Cognitive behavior therapy for health anxiety: systematic review and meta-analysis of clinical efficacy and health economic outcomes

ABSTRACT

Introduction: Health anxiety, also known as ‘hypochondriasis’, is a common, distressing and costly condition that responds to cognitive behavior therapy (CBT) but evidence pertaining to response and remission rates, treatment in routine care, therapist-guided Internet-delivered CBT (ICBT) and health economics has not been systematically reviewed.

Areas covered: In this systematic review and meta-analysis we searched PubMed, PsycINFO, and OATD (17/06/2019) for randomized controlled trials (RCTs) comparing CBT to non-CBT controls for health anxiety. Based on 19 RCTs, the pooled between-group effect on health anxiety was moderate to large (g = 0.79; 95% CI: 0.57–1.01; adjusted for publication bias: g = 0.62), with small to moderate effects on secondary symptoms and effects largely sustained 12–18 months after treatment. Moderators were control condition and recruitment path, but not treatment setting. The pooled CBT response rate was 66%, and the remission rate 48%. ICBT had effects comparable to face-to-face CBT. CBT for health anxiety is probably cost-effective, but with limited effect on the quality of life.

Expert opinion: CBT is a highly efficacious and probably cost-effective treatment for health anxiety. We recommend that ICBT is implemented more widely, and that health economic outcomes and ways of increasing response and remission rates are explored further.

KEYWORDS:

• Cognitive behavior therapy
• cost-effectiveness
• efficacy
• health anxiety
• hypochondriasis
• systematic review

1. Introduction

It has been estimated that 3.4% of the general population [1] and up to 20% of patients in medical clinics [2] suffer from clinically significant levels of health anxiety, also referred to as ‘hypochondriasis’. Persistent and excessive fear of, or preoccupation with, severe illness not only leads to suffering and functional impairment but is also predictive of a substantial increase in health-care consumption and societal costs [1]. An illustrative example of this is that primary care patients with health anxiety have been found to consume 41–78% more health care than primary care patients with a well-defined medical condition [3]. Such figures are particularly concerning as experts speculate that health anxiety may have increased substantially in recent years [2].

Health anxiety, though not always recognized as such, exhibits the hallmarks of an anxiety disorder [4]. That is, patients with health anxiety respond to a specific type of threat (e.g. physical sensations and health-related information) with subjective fear or anxiety, physiological arousal, heightened attention, and avoidance. Moreover, they are also prone to interpret ambiguous stimuli (e.g., bodily processes) as indicative of feared outcomes (e.g., dying from the serious disease) [5–11]. Similar to safety behaviors seen in panic disorder and rituals seen in obsessive-compulsive disorder, health-related behaviors intended to reduce health anxiety in the short term (e.g., frequent symptom-checking and reassurance-seeking) have been found to maintain or even exacerbate pathological fear over time [12–15]. Although many patients with health anxiety respond to antidepressant medication [16], most prefer psychological treatment [17] and as in the case of most anxiety and obsessive-compulsive spectrum disorders, cognitive behavior therapy (CBT) based on exposure and response prevention [18] or cognitive restructuring techniques such as behavioral experiments [19] is most widely used and researched. Meta-analyses have found that CBT for health anxiety is an efficacious treatment, with large and lasting effects on core symptoms of health anxiety as well as therapeutic effects on secondary symptoms of depression and general anxiety [20–25]. The availability of CBT for health anxiety is however still poor, and much recent work in the field has focused on exploring new means of administering CBT, such as via the Internet, to implement the treatment on a larger scale. Recent work has also incorporated health-economic outcomes so as to evaluate treatments effects in relation to costs.

Despite a relatively large body of evidence, there are notable gaps in the current literature when it comes to systematic reviews and estimation of pooled treatment effects of CBT on health anxiety across available studies. First, there are, to our knowledge, no published meta-analytic estimates of response and remission rates. Second, the relative effect of CBT for health anxiety when given in a research context versus routine care has not been investigated. Third, there are no published estimates, based on the entire study population, of whether Internet-delivered CBT produces similar reductions of health anxiety as compared to face-to-face CBT. Fourth, to our knowledge, health economic outcomes of CBT for health anxiety have never been systematically reviewed.

We set out to review the evidence pertaining to the clinical effects and economic impact of CBT for adults with health anxiety, as based on randomized controlled trials [26]. This was a systematic review where we used meta-analytic methods to estimate pooled effects of CBT vs. control conditions, response and remission rates, within-group effects, the relative effect of CBT for health anxiety in research versus routine care, and the relative efficacy of therapist-guided Internet-delivered CBT and traditional individual face-to-face CBT. We also reviewed outcomes pertaining to the question of cost-effectiveness and cost-utility.

2. Methods

2.1. Design & search strategy

In accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) [27], we conducted a systematic review and meta-analysis of efficacy and health economic outcomes from randomized controlled trials of CBT for adults with health anxiety. This was primarily based on search results from three databases, namely PubMed, PsycINFO, and the Open Access Theses and Dissertations (OATD) resource (www.oatd.org) which were all searched on 17 June  2019. Our overarching search strategy was to combine search terms for health anxiety (‘health anxiety’[All Fields] OR ‘hypochondriasis’ [MeSH Terms] OR hypochondria*[All Fields] and so on) with search terms for randomized controlled trials (‘Randomized Controlled Trial’[pt] OR ‘randomised controlled trial’ [All Fields] OR ‘randomized controlled trial’ [All Fields] and so on). We hoped that the inclusion of gray literature would allow us to identify outcomes that did not reach publication in a peer-reviewed journal. On June 28 we also conducted an additional search to identify separate publications reporting health economic outcomes. For this search, we combined search terms for health anxiety with search terms for health economics. See the supplement for full search strings. We also searched through the references of previous reviews in the field [20–25,28].

2.2. Selection of studies

We performed systematic de-duplication of search hits in EndNote X9 [29]. As to the main search for efficacy outcomes, all unique publications were assessed by independent raters (i.e. both authors) in three phases: In phase one, we based exclusions on titles. In phase two, for all publications deemed relevant by at least one rater in phase one, we based exclusions on abstracts (summaries). Last, in phase three all publications deemed relevant by at least one rater in phase two were assessed in full text. In the case of missing titles or abstracts, publications progressed to the next phase. In the last phase, the first author assessed the eligibility criteria sequentially so that all publications that were formally excluded on criterion b had passed criterion a, all that were excluded on c had passed criteria a + b, and so on. Whenever there was disagreement on the inclusion of a full text, the reason for the discrepancy was discussed, and a decision was made in consensus. As to the separate search for health economic publications, all search hits were reviewed by the first author only.

