Cognitive-behaviour therapy for health anxiety in medical patients (CHAMP): a randomised controlled trial with outcomes to 5 years

Article history

The research reported in this issue of the journal was funded by the HTA programme as project number 15/141/02. The contractual start date was in October 2008. The draft report began editorial review in May 2016 and was accepted for publication in February 2017. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HTA editors and publisher have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this report.

Declared competing interests of authors

Paul Salkovskis is Editor-in-Chief of Behavioural and Cognitive Psychotherapy, the official journal of the British Association for Behavioural and Cognitive Psychotherapy, but does not receive any money in connection with this work. Paul Salkovskis and Hilary Warwick developed cognitive–behaviour therapy for health anxiety but have no financial or other interests in its use. Helen Tyrer has written a book (Tyrer H. Tackling Health Anxiety: A CBT Handbook. London: RCPsych Press; 2013), and Peter Tyrer and Helen Tyrer have co-authored online teaching modules on the recognition and treatment of health anxiety for the Royal College of Psychiatrists. Peter Tyrer is Co-Editor of Personality and Mental Health, for which he receives an annual honorarium.

Copyright statement

© Queen’s Printer and Controller of HMSO 2017. This work was produced by Tyrer et al. under the terms of a commissioning contract issued by the Secretary of State for Health. 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.

Potential benefits of psychological treatment and previous studies

The failure to detect this serious pathology is perhaps less surprising when there is, or at least has been until very recently, a general belief that there is no adequate treatment. Pharmacological management is generally unsatisfactory, but psychological treatment in the form of cognitive–behaviour therapy (CBT) has been shown to be effective both in primary care8,9 and, more recently, in secondary care. 10 Although these trials suggested efficacy of this intervention, it is less clear if it has a significant impact on costs. Although the total costs were somewhat reduced in those receiving CBT in the only trial in secondary care, these manifested only in the 6 months after treatment had been completed. 10 It is also not known to what extent health anxiety is found in patients with existing medical disease and whether or not this can be successfully treated, as almost all published trials exclude those with definite physical illness.

The precursors to the cognitive–behaviour therapy for health anxiety in medical patients (CHAMP) trial were a series of research studies, mainly carried out at King’s Mill Hospital in north Nottinghamshire. The first of these studies established, in a genitourinary medicine clinic setting, that health anxiety was relatively common, that it persisted in the absence of detection or intervention and that it led to more consultations with doctors. 6 The second suggested that the prevalence of health anxiety was somewhat higher (around 12%) in medical clinics such as cardiology and endocrinology clinics,7 and the third was a randomised controlled trial of a modified form of CBT for health anxiety (CBT-HA) developed earlier by Salkovskis and Warwick11 versus standard care in patients attending a genitourinary medicine clinic. 10 The last of these was the essential precursor to the CHAMP study. Since the trial began in 2008, there have been many more published trials but, as explained in full later in this report, these have all been concerned with somewhat different populations.

Research objectives

The main objectives of the CHAMP trial were to examine the clinical value, cost and cost-effectiveness of the administration of CBT in health-anxious patients attending five medical clinics in secondary care. Specifically, we hypothesised that (1) between five and 10 sessions of health anxiety-directed CBT using the Salkovskis–Warwick model11,12 are more effective than standard care in reducing health-anxious symptoms measured by the Health Anxiety Inventory (HAI)13 1 year after randomisation to the trial; and (2) the costs of the CBT and the control are equivalent; the end point for this outcome is the cost of health service interventions at 2 years adjusted for baseline values.

The secondary hypotheses were that, compared with the control condition (single interview), CBT-HA would lead to significantly greater improvement in health anxiety at 3, 6, 24 and 60 months measured using the HAI,13 in self-rated generalised anxiety measured using the anxiety section of the Hospital Anxiety and Depression Scale (HADS),14 in quality of life using the EuroQol-5 Dimensions (EQ-5D) measure of health-related quality of life15 and in social functioning using the Social Functioning Questionnaire (SFQ),16 all at 6, 12, 24 and 60 months.

In addition, we tested a number of secondary hypotheses. The first of these was that health anxiety-focused CBT (CBT-HA) would be less effective in patients who had additional comorbid pathology in the form of obsessional symptomatology [measured using the Short Obsessive–Compulsive disorder Screener (SOCS)]. 17 We also hypothesised that personality status would have an impact on the effectiveness of treatment, and that those with dependent personalities [measured using the Dependent Personality Questionnaire (DPQ)]18 and other personality disorders19 [measured using the Quick Personality Assessment Schedule (PAS-Q)],20 which were subsequently converted into International Statistical Classification of Diseases, Eleventh Edition (ICD-11)21 personality disorder categories,22 would have a worse outcome with CBT-HA. We also expected that these comorbid disorders would be associated with increased costs.

The study also allowed prevalence estimates to be made for health anxiety in different age groups and in different medical clinics.

Patient and public involvement

We recognised at the outset of the study that one of the major problems in executing the trial was the relative lack of awareness of health anxiety in medical clinics. These clinics are busy places with clear tasks set for clinicians and, despite regular but fairly vague comments about the importance of mental pathology in patients who are attending, there is no requirement for mental health status to be assessed, with the possible exception of dementia.

With the help of a support group called Confederation of Health Anxiety Sufferers Supporting Increased Services (CHASSIS) we developed a strategy for identifying the patients who might have pathological health anxiety. In a previous study7 we found that many patients with existing, and often successfully treated, medical pathology had a high level of health anxiety, and so excluding such patients was felt to be a mistake.

This led to the two-way strategy of combining feedback from clinics about potentially unsuitable attendees (by consultation with clinical staff) with screening of attenders using the short form of the HAI. As broaching the subject of health anxiety with people who were attending the clinic ostensibly for medical follow-up and consultation only was potentially counterproductive, we asked our CHASSIS group of sufferers with health anxiety to advise on the best way of introducing the subject of health anxiety. We chose ‘worry about your health’ as the core component of this after this consultation.

The CHASSIS group has also been active in disseminating the results of the CHAMP trial, most recently (at the time of writing) by attending a meeting of the All Party Mental Health Group in the House of Commons in December 2014.

Trial summary

The study was a pragmatic randomised controlled trial with two parallel arms and approximately equal randomisation of eligible patients to an active treatment group of 5–10 sessions of CBT or to a control group. During the course of the baseline assessment an explanatory interview was given about the nature of health anxiety; this was the only specific health-anxiety intervention in the control group, but there is some evidence that a simple explanation is of benefit in its own right. 11

Settings

Patients attending cardiology, endocrinology, gastroenterology, neurology and respiratory medicine clinics in five general hospitals (King’s Mill Hospital, north Nottinghamshire; St Mary’s Hospital, London; Charing Cross Hospital, London; Chelsea and Westminster Hospital, London; and the Hillingdon Hospital, Middlesex) were considered for the study. A total of 107 consultants agreed to collaborate with the study and to allow their patients to be approached provided that they were not considered inappropriate for the study on account of the severity of their physical pathology.

Form of randomisation

After baseline assessment, randomisation was carried out by a computerised system [Open Clinical Data Management System (openCDMS), Centre for Health Informatics, University of Manchester, Manchester, UK] using block randomisation with no stratification in randomised blocks of four and six. The allocation was carried out by the UK Mental Health Research Network independently of CHAMP personnel.

Target population and procedure

Patients attending clinics of the collaborating consultants, apart from those who were specifically excluded, were approached while waiting for their outpatient appointments and, if they agreed, were given the short form of the HAI,13 a rating scale of 14 questions that takes 5–10 minutes to complete. Those that scored at least 20 points on the scale were offered the opportunity to take part in the trial and given an information sheet about the study. If the patients agreed immediately, they were given an information sheet about the study and baseline assessments were made, provided that they satisfied the inclusion criteria and the Diagnostic and Statistical Manual of Mental Disorders-Fourth Edition (DSM-IV) diagnosis of hypochondriasis [derived from the Structured Clinical Interview for DSM-IV (SCID-IV) interview]. 23 Those who were uncertain were approached later.

