A systematic review and meta-analysis of treatments for acrophobia

Acrophobia (irrational fear of heights) is a chronic disorder that may have a serious impact on people’s lives, inhibiting their ability to perform everyday tasks such as climbing a flight of stairs, standing near a balcony, or parking a car in a high-rise building, as well as interfering with recreational activities. Phobias are common in the community. Studies in developed countries that are similar to Australia have reported the prevalence of acrophobia,1-4 In the epidemiologic catchment area study, which comprised 20 000 participants across five sites in the United States,3 4.7% of participants fulfilled the criteria for a diagnosis of acrophobia. Comparable studies in Germany and Sweden have shown similar results.1,4 Additionally, 13% of patients presenting to their general practitioners in New Zealand have phobias, and 2.4% of these are situational phobias (eg, water, heights, flying).2 Our aim here was to assess the efficacy of the treatments for acrophobia by conducting a systematic review. This is the first systematic review on treatment for acrophobia. We have written it according to the PRISMA checklist.5

Methods

Protocol and registration

Details of the protocol, including the search strategy for this systematic review were registered on the international prospective register of systematic reviews, PROSPERO.6 We wished to examine all parallel-arm and crossover randomised controlled trials (RCTs), and controlled trials that met our participants, intervention, control and outcomes (the PICO model) eligibility criteria (anyone with acrophobia or a fear of heights, any psychological or medication intervention, control or waitlist, and outcome on questionnaire or behavioural avoidance test [BAT]). Eligible studies involved people with acrophobia or those with a fear of heights of any age. We wished to assess the effectiveness of any intervention against any comparison. We looked at studies from 1946 onwards, published, unpublished and in any language.

Information sources — the search

The search was conducted by the Cochrane Common Mental Disorders Group on 1 December 2015. It covered the Cochrane Common Mental Disorders Group — Specialised Register (CCMD-CTR; all years to date), Cochrane Central Register of Controlled Trials (CENTRAL; all years to date), PsycINFO (all years to date), MEDLINE (1950 to the present) and Embase (1980 to the present). Reports of trials were also sourced from international trials registers courtesy of the World Health Organization’s International Clinical Trials Registry Platform (ICTRP) and the United States National Institutes of Health registry and results database, ClinicalTrials.gov. The search terms were for RCTs involving participants with acrophobia or a fear of heights.

We attempted to contact the authors of all included studies and authors of ongoing studies in search of additional studies to include in the review, and we scanned the reference lists of included studies for any additional and relevant studies. We searched for grey literature studies that we identified and inter-library loan requests were placed where the abstracts were unavailable online. Two of us (B A and H W) selected the studies from abstracts. Studies were excluded if the they had fewer than 20 participants with acrophobia to avoid pilot studies and underpowered treatment studies. Full articles were obtained where there was disagreement or uncertainty.

The data were extracted in duplicate by H W and all the other authors. This included the PICOs, funding sources, dropouts, sources of bias and presence of a BAT. Data from crossover design trials were only extracted before the first crossover. We used the Cochrane Collaboration tool to assess the risk of bias (Table 8.5.a in the Cochrane handbook for systematic reviews of interventions7), which covers sequence generation; allocation concealment; blinding; incomplete outcome data (eg, dropouts and withdrawals); and selective outcome reporting and other categories of bias (we added sample size calculation).7 The extracted data were synthesised using RevMan 5.3.5 (the review management software used for preparing and maintaining Cochrane Reviews). If continuous data were reported without corresponding estimates of variance (standard deviations), we assigned these data a standard deviation by using the largest standard deviation available from intervention and control groups, respectively, from other studies (using the same measurement instruments) to be included in the same analysis, for the most conservative approach using RevMan 5.3.5 software.

Studies were added to meta-analyses (pooling) if any two or more studies using the same intervention reported the same outcomes. Pooled analyses were performed by computing standardised mean differences (SMDs) for continuous outcomes using a random-effects model to take the most conservative approach. We chose the SMD because it is the standard output in RevMan 5.3.5. We reported numbers needed to treat and relative risks. Measures of heterogeneity, including I² (which gives the percentage of variance in a meta-analysis that is attributable to study heterogeneity) were calculated (an I² > 50% suggests significant heterogeneity). No sub-group analyses were planned or performed.

Results

We identified 16 studies for the years 1973 to 2016 in 15 articles for inclusion in the review (the 1985 article by Marshall included two studies8). Appendix 1 (online at mja.com.au) summarises the search results.9-22 The most common reason why studies were excluded after reviewing full-text articles was because they had fewer than 20 participants with acrophobia. The specific reasons for study exclusion are available on request. At the time of the review, 15 studies were published10-22 and one was a conference abstract.9 Appendix 2 (online at mja.com.au) summarises the characteristics of included trials.

There were five broad categories of intervention: desensitisation (DS); in vivo exposure (IVE); virtual reality exposure (VRE); neurolinguistic programming (NLP);23 and VRE with medication. The medications studied were not primary treatments for phobias, but were intended to augment the cognitive processes underlying exposure therapy. There were variations within each of the broad groups with co-interventions, such as prolonged exposure therapy with coping self-statements8 and other interventions such as negative practice (compared against DS).18 The main interventions used in included studies are summarised in Box 1.