2.3. Eligibility criteria

1. Publications were required to be written in English so that they could be reliably assessed. In order to increase our chances of identifying more recent trials, we decided to include both peer-reviewed journal articles and gray literature in terms of dissertations and theses.

2. Studies were required to be randomized controlled trials that incorporated at least one arm that was not CBT (see criterion e). The control condition could, for example, be a waiting-list condition, antidepressant medication, or another psychological treatment. We also included cluster-randomized trials, i.e. trials where randomization took place at the level of groups rather than individuals. The impact of this was investigated in a sensitivity analysis.

3. Study participants were required to be adults, i.e. at least 18 years old.

4. Study participants were required to suffer from clinically significant health anxiety, i.e., either qualify for a formal diagnosis or score above a sensible cutoff on a psychometrically valid measure of health anxiety. As to diagnoses, we saw DSM-IV hypochondriasis and DSM-5 illness anxiety disorder as the prototypical health anxiety disorders but were willing to include studies were recruitment had been based on another diagnosis characterized by a persistent and excessive fear or, or preoccupation with, having or acquiring a serious health condition. RCTs that included participants based on the broad diagnosis of DSM-5 somatic symptom disorder [30] were included under the condition that recruitment clearly focused on those who were significantly worried about their health and feared having or developing a severe illness. We did not include RCTs where medically unexplained symptoms, heightened health-care consumption, or distress due to physical symptoms were the principal clinical problem because such phenomena can be present regardless of health anxiety [24,31].

5. Studies had to investigate the effects of cognitive behavior therapy for health anxiety. We strived to include only the typical form of CBT for health anxiety where a substantial portion of the treatment is devoted to cognitive restructuring (e.g., by means of behavioral experiments), exposure-based techniques (including response prevention), or a combination of the two. We did not include trials of third-wave CBT such as acceptance and commitment therapy or mindfulness-based cognitive therapy, as we saw these as distinct therapies with particular emphasis on psychological flexibility and mindfulness practice. We did not include generic stress reduction programs such as behavioral stress management or problem-solving therapy, which has little in common with the conventional CBT for anxiety disorders. We did not include explanatory therapy as this is a treatment where, in contrast with conventional CBT for health anxiety, the therapist is instructed to use repeated reassurance to achieve therapeutic effects. We also did not include transdiagnostic treatments as we wanted to determine the effect of best practice treatments rather than broader interventions that need to suit a variety of mental health conditions. We required treatments to be at least 1 week long, but as we were interested in the significance of CBT delivery formats we did not impose any further restrictions in this regard. In other words, we, for example, included RCTs regardless of therapist involvement, so that also pure self-help treatments were included in the review.

6. We required studies to report the outcome of the trial in such a way that at least one standardized between-group effect on health anxiety was either reported directly or could be indirectly approximated [32]. For the health economic review, we required studies to make inference on the cost-effectiveness or cost-utility of CBT for health anxiety relative to a control condition.

7. For each eligible trial, one publication, usually a peer-reviewed journal article detailing the primary efficacy outcome of the trial, was formally included. Secondary papers such as those detailing the outcome of long-term follow-up assessments were not formally included, but sometimes used to make inference on study outcomes.

2.4. Data extraction

Data were extracted and tabulated in electronic spreadsheets. One continuous health anxiety measure was used from each trial. In most cases, this was the primary outcome of the trial, usually a self-rated questionnaire. In cases where no primary outcome was defined, we chose the first reported health anxiety outcome with two exceptions. In one case [33], we chose a visual-analogue scale (‘time spent worried about health’) based on its acceptable construct validity, and that it was similar to the only other visual-analogue scale included in the review [34]. In the other case [35], the primary outcome was a clinician-rated measure, and in order to facilitate comparisons over studies we instead tabulated a self-rated (secondary) measure of health anxiety. Two studies did not report post-treatment standard deviations [36,37], which we then estimated as the baseline standard deviation multiplied by 1.25. One of these studies [37] reported continuous outcomes in a manner that required us to approximate effects on continuous self-rated health anxiety based on simulation, distributional assumptions and response rates (see the online supplement for details).

Regarding secondary outcomes, we tabulated measures of depression, general anxiety, functional impairment, and physical symptoms. In terms of dichotomous outcomes, we strived to tabulate all estimates of response (change beyond a clinically relevant criterion) and remission (no diagnosis or score below a cutoff for health anxiety). We also tabulated estimates of clinically significant improvement which combines (a) an improvement that is reliable given the test–retest reliability of the instrument with (b) remission in terms of a final score below the cutoff for health anxiety [38]. Several putative moderator variables were also extracted with regard to sample characteristics (e.g., mean age, percentage female), the characteristics of CBT (e.g., delivery format) and the design of the study (e.g., setting, control conditions). In the extraction of health economic outcomes, if several time frames were explored in the publication, we tabulated the endpoint that was presented as primary by the authors of the original study. We tabulated both results pertaining to cost-effectiveness (i.e., costs in relation to the effect on health anxiety) and cost-utility (i.e., costs in relation to the effect on health-related quality of life).

2.5. Assessment of risk of bias

We assessed study risk of bias based on four criteria advocated by the Cochrane collaboration [39]: (I) random sequence generation, (II) allocation concealment, (III) incomplete outcome data and (IV) selective reporting of outcomes. For each criterion, studies were given a rating of ‘high risk of bias’, ‘low risk of bias’ or ‘unclear’ (such as when there was no preregistered trial protocol so as to enable the assessment of selective reporting). We did not rate the Cochrane criteria ‘blinding of participants and personnel’ because such blinding is not possible for psychological treatments, or ‘blinding of outcome assessment’ because we wanted to focus on self-rated measures where it is debatable whether such should generate a rating of 'high risk of bias'. As to the ‘incomplete outcome data’ and ‘selective reporting of outcome’ criteria, we focused on those health anxiety outcomes that were included in the systematic review. In other words, these ratings are not to be regarded as quality indicators of each RCT as a whole.

2.6. Statistical analysis

2.6.1. Clinical efficacy

We used the free statistics software R 3.6.0 [40] with RStudio 1.2 and the metafor package [41] to conduct meta-analysis based on random-effects models fitted with the restricted maximum likelihood estimator. Effects on continuous outcomes such as health anxiety were analyzed in terms of Hedges’ g which is a standardized mean difference that is adjusted for sample size [42]. In this study, positive between-group values for g are in favor of CBT over the comparator, and negative within-group values for g correspond to an improvement in symptoms. For the g statistic, it is commonplace to interpret absolute numbers of 0.2 as a small effect, 0.5 as a moderate effect, and 0.8 as a large effect [43].