When each patient had satisfied the criteria for entry to the study, formal consent was obtained and baseline assessments completed. At the baseline assessment all patients received a standard explanation of the nature and significance of health anxiety. After completion of the assessments, the research assistant entered and registered each patient to an online system (openCDMS) and this automatically led to the appropriate randomisation. The study co-ordinator was then informed of the details of the treatment arm allocation. Equal allocation was made to either (1) the active treatment group – between 5 and 10 1-hour sessions of CBT from a psychologist or research nurse at the clinic, backed up by a short take home manual; or (2) the standard care group, which received the normal care in the clinic.

The patients were recruited over a 21-month period beginning in October 2008 and ending in July 2010, and were followed up for 5 years. Because there were so many clinics running simultaneously, we required additional assessment help in recruitment. This was provided by Clinical Studies Officers of the North London and East Midlands hubs of the Mental Health Research Network, and proved to be invaluable.

Inclusion and exclusion criteria

Patients suitable for inclusion were those who satisfied the criteria for excessive health anxiety outlined above and who (1) were aged between 16 and 75 years, (2) were permanently resident in the area, (3) had a sufficient understanding of English to read and complete the questionnaires and (4) gave written consent for the interviews and audio-taping of 50% of treatment sessions, and for access to their medical records. The presence of existing medical pathology, provided that it was not new and required further investigation, was not a bar to treatment in the study, and as in preliminary work we had found that many patients with existing pathology have a high level of health anxiety.

Some patients who would otherwise satisfy the inclusion criteria above were excluded if (1) they were felt by their consultants to have a level of continuing major pathology that was too severe for them to take part in the study; (2) they were currently being actively investigated for significant pathology suspected by the clinician and for whom CBT might confuse or cause distress; (3) they had progressive cognitive impairment; or (4) they were currently under psychiatric care.

Assessments

The following assessments were carried out at baseline only:

• personality assessment using the PAS-Q,20 followed by conversion to the ICD-11 personality levels22 but also including the questions from the hypochondriasis subsection of the full schedule19

• the SOCS17 (a set of seven questions that identify the likely presence of obsessive–compulsive disorder); and

• the DPQ,18 an assessment of dependent personality traits (this was included because both dependent personality and obsessional symptoms are associated conditions that may handicap or complicate treatment).

The remaining clinical assessments, in addition to the main health anxiety measure (HAI) were generalised anxiety and depression (measured with the HADS14), as these symptoms are common in patients with hypochondriasis, quality of life (EQ-5D)15 and social functioning (SFQ). 16 These assessments were given at baseline and at 6, 12, 24 and 60 months. To ensure the highest possible follow-up rate, patients were reminded of the next follow-up point at each assessment. The HAI was also given at 3 months by post or telephone with a research assistant.

Service use data for the economic evaluation were collected using the Adult Service Use Schedule (ADSUS).

Methods of overcoming bias

The independent central computerised system involved in randomisation (openCDMS) ensured no bias in allocation of treatments and also full baseline information, as it was only when all necessary data were entered that the patient was randomised. As this was a single-blind study, there was always the danger of disclosure of the form of treatment by the patient, therapist or other investigators in the study. This was minimised by (1) patients being asked by research assessors not to disclose their treatment to the research assessors, (2) the assessors and therapists working in different areas and (3) establishing a system whereby any assessor who was unwittingly informed about the nature of the treatment then ceased to assess that patient, and another research assessor was allotted for continued assessments.

Study interventions

Training and fidelity of intervention

Four of the applicants (PS, GS, DM and HS) were involved in the training of therapists in two all-day workshops, and also in the assessment of treatment fidelity, and HW, one of the originators of the treatment, also assessed the audio-taped interviews independently.

Approximately half of all treatment sessions were audio-recorded and tapes or discs of these sessions given to patients to help in their progress. Fidelity was tested using a health-anxiety modification of the Cognitive Therapy Rating Scale24 (Health Anxiety Version), in which all therapists were assessed before they were regarded as competent. Those who did not reach an accepted grade had further training until they achieved competence. Manuals for the training of therapists and to aid patients in their treatment were also prepared. This issue is important, as better treatment fidelity with CBT has been associated with better treatment effects. 25,26

Standard treatment arm

Patients allocated to the standard treatment arm continued their care in both primary and secondary care settings and were reminded that they would be followed up at regular intervals up to 5 years. No restrictions on treatment were given to this group as the study was a pragmatic trial and the follow-up period was long.

Sample size

We calculated sample sizes for both the primary outcome measure and the first secondary outcome measure, choosing the larger of the two for the study.

Based on the pilot study in a genitourinary medicine clinic,10 we assumed that the true difference in the change of HAI score between CBT and control at 2 years is 5.00 points and that the standard deviation (SD) for the change of HAI score at 2 years is 7.58 points. With these assumptions, the study needed 122 patients to detect the above difference with 95% power at a two-sided 5% significance level. Taking into account 20% dropout by 24 months, the sample size was estimated to be 152 patients. It is also fair to add that when the study was set up there was no knowledge on what was a minimally significant clinical difference in the HAI score, and even now there is little work on this subject, but an improvement of 2 points on the scale appears to be clinically important (see Chapter 5, Meaningful clinical difference in Health Anxiety Inventory scores).

There remains no agreed approach to calculate the sample size required for an economic evaluation, particularly in areas such as health anxiety, in which the willingness to pay for improvements in outcomes is unknown. Based on the pilot study, we considered that the CBT intervention would be cost-effective if it improved HAI score and that it was no more costly than the control treatment. The sample size calculation for the economic evaluation was therefore based on the total costs over 24 months being equivalent.

With a sample size of 186 per group, the study had 80% power to reject the null hypothesis that the costs of the CBT and control are not equivalent (when the difference in mean costs is ≈£150) in favour of the alternative hypothesis that the means of the two groups are equivalent, assuming that the expected difference in means is 0 and the common SD is 580 (from pilot study data).

The sample size was thus powered by the first secondary outcome of the CHAMP study, estimated as 466 patients, assuming a 20% dropout by 24 months, leading to a total of 372 completing treatment and equal randomisation between groups. The sample size calculation uses a one-sided (non-inferiority) test. Thus, with 466 patients, or fewer if the dropout rate was less, the study was adequately powered to both detect the assumed difference in the primary outcome and assess the equivalence in the secondary economic outcome between the CBT group and the control group.

Statistical analysis

All primary statistical analyses used the intention-to-treat principle using the statistical package SAS, version 9.3 (SAS Institute Inc., Cary, NC, USA). The numbers (with percentages) of losses to follow-up at 12, 24 and 60 months after randomisation will be reported and compared between the treatment arms with absolute risk differences [95% confidence intervals (CIs)]. Deaths were reported separately for each group. The Consolidated Standards of Reporting Trials (CONSORT) procedure was used for reporting flow through the trial.

The primary end point was analysed using a mixed model with time, treatment, and time × treatment interaction as fixed effects, baseline measurement as covariate and patient as random effect. The treatment differences at each time point together with the 95% CI were derived from the mixed model. Missing data were treated as missing at random in the mixed-model analysis. To assess the sensitivity of the result to missing values, the last observation carried forward strategy was used to compute the missing HAI score at the follow-up visits. Other assessments were analysed in a similar way. In addition, covariate-adjusted analysis was performed on the primary outcome analysis by mixed model, controlling for three pre-specified potential predictors for the primary end point (clinic type, site and age). In addition, the percentage of patients achieving normal levels of health anxiety (HAI score of ≤ 10 points) was compared using a generalised estimating equation model with visit, treatment, interaction between visits and treatment as fixed effect, baseline measurement of HAI score as covariate and patient as random effect (an exchangeable covariance structure).

