Antidepressant treatment with sertraline for adults with depressive symptoms in primary care: the PANDA research programme including RCT

Article history

The research reported in this issue of the journal was funded by PGfAR as project number RP-PG-0610-10048. The contractual start date was in March 2012. The final report began editorial review in September 2018 and was accepted for publication in June 2019. As the funder, the PGfAR programme agreed the research questions and study designs in advance with the investigators. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The PGfAR editors and production house have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the final report document. However, they do not accept liability for damages or losses arising from material published in this report.

Declared competing interests of authors

Rachel Churchill reports grants from the National Institute for Health Research (NIHR) (Programme Grants for Applied Research programme) during the conduct of the study. Simon Gilbody serves as deputy chairperson of the NIHR Health Technology Commissioning Board, but was not involved in the commissioning of this programme of research. Catherine Harmer reports personal fees from P1vital (Wallingford, UK), grants from UCB Pharma (Brussels, Belgium), grants and personal fees from Johnson & Johnson (New Brunswick, NJ, USA), and personal fees from H. Lundbeck A/S (Copenhagen, Denmark), Servier Laboratories (Neuilly-sur-Seine, France) and Pfizer Inc. (New York, NY, USA) outside the submitted work. Tony Kendrick reports grants from NIHR during the conduct of the study. Marcus Munafo reports grants and personal fees from Cambridge Cognition (Cambridge, UK) and personal fees from Jericoe Ltd (Bristol, UK) outside the submitted work. Tim Peters reports grants from NIHR during the conduct of the study. Howard Thom reports personal fees from Novartis Pharma AG (Basel, Switzerland), Pfizer Inc., Roche Holding AG (Basel, Switzerland) and Eli Lilly and Company (Indianapolis, IN, USA) outside the submitted work. Nicky Welton reports grants from NIHR during the conduct of the study; and she is the principal investigator on a Medical Research Council-funded project in collaboration with Pfizer Inc. Pfizer Inc. part funded a junior researcher on the project. The project is purely methodological using historical data on pain relief, and unrelated to this work. Nicola Wiles reports grants from NIHR during the conduct of the study. Glyn Lewis reports grants from University College London during the conduct of the study and personal fees from Fortitude Law (London, UK) outside the submitted work.

Copyright statement

© Queen’s Printer and Controller of HMSO 2019. This work was produced by Duffy et al. under the terms of a commissioning contract issued by the Secretary of State for Health and Social Care. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

2019 Queen’s Printer and Controller of HMSO

Background

Depression is the leading cause of disability globally. 1 Most people with depression or depressive symptoms are treated in primary care and antidepressants are often the first-line treatment. There were over 70 million antidepressent prescriptions in England in 2018. 2 There is much uncertainty about when people with depression might benefit from an antidepressant and concern that antidepressants are overprescribed. General practitioners (GPs) often feel under pressure to provide some treatment and have to make a difficult decision about whether or not an individual will benefit from an antidepressant. The existing guidelines use terms such as ‘mild’ or ‘moderate’ depression to guide prescription but do not specify what this means, and such guidelines are not based on empirical studies. Identifying the patients who are most likely to respond to an antidepressant would improve the management of depression in primary care as well as in other clinical settings. This would reduce the inappropriate prescription of antidepressants in those less likely to respond, and increase appropriate prescription in those more likely to respond.

The minimal clinically important difference

To make informed recommendations about when treatments are of benefit to patients, we must decide what constitutes a clinically important treatment effect. There is no consensus about the size of a ‘clinically important difference’ on continuous outcome scales. The National Institute for Health and Care Excellence (NICE) guideline group, which includes service users, suggested that this difference was three points on the Hamilton Rating Scale for Depression (HAMD),3 but did not provide any empirical justification. 4 We need to decide on a clinically important treatment effect before we can make recommendations about when selective serotonin reuptake inhibitors (SSRIs) are of benefit and to estimate the sample size of a study. There have been previous attempts to determine clinically important differences5,6 but these have relied on classifying people as ‘well’ or ‘ill’ and then calculating the differences in score between the groups. However, these metrics do not take into account the perspectives of the patient. We were interested in participants’ own views about improvement to determine a clinically important difference. Our approach was to ask people to rate their own improvement and then use this to calculate the difference in scores corresponding to a change. This method has been used in relation to quality-of-life scales but, as far as we are aware, not in relation to depression. 7 There has been work comparing the results of self-administered questionnaires with psychiatric diagnostic assessments. 8 Our approach was therefore in contrast to these methods and emphasises the patient’s perspective.