As to the analysis of controlled effects on health anxiety, our strategy was to first conduct an overarching meta-analysis of all controlled effects, and then conduct a series of sensitivity analyses based on putative moderators (study design [including setting], sample characteristics, and treatment format). We tested for moderation based on Q-tests and meta-regression, and present secondary meta-analyses based on the levels of the putative moderator. That is, for example, one estimate of the pooled controlled effect on health anxiety per type of control condition.

In order to pool baseline estimates of health anxiety and depression, and also assess the possible moderating effect of these variables, it was necessary to convert baseline means on various scales to common metrics. For all studies except one [44], we were able to estimate mean health anxiety scores on the 14-item and 18-item Health Anxiety Inventory (HAI-14, HAI-18 [45]) and the Illness Attitudes Scales (IAS [46]), based on z-score linking (linear equating [47]) using estimates from two clinical samples [48,49]. That is, when baseline means on the HAI-14, HAI-18 or IAS were not available, health anxiety means were converted to z-scores using means and standard deviations from the aforementioned two studies. These z-scores were then, in turn, converted to HAI-14, HAI-18 or IAS scores. For the purpose of moderation analysis, we chose the HAI-18 as our primary measure of baseline health anxiety as we wanted coefficients to be meaningful and the HAI-18 has been, if not the most widely used measure of health anxiety, at least one of the most widely used measures of health anxiety in the past decade. Linking of depression means to categories of ‘mild’, 'moderate' and severe’ was based on a published latent depression metric [50]. For two scales, i.e., the Montgomery-Åsberg depression rating scale – self-rated [51] and the Geriatric depression scale [52], this could not be done and common scoring guidelines for these scales were used instead.

In order to quantify and test for heterogeneity, we used the Q and I² statistics. I² can be interpreted as the proportion of the between-study variance that can be explained by true study differences in effects rather than random (i.e., sampling) error. This is measured in percent, and an I² value of 25% is typically seen as an indication of low heterogeneity, 50% as moderate heterogeneity and 75% as high heterogeneity [53].

Between-group effects on response and remission rates were analyzed in terms of odds ratios (ORs), i.e., the ratio of the odds in CBT and the odds in the control conditions. In this study, ORs > 1 are in favor of CBT over the comparator. We also report the numbers needed to treat-statistic (NNT), with is the inverse of the absolute risk difference, and can be interpreted as the estimated number of participants necessary to treat with CBT rather than the control condition to achieve one additional beneficiary outcome (such as remission).

We assessed publication bias based on standard error versus effect size funnel plot symmetry of the primary outcome, i.e., the post-treatment pooled controlled effect on health anxiety, in combination with Egger’s intercept test [54] and the Duval and Tweedie trim and fill procedure using the L0 estimator [55]. We also conducted a few post hoc tests for publication bias on secondary outcomes, see below.

2.6.2. Cost-effectiveness and cost-utility

Whereas cost-effectiveness refers to the relation of costs to effect on health anxiety, cost-utility refers to the relation of costs to effect on health-related quality of life. The overarching aim of including such outcomes in this review was to produce an overview of the relative cost-effectiveness and cost-utility of conditions that had been directly compared for health anxiety in an RCT. Whenever possible, we strived to present results from cost-effectiveness and cost-utility analyses based on the Incremental Cost-Effectiveness Ratio (ICER), which is ‘the difference in cost per difference in effect, for a particular pairwise comparison of [conditions], given a particular time frame, and, in terms of cost, given a particular stakeholder’s perspective‘ [56]. As to efficacy measures, for cost-effectiveness analyses we were willing to accept either continuous scales or dichotomous outcomes such as remission rates, as long as these were valid measures of health anxiety. We employed the same principle as for the rest of the review regarding control conditions in that we included comparisons of CBT to other conditions, but not randomized comparisons of CBT against CBT. As to time frame and stakeholder’s perspective, we included those analyses that were presented as primary. We did not include cost-effectiveness or cost-utility outcomes based on a clinic’s perspective that merely related to intervention costs.

3. Results

3.1. Search hits and study characteristics

We read 47 publications in full text and unanimously identified 17 RCTs for inclusion (κ = 0.91). Two additional trials were discussed due to disagreement. One case of disagreement concerned the adequacy of a particular cluster-randomized design [44], and the other case of disagreement concerned the adequacy of a including a very small trial in the review [57]. We ultimately included both studies, which resulted in 19 RCTs being part of the systematic review (Figure 1; Table 1). Thirteen RCTs were found via both PubMed and PsycINFO, 4 via PubMed only, 1 via OATD [57] and 1 was known by the authors [36]. We could compile health anxiety post-treatment outcomes from 2008 study participants with a mean age of 41.5 years (as based on data from 18/19 of the RCTs) and approximately two-thirds female (1334/2008 as based on proportions at baseline). The pooled estimated baseline CBT health anxiety means were HAI-18: 32.1 (range: 16.6 to 37.4), HAI-14: 26.2 (range: 14.1 to 30.4) and IAS: 64.3 (range: 33.2 to 74.1). Most studies were indicative of mild or moderate mean symptoms of depression. One-thousand and seven were enrolled in CBT and 1001 were allocated to one of the following control conditions: a waiting-list (9 studies), another psychological treatment (behavioral stress management, a course in problem solving, fortnightly psychoeducational material or short-term psychodynamic psychotherapy; 5 studies), treatment as usual (4 studies), antidepressant medication (fluoxetine or paroxetine; 2 studies), pill placebo (2 studies) or an attention control (participation in an online patient forum; 1 study).

Figure 1. Flowchart of the study search and selection process. CBT: cognitive behavior therapy; OATD: Open Access Theses and Dissertations; RCT: randomized controlled trial.

Table 1. Randomized controlled trials of cognitive behavior therapy for health anxiety.

