise issues that are of concern to them, and an inductive approach ensures that findings are related to the views articulated. The data were gathered from GPs and patients drawn from a large geographical area (nine PCTs). Using researchers from different professional and academic backgrounds is a recognised technique for increasing the trustworthiness of the analysis. 105

Limitations

Only HVs and GPs who were already involved with the RESPOND trial were interviewed and it seemed difficult to engage with London GPs even though they were signed up to the main trial. PCTs were invited to participate in RESPOND if they did not have a well-established pathway of care for PND within the PCT, and so participating PCTs may have poorer provision of services and attitudes of the health professionals working in these areas will reflect this. For this reason, the findings may not be representative of (even neighbouring) PCTs who may have developed a PND strategy and services for this group of patients. As respondents had limited experience of their women being in the trial, it proved difficult to fully explore acceptability of the RHV-delivered intervention from GP and HV perspectives.

Understanding postnatal depression

Women attributed a psychosocial aetiology to their symptoms in relation to the stresses of parenthood, such as changed relationships, reality not meeting expectations, and the birth of the child triggering memories of past events:

‘And then there was a whole set of issues about my relationship with my partner and how he was supporting me and these issues were all thrown up when my little girl was born and then when my little boy was born, it happened exactly the same. It was a real repeat, you know, and I’d done so much to try and prevent it going down the same route, but I just felt like we were going down the same groove.’ (B, ID28)

Some women described their feelings as a response to physical changes around childbirth, but suggested that they were susceptible to depression in some way:

‘I personally feel it…well, probably two factors…but I think perhaps it is some sort of chemical hormone or imbalance, you know, everything sort of all shifts it about and…and secondly, just the type of character that I am, and putting that pressure on myself the whole time.’ (L, ID16)

Women described insights into and awareness of their symptoms, often because they had suffered from depression in the past, although they suggested that the cause of PND might be different to the cause of previous episodes:

‘I’d had slight depression before, but this was quite different, I kind of knew why, whereas before it was kind of all suffering and something had triggered lots of other things, and here it was lack of sleep, and although I knew the reasons why I was feeling like I was feeling, but I couldn’t stop it.’ (B, ID3)

Health professionals also attributed a psychosocial aetiology to PND and demonstrated ambivalence about the status of PND as a separate condition as compared with depressive illness at other times in a woman’s life:

‘I can certainly give you a list of things that would put women at risk, but, you know, clearly doesn’t always result in postnatal depression. So a previous history of mental-health problems or depression, unfulfilled expectation, difficult birth, wrong sex, partner unsupportive…but, I think there’s quite a large proportion where there appears to be no risk factors.’ (B, HV2)

Thus, both women and health professionals viewed the cause of PND as multifactorial and often a social response to birth.

Making the diagnosis

As discussed in the previous chapter, GPs and HVs described a reliance on instinct or clinical intuition, which would alert them to the possibility of PND, rather than using formal screening instruments or actively seeking out symptoms of depression:

‘I think any kind of flatness…it’s a difficult thing to explain, isn’t it?…You can just tell by having a conversation…just chatting to them.’ (B, HV1)

The GPs described difficulties in using the label for PND with women, particularly referring to the stigma that they perceived was felt by women, and the effect of this on the consultation:

‘I mean, if they deny that they have got a problem but are still in tears, it becomes very difficult, because you can’t treat somebody if they don’t accept that there’s something to treat.’ (B, GP1)

Other GPs, however, described consultations where the woman was happy to accept the label:

‘…and equally others will just come in and say “my husband said I’ve got to get this sorted out, and I need a tablet to calm me down” or whatever. You get the whole spectrum, really.’ (M, GP2)

Health visitors did not describe such dilemmas within their consultations with women, although they recognised the reluctance women might feel to accept the label of depression and the anticipated treatment:

‘But yes, I’ve found there are many clients who don’t want to take medication, because they do think there is a stigma attached.’ (L, HV5)

Some GPs described a reluctance to use the term ‘postnatal depression’ because they felt that symptoms would recover without formal interventions, because of a lack of services or referral options, and the feeling that antidepressants were the only treatment option available:

‘I don’t want to medicalise it too much really. I think it needs to be a sort of informal sort of network because I do think most of the time people do recover from it if they are just given some support rather than medication.’ (M, GP8)

‘I mean, it’s best if it’s a multiple approach rather than just drugs. Unfortunately that’s all we can offer.’ (L, GP1)

Health professionals described a variety of difficulties making the diagnosis of PND; HVs because diagnosis was not felt to be within their remit, and GPs because of the perception that they had limited management options to use if they used the label of PND.