Measurement of depression

Another barrier to the development of evidence-based guidelines for antidepressant prescription in primary care is the lack of a standardised measure of depressive symptoms that is easily implemented. Existing studies on antidepressants mostly use rating scales, such as HAMD, that are difficult to standardise, are designed for clinician administration and require training. It is extremely unlikely that these scales would ever be used in primary care and this has never been proposed, to our knowledge. Shorter self-administered scales, such as the Patient Health Questionnaire-9 items (PHQ-9),8 have been used in UK primary care and in the NHS Improving Access to Psychological Treatment services. However, it is widely thought that the PHQ-9 does not provide sufficiently good data to guide prescribing even if it is useful as an outcome measure. The NICE CG90 depression guideline9 explicitly recommends not using the PHQ-9 and similar scales alone for the purpose of guiding treatment.

Short questionnaires, such as the PHQ-9, are unlikely to give accurate information to guide prescription, but evidence suggests that they do perform well as a measure of outcome or change. 10 However, this symptom-based approach to measuring outcome has been challenged by the growing literature about ‘recovery’, largely from the perspective of people with psychosis. Two areas highlighted in this literature include the concepts of hope and empowerment, the notion that the patient feels able to change. 11 We are aware of only a limited literature concerned with ‘recovery’ in people who have experienced depression. 12 Malpass et al. 13 have also interviewed people recovering from depression in relation to change on the PHQ-9 and did identify some areas that were not well covered in that questionnaire. These included anxiety along with ‘awareness’ and ‘ability to make changes’.

There is still some uncertainty about how antidepressants work but Harmer et al. 14 have found that antidepressants lead to changes early on in the processing of emotional information. Beck et al. 15 first proposed that negative interpretations, beliefs and memories play a key role in depression and developed cognitive–behavioural therapy. Subsequently, the cognitive neuropsychological model of depression has proposed that lower-level changes in emotional processing play a causal role in the genesis of symptoms and precede changes in depressive symptoms. 14 Simple tests, such as face recognition and word recall, could be used alongside symptom measures to investigate changes in depression and complement measures of symptoms. Existing studies have been small and used case–control designs that are prone to selection bias. If we could confirm that emotion-processing tasks were related to depressive symptoms, future research could see if the response to emotion-processing tasks could be used to predict clinical response. We therefore also investigated associations between depressive symptom severity and emotional processing and how this might change in recovery.

There are a wide range of existing outcome assessments of depressive symptoms, for example HAMD, PHQ-9, Beck Depression Inventory, version 2 (BDI-II)16 and the Hospital Anxiety and Depression Scale (HADS). 17 It would be useful to know how these scales inter-relate so that the results of clinical trials can be compared with each other. 18,19 If we could ‘map’ the scores on the scales against each other, this would help with the interpretation of existing data and in the application of existing and future results to clinical practice.

What factors are associated with response to antidepressants?

It has been proposed that antidepressants are more effective for patients with more severe depression, but the evidence for this is inconsistent20–25 and recent large studies of individual patient data suggest no influence of depression severity. 23–25 On the other hand, a systematic review of patients with depressive symptoms not meeting diagnostic criteria26 did not find evidence that antidepressants were effective. However, the majority of the existing trials were not designed to investigate this hypothesis and exclude patients who are below a certain severity threshold. 27,28 A more general criticism is that the current evidence is dominated by trials performed for regulatory purposes so it is difficult to generalise their results to patients currently receiving treatment, mainly provided in primary care.

To help GPs decide whether or not to prescribe SSRIs, it is important to compare a SSRI with a placebo. This becomes more rather than less important in less severe depression as the differences between placebo and active medication are likely to be smaller than for more severe depression. 20 By ‘response’ to antidepressants, we refer to the difference between the antidepressant and the placebo.

The two main diagnostic manuals, International Classification of Diseases, Tenth Edition (ICD-10),29 and Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV),30 have similar, but not identical, diagnostic criteria for a depressive episode (ICD-10) and major depression (DSM-IV). DSM-IV also has a category of minor depression that requires fewer symptoms. As a rule, GPs do not use these diagnostic criteria and there is currently little empirical evidence that meeting either of these diagnostic criteria can alone indicate whether or not antidepressants will be beneficial. There is a consensus that depressive symptoms can be viewed along a continuum of severity, and, in population terms, ‘subthreshold’ symptoms are still a public health concern. 31,32 It is therefore important to assess the severity of symptoms along this continuum. People who do not meet diagnostic criteria might still benefit from antidepressant treatment, or the converse.