Year (country)	N	Health anxiety eligibility criterion	Main outcome in meta-analysis	CBT format	Control condition	Ref
1996 (UK)	32	DSM-III-R hypochondriasis	VAS ‘time spent worried about health’	Individual face-to-face	WLC	[33]
1998 (UK)	60	DSM-III-R hypochondriasis	VAS ‘time seriously worried about health’	Individual face-to-face	i. BSM ii. WLC	[34]
2001 (Netherlands)	56	DSM-IV hypochondriasis	IAS health anxiety subscale	Individual face-to-face	WLC	[58]
2004 (USA)	182	≥150 on the WI-14 (Likert-5) + SSI	WI-14, Likert-5 version	Individual face-to-face	TAU	[69]
2007 (Netherlands)	43	DSM-IV hypochondriasis	GIAS	Group face-to-face	Problem solving course	[44]
2007 (Netherlands)	109	DSM-IV hypochondriasis	WI-14, Likert-4 version	Individual face-to-face	i. ADM ii. Pill placebo	[64]
2008 (UK)	43	≥20 on the HAI-14	HAI-14	Individual face-to-face	TAU	[70]
2011 (Denmark)	76	ICD-10 hypochondriasis	HAI-14	Mixed face-to-face	i. PDT ii. WLC	[65]
2011 (Sweden)	81	DSM-IV hypochondriasis	HAI-64	Guided SH Internet	Attention control	[63]
2012 (Canada)	14	≥8 on the WI-14 (yes/no)	HAI-18	Guided SH bibliotherapy	WLC	[57]
2013 (Canada)	57	≥8 on the WI-14 (yes/no)	WI-14, yes/no version	Individual face-to-face	WLC	[59]
2014 (UK)	444	DSM-IV hypochondriasis	HAI-14	Individual face-to-face	TAU	[60]
2014 (Sweden)	151	DSM-IV hypochondriasis	HAI-64	Guided SH Internet	Guided SH Internet BSM	[66]
2015 (Germany)	84	DSM-IV hypochondriasis	IAS	Individual face-to-face	WLC	[35]
2016 (Sweden)	127	DSM-5 SSD or IAD	HAI-64	i. Guided SH Internet ii. Pure SH Internet iii. Pure SH bibliotherapy	WLC	[48]
2017 (USA)	142	DSM-IV hypochondriasis	WI-14, Likert-5 version	Individual face-to-face	i. ADM ii. Pill placebo	[37]
2017 (Denmark)	82	≥21,4 on the WI-7 (Likert, 0–100)	WI-7, Likert version	Group face-to-face	WLC	[36]
2018 (Australia)	69	DSM-5 SSD or IAD	HAI-18	Guided SH Internet	Psychoeducation	[67]
2019 (UK)	156	≥18 on the HAI-14	HAI-14	Individual Internet/phone	WLC	[61]

Note that the sample sizes and health anxiety outcomes listed here correspond to the estimates included in the review. This is not necessarily the same as the total sample size and primary outcome of the clinical trial itself. ADM: antidepressant medication; BSM: behavioral stress management; GIAS: Groningen Illness Attitude Scale; HAI: Health Anxiety Inventory; IAD: Illness Anxiety Disorder; IAS: Illness Attitude Scales; PDT: psychodynamic short-term therapy; SH: self-help; SSD: Somatic Symptom Disorder; TAU: treatment as usual; UK: United Kingdom; VAS: visual-analogue scale; WI: Whiteley Index; WLC: waiting-list condition.

3.2. Post-treatment effects and moderators

As to the primary outcome (Figure 2), the pooled between-group effect at post-treatment was large (g = 0.79, 95% CI: 0.57–1.01) with a high level of heterogeneity (I² = 83%; Q = 136, df = 27, p < .0001). In sensitivity analyses, the outcome was similar without the two cluster-randomized trials (g = 0.84, 95% CI: 0.62–1.07; I² = 82%) and without the two trials that used visual-analogue scales to measure health anxiety (g = 0.75, 95% CI: 0.52–0.99, I² = 84%). Secondary outcomes at post-treatment, i.e., between- and within-group effects on health anxiety, depression, general anxiety, physical symptoms, and disability are presented in Table 2. Analyses of categorical putative moderators of the primary outcome are displayed in Table 3. As shown in Table 3, the recruitment path and the type of control condition significantly moderated the effect of CBT for health anxiety. As to continuous moderators of the between-group effect on health anxiety, we saw no significant effect of participant mean age (k = 27, est = 0.00, p = .816), sample percentage of women (k = 28, est = 0.01, p = .316), the number of treatment sessions or modules (k = 28, est = 0.04, p = .186), or baseline level of health anxiety (k = 27, est = 0.02, p = .436). In summary, the primary pooled between-group effect on health anxiety was large and moderated by recruitment path and control condition.

Figure 2. Forest plot of all comparisons (k = 28) of CBT and control conditions. CBT: cognitive behavior therapy; TAU: treatment as usual; WLC: waiting-list control.

Table 2. Pooled post-treatment between-group effects and pre-treatment to post-treatment within-group effects of cognitive behavior therapy for health anxiety.

				Pooled between-group effects
	Pooled within-group effects			vs any comparator			vs other psychological			vs treatment as usual			vs waiting-list
	k	g (95% CI)	I^2	k	g (95% CI)	I^2	k	g (95% CI)	I^2	k	g (95% CI)	I^2	k	g (95% CI)	I^2
Health anxiety	24	1.76 (1.51–2.01)	81%	28	0.79 (0.57–1.01)	83%	5	0.64 (0.06–1.23)	84%	4	0.54 (0.29–0.79)	60%	14	1.08 (0.86–1.30)	44%
Depression	22	0.74 (0.63–0.85)	16%	25	0.41 (0.25–0.58)	66%	5	0.12 (−0.26–0.51)	65%	4	0.28 (0.14–0.41)	0%	13	0.53 (0.30–0.76)	46%
General anxiety	20	0.79 (0.65–0.93)	41%	22	0.44 (0.27–0.62)	68%	5	0.15 (−0.38–0.68)	82%	4	0.31 (0.18–0.45)	0%	10	0.58 (0.35–0.81)	37%
Physical symptoms	9	0.63 (0.45–0.81)	20%	9	0.40 (0.23–0.57)	15%	1	Not applicable		2	Not applicable		6	0.36 (0.11–0.61)	0%
Disability	9	0.45 (0.19–0.71)	78%	9	0.26 (−0.02–0.54)	81%	2	Not applicable		4	0.05 (−0.41–0.50)	88%	3	0.61 (0.32–0.90)	0%

Note that k stands for the number of pairwise group comparisons rather than the number of randomized controlled trials.

Table 3. Subgroup and sensitivity analyses based on putative moderators of the pooled controlled effect of cognitive behavior therapy on health anxiety: all comparisons.