Disclosure and hindering disclosure

Women described making a conscious decision about whether or not to disclose their feelings to their GP or HV. Some women cited their own personal barriers to being able to talk to their GP:

‘Again, GPs, I don’t think, er, well I personally couldn’t talk to my GP.’ (M, ID16)

Other women mentioned characteristics of GPs which inhibited their ability to disclose, such as the GP being perceived as not willing to listen:

‘…he [GP] could have listened. Again, I think they could have done that at least, they could have listened…’ (M, ID22)

Other women described system factors which made them reluctant to attempt to approach their GP to discuss how they were feeling:

‘…wouldn’t go to the doctors because you can never get an appointment and it’s crap. They always treat you like there’s something else wrong and why are you wasting his time…I wouldn’t have gone [to the doctors] even if I’d been dragged kicking and screaming.’ (M, ID24)

Many women described a fear of disclosure because of how they would be perceived by their HVs, such as being a bad mother, and others described fear of having their children being referred to social services:

‘…with my health visitor, I, I try not to, try not to let too much out because then she won’t think I am a bad mum, if you see what I mean, so I tend not to let too much out with the health visitor.’ (B, ID2)

Women also questioned the role of the HV and who she was there for:

‘…what is the health visitor there for? Is she there for the welfare of the child, or is she there for the welfare of the mother, or both?’ (M, ID24)

‘No, I don’t think I discussed that with the health visitor. I was more concerned with her [baby] at the time. So I just assumed health visitors were for looking after her [baby] needs, so if she had a rash or was throwing up, or whatever, things that I wouldn’t necessarily know what to do about. Whereas your own needs you don’t tend to think about them in the same way. So there’s known and unknown, and she [the HV] was definitely an unknown’ (L, ID12)

Other women described a fear of consulting their GP as they anticipated that the only treatment that would be offered would be antidepressants, which they did not feel was an acceptable treatment option:

‘That’s all they have, GPs, and I just didn’t want to go onto antidepressants, because obviously I’ve heard people get addicted to them and then you’re stuck on them and you have a vicious circle.’ (M, ID24)

‘My concern is that I will just get addicted and it will change my personality.’ (B, ID1)

This view might be reinforced by the women’s HVs because the HVs who were interviewed described GPs’ role in the management of PND as being limited to the prescribing of antidepressants:

‘The GPs…and I think they’re even worse than we are because they…from experience of women I’ve visited, they just write a prescription, put them on antidepressants…’ (HV, M2)

‘…sometimes the GPs are, they don’t have a very sympathetic attitude to postnatal depression, let’s say. And so I would imagine it puts the mothers off going to see them actually.’ (HV, L5)

A few women did talk about why they felt able to discuss their symptoms with the GP and cited factors such as having been to their GP with depression or PND in the past and recognising the symptoms. Women who had previous experience of antidepressants were more accepting of this treatment being offered to them again:

‘I thought, well, I’ll try them and you know, it did help a bit last time, not, you know, it wasn’t fantastic, but it did help a bit so I thought well, okay, I’ll try them again.’ (B, ID6)

Most importantly, those women who did feel comfortable seeking help from their GP described having a good relationship with him/her, making discussion of depression possible:

‘I don’t go to the doctor that often but, well [names child] was with [names doctor] so I’ve been quite a few times with having colds and stuff so I knew him quite well, so it was quite easy to go and say “look, I’m just, I’m not feeling right at the moment”.’ (M, ID25)

Some GPs described strategies used to facilitate disclosure and offer women ongoing support:

‘Once you kind of know they’re in distress you don’t just give them one session, you ask them to come back always…you get them to come back two weeks later to see how they’re doing.’ (L, GP1)

In addition, a few of the HVs described exploring depressive symptoms as a routine part of their interactions with women:

‘…but I think most people have actually heard about it now and don’t find it unusual that you ask them about their mood and how they are feeling.’ (B, HV2)

Most HVs, however, suggested that there was limited value in identifying women with PND because all they could do was to refer a woman to the GP and they viewed the GP’s role as limited to the prescribing of antidepressants:

‘So, there’s almost an ethical dilemma of, well, is there any point in identifying them if you can’t do anything with them other than send them to the GP for antidepressants, which isn’t good, you know.’ (M, HV5)

This view pervaded the HV transcripts and was reported by the HVs to be a major reason why they did not encourage women to disclose their feelings to their GP.