There is evidence that antidepressants can be effective for people with dysthymia. 33,34 Dysthymia is a US term used in DSM-IV but not in ICD-10. It describes depressive symptoms of long duration (≥ 2 years) but not meeting the diagnostic criteria for depression. NICE guidelines19 recommend SSRIs for persistent subthreshold depressive symptoms but give no guidance on the duration of the persistence. We proposed, as adopted by current UK guidelines,19,35 that symptom severity and duration of symptoms are two separate dimensions that both might help to predict response to antidepressants.

The PANDA research programme

The overall aim of the PANDA (What are the indications for Prescribing ANtiDepressAnts that will lead to a clinical benefit?) research programme was to provide GPs or other clinicians with improved guidance about when antidepressants are most likely to result in a clinical benefit for patients with depressive symptoms in primary care and be cost-effective for the NHS. The main hypothesis was that response to antidepressants (compared with placebo) would increase with both the severity and duration of depressive symptoms. The programme aimed to investigate whether or not there are thresholds of severity and duration above which a clinically important response to antidepressants is most likely. In line with growing emphasis on patient-centred care, ‘clinically important’ was defined as a reduction in depressive symptoms large enough that patients would detect feeling better. An additional aim was therefore to establish, for the first time, the reduction in depressive symptoms on self-administered depression questionnaires that is required for patients to feel better. To assess severity and duration, we used a detailed, self-administered, computerised assessment that could be easily implemented in primary care. Finally, we also wanted to investigate how patients interpreted the questions in the current self-administered scales, to investigate how that related to self-reported improvement and to investigate the relationship between depressive symptoms and the underlying cognitive biases that are characteristic of depression.

A new randomised controlled trial (RCT) was required to provide this information, despite the wealth of data available from previous placebo-controlled studies that have largely been carried out by the pharmaceutical industry and are comprehensively summarised in Cipriani et al. 28 First, the existing data were mostly of a poor quality and were carried out decades ago for regulatory purposes rather than to study clinical indications. Cipriani et al. 28 noted that 78% of trials were industry funded and that, overall, trials were of poor quality and small, with a mean sample size of 224 (across arms). In addition, 82% of trials were also at moderate or high risk of bias and the larger more recent placebo-controlled trials reported smaller effect sizes, perhaps reflecting more rigorous methods. Second, existing results on the influence of depression severity on antidepressant response are in terms of HAMD scores. This is not useful in primary care because of the training required. Even though we should be able to calculate equivalent scores on measures such as the PHQ-9, more detailed assessments would provide a better predictor of response than brief questionnaires. Third, existing data are unlikely to generalise to the population currently receiving antidepressants in primary care in the UK or in other countries. Most of the trials excluded people at lower levels of severity and the recruitment methods are usually unknown. Finally, we are not aware of any existing data from previous studies on the influence of duration of illness.

The objectives of the PANDA research programme

The PANDA research programme consisted of three phases; the aims of each phase and the title of corresponding PANDA papers are listed below.

Phase 3: randomised controlled trial

Aim 3: to investigate the severity and duration of depressive symptoms that are associated with a clinically important response to sertraline in people with depression and whether or not these factors are associated with the cost-effectiveness of sertraline.

• See Appendix 10: Salaminios et al. 43

• See Appendix 11: Lewis et al. 44

• See Appendix 12: Hollingworth et al. 45

The inter-relationships between the three phases are summarised in Figure 1.

FIGURE 1.

The inter-relationships between the three phases.

Changes to the programme

1. In our research proposal, we originally had aimed to carry out a systematic review and individual patient data meta-analysis to investigate the relationship between the severity of depressive symptoms and the response to antidepressants. However, after the research programme was funded, an individual patient data meta-analysis was published by Gibbons et al. 23 using data provided by the pharmaceutical industry and there have been more recent replications. 24,25 Gibbons et al. 23 analysed all the placebo-controlled studies of the antidepressants fluoxetine and venlafaxine that were sponsored by the relevant pharmaceutical company and had therefore obtained a complete and unbiased sample of the placebo-controlled trials. As a result, we decided to abandon the original proposal for a systematic review in this area based on literature searching and approaching authors for individual patient data.