Putative moderator	p	Subgroup	k	g	95% CI	I^2
Pooled total			28	0.79	0.57–1.01	83%
Study country of origin	.487	Europe	21	0.83	0.60–1.05	78%
		Not Europe	7	0.71	0.09–1.32	90%
Participant recruitment path	.042	Advertisements	5	1.20	0.90–1.50	0%
		Mixed	10	0.52	0.14–0.89	87%
		Routine care	11	0.89	0.59–1.18	74%
Main inclusion criterion	.864	Diagnosis	21	0.80	0.54–1.07	84%
		Cutoff	7	0.71	0.33–1.09	74%
Mean level of depression	.273	Moderate	9	0.73	0.40–1.06	79%
		Mild	12	1.05	0.78–1.33	54%
		Below cutoff	3	0.99	0.20–1.79	90%
CBT setting	.792	Research/university	17	0.72	0.41–1.03	85%
		Routine care	6	0.77	0.38–1.16	81%
CBT use of therapist	.111	Individual therapy	17	0.69	0.42–0.96	83%
		Guided self-help	5	1.20	0.61–1.79	83%
CBT delivery medium	.224	Face-to-face	20	0.71	0.45–0.98	83%
		Internet	5	1.07	0.58–1.55	81%
Control condition	.032	Psychological	5	0.64	0.06–1.23	84%
		Treatment as usual	4	0.54	0.29–0.79	60%
		Waiting-list	14	1.08	0.86–1.30	44%
Cochrane: I (RSG)	.382	Low risk of bias	20	0.73	0.47–0.99	86%
		Unclear risk of bias	8	0.95	0.54–1.37	69%
Cochrane: II (AC)	.720	Low risk of bias	16	0.83	0.59–1.08	79%
		Unclear risk of bias	11	0.79	0.32–1.26	85%
Cochrane: III (IOD)	.454	Low risk of bias	14	0.93	0.68–1.19	72%
		High risk of bias	11	0.80	0.38–1.22	88%
Cochrane: IV (SRO)	.493	Low risk of bias	13	0.93	0.66–1.19	80%
		Unclear risk of bias	13	0.80	0.47–1.12	73%

Note that, due to missing data and substrata where k < 3, the number of total comparisons (k) differed over the sensitivity analyses. Note also that in some cases, more than one comparison was used from one and the same clinical trial. AC: allocation concealment; CBT: cognitive behavior therapy; IOD: incomplete outcome data; RSG: random sequence generation; SRO: selective reporting of outcomes.

3.3. Risk of bias

Focusing on the health anxiety outcomes used for the present review, we assessed all RCTs based on four criteria of the Cochrane collaboration’s tool for assessing risk of bias [39]. Overall, the risk of bias appeared to be relatively low (Figure 3). The primary threat in this regard was incomplete outcome data, i.e., high levels missing data at the post-treatment assessment and to some degree the use of inadequate modeling techniques. As illustrated in Table 3, there were no statistically significant effects of risk of bias ratings on the primary pooled between-group effect size on health anxiety.

Figure 3. Cochrane risk of bias ratings for included randomized controlled trials (N = 19). Please note that these ratings primarily concern the health anxiety outcomes that were included in the primary meta-analysis rather than each included trial as a whole.

3.4. Publication bias

We assessed the possible influence of publication bias on the primary meta-analysis of between-group effects at the post-treatment assessment. Based on Egger’s test, there was evidence of funnel plot asymmetry (z = 3.78, p = .0002). This was verified by the Duval and Tweedie procedure, and after imputation of six additional studies on the left side of the graph (see the online supplement), the pooled post-treatment effect size decreased but was still in the upper moderate range (from g = 0.79 to g = 0.62, 95% CI: 0.39–0.85). In summary, there were signs of publication bias which probably inflated the primary estimate by approximately 0.17 units.

After concluding that there was a high probability of publication bias, we also performed post hoc analyses to assess the possible impact of publication bias on between-group estimates against treatment as usual and waitlist controls specifically (see Table 2). Bias appeared to be driven largely by the asymmetry of the waiting-list controlled trials. Based on the Duval and Tweedie procedure, four studies were imputed on the left side of the waiting-list plot (see the online supplement) but the waiting-list controlled effect on health anxiety remained large (from g = 1.08 to g = 0.96, 95% CI: 0.75–1.17). There was no need to impute studies for the meta-analysis of treatment as usual trials.

3.5. Responder and remission rates

We were able to pool post-treatment rates of response (change beyond a clinically relevant criterion) from five studies, remission (no diagnosis or a score below cutoff) from four studies and clinically significant improvement [38] from four studies. The criteria for response corresponded to a reduction in health anxiety of 19–29%. Three studies of remission were based on cutoff scores on a psychometric instrument [60–62], and one was based on a clinical diagnosis of DSM-IV hypochondriasis [63]. The pooled responder rate was 66% (k = 7, 95% CI: 55–78), the pooled remission rate was 48% (k = 5, 95% CI: 28–67) and the pooled rate of clinically significant improvement was 51% (k = 7, 95% CI: 45–57). Thus, approximately two-thirds of patients in CBT were responders and about half of the patients achieved remission.

Regarding between-group effects, for a response, the OR was estimated at 3.80 (k = 6, 95% CI: 1.13–12.81) and the NNT was 3 (p = .026). The corresponding effect was slightly smaller and not significant against active control groups only (k = 3, OR = 2.75, 95% CI: 0.05–155.14; NNT = 4, p = .352). In the case of remission rates, we estimated the OR at 5.17 (k = 3, 95% CI: 0.89–30.20) and the NNT at 3 (p = .077). Last, for clinically significant improvement, the OR was estimated at 7.74 (k = 6, 95% CI: 2.00–30.02) and the NNT at 3 (p < .0001). In summary, it appears to be necessary to assign approximately three patients to CBT rather than the control conditions in order to achieve one additional case of response, remission or clinically significant improvement.

3.6. Long-term follow-up

Fourteen studies reported outcomes from follow-up assessments approximately 6 months after treatment termination (M = 6.0, SD = 0.6). There was a significant increase in health anxiety from post-treatment to follow-up, but the pooled within-group effect size was small (k = 18, g = 0.11, 95% CI: 0.00–0.21; I² = 8%). The follow-up between-group effect size was also small but statistically significant (k = 10, g = 0.33, 95% CI: 0.14–0.52; I² = 56%). The pooled between-group effect vs. treatment as usual was significant and in the small to medium range (k = 4, g = 0.43, 95% CI: 0.29–0.57; I² = 0%) but the effect vs. other psychological treatments was not significant (k = 4, g = 0.31, 95% CI: −0.05–0.67; I² = 51%). Six studies reported outcomes from follow-up assessments approximately 12 to 18 months after treatment termination (M = 14.0, SD = 3.1). The pooled within-group effect from post-treatment was small and non-significant (k = 8, g = −0.03, 95% CI: −0.28–0.21; I² = 59%), and so was the pooled between-group effect (k = 4, g = 0.19, 95% CI: −0.04–0.43; I² = 29%). Thus, although there was no significant pooled between-group effect versus other active treatments, the post-treatment health anxiety levels were maintained or only marginally increased to follow-up.