How the system of care hinders disclosure

Both GPs and HVs described the current systems of care as hindering disclosure of symptoms of PND:

‘You know, we’re not user-friendly in the health services…say someone is de-motivated because of low mood, then they ring for an appointment and they can’t get through and then we ask patients “do you really need to be seen today?”…they have to jump through hurdles.’ (B, GP6)

‘They know it’s a hurried environment, and they know it’s hardly the environment for them to give you clues or confide in you that they’re depressed.’ (M, HV5)

Some health professionals described consciously inhibiting disclosure so as not to be placed in this position citing lack of continuity of care as the reason:

‘Easier not to ask, if I’m not going to see her again.’ (L, GP1)

‘For the vast majority of them we don’t (see them again), we’re not able to offer even a routine follow-up visit, even if it’s their first baby. We explain the situation to them in terms of staff and resources, and we encourage them to come to clinic or to phone us, and we explain that we do visit some families at home and offer them extra support…however, in terms of listing the priorities I would say the postnatal depressed ones aren’t high up on the agenda. When we’ve got much more prioritised.’ (M, HV5)

‘…but I think they used to get to know us, and we used to get to know them and obviously if they know someone they’re more likely to sort of be forthcoming with any problems aren’t they? Whereas now they, they probably don’t get that input so they’re probably less likely to come forward with things.’ (M, HV4)

A few of the HVs described a resistance to being involved in providing care for women with PND, and this reluctance was made easier by the move to corporate working:

‘Working corporately, we all work between the GPs and as work comes in we allocate families…I mean the families aren’t becoming reliant on you, that’s the good thing.’ (M, HV1)

‘You cannot emotionally and mentally prop somebody up for years and years…sounds awful, doesn’t it?’ (B, HV4)

Some women suggested that neither their GP nor their HV could do very much, which resonates with the views of HVs on the GP role:

‘There is nothing else available. No GP, no health visitor; they’re there but not in a helpful sense, sort of like.’ (M, ID22)

‘…because I was very struck by the health visiting that I’d had so far. I hadn’t felt like my needs had been met at all.’ (M, ID27)

‘The doctors are like, always helpful, but I think they’re limited as to what they can do, which is what I thought, you know…that they can’t actually do that much…’ (B, ID6)

Both GPs and HVs described organisational factors which precluded the facilitation of the disclosure of depressive symptoms in their postnatal women patients. Difficulties providing continuity of care were frequently cited. Women talked less about system factors, but rather suggested that neither the GP nor HV had much to offer. The development of long-term relationships which may facilitate disclosure may be impeded by the systems within which health professionals currently work.

Primary outcome

Two hundred and fifty-four women with PND were randomised to receive either antidepressants or listening visits from a HV. At 4 weeks, women with PND were more than twice as likely to have improved, i.e. EPDS < 13, if they had been randomised to antidepressants compared with women randomised to listening visits, which started after 4 weeks follow-up, i.e. after receiving GSC for 4 weeks (primary ITT, 45% versus 20%; odds ratio 3.4, 95% CI 1.8 to 6.5, p < 0.001). At 18 weeks, the proportion who improved was 11 percentage points higher in the antidepressant group than in the listening visits group (62% versus 51%), although from the logistic regression analysis there was no statistical evidence of benefit for one group compared with the other (primary ITT, odds ratio 1.5, 95% CI 0.8 to 2.6, p = 0.19). The trial has therefore provided answers to the two main questions posed. Antidepressants offer significantly greater improvement compared with GSC at 4 weeks, and at 18 weeks women randomised to antidepressants still seem to have an increased chance of improvement compared with those randomised to listening visits although this difference is no longer significant.

At 4 weeks, the secondary analyses of the primary outcome made virtually no difference to the size of the difference between the two groups. The increased magnitude of the differences in the CACE models is the result of patient choice entering after randomisation, where those who start to improve spontaneously or improve early on the allocated treatment may make a choice to wait and see whereas those not improving tend to add treatments. Using a 4-week end point is by current standards very early in treatment and if randomised allocation had continued unaltered to 8 or 12 weeks there may have been a further increase in difference – but this was not possible in a pragmatic trial that included an arm with no active treatment in the first instance. However, GPs can be reassured that prescribing antidepressants about 10 weeks postnatal gives a much better chance of improvement over the next 4 weeks than GSC.

It is not possible from these separate analyses to address directly the issue of the evidence for a change through time in the differences between the groups, but the interaction test from the repeated measures logistic regression analysis does indicate weak evidence that this reduction in the odds ratio is greater than would be expected by chance (p = 0.032) This supports the finding that the benefit of taking antidepressants was larger at 4 weeks than at 18 weeks when the comparison was between antidepressants and listening visits (if needed) versus listening visits. Unlike many trials, the comparison has changed between the two follow-up points. Overall, there is evidence of a difference between the groups in favour of the antidepressant group of about 25 percentage points at 4 weeks which is reduced at 18 weeks. There is no statistical support for a benefit of antidepressants at 18 weeks, but the confidence intervals cannot rule out a clinically important benefit.