2. With the resources that had originally been earmarked for the individual patient data meta-analysis, we carried out an analysis of existing data to begin our study of the MCID (aim 1a and see Appendix 1) and a value-of-information study (aim 1c and see Appendix 3). Previously published economic models including that used in the NICE CG9019 did not consider treatment by baseline severity and none modelled depression severity itself, so this required the development of an economic model in which severity of depression was part of the decision-making process.

3. We also included some simple emotion-processing tasks in the PANDA cohort study. These tasks are influenced by antidepressant medication and might indicate changes that occur during recovery from depression that are not currently assessed by self-administered questionnaires. These tasks therefore extended aim 2b and the relationship with depressive symptoms and changes in symptoms are reported in Appendices 6 and 7.

4. When originally devised, the primary aim of the RCT was to investigate the severity and duration of symptoms associated with a clinically important response to sertraline, as stated in the protocol paper43 and on the International Standard Randomised Controlled Trial Number (ISRCTN) registry. However, it was apparent towards the later stages of designing the RCT and in formulating the detailed analysis plan46 (uploaded before any analyses were performed to http://discovery.ucl.ac.uk and approved by the Trial Steering Committee) that we would have insufficient statistical power to estimate plausible interaction effects that would allow us to investigate those aims. Our power calculation and primary analysis (as stated in the analysis plan46) are therefore based on a primary aim to examine the clinical effectiveness of sertraline versus placebo. Interactions between severity and duration at baseline and treatment response were planned as exploratory.

5. In the original funding application, we used citalopram as our choice of antidepressant. However, during the set-up stage, new guidance was released informing clinicians that citalopram can prolong the QT interval, especially in higher doses. Although the risk was low, we decided to change the study medication to sertraline, another commonly prescribed SSRI that is no longer under patent. There are very few pharmacological differences between the SSRIs and we believe that the results of our study can be applied to all SSRIs.

Aim 1a: to use existing data to estimate minimal clinically important difference from the patient perspective

Aim 1b: ‘mapping’ the relationship between different depression scales

Aim 1c: to assess the value of information from carrying out a randomised controlled trial of antidepressants in depression of mild severity

Aim 2a: estimating a clinically important difference in commonly used self-administered questionnaires for depressive symptoms

Methods for data collection

The PANDA cohort consisted of patients who had presented to UK primary care surgeries with depressive symptoms or disorder, or depressed mood, during the previous year. Participants had a range of depressive symptoms and were recruited from one population, reducing selection bias. Overall, 7721 patients were sent an information letter in the post and 1470 (19%) replied (Figure 2). Of these, 821 were willing to be contacted, 23 (3%) of whom were ineligible. The remaining 798 were contacted to arrange an interview and 563 consented to take part in the cohort study. Data on our measures were collected at four time points. At time 1, 559 people provided data (four could not be contacted), with corresponding figures at follow-up at 2, 4 and 6 weeks of 476 (85%), 443 (79%) and 430 (77%), respectively. For this analysis we used data from 400 participants who gave complete data on all follow-ups.

FIGURE 2.

The Consolidated Standards of Reporting Trials (CONSORT) flow diagram: PANDA cohort study.

The participants were asked to rate their own improvement using a GRC question. The GRC was assessed by asking patients ‘compared to when we last saw you 2 weeks ago, how have your moods and feelings changed?’ Response options were ‘I feel a lot better’ (1), ‘I feel slightly better’ (2), ‘I feel about the same’ (3), ‘I feel slightly worse’ (4) and ‘I feel a lot worse’ (5). These ratings were compared with the changes in the score on the self-administered questionnaires BDI-II, PHQ-9 and GAD-7. This enabled us to calculate the change in scores (on the questionnaires) that corresponded to an improvement in patients’ GRC. To assess the reliability of the GRC, the question was completed twice by the participants at each time point. The participants completed the Clinical Interview Schedule – Revised (CIS-R)61 at baseline only. The data were used to derive the diagnosis of depression and its severity.

Aim 2b: to investigate the changes reported by patients as they recover from depression

We carried out three studies to investigate this aim. First, we carried out a qualitative investigation of the meaningfulness of the PHQ-9 in determining meaningful symptoms of low mood. We also examined the processing of emotional information and how this varied with depressive symptoms and over time. Emotion-processing is a key abnormality in depression and influenced by antidepressant medication. 14 The second study investigated the variation of emotional face recognition in relation to depressive symptoms. Finally, the third study examined variation in recall of socially rewarding information according to depressive symptoms.