Subgroup analyses were also carried out to determine if there was heterogeneity in any of the main groups identified for analysis.

Primary outcome

The scores on the HAI showed greater improvement with CBT-HA than with standard care at all times of testing, with maximal differences at 6 months (Table 2 and Figure 3).

FIGURE 3.

Change in HAI scores between CBT-HA and standard care (treatment as usual) over the 5-year period. Note: the change in HAI scores between the two treatments at 12 months is the primary outcome. Adapted from the Lancet, vol. 383, Tyrer P, Cooper S, Salkovskis P, Tyrer H, Crawford M, Byford S, et al. Clinical and cost-effectiveness of cognitive behaviour therapy for health anxiety in medical patients: a multicentre randomised controlled trial, pp. 219–25. Copyright (2014),28 with permission from Elsevier.

Secondary outcomes

Generalised anxiety and depression

Generalised anxiety and depression as recorded using the HADS anxiety (HADS-A) and depression (HADS-D) sections also showed significant differences between the two treatments over the 5-year period, with those for generalised anxiety (HADS-A) being maximal at 6 months (Table 4 and Figure 4) and those for depression (HADS-D) being maximal at 5 years (Table 5 and Figure 4).

TABLE 4 - Change in HADS-A scores over 5 years

Visit	Observed value				Mixed-model analysis
	CBT-HA		Standard care		Treatment difference (95% CI)	p-value
	n	Mean score (points) (SD)	n	Mean score (points) (SD)
Baseline	219	12.56 (3.74)	225	12.25 (3.88)
Improvement from baseline
6 months	197	2.74 (4.41)	204	1.46 (3.89)	1.30 (0.48 to 2.11)	0.0019
12 months	194	2.80 (4.40)	193	1.67 (4.04)	1.02 (0.19 to 1.85)	0.0165
24 months	190	3.33 (4.57)	183	2.07 (4.35)	1.01 (0.17 to 1.86)	0.0186
5 years	150	5.56 (5.72)	158	4.54 (5.48)	0.71 (–0.20 to 1.61)	0.1245
Overall					1.01 (0.38 to 1.64)	0.0018

TABLE 5 - Change in HADS-D scores over 5 years

Note

A negative score indicates an increase in depressive symptoms from baseline and this was shown in the standard care group at 5 years.

Visit	Observed value				Mixed-model analysis
	CBT-HA		Standard care		Treatment difference (95% CI)	p-value
	n	Mean score (points) (SD)	n	Mean score (points) (SD)
Baseline	219	9.12 (4.34)	225	8.83 (4.58)
Improvement from baseline
6 months	197	2.74 (4.41)	204	1.46 (3.89)	0.75 (–0.09 to 1.58)	0.0797
12 months	194	2.80 (4.40)	192	1.67 (4.04)	0.75 (–0.10 to 1.60)	0.0821
24 months	189	3.33 (4.57)	181	2.07 (4.35)	0.63 (–0.23 to 1.49)	0.1521
5 years	150	5.56 (5.72)	157	4.54 (5.48)	1.41 (0.48 to 2.33)	0.0029
Overall					0.88 (0.25 to 1.52)	0.0065

FIGURE 4.

Changes in the anxiety and depression sections of the HADS over 5 years. (a) HADS anxiety total score; and (b) HADS depression total score.

Social functioning

There was very little difference between the scores on the SFQ between the two groups over the 5-year period. The scores at baseline indicated a moderate degree of social dysfunction (the normal population average is 4.6 points and a mean score of 7.7 points is found with patients attending general practice with common mental disorders. 13 The higher than expected SFQ scores were partly explained by comorbid medical pathology but also by personality status (see Table 9). Over the course of the study there was a modest improvement in social functioning in both groups (Table 6).

Influence of obsessional symptoms on outcome

Scores on the SOCS scale were dichotomised at the level of 6 points, as recommended by the original and subsequent authors. 17,29 The results showed that patients with higher SOCS scores had somewhat lesser improvement in HAI scores with CBT-HA over the 5-year period (Figure 5) and that at 5 years the difference between CBT-HA and standard care was not significant (see Table 13).

FIGURE 5.

Mean change in HAI score by initial scores on the SOCS, with scores dichotomised at the recommended score of 6. (a) SOCS score of ≤ 6 points; and (b) SOCS score of ≥ 6 points.

Influence of personality status on outcome

Personality assessment was carried out using the PAS-Q,20 derived from its parent instrument,30 which records both the severity and the type of personality disorder with an algorithm to identify the International Statistical Classification of Diseases, Tenth Edition (ICD-10)31 personality disorders, including dependent personality disorder. SC trained all raters in advance on a standard course. During the progress of the study the Working Group for the Reclassification of Personality Disorder in ICD-11 completed its initial work on a classification based on severity criteria (April 2010). 22 The PAS-Q data were subsequently reclassified to the ICD-11 severity equivalents. Good reliability between assessors in this exercise was achieved. 32

Using the ICD-11 classification, only 63 (14.2%) had no personality dysfunction but 197 (44.3%) had personality difficulty (a subthreshold condition not qualifying for disorder). Only three people assessed had severe personality disorder and so they were included with the moderate group. No differences in patient characteristics at baseline were identified and there was an even spread of male/female and a similar age profile between the ICD-11 personality groups (Table 7). However, there were significant differences in symptoms of health anxiety and generalised anxiety, depression and social functioning at baseline; participants with moderate to severe personality disorder had significantly higher scores than those with no personality disturbance (see Table 7). There were no differences in total cost at baseline apart from somewhat greater costs for each increment of personality severity.

The outcome data over follow-up by ICD-11 classification are detailed in Table 8 and Figure 6. Contrary to our hypotheses, the results show that those with no personality dysfunction showed no benefit from CBT-HA at any time point in the study; overall standard care was superior (p < 0.05). For all other groups the picture was different. For participants with personality difficulty and mild personality disorder, there was evidence of strong gains from CBT-HA at all time points compared with standard care (p < 0.001), and these were maintained at 5 years, especially in those with mild personality disorder. For participants with moderate and severe personality disorder, the initial benefit was not retained at 2 years, resulting in a weaker relationship over follow-up (p < 0.05), and by 5 years benefit was lost. Improvement in social function was similar in all groups except in those with no personality dysfunction (p < 0.02 in favour of standard care) and in those of mild personality disorder at 5 years, whose SFQ scores were significantly lower in the CBT-HA group (Figure 7). Clinical symptomatology increased and social dysfunction was greater with each increment of personality pathology, and although the results were most marked in those with health anxiety they were also found with generalised anxiety and depressive symptoms (see Table 7).

FIGURE 6.

Differential outcome of patients over 5 years treated by CBT-HA and standard care groups separated by ICD-11 personality status at baseline (see Table 8 for significance of differences). (a) No personality dysfunction; (b) personality difficulty; (c) mild personality disorder; and (d) moderate/severe disorder.

FIGURE 7.

Change in social functioning separated by ICD-11 personality status in CBT-HA and standard care groups over a 5-year period. (a) No personality dysfunction; (b) personality difficulty; (c) mild personality disorder; and (d) moderate/severe disorder.

Costs were lower in patients treated with CBT-HA for all levels of personality disturbance apart from moderate and severe personality disorder, when costs were greater (Table 9).