3.7. Efficacy of individual face-to-face CBT

We conducted a subgroup analysis of individual face-to-face CBT as this is arguably the gold standard CBT format for health anxiety. Individual face-to-face CBT was used in 11/19 RCTs, where the mean length of treatment was 10.6 sessions (SD = 4.3). The within-group effect on health anxiety was large (k = 14, g = 1.89, 95% CI: 1.51–2.27; I² = 85%), and so were the within-group effects on depression (k = 13, g = 0.78, 95% CI: 0.60–0.97, I² = 41%) and general anxiety (k = 11, g = 0.79, 95% CI: 0.54–1.04; I² = 64%). At post-treatment, the between-group effect on health anxiety was large (k = 18, g = 0.79, 95% CI: 0.51–1.06; I² = 82%), and the between-group effects on depression (k = 16, g = 0.47, 95% CI: 0.26–0.68; I² = 64%) and general anxiety were small to moderate (k = 14, g = 0.47, 95% CI: 0.27–0.66; I² = 54%). At the primary endpoint, individual face-to-face CBT was found to be superior to pill placebo in one trial [64] and to another psychological treatment (behavioral stress management and short-term psychodynamic psychotherapy) in two trials [34,65]. There was a small and non-significant pooled within-group effect on health anxiety from post-treatment to follow-up 6 months after treatment (k = 10, g = 0.09, 95% CI: −0.06–0.25; I² = 19%) and also to follow-up 12 to 18 months after treatment (k = 4, g = −0.24, 95% CI: −0.63–0.16; I² = 70%).

3.8. Efficacy of therapist-guided Internet-delivered CBT

We also conducted a subgroup analysis of therapist-guided Internet-delivered CBT which, apart from individual face-to-face CBT, was the only CBT format to be used in more than two RCTs. In therapist-guided Internet-delivered CBT, the patient works with the same types of behavior changes and strategies as in conventional CBT (e.g., exposure and response prevention) but the treatment content is administered in text form via an online platform rather than physical one-to-one sessions with the therapist. Therapist-guided Internet-delivered CBT was used in 4/19 RCTs, where the mean number of treatment modules was 10.5 (SD = 3). The within-group effect on health anxiety was large (k = 4, g = 2.01, 95% CI: 1.75–2.27; I² = 6%) and so were the within-group effects on depression (k = 4, g = 0.81; 95% CI: 0.55–1.07; I² = 30%) and general anxiety (k = 4, g = 0.85, 95% CI: 0.64–1.06; I² = 0%). At post-treatment, the between-group effect on health anxiety was large (k = 4, g = 1.09, 95% CI: 0.48–1.70; I² = 86%), the between-group effect on depression was moderate but not statistically significant (k = 4, g = 0.47, 95% CI: −0.03–0.96; I² = 81%) and the effect on general anxiety was moderate (k = 4, g = 0.62, 95% CI: 0.19–1.04; I² = 73%). At the primary endpoint, two trials [66,67] found therapist-guided Internet-delivered CBT to be superior to another psychological treatment (fortnightly provision of psychoeducational material and therapist-guided Internet-delivered behavioral stress management). There was a small and non-significant pooled within-group effect on health anxiety from post-treatment to follow-up 6 months after treatment (k = 3, g = 0.16, 95% CI: −0.08–0.39; I² = 0%). Two trials reported outcomes at 12 months follow-up [56,68], and the mean level of health anxiety was sustained in both cases.

3.9. Protocol integrity of face-to-face and guided Internet-delivered CBT

Over and above the question of efficacy, we also compared face-to-face CBT and therapist-guided Internet-delivered CBT with regard to protocol integrity in terms of missing data rates and adherence to the treatment. At post-treatment, the missing data rate in therapist-guided Internet-delivered CBT did not differ from that in face-to-face CBT (8%, 95% CI: −3–18, I² = 93%, k = 4 vs. 15%, 95% CI: 8–22, I² = 87%, k = 12; p = .291). The pooled percentage of completed modules in therapist-guided Internet-delivered CBT was 74% (95% CI: 61–87, k = 4). As to face-to-face CBT, only 1 RCT defined an a priori number of sessions that patients were encouraged to complete. In this study, patients completed approximately (based on the strata of page 1466: 4.5/6 =) 75% of the sessions [69].

3.10. Cost-effectiveness and cost-utility

Due to the scarcity of studies and different methods used we did not pool cost-effectiveness estimates using meta-analysis. Six RCTs reported health economic outcomes in terms of cost-effectiveness and cost-utility [56,60,61,68,70,71], see Table 4. Three of these were based in Sweden and concerned therapist-guided Internet-delivered CBT and other minimal-contact CBT formats. The remaining three were based in the UK, of which two concerned face-to-face CBT and one concerned remote CBT delivered via online conference software or telephone. The overarching conclusion of the three Swedish studies was that therapist-guided ICBT is likely to be a cost-effective treatment both when compared to passive controls and when compared to Internet-delivered behavioral stress management, at least in the short term. Treatment effects on health-related quality of life are however limited, and clear evidence of cost-utility was only seen in the case of unguided ICBT when compared to no treatment. Unguided ICBT was associated with a net societal saving, and as so was therapist-guided ICBT in one out of three studies. As to the two UK studies of face-to-face CBT, a strength of these studies was the longer time frame (12 to 24 months). Face-to-face CBT was found to be cost-effective in comparison to treatment as usual but, as in the case of ICBT, to have little effect on health-related quality of life. Notably though, cost-utility was seen in the RCT that compared remote CBT to treatment as usual. The effect on quality of life was still relatively small, but remote CBT was associated with a net societal saving.

Table 4. Cost-effectiveness and cost-utility of cognitive behavior therapy for health anxiety.

Country (Year)	Perspective	N in analysis, conditions	Time frame	Efficacy outcome	ICER, probability of CBT C-E or C-U at WTP = 0	Ref
Sweden (2013)	Societal C-E + C-U	N = 81, i. Guided ICBT ii. Attention control	3 m	C-E: DSM-IV hypochondriasis C-U: Utility based on the EQ-5D	Cost-effectiveness: -784/0.63 = -£1244, 64% Cost-utility: -784/0.12 = -£6533, 67%	[68]
Sweden (2016)	Societal C-E + C-U	N = 158, i. Guided ICBT ii. Internet-delivered behavioral stress management	3 m	C-E: Clinically significant improvement C-U: Utility based on the EQ-5D	Cost-effectiveness: 310/0.14 = $2214, approximately 10% Cost-utility: 310/0.031 = $10,000, approximately 10%	[71]
Sweden (2018)	Societal C-E + C-U	N = 132, i. Guided ICBT ii. Unguided ICBT iii. BIB-CBT iiii. WLC	3 m	C-E: Clinically significant improvement C-U: QALYs based on the EQ-5D	Cost-effectiveness: Guided ICBT: 208/0.50 = £416, approximately 35% Unguided ICBT: -57/0.42 = £-134, approximately 50% BIB-CBT: 128/0.47 = £273, approximately 40% Cost-utility: Guided ICBT: 208/0.0020 = £103,048, approximately 35% Unguided ICBT: -57/0.0034 = £-16,951, approximately 50% BIB-CBT: 128/0.0053 = £24,245, approximately 40%	[56]
UK (2008)	Health-care C-E	N = 59, i. CBT ii. TAU	12 m	Score on the 18-item Health Anxiety Inventory	277/8.4 = £33, not reported	[70]
UK (2014)	Health-care C-E + C-U	N = 342, i. CBT ii. TAU	24 m	C-E: Score on the 18-item Health Anxiety Inventory C-U: QALYs based on the EQ-5D	Cost-effectiveness: 156/2.79 = £56, approximately 45% Cost-utility: 156/0.011 = £14,182, approximately 45%	[60]
UK (2019)	Extended health care C-U	N = 156, i. Remote CBT ii. TAU	12 m	QALYs based on the EQ-5D	−1064/0.07 = £-15,200, high but not reported	[61]