Aim 2c: to investigate disagreement between self-reported improvement and changes in the scores on depressive symptom questionnaires

We used quantitative and qualitative methods to identify those aspects of recovery that are currently missed by questionnaires.

Research aims

The aim of the RCT was to investigate the severity and duration of depressive symptoms that are associated with a clinically important response to sertraline and cost-effectiveness (compared with placebo) in people with depression who present to primary care. The main hypothesis was that response to antidepressants would increase with both severity and duration of depressive symptoms. Our primary analysis was a comparison between sertraline and placebo at 6 weeks.

Methods for data collection

We used broad and pragmatic inclusion criteria, recruiting people who had sought treatment for depressive symptoms of any severity or duration in primary care. The key entry criterion was that GPs and/or patients were uncertain about the potential benefits of an antidepressant and we did not set any severity or duration thresholds as exclusions. Patients were recruited from primary care surgeries in four UK sites (i.e. Bristol, London, Liverpool and York) and identified by GPs, who either invited patients during a consultation or conducted a database search and then sent an invitation in the post. Participants were randomised to 100 mg of sertraline or an identical placebo and followed up at 2, 6 and 12 weeks.

Figure 3 is a flow diagram of the progress through the trial.

FIGURE 3.

The Consolidated Standards of Reporting Trials (CONSORT) flow diagram: PANDA RCT.

Analysis

The primary outcome was the PHQ-9 at the 6-week follow-up. Interactions terms of a realistic size, that are smaller than the main effect, require very large sample sizes for adequate power. As a result, we modelled the treatment effect on log-transformed PHQ-9 scores (continuous outcome) using an intention-to-treat analysis. The exponentiated regression coefficient is the proportional (or percentage) change in PHQ-9 scores between randomised groups. Evidence of a treatment effect using a proportional model implies that the treatment effect expressed as a mean difference would increase with severity. In sensitivity analyses we fitted an additive model using absolute depression scores (non-logged PHQ-9) and calculated an interaction between treatment allocation and baseline CIS-R depression severity score. However, we expected the power of this analysis to be low.

Secondary outcomes at 2, 6 and 12 weeks were depressive symptoms and remission assessed using the PHQ-9 and the BDI-II, generalised anxiety disorder symptoms, mental and physical health-related quality of life and self-reported global improvement. We used linear multilevel models for repeated measures of continuous secondary outcomes at 2, 6 and 12 weeks (PHQ-9, BDI-II, GAD-7 and Short Form questionnaire-12 items physical and mental health-related quality of life). Logistic multilevel models were calculated for repeated measures of binary secondary outcomes at 2, 6, and 12 weeks (remission on the PHQ-9, the BDI-II and feeling better on the GRC scale).

We undertook a cost-effectiveness analysis from the perspective of the NHS and Personal and Social Services alongside the PANDA RCT. Quality-of-life data were collected at baseline and 2, 6 and 12 weeks post randomisation using EQ-5D-5L, from which we calculated QALYs. Costs were collected using patient records and from resource use questionnaires administered at each follow-up interval. Differences in mean costs and mean QALYs and net monetary benefits were estimated. Our primary analysis used net monetary benefit regressions to identify any interaction between the cost-effectiveness of sertraline and subgroups defined by baseline symptom severity (0 to 11; 12 to 19; ≥ 20 on the CIS-R) and, separately, duration of symptoms (greater or less than 2 years’ duration). A secondary analysis estimated the cost-effectiveness of sertraline versus placebo. In sensitivity analyses, we (1) performed a complete-case analysis to check the robustness of our findings to missing data, (2) examined the impact of excluding costs that were not judged to be directly related to the treatment of depression and (3) excluded all secondary care costs from total NHS and Personal and Social Services costs to assess whether or not our findings were robust to infrequent but expensive hospitalisations.

Key findings

We found no evidence that the antidepressant sertraline reduced depressive symptoms at 6 weeks. In the sertraline group, PHQ-9 scores were 5% (95% CI –7% to 15%; p = 0.41) lower than those in the placebo group. In the sensitivity analyses using additive models, there was no evidence of an interaction with severity or duration of depressive symptoms with treatment effect, but these analyses would have lacked statistical power.