Effect of personality: dependent personality traits

Dependent personality status was also assessed dimensionally in the same way as the ICD-11 severity levels (range 0–24) using the DPQ. 18 A previous epidemiological study in general practice33 had identified the range of scores associated with disability with 0–6 (no personality dysfunction), 7–11 (personality difficulty), 12–16 (mild personality disorder) and ≥ 17 (moderate personality disorder) linking to ICD-11 severities. The DPQ does not allow severe dependent personality disorder to be identified because none of its questions address self-harm or aggression, even though these may sometimes be present in association with dependent personality. 34

As with the main study using ICD-11 diagnostic criteria for all personality disorders, most patients in the study had some degree of dependent personality disturbance at baseline, with women having more dependent traits than men (Table 10). The generalised anxiety and depression scores were higher in those who were more dependent, although this was not the case for the HAI scores, and there was significantly greater social dysfunction in higher scorers (see Table 10). The mean number of treatment sessions was higher in those with more dependent personality dysfunction (p < 0.005) (Table 11).

Of the 444 patients, 47 had an ICD-10 diagnosis of dependent personality disorder, seven with dependent disorder alone and 18 in conjunction with other personality disorders. Agreement between the DPQ and the PAS-Q groups was significant but relatively low (weighted kappa 0.16; p < 0.01).

Contrary to the initial hypotheses, patients treated with CBT-HA who had greater levels of dependence showed superior improvement in health-anxiety scores to those receiving standard care at all times of testing, and this was equally strong after 5 years (Table 12). The score differences were of greater magnitude in those with moderate dependent personality disorder, among whom, at 5 years, the mean score of 13.9 was fairly close to the normal level in the population (around 11)18 (see Table 12). Generalised anxiety and depression scores (HADS) were also significantly more improved in those who received CBT-HA than in those who received standard care. Social function showed no significant differences between the allocated treatments separated by dependence groups (see Table 12).

Differences in outcome by site, clinic and age

These outcomes were considered relevant in the covariate-adjusted analyses and are summarised here, but it is important to emphasise that there were no particular hypotheses linked to these outcomes and randomisation did not take account of these. The five sites covered a mix of urban, suburban and rural areas, including two main teaching hospitals (St Mary’s Hospital and Charing Cross Hospital), two hospitals linked to teaching hospitals (Hillingdon and Chelsea and Westminster) and one district general hospital (King’s Mill Hospital). Four of the clinics concerned (cardiology, endocrinology, gastroenterology and respiratory medicine) had been identified previously as having at least 12% of attenders with a high level of health anxiety,7,35 and during the course of the study another clinic type, neurology, was added.

The differences between sites are shown in Figure 8. In general, the non-teaching hospitals showed the largest differences between the patients receiving CBT-HA and those receiving standard care. There is no obvious explanation for this difference except in terms of therapist type (see Subgroup analyses). There were also differences in HAI outcome by clinic. Patients seen in cardiology clinics showed the largest differences between those who received CBT-HA and those receiving standard care, maximal at 5 years (Figure 9), and respiratory medicine patients showed the least differences between groups (with no benefit after the first 6 months of treatment).

FIGURE 8.

Differences in outcome by HAI scores separated by site. (a) Chelsea and Westminster; (b) Charing Cross Hospital; (c) Hillingdon Hospital; (d) King’s Mill Hospital; and (e) St Mary’s Hospital.

FIGURE 9.

Differences in outcome by HAI scores separated by clinic type. (a) Cardiology; (b) endocrinology; (c) gastroenterology; (d) neurology; and (e) respiratory medicine.

Older patients derived greater benefit from CBT-HA compared with standard care than younger patients. This was not an expected finding and possible reasons for it are discussed later in this report.

Subgroup analyses

There were several subgroup analyses that showed generally similar findings. The subgroup analysis to examine the heterogeneity of the variables of interest is shown for the primary outcome, the HAI, in Table 13. There were no significant interactions, suggesting that the data were consistent and robust. The only variable approaching heterogeneity was the study site. King’s Mill Hospital had a better outcome than the other sites over the period of the study; this was also the site where the nurses were the main therapists.

Outcome by therapist type

When the CHAMP trial started it was hoped that sufficient excess treatment costs might be available to fund dedicated therapists at all sites in the study; this was not possible, but at one site (King’s Mill Hospital) a subvention grant allowed two general nurses (SM and YL-S) to be employed for approximately 80% of their time in giving CBT-HA. In other centres somewhat less funding was obtained for other therapists, mainly trainee psychologists, but also other health professionals, including a dietitian, to be employed on a mainly part-time basis.

Apart from the nurses and the dietitian, the other therapists had at least some knowledge of the principles of CBT before the trial began and some had practised, but none had been trained in, CBT-HA. All therapists were involved in the initial 2-day workshop and, once trained at this standard level, received supervision from established trainers at each site.

The results, also reported elsewhere,36 showed that although the patients treated by all three therapist groups (nurses, assistant psychologists and graduate workers) showed superior HAI outcomes after 6 months, the nurses group showed the largest benefit across the initial 2-year period (Figure 10 and Table 14). After 2 years, 23.5% of patients in standard care had improved to a score of 15 on the HAI (mild but not pathological health anxiety), whereas the proportions in the three therapist CBT-HA groups were 28.9% (assistant psychologists), 34% (graduate workers) and 44.3% (nurses) (Table 15).

FIGURE 10.

Mean differences in HAI scores between the outcomes of 219 patients with health anxiety treated by nurses (n = 66), graduate workers (n = 66) and assistant psychologists (n = 87) over 2 years. Reprinted from the International Journal of Nursing Studies, vol. 52, Tyrer H, Tyrer P, Lisseman-Stones Y, McAllister S, Cooper S, Salkovskis P, et al. Therapist differences in a randomised trial of the outcome of cognitive behaviour therapy for health anxiety in medical patients, pp. 686–94. Copyright (2015),36 with permission from Elsevier.

Perspective

The economic evaluation took a health and social care perspective, which included the costs of the CBT intervention, other health-care costs and other community health and social services. In addition, we completed the sensitivity analysis from a societal perspective, to include all resources in the health and social care perspective plus productivity losses.

Identification of resources

The calculation of costs was separated into three stages: identification, measurement and valuation of resources.

We identified relevant resources based on the results of our pilot study10 and in discussion with study clinicians and patient representatives. Resource use was collected in the following domains:

• delivery of CBT-HA

• use of NHS secondary care services

• inpatient stays

• day-case procedures

• outpatient appointments

• accident and emergency attendances

• use of NHS primary care services

• General practitioner [(GP) in practice, at home or by telephone]

• community nurse (practice nurse, district nurse, health visitor, midwife)

• community mental health services

• community medical services (walk-in clinic)

• community medical professional (physiotherapist, chiropodist)

• use of all medication

• use of social care and voluntary sector services

• social worker

• voluntary sector advice worker

• housing services

• service-provided accommodation

• productivity losses measured as absenteeism.

Measurement of resources

The identified resources were measured using the following methods.

The CBT-HA therapists recorded information on attendance and non-attendance at treatment sessions together with the duration of each session. These data were extracted for analysis from trial records.

The computerised records of the host hospital trusts (Sherwood Forest Hospitals NHS Foundation Trust, Imperial College Healthcare NHS Trust, Chelsea and Westminster Hospital NHS Foundation Trust and The Hillingdon Hospitals NHS Foundation Trust) were examined and accident and emergency attendances, inpatient stays, day-patient procedures and outpatient appointments were extracted for analysis by hospital trust-based information specialists co-ordinated by AP.

Other service use data for the economic evaluation were collected using the ADSUS, based on previous economic evaluations in adult mental health populations37 and on the data collected in the pilot study. 10 The ADSUS was completed at baseline, at 6, 12 and 24 months and at the 5-year follow-up. At baseline, the ADSUS covered the previous 6 months and at each of the follow-up interviews, service use since the previous interview was recorded so that the entire period from baseline to final follow-up was covered. The ADSUS records the number and duration of contacts with a range of health and social service professionals and medications taken.