Secondary outcomes

When the EPDS was considered as a continuous score, the difference in EPDS from baseline to 4 weeks between those randomised to antidepressants and those randomised to listening visits was about 2 points lower in the antidepressant group than in the listening visit group, with a margin of error of about 1 point. The point estimate corresponds to a standardised difference of about 0.4 SD which is very similar to differences found between antidepressants and placebo. Although the mean EPDS score at 18 weeks was still lower for those in the antidepressant group in all analyses performed, the difference between the groups had reduced considerably and the scores for both groups appeared to be converging. For both the primary ITT comparison and the secondary analyses, the differences between the means of the randomisation groups reduced to less than 1 point by 18 weeks, the p-values all indicated lack of evidence of differences beyond chance and the 95% CI spanned zero. The interaction test from the repeated measures regression analysis led to only marginal evidence of a change in the difference between the groups over the two follow-up times (p = 0.070). The average effect over the follow-ups of – 1.4 (95% CI – 2.4 to – 0.3, p = 0.013) should also be treated with caution for similar reasons to those given above for the primary ITT analysis.

For the other (continuous) secondary outcomes, the results for the SF-12 mental component score were very similar to those for the EPDS score, especially in its continuous version. In particular, the benefit in scores for the antidepressants group was observed at 4 weeks for the ITT and all secondary analyses, whereas no such differences were apparent at 18 weeks. The lack of any corresponding differences between groups for the SF-12 physical component score and the EQ-5D (which only includes one question about psychological symptoms) suggests that any effects are specific to mental-health quality of life.

For the MAMA attitudes to the baby subscale and the GRIMS partner relationship scale there was very little difference apparent between the various groups, although the suggestion that those who received both interventions by 18 weeks have the worst outcome is consistent with the other outcomes and presumably reflects strong selection effects by those who are not improving on their initial clinical management.

Other secondary analyses

Although there was a reasonably small amount of attrition overall in the context of this study (approximately 15% by 4 weeks and 20% by 18 weeks) there was evidence of differential attrition rates across the randomisation groups, with fewer lost to follow-up in the listening visits group (p = 0.090 and p = 0.015 at 4 weeks and 18 weeks, respectively). The sensitivity analyses involving multiple imputation, however, indicated that this had no appreciable effect on the results. Only 22 women had missing EPDS scores at both follow-up times and the differences between the groups were similar in direction (if not magnitude) at both time points. Furthermore, baseline EPDS was strongly associated with outcome and used as a covariate in the ITT analysis. It would appear that there was little impact of attrition on the observed results.

Reflecting their low power, overall the prespecified subgroup analyses gave only weak evidence of differential effects according to baseline severity of depression. The suggestion of greater effects for those more severe at baseline is both generally plausible and consistent with the pattern of findings for the binary and continuous version of the primary outcome.

Economic evaluation

Whereas the economic evaluation suffered from having originally been set up to report at 44 weeks and from a significant amount of missing data (particularly that relating to secondary care resource use) there is no clear evidence for differential resource use between intervention groups for either mothers or babies at 4 or 18 weeks. There are one or two findings that deserve consideration but all must be treated with caution because of the small sample size as well as the analysis using the original randomised groups with no adjustment for switching groups or adherence. By 4 weeks more women in the AD group had seen their GP than those in the listening visits group – possibly to receive their prescription but it might have been anticipated that women receiving GSC only, in those 4 weeks, might have visited their GP at least as often. By 18 weeks more women in the listening visits group had consulted so although most of these would have been receiving visits from the RHV, some will have been seeing the GP for a prescription for antidepressants. There were remarkably few consultations with the PHV for women but more surprising was that only about 50% of babies were seen by PHVs in either of the 4-week periods. There were no obvious differential patterns of use of either other primary/community-care or secondary-care facilities with on average no more than 20% of either women or babies attending outpatients or Accident and Emergency departments. The greatest resource utilisation by postnatal women is clearly in the primary care setting.

Statistical considerations

It is useful to compare the observed percentage improvement with those assumed in advance as worth detecting in the sample size considerations. At 4 weeks the difference of 25% exceeded that specified in the original power calculations of 15%, and was similar to that accepted for the revised calculations (20–22.5 percentage points). However, the power calculations assumed that 40% would improve in the listening visit (GSC) group at 4 weeks, and in the antidepressant group, 55% in the original and 60–62.5% in the revised calculations. The initial assumptions were therefore reasonably close for the antidepressant group but (quite considerably) overestimated the improvement in the listening visit (GSC) group. A meta-analysis of psychotherapy studies that randomised to a waiting list control group, not dissimilar to our GSC arm, found a similar percentage, 20%, improved in the short term. 106 At 18 weeks, the original assumption of 50% improvement in the listening visits group was very close to that observed; at this follow-up, therefore, the lack of clear evidence of differences reflects the smaller difference observed (11 percentage points) than that either specified as clinically worth detecting in the original power calculations (15 percentage points) or that for which there was in the event adequate power (just over 20 percentage points; see Chapter 2, Sample size). However the confidence intervals are quite broad at 18 weeks and so an important difference cannot be excluded.