Of the secondary outcomes, there was strong evidence that sertraline reduced anxiety symptoms (GAD-7 score reduced by 17%, 95% CI 9% to 25%; p < 0.000046) and improved mental but not physical health-related quality of life as well as self-reported global improvement. There was weak evidence that depressive symptoms were reduced by sertraline at 12 weeks for both the PHQ-9 and the BDI-II. Given our findings, we also investigated whether or not the treatment effect for anxiety symptoms was influenced by baseline severity. We found no evidence that the effect of sertraline on anxiety symptoms varied according to the severity of anxiety or depressive symptoms. The number needed to treat in order to feel better according to our self-reported global improvement question was 8.5 (95% CI 5.2 to 22.1) people at 6 weeks and 6.4 (95% CI 4.6 to 10.3) people at 12 weeks.

There was no evidence of an association between the baseline severity of depressive symptoms and the cost-effectiveness of sertraline. Compared with patients with low symptom severity, the expected net benefits in patients with moderate symptoms were £64 (95% CI –£312 to £441) and the expected net benefits in patients with high symptom severity were –£51 (95% CI –£389 to £287). Patients who had a longer history of depressive symptoms at baseline had lower expected net benefits from sertraline than those with a shorter history; however, the difference was uncertain (–£132, 95% CI –£431 to £167). In the secondary analysis, patients treated with sertraline had higher expected net benefits (£118.37, 95% CI –£23.39 to £260.14) than those in the placebo group. Sertraline had a high probability (> 90%) of being cost-effective if the health system was willing to pay at least £20,000 per QALY gained.

Limitations

We had broad inclusion criteria and some participants had very few symptoms. This may have reduced the treatment effect, but our methods of analysis using a proportional approach should have helped to take account of this. There was attrition of nearly 20% by 12 weeks, although this did not differ by study arm, and when we investigated the impact of missing data, this did not appear to explain the findings. We had limited statistical power to explore interactions between treatment response and symptom severity or duration. It is possible that subgroups in which sertraline was more cost-effective might have become more evident with a larger sample size or longer follow-up.

Inter-relation with other parts of the programme

The RCT did not find evidence of an early antidepressant effect of sertraline on depressive symptoms in the population studied. There was, however, evidence that sertraline reduced anxiety symptoms and was more likely to lead to a clinically important benefit. The results of the trial and those from the MCID can be used to provide some initial guidance about the likely cost-effectiveness of sertraline and other SSRI antidepressants in primary care.

Conclusion

We chose the PHQ-9 as our primary outcome in the PANDA trial. The PHQ-9 is widely used in primary care and there is evidence that it is better at detecting changes in depressive symptoms after treatment than other measures. 37,55 It also avoids the observer bias that affects clinician-rated HAMD and MADRS scores. However, our qualitative research identified a number of reasons for disagreement between the PHQ-9 and the self-reported GRC, as well as indicating that up to 50% of patients might show disagreement between self-reported change and the results of questionnaires. Our findings indicate that the processes and motivations behind completing the PHQ-9 are complex and influenced by ongoing physical, social and emotional issues. Our findings suggest that PHQ-9 and, by implication, other self-administered questionnaires should not be used alone to assess improvement or deterioration. Their use should be supplemented with further clinical assessment and the use of more open-ended questions.

The MCID is the smallest change in symptoms that is considered clinically worthwhile by the patient. Our MCID research in phases 1 and 2 has enabled us to develop appropriate analytical methods for estimating the MCID and to provide values for the MCID in a primary care population for the first time. We can apply our MCID estimates to the results of the PANDA trial, but we acknowledge that our estimates of MCID are still uncertain.

When we initially formulated our research questions we assumed that the treatment effect varied according to depression severity but that the MCID was a fixed value, irrespective of depression severity. Our results, if anything, now point in the opposite direction, at least when we estimate treatment effects and MCID using a proportional approach. Our results suggest that sertraline has a similar (proportional) effect size over the whole range of depression (and anxiety) severity and it is the MCID that changes and gets larger, proportionally, at lower levels of severity. We have found that, at higher levels of severity, the proportional approach works better for the MCID. The proportional approach also has attractions when analysing clinical trial data. It avoids the assumption that the same absolute treatment effect is observed regardless of whether a person scores 5 or 25 on a scale. This seems unlikely and a proportional reduction approach appears more plausible as well as providing a better statistical fit to the data.