The number of days taken off work because of ill health was collected with the relevant questions from the World Health Organization Health and Work Performance Questionnaire. 38

Valuation of resources

The cost of the CBT-HA intervention was estimated using the bottom-up approach set out by the Personal Social Services Research Unit at the University of Kent. 39 The salary cost for the therapist was based on that of an experienced nurse or therapist (Agenda for Change Band 6),40 onto which employer costs (National Insurance and pension contribution) were added. Overhead costs were then added to include building costs, administrative and managerial costs and capital costs. An hourly cost was calculated on the working time assumptions set out in the Unit Costs of Health and Social Care,41 and then weighted to account for the therapist spending time on non-patient-facing activities. In addition, the final cost per hour included an allowance for the cost of time spent by the therapist in supervision and the cost of supervisor time.

For each type of other service use, an appropriate unit cost was identified. All unit costs were for the financial year 2009–10 and costs with their sources are listed in full in Table 16. When necessary, unit costs were inflated to 2009–10 costs using the Hospital and Community Health Services inflation index or the Retail Price Index as appropriate. 41 Costs in the second, third, fourth and fifth years were discounted at a rate of 3.5% per annum as recommended by the National Institute for Health and Care Excellence (NICE). 44

The cost of productivity losses was calculated using the human capital approach, when each day off work is valued at the daily salary for that individual. 45

Calculation of quality-adjusted life-years

Quality-adjusted life-years (QALYs) were calculated on the basis of the EQ-5D health state classification instrument46 and the health states were assigned a utility score using responses from a representative sample of adults in the UK. 47 QALYs were then calculated as the area under the curve defined by the utility values at baseline and at the 6-, 12-, 24- and 60-month follow-ups. We assumed that changes in utility score over time followed a linear path. 48 QALYs in the second to fifth years were discounted at a rate of 3.5% as recommended by NICE44 and all analyses were adjusted for baseline utility scores to take into consideration the impact of baseline differences on the area under the curve. 49

Data analysis

For the economic evaluation the base case was a complete-case analysis.

We report the mean average resource use by randomised groups at 24 months’ and 5 years’ follow-up, as well as the percentage of people in each randomised group who have had at least one contact with each service. No between-group statistical comparisons are made in order to avoid the problems of multiple comparisons and because the focus of the economic evaluation is on cost and cost-effectiveness.

Our hypothesis was that the cost of the CBT intervention would be recouped through a reduction in the use of health and social care services. We tested this through a non-inferiority (equivalence) test, with a pre-specified equivalence test of £150. The equivalence was declared between active and control groups if the 95% CI fell between –£150 and £150. These analyses were adjusted for baseline differences in cost.

Cost-effectiveness was considered for two outcomes: QALYs and the HAI score. Cost-effectiveness was initially explored through the calculation of incremental cost-effectiveness ratios (ICERs), a ratio of differences in cost by differences in outcomes between the CBT-HA and standard care groups. ICERs are a useful decision-making tool when there is no uncertainty around the costs and outcomes used in the ratio. Here, the ICERs are based on four sample means; therefore, the uncertainty around the ICER was explored.

Cost-effectiveness planes were produced by the production of bootstrapped regressions (5000 replications) of study group on total costs and HAI score or QALYs, with covariates for baseline costs or outcomes as appropriate. The coefficients of group differences were then plotted on to a graph.

Replications in the north-east quadrant suggest that CBT-HA is more effective and more costly than standard care, replications in the south-east quadrant suggest that CBT-HA is more effective and less costly than standard care, replications in the south-west quadrant suggest that CBT-HA is less effective and less costly than standard care and replications in the north-west quadrant suggest that CBT-HA is more costly and less effective than standard care.

Knowledge of uncertainty around incremental cost-effectiveness is not sufficient for decision-making, which will depend on the how much society is willing to pay for improvements in outcomes. Cost-effectiveness acceptability curves (CEACs) were therefore constructed, and these show the likelihood that CBT is cost-effective for different values a decision-maker is willing to pay for improvements in outcome. 50 All analyses were adjusted for baseline costs and for baseline HAI and EQ-5D scores as appropriate.

We completed a number of sensitivity analyses. First, we used the Stata^® (StataCorp LP, College Station, TX, USA) impute command (a regression-based single imputation) to impute missing costs at 24 months and 5 years. Second, we undertook an analysis of differences in costs, including the estimates of the cost of productivity losses.

Results

At 24 months, full ADSUS and hospital records data were available for 172 participants in the CBT-HA group and 170 in the standard care group (77% of those randomised). At 5 years, full ADSUS and hospital records data were available for 149 participants in the CBT-HA group and 157 in the standard care group (69% of those randomised). For total costs over 5 years we needed data from the 24-month and 5-year interviews and there were 141 participants in the CBT-HA group and 133 in the standard care group who met this criterion (62%). We were able to collect hospital service use for 442 randomised participants (99%).

We examined the cost data in order to examine the impact of highly influential observations, which, as defined by Weichle et al. ,51 were those that were above the 99th centile and when removed would have resulted in major changes to the results. Following this, two observations were identified and were removed from the main analysis as recommended.

Resource use

Service use over 24 months’ follow-up is detailed in Table 17. There were no between-group differences in service use and no evidence that CBT-HA reduced service use over this period, as the use of some hospital and community services remained high.

Use of service-provided accommodation such as hotels, refuges and emergency accommodation (e.g. bed and breakfast) was low, being accessed by 2–4% of the sample over the 24 months.

Hospital outpatient appointments were attended by 99% of the study participants and the average number of appointments was 13. Around one-third of the study participants were admitted as inpatients and the average length of stay among the whole sample was between two and three nights. Attendance at accident and emergency departments was low.

The use of GPs and community medical services such as community nursing, physiotherapy, pharmacy and walk-in centres was common. Almost all participants (up to 98%) had at least one contact with a GP and the average number of appointments was between 13 and 16. Interestingly, the use of community mental health services was low; only 2–5% of participants had any contact with a community mental health professional.

In some areas, the use of services between 2 and 5 years continued the pattern set at 24 months, as detailed in Table 18. For example, around one-third of the sample in both groups had an inpatient stay. However, there was a 20% drop in the percentage of participants having at least one outpatient appointment and, as a yearly average, the number of appointments was lower. Similarly, over the 2–5-year period there were, on average, only five GP appointments, compared with 14 over the first 24 months.

Over the 5-year follow-up period, almost all inpatient stays were medical admissions and there were no admissions to mental health units. Similarly, almost all non-trial outpatient appointments were to medical specialties rather than to a mental health professional. In common with the numbers at 24 months, the use of community mental health services was very low.

Total cost

The mean cost of the CBT-HA intervention in the intervention group was £421.51 per patient (range 0–2383) for an average of six sessions.

Table 19 shows total costs per patient over 24 months’ follow-up. For all cost categories, total costs were lower for the CBT-HA group than for the control group. Overall, total health and social care costs, including the cost of the intervention, were lower in the CBT-HA group (mean £7314) than in the control group (mean £7727). In analyses adjusted for baseline cost, however, the adjusted mean difference between the two groups was £156 (95% CI –£1446 to £1758; p = 0.848). Although equivalence was not achieved, there was no evidence of a significant difference in cost between the CBT-HA and standard care groups.

The total cost for the 5-year follow-up is detailed in Table 20. The second row of the table reports costs in years 3, 4 and 5 as £5055 in the CBT-HA group and £4835 in the standard care group. The adjusted mean difference in cost is –£557, thus rejecting the null hypothesis that costs are equivalent at follow-up. Taking the costs between randomisation and 24 months and 2 and 5 years together gives the total costs over 5 years’ follow-up. These are also listed in Table 19. In total, costs were £12,591 in the CBT-HA group and £13,335 in the standard care group. The regression adjusted for baseline costs also showed that these costs were not equivalent [coefficient = 545, 95% CI (–2375 to 3466); p = 0.713]. The direction of the difference in costs changes when the adjustment is made for baseline costs.