In a trial of this nature, the definition of a ‘plausible, clinically important difference’ is always debatable and (as can be seen above) cannot be determined, at present, with any confidence. It is also sometimes argued that small differences become more important for common conditions, and depression including PND is very common and leads to a substantial public-health burden. In RESPOND, the attained sample size was smaller than planned, so the detectable differences clearly increased. However, for what was considered at the outset to be an acceptable power (at least 80%), the attained sample size was adequate to detect a 20 percentage point difference, which is itself clearly clinically important. Moreover, on a number of grounds the sample size calculations could be argued to be conservative. First, although in the analysis the formal primary outcome was the binary version of improvement to an EPDS score of < 13, the continuous EPDS score was a secondary outcome and should have greater statistical power. Second, evidence from the literature reveals that we might have been overly conservative in our original assumptions for the binary version. We had assumed about 55% would have ‘improved’ on antidepressants at 4 weeks compared with 20% receiving no active treatment, whereas data from waiting list control groups similar to our supportive care only resulted in 20% having improved. 106 Moreover, the entry criteria to our study were fairly stringent and in line with NICE guidelines, meaning that randomised participants were more likely to benefit from an intervention, which in turn increases the magnitude of plausible differences.

Qualitative findings

Methodological issues

The practices in each centre were approached with regard to participating in the study on the basis of having a sufficient number of births (about 100) each year, to result in about 21 practices being recruited in each centre. This approach worked well in Bristol and London but in Manchester, practices tended to be smaller, including some single-handed practices, resulting in 35 practices being recruited to ensure a sufficient number of births. This made monitoring of practice activity by the Manchester research associate much more difficult because of both the wider geographical area covered and the number of practices. In each practice, a member of the administrative staff (clerical assistant) was identified as the main liaison with the centre research associate, who would be responsible for collating the practice birth data and on a regular basis, usually weekly, sending information packs out to women who were between 5 and 6 weeks postnatally. The research associate, on the basis of responses received, was then able to ask the clerical assistant to send reminders to the non-responders.

The requirement by ethics committees for patients to be contacted by the clinical team rather than the research team does require the development of a positive relationship between the two teams so that the research team can be sure that letters are being sent out on their behalf as requested. Although NHS support costs are provided to practices to undertake these roles, not all practices recognise the vital and responsible role they are playing in the research process and as the pressure of providing the front-line service for the NHS becomes inexorably greater, research activities are sometimes forgotten. This is demonstrated by the discrepancies between the numbers of invitations sent out by practices and the total births recorded (Table 2). In addition, the timing of sending out the invitations varied depending on how often the clerical assistant found time to undertake this task and this impacted on exactly how long after the birth women received their invitation. This also varied depending on how quickly hospital and PCT notifications of births were sent to practices. However, overall, 95% of women who had been recorded as being eligible for the study by practices received an invitation. This varied between 98% in Manchester and 89% in London. These differences may be explained by the list sizes in Manchester being systematically smaller than those in London and Bristol; also, in London the population mobility is very high, which may have resulted in women moving from their registered address soon after the birth. At PCT level, where the requirement was to provide researchers with regular birth data for each practice, some found this difficult to do on a monthly basis which meant that in some instances the denominator for total births per practice is not secure (Table 2).

During the course of the study, a number of PCTs moved from a system of practice-based HVs to corporate working, which does not provide for the same degree of either practice-linked or even family-linked working by HVs. As time went on, this made it harder to engage PHVs with the study in terms of both encouraging practices and women to participate. PHVs and GPs were vital to the running of the study. In particular, they were responsible for ensuring that the exclusion criteria for the study were adhered to. On the whole the notification system worked well but there were instances where women were not excluded who should have been (for example the two women whose English language skills should probably have made them ineligible). PHVs were also key in disseminating information about the study to pregnant and postnatal women. They gave out leaflets in antenatal classes and carried information packs with them on their postnatal home visits. GPs were less involved on a day-to-day basis and the study was designed so as to cause minimal extra work for primary care professionals. One of their major roles was to assess the real risk to women of entering the study who had revealed some degree of suicidal ideation – an answer of ‘Sometimes’ or ‘Yes, quite often’ to question 10 of the EPDS ‘The thought of harming myself has occurred to me’. The MREC had made it a condition that women who expressed these thoughts had their suitability for the trial reviewed by their GP, mainly to ensure that severely ill women did not receive listening visits when antidepressants might have been the appropriate treatment. This process worked well and in the event very few women were excluded by GPs on the grounds that they were at a substantial risk. GPs and PHVs were also responsible for helping to ensure adherence to protocol. Hence, in the first 4 weeks we asked GPs not to prescribe antidepressants to women in the listening visits group unless necessary and we asked PHVs not to offer formal listening visits to women in either group. In the end GPs seemed to respond by excluding women who they thought might not benefit sufficiently if they were randomised to listening visits and PHVs were able to continue to provide their usual care with referral to the GP if they believed a more formal assessment was needed. It is worth noting that by the time the home visit to assess eligibility took place 94 women were already having some form of treatment for PND.