Contrary to our initial hypothesis in the PANDA trial, we found no evidence of a clinically important effect of sertraline on depressive symptoms. We found a 5% reduction in the sertraline group at 6 weeks, and this is considerably smaller than the MCID estimates for the PHQ-9 we have obtained. We cannot exclude the possibility that sertraline led to a clinically important improvement at 12 weeks, as we found a 13% (95% CI 3% to 21%) reduction in PHQ-9, but this is still well below our MCID. In contrast we found strong evidence that sertraline reduced anxiety symptoms on the GAD-7, with a reduction of 21% (95% CI 11% to 30%) at 6 weeks. This is consistent with some of our estimates of the MCID for GAD-7 (see Appendix 4) and suggests that this change is clinically important. We found insufficient evidence of an interaction between the cost-effectiveness of sertraline and severity or symptom duration that GPs could use to efficiently target prescribing. There was no evidence of a substantial treatment effect of sertraline on quality of life, as measured by the EQ-5D-5L. However, sertraline is an inexpensive intervention that has a high probability of being cost-effective compared with placebo across primary care patients with depression or low mood.

Our MCID estimates are based on an average within-person change related to improvement; however, a RCT compares groups. Application of our results has therefore required a counterfactual argument in which researchers compare the same individual(s) who receive placebo but who might have received the active treatment. The MCID estimated from a within-person calculation can then be applied to the between-group differences in a clinical trial.

Our estimates of MCID could be used to guide decisions about whether or not a treatment will benefit an individual. For this, one needs to be able to predict the likely score for that person on the proposed outcome measure were they not to receive that treatment. In other words, we need to know the likely value of an individual’s PHQ-9 or GAD-7 score at follow-up, 6 or 12 weeks later, if they were to receive a placebo. The expected value of the placebo at follow-up can then be used to determine the appropriate initial value for the MCID and thus decide if the proportional reduction expected from a treatment would be larger than the MCID for such a person. For this to be feasible, we would need more precise estimates of MCID and also the ability to predict future scores of patients on the basis of their clinical and other characteristics.

Finally, we can make some very approximate estimate of what proportion of participants in the PANDA cohort would probably have benefited from treatment. From our results in Appendix 4, it is clear that those with a GAD-7 score of 3 at 6 weeks have a MCID of about 50%. It is highly unlikely that those individuals would have experienced any benefit from sertraline. In the PANDA RCT, about 30% of participants scored ≤ 3 at 6 weeks. However, we cannot conclude this with any confidence at this stage. We do not know the distribution of symptoms in those receiving antidepressants in the UK and our estimates of MCID are approximate. Our overall results are reassuring in indicating that, on average, patients in the PANDA RCT are benefiting from sertraline. However, it is probable that a substantial proportion or patients receiving antidepressants are not experiencing any individual benefit. For clinicians to be confident about recommending treatment to patients, we need accurate information on individualised treatment effects and the outcome without treatment as well as MCID. Of course, any recommendations for treatment will also have to take account of any risks and adverse effects that result from the treatment as well as patient preference.

Recommendations for future research

Our finding that sertraline seems to be effective for anxiety but not depressive symptoms has a potential implication for understanding the mechanisms of antidepressant treatment as well as the clinical benefit that patient’s will experience.

Implications for practice and any lessons learned

The PHQ-9 and similar self-administered questionnaires should not be used alone in assessing improvement or deterioration. It is important to supplement such standardised measures with a clinical assessment.

Sertraline is effective and cost-effective in reducing anxiety symptoms such as worry and restlessness in the first 6 weeks of treatment in people who present with depressive symptoms. Any effect on depressive symptoms takes longer to emerge; although an improvement in anxious symptoms in someone presenting with depressive symptoms could lead to a clinical benefit. Patients who present to primary care with depressive symptoms have a wide range of severity of symptoms. Overall, this population is likely to benefit from SSRI antidepressants. Our findings support the prescription of SSRI antidepressants in a wider group of participants than previously thought, including those who do not meet diagnostic criteria for depression or generalised anxiety disorder, especially when anxiety symptoms such as worry and restlessness are present.