Outcomes

The EQ-5D tariff over the 5-year follow-up is detailed in Table 21 and shown graphically in Figure 11. Utility scores are very similar between groups over follow-up. When QALYs were calculated, these were 1.09 in both the CBT-HA group and the standard care group at 24 months, and 2.75 in the CBT-HA group and 2.64 in the standard care group at 5 years; this small difference at 5 years is not significant.

FIGURE 11.

Mean EQ-5D tariff by randomised group.

The EQ-5D visual analogue scale (VAS) scores are also presented in Table 21 and Figure 12. Again the results are generally similar between groups and, while there are some differences in VAS score at 24 months, the difference is not significant and not sustained at 5 years.

FIGURE 12.

Mean EQ-5D VAS by randomised group.

Cost-effectiveness analysis

In adjusted analyses, the ICER for QALYs was £14,169 (the difference in cost, £156, divided by the difference in QALYs, 0.011), which highlights that additional QALYs can be gained only with additional expenditure. The scatterplot and CEAC in Figures 13 and 14 demonstrate that there is substantial uncertainty around the mean values used in the ICER. In Figure 13, the bootstrapped replications of costs and outcomes fall almost equally across all four quadrants of the cost-effectiveness plane, suggesting that there is no clear evidence on the cost-effectiveness of either CBT-HA or standard care. The CEAC in Figure 14 suggests that CBT-HA is not cost-effective for the usually accepted values that a decision-maker would be willing to pay for improvements in outcomes.

FIGURE 13.

Cost-effectiveness plane showing bootstrapped cost and QALY pairs at the 24-month follow-up.

FIGURE 14.

Cost-effectiveness acceptability curve showing the probability that CBT-HA is more cost-effective than standard care in terms of QALYs at the 24-month follow-up.

The ICER for the HAI, adjusted for baseline costs, was £55.86 (the difference in cost, £156, divided by the difference in HAI score, 2.79 points). The adjusted scatterplot in Figure 15 shows replications in the north-west and south-west quadrants. Here, these quadrants represent better outcomes in the CBT-HA group because a higher score on the HAI denotes worse anxiety. There is clear evidence that CBT-HA improves outcomes, but the picture regarding costs is uncertain. The CEAC in Figure 16 reflects this uncertainty, demonstrating that, at a willingness to pay for an improvement in HAI score of zero points, there is only a 44% probability of CBT-HA being cost-effective. However, if willingness to pay per point change in HAI score increases to values over £53, there is a > 50% probability that CBT-HA is cost-effective.

FIGURE 15.

Cost-effectiveness plane showing bootstrapped cost and HAI score pairs at the 24-month follow-up.

FIGURE 16.

Cost-effectiveness acceptability curve showing the probability that CBT-HA is more cost-effective than standard care in terms of health anxiety at the 24-month follow-up.

Over 5 years, costs are, on average, lower in the CBT-HA group than in the standard care group and outcomes in terms of QALYs and HAI scores are better. In these cases ICERs are not the most useful decision-making tool, as negative values can produce a positive ICER; we therefore leave them to one side in order to concentrate on the scatterplots and CEACs presented in Figures 17–20. The scatterplot in Figure 17 shows that the bootstrapped replications mainly fall to the right of the y-axis in the north-east and south-east quadrants, where outcomes are better but costs are either higher or lower. This finding is reflected in the CEAC for QALYs in Figure 18, which shows that there is little evidence that CBT-HA is cost-effective in terms of QALYs.

FIGURE 17.

Cost-effectiveness plane showing bootstrapped cost and QALY pairs at the 5-year follow-up.

FIGURE 18.

Cost-effectiveness acceptability curve showing the probability that CBT-HA is more cost-effective than standard care in terms of QALYs at the 5-year follow-up.

FIGURE 19.

Cost-effectiveness plane showing bootstrapped cost and HAI score pairs at the 5-year follow-up.

FIGURE 20.

Cost-effectiveness acceptability curve showing the probability that CBT-HA is more cost-effective than standard care in terms of health anxiety at the 5-year follow-up.

For the HAI score at 5 years, the bootstrapped replications in Figure 19 fall between the south-west and north-west quadrants (again these quadrants represent better outcomes in the CBT-HA group because a higher score on the HAI denotes worse anxiety). The related CEAC in Figure 20 generally supports the cost-effectiveness of CBT-HA, particularly for willingness-to-pay values for a unit change in HAI scores of > 100 points.

Sensitivity analysis

Imputation of missing data and the inclusion of productivity losses did not alter the finding that costs were not equivalent, as detailed in Table 22.

Identification and classification of health anxiety

When the CHAMP trial started in 2008, health anxiety had uncertain status in international diagnostic systems. It was regarded as a synonym for hypochondriasis, although there was a suspicion among researchers that it was not exactly the same condition. Because of this uncertainty we also required patients to satisfy the diagnostic requirements of the DSM-IV for hypochondriasis, and so all who progressed to randomisation had this formal label, which placed patients in the somatoform disorder group.

Since then there have been new international classifications developed and a new diagnosis, illness anxiety (diagnostic code 300.7), was introduced to the fifth American classification Diagnostic and Statistical Manual of Mental Disorders-Fifth Edition. This is very similar to health anxiety, as it incorporates excessive concern and monitoring of health, fear of developing or having a physical disease, bodily checking and high levels of anxiety, but an absence of marked somatic symptoms. 57 This last requirement serves to separate illness anxiety from what used to be called ‘medically unexplained symptoms’, but have been relabelled somatic symptom disorders. Because of the apparent close links between illness anxiety and these other conditions, illness anxiety is placed in the somatic symptom group, but paradoxically has introduced the requirement that ‘marked somatic symptoms’ must be absent. In the experience of therapists in the CHAMP trial, somatic symptoms of anxiety are very common.

The ICD-11 classification group has chosen differently; it has opted to place health anxiety in the obsessional group of disorders, as there is a degree of overlap between the excessive checking behaviour of the health-anxious patient and obsessional behaviour,58 and this is also noted in the CHAMP study. There is considerable dissatisfaction with the somatic group of psychiatric disorders, even after these revisions,59 and the CHAMP trial results suggest that health anxiety is primarily an anxiety disorder, not a somatic one. This suggestion is made as the results of the HADS-A scale paralleled those of the HAI throughout the 5 years of study, and when health anxiety reduced so did generalised anxiety. A similar recommendation that health anxiety is fundamentally an anxiety disorder has also been made on the basis of its links to other disorders. 60

Representativeness of sample

It could be argued that, as < 10% of those who were originally identified with a high level of health anxiety took part in the trial, those who did participate were unrepresentative. This cannot be answered definitively, as we had no data apart from the HAI scores on those who did not participate, but we were not surprised by the relatively low uptake of invitee participation in the trial. Patients were approached at ordinary outpatient clinics and were not seeking any form of intervention for health anxiety. As they were attending medical clinics, it is reasonable to suppose that patients were seeking medical advice and not expecting to be invited to take part in a psychological treatment.

If we had adopted other ways of recruiting, such as asking consultants to refer patients, we do not think that we would have recruited so many participants. Although the staff in some hospitals are conscious of the importance of health anxiety – several were noted in our study – they are currently in a minority, and most medical staff do not regard the psychological aspects of their patients as a priority when their main tasks are to treat and exclude organic disease. In this respect, the study population was very different from those in other recent trials of internet-delivered or group treatment of health anxiety,52–55 as clearly those populations recognised the problem of health anxiety and were seeking help for it.