Recruitment

The consent process for the study was in two parts. This gave women the option to participate in the first screening phase with the use of their data on an anonymous basis but to opt out of the second eligibility (for the trial) phase which involved a home visit and the potential to be randomised to antidepressants. As the consent form specifically stated that women could withdraw from the study at any time without giving a reason, it is not possible to define the reasons for declining further involvement. Hence, 155 (3.6%) did not complete the initial EPDS, and 113/3184 (3.5%) of those scoring < 11 and 181/989 (18.3%) of those scoring ≥ 11 declined to participate further. It is therefore possible that we lost a disproportionate number of women who were likely to fulfil diagnostic criteria for PND at the eligibility phase of the study (see above). Centre differences were again notable among the 969 women who were high scorers on the initial EPDS, ≥ 11, with women from Manchester less likely to have dropped their EPDS score to < 13, more likely to already be on treatment from their GP or HV and have had treatment for depression in the past.

At the time of requesting a home visit, 55 women were known to have improved such that an interim EPDS score was < 8. This method of reducing false positives before the eligibility home visit did not prove to be efficient and was dropped after the first 272 women. On the other hand, 88 women were already on treatment and therefore not eligible for the trial. We tried to encourage GPs and HVs in participating practices to desist from offering treatment to women who had screened positive on the EPDS but despite being able to offer a very tight timetable for randomisation, we lost these 88 women (9%). Home visits were therefore undertaken with 628 women of whom: 298 (47%) were no longer eligible for trial as their EPDS score was < 13, 54 (8.6%), although scoring ≥ 13 on the EPDS, did not have an ICD-10 diagnosis of depression, a further six had commenced treatment for PND, seven were excluded by their GP or HV (all of whom had ICD-10 depression on the CIS-R) and one was past the 26-week limit for entry to the trial. This resulted in 262 women being eligible for the trial, of whom eight refused to be randomised.

A great deal of research has been undertaken regarding the sensitivity and specificity of the EPDS as a screening instrument for PND. 112 Many studies have found the EPDS wanting in terms of satisfactory psychometric properties to support its routine use as a screening instrument. In this respect an interesting finding in this study is that of the 54 women who were high scorers on the eligibility EPDS, but did not have depression, 35 of them had other ICD-10 diagnoses, mainly some form of anxiety. There has been a tendency over the years to consider all postnatal illness under the umbrella term PND. But it may well be that a case could be made for its use if it was used to screen for mental illness more generally and more attention was paid to the characteristics of the patient’s symptoms including anxiety. In addition, of the 298 women who were low scorers, < 13, on the eligibility EPDS, 66 of them completed a CIS-R and of these, 19 had an ICD-10 diagnosis, mainly in the anxiety–depression spectrum.

There are one or two further comments to be made about recruitment. We had initially aimed to recruit 468 women – assuming a 50% participation rate and a 10% prevalence of PND. Our recruitment figures were slightly over ambitious – although the London centre did meet this target. As a result of the very different practice demographics, the Manchester centre had to recruit 35 rather than the originally planned 21 practices. This resulted in reduced monitoring time by the research associate and possibly less commitment from practices in terms of adhering to the protocol. Information fed back to us by practice HVs led us to believe that the detail in the initial invitation pack, which had been requested by the MREC, was off-putting to a lot of women. With the agreement of the MREC we simplified the participation procedure, which ensured that we met our revised target sample size of 248. We also overestimated the prevalence of PND – we used the lower end of the widely quoted 10–15% range6 for our sample-size calculation. In the event, the prevalence in the women available to enter the trial was only 6.5%. However, the 10–15% figure largely comes from studies that have not used a diagnostic instrument and when these more robust criteria are used the prevalence drops to about 7%. 6 We used the EPDS followed by the CIS-R to ensure that women eligible for the trial met diagnostic criteria for ICD-10 depression, which is the morbidity level recommended by the NICE guidelines for the use of antidepressants. 53 In addition we ‘lost’ quite a large number of women who would have been eligible for the trial who were already on treatment (94) or excluded by their GP (seven, all of whom had ICD-10 depression on their CIS-R). If all these women had been eligible and recruited, the prevalence of PND would have been 8.7%. Finally, there were 35 women whose EPDS score was ≥ 13, but whose CIS-R determined an ICD-10 diagnosis other than depression. It is likely that in studies that have not used a second-stage instrument these women might have been described as suffering from PND and in this case our overall prevalence would have been 9.6% – not so far from the 10% estimate in our protocol. It is well known that different assessment methods can lead to different estimates for the prevalence of depression. The CIS-R was used in the UK Psychiatric Morbidity surveys carried out in 1993 and 2000. Among women of childbearing age, the prevalence of depression in those surveys was between 3 and 4%. 2 It is possible that the higher prevalence reported here resulted from response bias, as only about 40% responded to the initial request. The information sheet was asking for people to join a trial on PND so many women who did not feel depressed may not have responded.