Note that the probability of cost-effectiveness (or cost-utility) at the willingness to pay level of zero is a sensitivity analysis of (cognitive behavior therapy vs. comparator) net costs only. Only studies that assessed indirect costs relating to productivity losses are here reported as using a societal perspective. BIB-CBT: cognitive behavioral bibliotherapy; CBT: cognitive behavior therapy; C-E: cost-effectiveness; C-U: cost-utility; EQ-5D: EuroQol-5; ICBT: Internet-delivered cognitive behavior therapy; ICER: incremental cost-effectiveness ratio; QALYs: quality-adjusted life years; TAU: treatment as usual; WLC: waiting-list; WTP: willingness to pay.

4. Discussion

4.1. Main findings

This study investigated the clinical efficacy and cost-effectiveness of CBT for health anxiety. We conducted a systematic review and meta-analysis of the available literature and included 19 randomized controlled trials with post-treatment outcome data from a total of 2008 participants. In relation to control conditions, CBT leads to large reductions of health anxiety and small to moderate effects on depression, general anxiety, and physical symptoms. Though pooled remission rates were associated with notable uncertainty, we estimate that two-thirds of participants in CBT respond to treatment and about half are in remission post-treatment. We also found that the effect of CBT on health anxiety is sustained at longer-term follow-up. Six studies, of which four tested remotely delivered CBT, reported cost-effectiveness data suggesting that the treatment is likely to be cost-effective. This comprehensive review of the available research adds to the body of knowledge showing that CBT is highly effective in the treatment of health anxiety.

As shown by the present study this research field has moved forward at a rapid pace. The 19 included randomized trials have been published in just about two decades, making CBT the clearly most well-researched treatment for health anxiety. In line with previous meta-analyses [23,24] the pooled primary outcome effect size was large, with some factors moderating this effect. CBT was superior to waitlist controls as well as treatment as usual and other psychological treatments, which indicates that there are effects of CBT also above and beyond those of receiving attention and support from a clinician. Nonetheless, the pooled effect in studies using waiting-list controls was nearly twice as large as that based on studies with active controls. Another moderator of the pooled between-group effect was the recruitment path, i.e., whether patients were recruited via advertisements, routine care or both. The largest pooled between-group effect was seen when patients had been recruited via advertisements. However, rather surprisingly, the gradient in effect size was such that studies that accepted routine care referrals (only) reported a larger pooled between-group effect than studies that employed a mixed recruitment strategy. We therefore suspect that the effect of recruitment path was explained by the other significant moderator, i.e., the use of different control groups. This is since all group comparisons based on sampling via advertisements (k = 5) used waiting-list controls, and approximately 82% of comparisons based on routine care sampling (k = 9/11) used treatment as usual (k = 3) or waiting-list controls (k = 6). In contrast, only 40% of comparisons based on mixed sampling (k = 4/10) were based on treatment as usual (k = 1) or waiting-list controls (k = 3). Last, given the limited statistical power of our tests, we wish to briefly mention baseline depression as a possible moderator that is worth studying further. The formal test was not significant for this variable, but the gradient in effect size was notable and in the expected range.

To our knowledge, this study is the first to estimate pooled response and remission rates in CBT for health anxiety. We believe that such dichotomous measures of treatment effects are important for both clinicians and patients. Most patients are probably little helped by the information that the pooled between-group Hedges' g effect size of CBT for health anxiety is 0.79 (or 0.62 after adjustment for publication bias). However, knowing that approximately two-thirds of patients respond to CBT (which the present study showed) can give a clearer picture of what to expect from the treatment. Although the pooled responder and remission estimates indicate that CBT is an effective treatment, they also show that there is room for improvement. It is also important for researchers to collect more data on response and remission rates in general. This is particularly true of remission rates where the 95% confidence interval for CBT spanned all the way from 28% to 67%.

As in a meta-analysis from 2014 [23], this study showed that the controlled effect of CBT was smaller at long-term follow-up than immediately post-treatment. Due to the fairly stable within-group effects in CBT we interpret this result not as an indicator of CBT having only short-term effects, but as a ‘catching up’ effect in the control conditions at long-term follow-up. It is yet unclear to what degree this ‘catching up’ effect is to be understood as inherent to the natural course of health anxiety, and to what degree there are similarities between individuals who improve after treatment termination and those who improve spontaneously (without treatment).

Regarding the health economic effects of CBT for health anxiety, the results of the present study suggest that CBT probably is cost-effective compared to both passive and active controls, whereas the treatment’s cost-utility is less certain. In other words, CBT leads to large reductions of health anxiety compared to no treatment or other active treatments, while incurring relatively low or no net costs. The treatment’s effect on health-related quality of life, however, seems to be smaller, probably owing to the fact that widely used measures of quality of life and disability that weigh in physical functioning are not affected as much in anxiety disorders [72]. Our overall interpretation is that, from a health economic perspective, it is probably reasonable for both society as a whole and the health-care system to offer CBT for health anxiety instead of no treatment or treatment-as-usual. It should, however, be pointed out that only six studies reported health economic outcomes, of which the three that concerned therapist-guided Internet-delivered CBT were conducted by the same Swedish research group, and both studies that concerned face-to-face CBT were conducted by the same British research group. Thus, more health economic research is needed, and preferably by more research groups, before firm conclusions can be drawn in this regard.