Differences from recent studies

Since the CHAMP trial was started in 2008 there has been one meta-analysis and five other trials published of CBT and related psychological treatments in health anxiety,52–56,61 all of which showed positive results in favour of the CBT-linked psychological intervention. There were some important differences between the populations treated in these studies and those in the CHAMP trial that need to be highlighted. All of these other publications described populations that had consulted or responded to advertisements from professionals who were providing treatment for health anxiety (i.e. the patients understood fully that they had health anxiety and wanted treatment for it). Patients in the CHAMP study were not seeking treatment and > 90% of those who screened positively for health anxiety (HAI score of ≥ 20 points) did not proceed to randomisation, largely because of refusal rather than exclusion criteria. This is not really surprising, as the people approached in medical clinics had a predominantly medical consultation mindset, with the expectation that, whatever the nature of any worries that might beset them, these would be answered by doctors in the clinic, not external psychological sources.

The mean scores for the HAI (≈25 points at baseline) in CHAMP were also considerably lower than the equivalent scores in these other studies (≈40 points), and this explains why many of these other studies refer to ‘severe health anxiety’ in their titles. The population in CHAMP was not a highly handicapped one (except when there was associated significant medical pathology), and it is possible that the long-term benefits shown with CBT-HA in this group were achieved because of this lesser handicap. Put another way, once people had ‘seen the light’ and realised that health anxiety was an important part of their problem, it did not require a great deal of intervention to make this benefit semi-permanent. The additional evidence that this was achieved with benefits to generalised anxiety and depression over 5 years shows the degree of overlap between health anxiety and other common mental disorders.

This supposition is also supported by the improvement shown in the patients allocated to standard care. Over the course of 5 years the maximal improvement (60%) was shown in the first 3 months (see Figure 3), suggesting that even the relatively short and bland explanation of health anxiety during the baseline assessment was enough to give a degree of insight into the subject and yield some benefit.

Meaningful clinical difference in Health Anxiety Inventory scores

It is not known what constitutes a meaningful clinical difference (MCD) in scores on either the long or short form of the HAI. In very large trials it is possible to find statistically significant differences that are not clinically meaningful because the differences are so small. In our original sample size calculations we suggested that a difference of 5 points in treatment groups was meaningful, but this was based on limited evidence.

Previous studies of anxiety using the HADS have suggested that a reduction in the HADS-A score of 0.7 points or a reduction in the total scale score of 1.32 points is clinically meaningful62,63 and, if we take a conservative estimate of 1 point for a MCD for the HADS-A and 0.7 points for the HADS-D, this roughly equates to a MCD of 2 points on the HAI. On this basis, the differences between treatments at all assessment points after baseline suggest that CBT-HA is clinically effective as well as statistically superior to standard care.

Relevance of findings of personality status

The findings with regard to personality status were of particular interest in this study because they contradicted previous robust findings with regard to the outcome of depression and, to a lesser extent, anxiety. 62–65 The main studies, backed up by a substantial meta-analysis, showed that patients with comorbid depressive and personality disorder had an odds ratio of > 2 in having a worse symptomatic outcome than those who had depression alone. 64,65 It was expected that these findings would extend to health anxiety and a previous cohort study had found that those with hypochondriacal personality characteristics, possibly of sufficient severity to be called hypochondriacal personality disorder, had a much worse outcome than those without this disorder. 66

All studies in this subject carried out previously were with patients who had generalised anxiety or depression. The CHAMP study findings of greater improvement with CBT-HA than standard care, extending over 5 years, in those with personality difficulty and mild personality disorder, findings not shown with other personality groups, are striking. The additional evidence that this difference in improvement also extended to social functioning, and that standard care led to better social function in those with no personality dysfunction, adds to the impact of these findings.

However, there was an important difference between the assessment of personality in CHAMP and those of other studies. Currently personality disorder is diagnosed dichotomously in classification systems: patients either do or do not have personality disorder. The ICD-11 system of classification67 uses a single dimension of classification consistent with research evidence; of the two levels of personality disturbance that showed persistent preferential improvement with CBT-HA, one, personality difficulty, is classified as a ‘non-disorder’, and the other, mild personality disorder, is included among the personality disorders. Using the older dichotomous system of classification, the differences between outcomes would have been attenuated, but not to the extent that the previous findings of a poorer outcome in personality-disordered patients would be replicated.

The high proportion of patients with at least some level of personality disturbance (80%) may appear surprising but is also consistent with other data. 68–71 People with high levels of health anxiety and, to a lesser extent, generalised anxiety, often have personalities that are strong in avoidant, anxious and dependent characteristics, and this impairs both relationships and functioning, and adds to symptomatology (see Table 17). The evidence that this does not handicap patients’ responses to CBT-HA, in either the short or longer term, is encouraging. Whether or not the difference in response by personality level is sufficiently strong to suggest that personality assessment should be undertaken in ordinary practice is beyond the scope of this study, but there are cogent arguments in its favour. 70,71

It is not clear why those with no personality dysfunction did not respond to CBT-HA or why this treatment seemed to impair social functioning. One possible explanation is that this group had appropriate rather than pathological anxiety, and so the introduction of CBT-HA was regarded not only as unnecessary, but also as irritating or annoying.

Implications for health services

Our findings indicate that excessive health anxiety is common and, if untreated, tends to persist, which supports previous evidence from several sources. 6,72,73 Therefore, it appears that the natural course of health anxiety is quite different from that of generalised anxiety disorder, which tends to pursue a fluctuating path depending on circumstances. Possible reasons for health anxiety persisting without much improvement include the reinforcing aspects of reassurance and medical consultation, and these tend to continue in the absence of any appropriate management.

At the time of writing, in most parts of the UK this management can be accessed only through the Improving Access to Psychological Therapies (IAPT) programme but, although practitioners in the services were initially trained to have skills in CBT for depression and anxiety, the IAPT services have extended these to include health anxiety. However, the numbers potentially benefiting from therapy, possibly one in five of all medical outpatients, are far beyond the capacity of an external service to serve. It may appear too strong a claim, but the number of people with major health-anxiety concerns in general hospitals is both sufficiently large and undetected to be called a silent epidemic. 74 There is a growing realisation that better integration of medical and mental health services is necessary in general hospitals75 and that it would be a great advantage to develop skills within the general hospital to identify and treat patients with health anxiety, alongside the provision of normal care. The evidence from the CHAMP study that general nurses, when appropriately trained, were highly suitable practitioners of CBT-HA, suggests that an obvious way forward would be to train interested nurses in general hospitals in the principles of treatment and to raise awareness of the nature of health anxiety among their fellow professionals.

Comment was made after the first major CHAMP publication that the intervention could not be regarded as cost-effective because of the high degree of screening needed to recruit patients. 76 We argue that if health-anxiety literacy in medical clinics were greater, there would be no need for screening, and until this time comes it would be perfectly reasonable for the HAI or a similar scale to be given to patients when first registering at a medical outpatient clinic, at very low cost. In this way, patients who present recurrently with anxiety about problems that appear to have no physical cause would be identified in the clinic and an appropriate response made.

Long-term benefit of cognitive–behaviour therapy for health anxiety

The benefit of CBT-HA in reducing health anxiety appears to be long lasting. This finding was unexpected, not least as most interventions for anxiety and depression using CBT tend to diminish over time and are seldom present over 1 year after completing therapy. 77–81 There is no doubt that CBT is effective in treating both generalised anxiety and depression (these benefits often go together),80 but the evidence that these benefits extend beyond 1–2 years is very weak, partly because so many trials have waiting list controls that curtail comparison of groups over long periods. 79,80 There is one exception to this: the CoBalT study of CBT for depression showed continuing benefits of CBT 40 months after randomisation. 81 There is no equivalent for generalised anxiety.