It is important to look at the demographic characteristics of the women who were confirmed as suffering from PND and were therefore randomised to the trial compared with the demographic associations noted in the literature. In this study, women with PND who were randomised were less likely to be married or living with a partner and tended to have less social support. They were also more likely to report previous antidepressant treatment, to have fewer qualifications, and to be in routine occupations. This pattern of associations looks very similar to that described in Chapter 1 (see Epidemiology of postnatal depression). We cannot however assume that our findings can be extrapolated to women from other ethnic groups who formed only a small minority of the trial population.

Randomised controlled trials evaluating antidepressants

The interpretation of the study findings in the light of previous research is difficult because there are very few similar trials with which to compare the results. There are only three previous RCTs evaluating the use of antidepressants in the management of PND, one in the UK and two in the USA; only two of these compared an antidepressant with a psychological therapy: fluoxetine or placebo with cognitive behaviour counselling (either one or six sessions)54 and paroxetine compared with paroxetine plus CBT. 56 In the UK study, a similar methodology to the current trial was employed using the EPDS and the CIS-R albeit with different thresholds for eligibility. The antidepressant was constrained to fluoxetine and the counselling was based on the CBT model. Significant improvement was seen in all four treatment groups. The improvement with fluoxetine was significantly greater than with placebo and the improvement after six sessions of counselling was significantly greater than after a single session. However, there did not seem to be a significant improvement for women in receiving counselling (six sessions) as well as fluoxetine. Many women were noted to be reluctant to take an antidepressant for PND. The study by Misri et al. 56 was very small, randomising only 35 women who had postpartum depression and comorbid anxiety. They found no difference between the two arms with about 60% having at least a 50% reduction in EPDS score at 12 weeks post randomisation. This trial, albeit lacking in power, confirmed the findings of the UK trial,54 in that adding CBT to an antidepressant does not seem to increase effectiveness and would certainly be more costly. The final trial compared sertraline and nortryptyline55 an SSRI and a tricyclic antidepressant. There were no differences between the two classes of drug, 46% versus 56% responding at 4 weeks, but there were more dropouts with the SSRI. The methodological quality of these three trials was regarded as poor in the recent NICE guidance. 45 None of these trials took a pragmatic approach to the choice of antidepressant nor did they compare antidepressants with GSC. Our ITT analysis found a significant, 25%, improvement in the antidepressant arm compared with usual care at 4 weeks (45% versus 20%), which at 18 weeks had reduced to 11% (non-significant difference) when compared with listening visits (62% versus 51%). The pragmatic nature of this study makes it hard to draw direct comparisons with previous studies but there is good evidence for the effectiveness of antidepressants at both follow-ups, even though the superiority of antidepressants at 4 weeks is attenuated by 18 weeks.

The evidence for the treatment of moderate depression with antidepressants in women outside the postpartum period is well made53 and so it is not surprising that antidepressants have been found to be effective in this trial and others. Given the lack of evidence for a distinct hormonal or other aetiology for PND, its responsiveness to antidepressants is in line with it being a typical depressive disorder. There are a number of possible mechanisms whereby gender may influence treatment response. Drugs with effects on the serotonergic system may be relevant for younger women because serotonergic agents have demonstrated efficacy in disorders such as premenstrual dysphoric disorder. 113 Second, the presence of atypical depression, for example with weight gain rather than loss, may modify treatment responsivity and women are more likely to present with atypical depressive symptoms. 114 Another explanation is that female reproductive hormones may play a permissive or inhibitory role in antidepressant activity. For example, oestrogen may enhance serotonergic activity. 115 There has been one trial finding some evidence in favour of oestrogen as a treatment for PND. 116 However, PND is different, at least in some ways, in that it occurs at a particular time and has repercussions not only for the mother but also for her infant. 27,31,38 The impact on the infant may be the result of biological as well as psychosocial mechanisms. The other issue that makes PND and considerations about its treatment with antidepressants special is the need to consider risk to the infant if the mother is breastfeeding. Several studies have been undertaken to address this risk – the overall conclusion being that most antidepressants are safe for infants, the exceptions being fluoxetine (because of its long half-life), citalopram and doxepin. 59