4.2. Strengths and limitations

Central strengths of the present study were: (a) that we searched the largest databases for published peer-reviewed articles (PubMed and PsycINFO) as well as a database covering theses, (b) that inclusion of studies were conducted based on independent assessments by two raters with high interrater agreement, (c) the comprehensive sensitivity and moderator analyses, and (d) that risk of bias was systematically assessed. As to limitations, Egger’s test of publication bias indicated an asymmetric funnel plot, which suggests that there ought to have been a few more studies with lower than average effect sizes. When modeling imputation of such studies the pooled standardized effect size on the primary outcome dropped by 0.17 units, indicating that potential publication bias probably had a small impact on our reported estimates. We also found substantial heterogeneity in several of our analyses. As judged from our moderator analyses, this is probably to some extent accounted for by the use of different control groups. As shown in Table 3, the heterogeneity was small to moderate (44%) in pooled comparisons to wait-list controls, but large when comparing CBT to other psychological treatments (84%). Our assessment of the risk of bias indicated that a fairly large proportion of the included studies had non-negligible missing health anxiety data at post-treatment. The moderator analyses, however, showed that risk of bias ratings were unrelated to effects on health anxiety, which suggests that missing data did not inflate the effect size estimates of this meta-analysis. Lastly, the identification of publications pertaining to cost-effectiveness was conducted by the first author only, which means that there was more room for error in this process than the identification of primary clinical outcomes.

4.3. Expert opinion

As to the significance of treatment formats, one of the most intriguing results was that there was no significant effect difference between studies testing face-to-face treatment and those testing Internet-delivered treatment. A challenge in the implementation of CBT for health anxiety in routine care settings is the fact that many patients with health anxiety are found in medical clinics where clinicians have little or no training in administering psychological interventions. In this context, therapist-guided Internet-delivered CBT has two key advantages. First, less therapist time is needed per patient, which makes it possible for relatively few clinicians to help a relatively large number of patients. Second, as treatment is delivered remotely, clinics can be centralized. This means that there is no need for each and every medical clinic to invest in information technology infrastructure such as web servers and technical support. At the very least, costs can be shared. Also, not only is it possible for patients from clinics in vastly different parts of a country (or the world) to be referred to the same online treatment platform, but it is also possible to have therapists working in different parts of a country (or the world) administering the treatment. Thus, as long as patients have access to the Internet, even small medical clinics with little resources for conducting psychological treatment situated in sparsely populated areas have the potential to refer their patients to Internet-delivered CBT for health anxiety. That being said, before concluding that Internet-delivered CBT for health anxiety is non-inferior to face-to-face CBT, randomized trials directly comparing these two delivery formats need to be conducted. Nonetheless, as Internet-delivered CBT has been found to be effective in four randomized trials carried out by two independent research groups, we view this therapy to be a viable treatment option, at least for the majority of patients with health anxiety.

In the implementation of remote treatment for health anxiety, there is also the question of assessment. In order to establish that the patient suffers from health anxiety, one has to verify that the patient’s fear or preoccupation with severe illness is indeed persistent, excessive, and clinically significant. In doing so, it is necessary to consider other psychiatric diagnoses such as panic disorder, generalized anxiety disorder, and obsessive-compulsive disorder. Moreover, it is also common practice to assess potential barriers for treatment such as other medical or mental health conditions (e.g., severe depression), functional impairment (e.g., severe dyslexia), psychosocial factors (e.g., homelessness), and technical boundaries (e.g., lack of Internet access) in order to help the patient accordingly. We are of the opinion that, not least for reasons of safety, pre-treatment assessment should preferably be implemented as part of any routine care program for health anxiety. In most cases, it is perfectly feasible to assess patients remotely, for example, via telephone, as was done in the studies of therapist-guided Internet-delivered CBT under review.

More high-quality studies testing prognostic factors, treatment components, and putative mechanisms will probably play an important role for making CBT even more effective. There is a need for researchers to move beyond studies where baseline variables such as health anxiety and depression are identified as predictive of within-group change, to experimental designs where such predictors are manipulated so as to hopefully achieve added therapeutic effects. That is, what would, for example, be the significance of adding behavioral activation as an intervention for patients with depression, or motivational interviewing for patients with low levels of motivation, before standard CBT for health anxiety is commenced? As there was substantial variance in how responder and remission were operationalized in the studies in this review, an important task for researchers in this field is also to agree on how to define these constructs to begin with (see for example [73] for how this has been done in obsessive-compulsive disorder research). Further research is needed to help those individuals who do not respond to standard treatment. Importantly, given that very little is known about the non-responder group and possible mechanisms behind their apparent failure to benefit from treatment, non-responders are not to be regarded as resistant to treatment. Therefore, study designs evaluating new practices to help non-responders further should preferably estimate the added value of a new or added intervention in relation to standard CBT. One way of doing this could be to conduct clinical trials where non-responders, or patients at risk of being non-responders, are randomized to either continue with (or resume) the standard treatment, or be crossed over to an alternative (or enhanced) intervention.

Briefly, we also wish to mention that there are several psychological treatments for health anxiety that have been empirically evaluated but which were not included, or only briefly mentioned, in the present study. Examples of such treatments include (in alphabetic order): acceptance and commitment therapy, attention training therapy, behavioral stress management, explanatory therapy, metacognitive therapy, mindfulness-based cognitive therapy, and problem-solving therapy. However, the evidence base for the efficacy of each of these treatments is considerably smaller than that for traditional CBT, and less is known, for example, about effect sizes and long-term efficacy. In developing psychological treatments for health anxiety further, there is a need for more direct comparisons between active treatments, and also components from promising treatments, where traditional CBT may be viewed as the benchmark intervention.

4.4. Conclusions

The results of this systematic review and meta-analysis indicate that CBT is a well-researched and highly effective treatment for health anxiety. The treatment yields improvements that are sustained in the long-term and it is probably cost-effective in comparison to both active and passive controls. CBT is effective also when delivered via the Internet, which is a mode of delivery with strong potential to improve access to effective psychological treatment for health anxiety.

Article highlights

• Cognitive behavior therapy (CBT) is an efficacious treatment for health anxiety.

• Evidence pertaining to responder and remission rates, effects in routine care, new CBT delivery formats, and health economic outcomes has not been systematically reviewed.

• About two-thirds of patients with health anxiety respond to CBT, and every other patient achieves remission.

• Between-group effects are smaller when CBT is compared with other active conditions such as other psychological treatments than when it is compared to waiting-list conditions.

• Therapist-guided Internet-delivered CBT appears to be an efficacious treatment format, and has been found to be superior to two active controls by independent research groups.

• CBT for health anxiety is probably a cost-effective treatment, but its effects on generic measures of health-related quality of life are small.

Declaration of interest

The authors have written books about CBT for health anxiety for which they have received royalties. The authors have no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed.

Reviewer disclosures

A reviewer on this manuscript has disclosed that they were the senior author in one of the larger trials. Peer reviewers on this manuscript have no other relevant financial relationships or otherwise to disclose.

Supplemental Material

Supplemental data for this article can be accessed here.

Additional information

Funding

This work was funded by Karolinska Institutet and Region Stockholm which are public institutions that did not take part in the design, conduct or publication of the study.