The evidence that these benefits were maintained with gains at 2 years being just as strong at 5 years deserves discussion. There could be at least three reasons for this: first, the failure of those in standard care to improve beyond the initial gains at 3 months could have been a consequence of their continued concerns and consultations about their health, which only reinforced their symptoms. Second, in those who received CBT-HA in this particular population, the acquisition of insight into the presence of health anxiety, together with a relatively straightforward set of learnt procedures to overcome it, may have been a powerful restitutional force that kept people well (other general advice, including avoidance of internet browsing for health problems, may also have been of benefit). The third reason relates to psychological help-seeking – in most studies of psychological treatment for anxiety and depression the patients are seeking treatment, and if they do not receive it from one source they get it from another. In the words of a study of collaborative care, those in control populations ‘catch up’ on those who have more specific interventions over the course of time82 because they eventually receive the care that is needed.

The population with health anxiety in CHAMP was not help-seeking and so ‘catching up’ was not an issue for them. Before being involved in the CHAMP study, patients felt that it was quite in order for them to continue to express concerns about their health to their physicians and no further intervention was necessary. Therefore, there was no attempt from those in the control group to seek further assistance and, indeed, there was no evidence of crossover to CBT-HA or other related therapies in the standard care group over the 5-year period.

In view of the maintenance of benefit of CBT-HA after 5 years, the project group is now carrying out an 8-year follow-up. This will also examine costs again to see if the continued savings between 2 and 5 years are extended.

Interpretation of the economic evaluation

In terms of cost-effectiveness, our data suggest that the picture is more complex than our original hypothesis that successful CBT-HA would reduce health anxiety and, therefore, unnecessary service use would result in lower costs. The results of the cost-effectiveness analysis therefore require careful interpretation. Our original hypothesis was that savings would offset the additional costs of the CBT intervention elsewhere in the health system, resulting in equivalent costs between groups. We found evidence of offset but did not find hypothesised equivalence, nor did we find any noticeable or significant difference in cost between groups, except in personality status. As a consequence, it is right to state that the CHAMP trial did not find clear evidence of cost-effectiveness. These issues will be considered in the following section.

The inability to demonstrate equivalence and cost-effectiveness has its root in two key characteristics of the data and the patient group: that there is considerable variation in costs and that the costs are multifaceted. Variation in costs is not unusual in cost data; however, the variation in this group was substantial enough that it was necessary to remove two outliers from the main analysis as they met the criteria for influential outliers. The multifaceted nature of the costs needs to be unpacked a little. If we take costs back to service use, it allows us to try and explain what exactly this phenomenon might be. Table 17 shows that levels of hospital and GP use were considerable, and that use of these services continued in both randomised groups throughout follow-up. Given that we believe that the patient group is one with substantial medical pathology, it is perhaps unsurprising that use of these medical services remained high despite a noted and consistent reduction in health anxiety in the CBT-HA group. Thus, any reductions in service use as a result of changes in behaviour following successful CBT-HA are lost in the noise of all other medical service use.

A previous study of the cost-effectiveness of CBT-HA compared with control treatment was carried out in Sweden using internet-guided treatment. 83 This showed that the total costs were only a little higher (< 15%) than those of health care in this population, suggesting that future studies could rely on health-care costs alone.

There are a number of limitations of the economic evaluation, which resulted from both the duration of the analysis and the patient group. In terms of the duration of the analysis, collecting service use data over long periods generates a number of challenges. For the follow-up interviews up to 12 months, the ADSUS asked participants to recall their service use only for the previous 6 months. At the 24-month interview recall was for 12 months, and at the 5-year interview recall was for 3 years. We acknowledge that the accuracy of recall over long periods is limited and we sought to limit the impact of this in two ways. First, the list of services that we asked participants to recall was short and limited to services provided in the community. Second, we did not ask participants to recall use of hospital services and instead extracted these from hospital trust computer databases. In addition, we were able to extract hospital use data from the trial host hospital trust databases only, and it is likely that participants accessed services from other hospitals from which we were not permitted to access information. It is important to note that there is nothing to suggest that recall bias is not equally distributed across the whole cohort. If this were so, the randomised group differences are valid. However, recall bias may provide an explanation for the apparent reduction in community service use at 5 years.

Another important limitation was the choice of outcome measure for the economic evaluation. Our results suggest that it is likely that CBT-HA is cost-effective in terms of health anxiety and that the probability of it being cost-effective increases with positive values on the willingness to pay for an improvement in outcome. The difficulty with this is that the willingness-to-pay value for a change in HAI score is unknown, so it is therefore not possible to make clear statements in terms of policy-making. Regarding the economic outcomes, there were no noticeable improvements in quality of life as measured with the EQ-5D. One explanation for the inability of the EQ-5D to pick up the improvement in the CBT-HA group that was found with the HAI outcome links back to a key characteristic of the patient group; rather than being a group of individuals with health anxiety and medically unexplained symptoms, they are in many cases individuals with health anxiety and substantial medical pathology. These problems would inevitably have an impact on an individual’s quality of life. The EQ-5D contains only one domain on depression/anxiety and it is likely that study participants would continue to score highly in the other domains of the EQ-5D, making any differences in anxiety as a result of the intervention difficult to capture. It is also possible that the health-anxiety intervention did not have an impact on depression and, thus, there could be no impact on the EQ-5D scores in this way either. Future studies in health anxiety should consider a different measure of quality of life.

It does remain possible and important, however, to reach conclusions based on the evidence we have collected on the costs of health anxiety in medical patients. The first is that, in terms of the HAI score as an outcome, CBT-HA has a > 50% probability of being cost-effective for almost all willingness-to-pay values. The second is that, counter to this, there is no evidence that CBT-HA is cost-effective in terms of QALYs as measured using the EQ-5D.

Contributions of authors

The trial was initiated by Peter Tyrer and Helen Tyrer, who, with Paul Salkovskis, Michael J Crawford, Simon Dupont, John Green, David Murphy, Steven Reid, Sarah Byford and Barbara Barrett, designed the structure of the trial.

Duolao Wang and Barbara Barrett were involved primarily in developing the statistical analysis plan.

Sylvia Cooper was the trial co-ordinator and organiser of the recruitment strategy.

Georgina Smith, Rachel Evered, Yvonne Lisseman-Stones, Sharon McAllister and Jessica Nagar were the main therapists.

Rahil Sanatinia co-ordinated the acquisition of personality data for ICD-11 conversion.

Sharandeep Bhogal, Shaeda Nourmand, Valentina Lazarevic, Gemma Loebenberg, Stephanie Kings, Antoinette McNulty, Kofi Kramo, Katherine Whittamore and Gemma Walker were the main research assistants.

Aaron Philip assisted with the co-ordination and checking of hospital data, Hilary Warwick helped with assessment of treatment fidelity, and Sarah Byford and Barbara Barrett carried out the economic analysis.

All authors read and approved the final manuscript.

Publications

Tyrer P, Cooper S, Salkovskis P, Tyrer H, Crawford M, Byford S, et al. Clinical and cost-effectiveness of cognitive behaviour therapy for health anxiety in medical patients: a multicentre randomised controlled trial. Lancet 2014;383:219–25.

Tyrer H, Tyrer P, Lisseman-Stones Y, McAllister S, Cooper S, Salkovskis P, et al. Therapist differences in a randomised trial of the outcome of cognitive behaviour therapy for health anxiety in medical patients. Int J Nurs Stud 2015;52:686–94.

Sanatinia R, Wang D, Tyrer P, Tyrer H, Crawford M, Cooper S, et al. Impact of personality status on the outcomes and cost of cognitive-behavioural therapy for health anxiety. Br J Psychiatry 2016;209:244–50.

Tyrer P, Wang D, Tyrer H, Crawford M, Cooper S. Dimensions of dependence and their influence on the outcome of cognitive behaviour therapy for health anxiety: randomized controlled trial. Pers Ment Health 2016;10:96–105.

Data sharing statement

The full data set for this study can be obtained from the corresponding author.

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, NETSCC, the HTA programme or the Department of Health. 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 HTA programme or the Department of Health.