Randomised controlled trials evaluating psychosocial interventions

There is much more evidence regarding the effectiveness of different psychological and psychosocial interventions for PND. 62 In this systematic review there was significant heterogeneity among the four trials evaluating non-directive counselling; however, overall there was evidence of significant benefit in reducing depressive symptomatology both immediately post-treatment and at 1 year. In two trials that compared non-directive counselling (listening visits) with CBT there was no difference at the final assessment. 71,72 In the first of these trials, which had several methodological flaws, there was no difference between the proportions free of depression as assessed by the Structured Clinical Interview for DSM-III-R (SCID)117 at 4.5 months (54% with non-directive counselling and 57% with CBT). 71 In the second trial, at 6 months, there was no difference in mean EPDS score between the women who received CBT and those who received person-centred counselling. 72,72 Non-directive counselling is a form of counselling that is based on the understanding that, in many situations, people can resolve their own problems without being provided with a solution by the counsellor. In particular, the counsellor’s role is to encourage the person to express their feelings but not suggest what decision the person should make. By listening and reflecting back what the person reveals to them, the counsellor helps them to explore and understand their feelings. With this understanding, the person is able to make the decision that is best for them. This treatment modality may be an important option for mothers with mild to moderate postpartum depression. Previous trials have demonstrated the feasibility of population-based screening and the application of home visiting using trained health professionals. The results from this study add to the evidence base for the effectiveness of HV-delivered listening visits, even for women with moderate to severe depression, where at 18 weeks the improvement in women who received listening visits was very similar to that in women who received antidepressants.

The qualitative work demonstrated that many women viewed listening visits as a preferred intervention and some expressed scepticism about the value of antidepressants. However, the study has demonstrated that antidepressants are an effective treatment for depression, at least in the short term, despite these rather negative attitudes.

Entering trial

• How did you hear about RESPOND?

• Why did you decide to take part?

• How did you feel about being asked to take part in a trial about PND?

• How did you feel when you found out that you were eligible to take part in the trial?

Diagnosis

• How did you feel about being diagnosed as having PND? Did you think you would be? Prompt surprise, relief, agreement with diagnosis.

• How were you feeling at the time? Why do you think you felt like this?

• What care were you receiving at this time? (GP, HV etc.) Did you have any other forms of support (family, friends, clinics, groups)?

Randomisation

• How was it decided what type of care you got in the trial? Why was it decided like this? Prompt for understanding of trial aims and process.

• Did you think you were more likely to get one type of care rather than another? Why? Was it down to chance?

Treatment: expectations

• In what ways did you think each of the treatments might help you?

  1. – Antidepressants: prompt for understanding, expectations, prior experience, concerns including breastfeeding and antidepressants, side effects, stigma.

  2. – Listening visits: as before and include time, relationship.

• What type of care were you hoping to get? Why?

• What type of care were you allocated first? How did you feel about this? Prompt for acceptance, disappointment, preferred treatment etc.

• What treatments have you received since being in RESPOND?

Comparing care (all women)

• Thinking about all the care that you have received during the trial, what has been most useful and why?

• How are you feeling now?

Experiences of the trial

• What were your experiences of being involved in the RESPOND trial?

• Negative and positive?

• Are there any ways you think the trial could have been improved?

• Is there anything else you would like to say about postnatal depression, your experiences of a particular treatment, views on treatments available and/or the trial?

Entering trial

• How did you hear about RESPOND?

• How did you feel about being asked to take part in a trial about PND?

• Why did you decide to return the initial questionnaire?

• How did you feel when you found out that you were eligible to take part in the trial?

Diagnosis

• How did you feel about being diagnosed as having PND? Did you think you would be? Prompt surprise, relief, agreement with diagnosis.

• How were you feeling at the time? Why do you think you felt like this?

• What care were you receiving at this time? (GP, HV etc.) Did you have any other forms of support (family, friends, clinics, groups)?

Randomisation

• How was it decided what type of care you got in the trial? Why was it decided like this? Prompt for understanding of trial aims and process?

• Did you think you were more likely to get one type of care rather than another? Why? Was it down to chance?

Declining

• Why did you decide not to take part in the RESPOND trial? Prompt for felt better, concerns and expectations about treatment, the attitudes of others.

• Is there anything that would have encouraged you to take part in the trial?

Treatments

• What did you know about the different treatments offered in the trial?

• What ways do you think the treatments might have been helpful for you at that time? What ways do you think they might have been unhelpful?