Patient and public involvement

Paul Lanham, a service user and a co-applicant, was also a member of the independent steering committee during phase 3 of the programme and was involved in PANDA for over 6 years. Paul Lanham and Derek Riozzie have also contributed to PANDA annual meetings where all co-applicants and researchers discuss progress, review the protocols and discuss any findings. They made important contributions to the discussion and influenced the interpretation of the results and decisions about study design. All study documentations have been revised and commented on by Paul, Derek and the user group co-ordinated by Derek at Liverpool University. In addition, we enlisted the support of the North London Service Users Research Forum (SURF). The SURF was co-founded in 2007 by service users and clinical academic psychiatrists at University College London to provide meaningful consultation on research. It has 12 members with mental health problems. Since 2007, it has consulted on > 50 projects and SURF members have been invited to join steering/management groups on many of these. As a result, the group is very experienced and confident about the advice and input they provide; their comments on the trial paperwork have been invaluable. The letter templates, patient information sheets and the questionnaire were amended to reflect the patient and public involvement (PPI) feedback. We also consulted on the protocol concerning self-harm or risk of self-harm, which we used if patients reported this in the course of the cohort and RCT. Having close involvement of the PPI for the duration of the programme (i.e. over 6 years) has been invaluable for its success. It has also enabled us to build on it and we have recruited a PANDA RCT participant to represent PPI on a different depression trial: ANTidepressants to prevent reLapse in dEpRession (ANTLER). 64 We plan to carry on using the services users’ comments in the design, documentation and analysis of any future studies.

Ethics approval and sponsorships

For the PANDA cohort study, ethics approval was obtained from National Research Ethics Service (NRES) Committee South West-Central Bristol. The University of Bristol acted as sponsor for the study.

The PANDA RCT was approved by NRES Committee East of England – Cambridge South (reference number 13/EE/0418). The Joint Research Office, University College London acted as sponsor for the RCT.

Contributions of authors

Larisa Duffy was the programme manager and with Gemma Lewis was responsible for drafting the report.

Larisa Duffy, Gemma Lewis, Anthony Ades, Ricardo Araya, Jessica Bone, Sally Brabyn, Katherine Button, Rachel Churchill, Tim Croudace, Catherine Derrick, Padraig Dixon, Christopher Dowrick, Christopher Fawsitt, Louise Fusco, Simon Gilbody, Catherine Harmer, Catherine Hobbs, William Hollingworth, Vivien Jones, Tony Kendrick, David Kessler, Naila Khan, Daphne Kounali, Paul Lanham, Alice Malpass, Marcus Munafo, Jodi Pervin, Tim Peters, Derek Riozzie, Jude Robinson, George Salaminios, Debbie Sharp, Howard Thom, Laura Thomas, Nicky Welton, Nicola Wiles, Rebecca Woodhouse and Glyn Lewis contributed to the constituent papers in the appendices. The papers included as appendices each have its own lists of authors.

Anthony Ades, Ricardo Araya, Rachel Churchill, Tim Croudace, Christopher Dowrick, Simon Gilbody, William Hollingworth, Tony Kendrick, David Kessler, Paul Lanham, Alice Malpass, Tim Peters, Jude Robinson, Debbie Sharp, Nicky Welton, Nicola Wiles and Glyn Lewis were responsible for the original proposal and for securing funding.

Glyn Lewis was chief investigator of the programme and had clinical responsibly for the RCT recruitment at the London site.

All authors have provided substantial contributions to the conception and design of the PANDA programme and interpretation of data and had input into drafting the report and/or revising it critically for important intellectual content. All authors have given final approval of the version to be published.

Data-sharing statement

All data requests should be submitted to the corresponding author for consideration. Access to available anonymised data may be granted following review.

Patient data

This work uses data provided by patients and collected by the NHS as part of their care and support. Using patient data is vital to improve health and care for everyone. There is huge potential to make better use of information from people’s patient records, to understand more about disease, develop new treatments, monitor safety, and plan NHS services. Patient data should be kept safe and secure, to protect everyone’s privacy, and it’s important that there are safeguards to make sure that it is stored and used responsibly. Everyone should be able to find out about how patient data are used. #datasaveslives You can find out more about the background to this citation here: https://understandingpatientdata.org.uk/data-citation.

Disclaimers

This report presents independent research funded by the National Institute for Health Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, CCF, NETSCC, PGfAR or the Department of Health and Social Care. If there are verbatim quotations included in this publication the views and opinions expressed by the interviewees are those of the interviewees and do not necessarily reflect those of the authors, those of the NHS, the NIHR, NETSCC, the PGfAR programme or the Department of Health and Social Care.