While NLP has characteristics of imaginal exposure, we chose to consider it separately as it is not described in the literature as imaginal exposure and it has very brief intervention durations. There were three studies of VRE with the addition of medication at different times during therapy. Comparison groups for VRE with medication studies varied widely from waitlist controls to participants receiving VRE and other treatments.

The methods used to measure outcomes in the studies also varied. All used questionnaires, and most also used a BAT. Certain standard questionnaires were commonly used. One example is the Acrophobia Questionnaire (AQ), which describes 20 situations and assesses levels of avoidance (scores, 0–3) and anxiety (scores, 0–6). This scale is a widely used measure of acrophobia and has been reported to have adequate retest reliability and validity.21,24 Another commonly used instrument was the Attitudes Towards Heights Questionnaire (ATHQ), which includes six heights situations and assesses attitudes towards these situations on a scale of 0–10. The internal consistency and validity of this questionnaire are reported to be “acceptable”.21,25

BAT involved exposing clients to a real-life height situation and gauging their anxiety, often using the subjective units of discomfort scale (SUDS), which provides a measure of the clients’ current anxiety or discomfort.26 Each study that used a BAT chose one that was suitable to its own locality; there appears to be no standard BAT. One study used the Heights Interpretation Questionnaire, which is validated against actual heights.

Most interventions (including DS therapy, VRE therapy, negative practice and IVE therapy) tested against a waitlist control group (ie, no intervention) were shown to be superior.10,14,18,20,27 However, when the control group was one receiving another intervention, the trend was towards equivalence. NLP was trialled in a single study, which showed it to be superior to a meditation-based control.9 Recent studies have also examined the use of medication therapies in conjunction with VRE therapy for acrophobia.12,19,21 One such study has shown a benefit for the use of hydrocortisone in conjunction with VRE, and another showed a benefit over placebo of administering d-cycloserine before exposure to VRE.12,19 No benefit has been shown for giving d-cycloserine after VRE.21

The results of our meta-analytic pooling are shown in Appendix 3 (online at mja.com.au). IVE and DS were effective in the short term when pooled. We use short term to mean outcomes measured in the immediate post-treatment period. IVE was not effective in the longer term follow-up analysis (follow-up for IVE studies ranged from 1 to 9 months, with a median follow-up of 1 month). VRE therapy was effective when measured with the ATHQ scale, but not when measured with the AQ. DS and VRE had no reported follow-up data. Studies comparing their intervention with another similar intervention rather than using a control group was a major limiting factor. Most comparisons of interventions showed equivalence of effect. Also, two studies failed to report the numbers of participants in their intervention and comparison groups, reporting only the total number of participants. In these cases we estimated the numbers of participants in groups by dividing the total number of participants by the number of study groups to perform the meta-analysis.8,18

Nine studies included an actual BAT,11,12,14,16-18,22,28 and two studies included a virtual BAT (where the height exposure was in a virtual environment, rather than a physical one).12,21 Many of these studies also required clients to be unable to complete the BAT successfully (ie, reach the highest fear stimulus without a high degree of anxiety) to participate in the study. The findings for the BAT were the same as those found by the questionnaires for all but two studies. These were: Krijn (2004),14 where the BAT showed superiority of VRE over waitlist controls while the AQ showed equivalence; and de Quervain (2011),12 where VRE plus hydrocortisone therapy showed equivalence with VRE plus placebo on the BAT, but showed superiority on the AQ.12,14

Only three studies reported dichotomous outcomes.9,22,27 The number needed to treat (NNT) for improvement with guided mastery DS versus DS alone was 2.6 (95% CI, 1–15) participants.22 For IVE therapy versus waitlist control, the NNT to show improvement was 2.8 (95% CI, 1.5–16.5). When oppositional actions (Box 1) were added to the guided mastery, the NNT to show improvement was 1.427 (95% CI, 1.0–2.2), and for the rapid phobia NLP, the NNT to show improvement was 6.0 (95% CI, 2.8–35.5).9 The forest plots of our comparisons are shown in Box 2, Box 3 and Box 4 and in Appendix 4 (online at mja.com.au), and the findings they illustrate are summarised in Appendix 3 (online at mja.com.au).

The assessment of risk of bias is shown in Appendix 5 (online at mja.com.au). Most articles did not report enough information in their methods to enable us to make an adequate assessment of bias. Most did not describe the randomisation process or the attempts to conceal randomisation, the methods for blinding outcome, whether or not an outcome was primary (on which the sample size calculation was conducted) or secondary, and how resentful demoralisation was managed (those in the control group being upset at not being in the intervention group).29 Only three studies reported their sample size calculation.9,21,27

Discussion

The results of our pooled analysis suggest that DS, IVE, and VRE exposure are probably effective treatments, at least in the short term. Single studies showed that NLP, negative practice, and variations on IVE therapy trialled with oppositional actions, coping self-statements, and exposure with guided mastery were superior to comparison interventions. The case for the effectiveness of these interventions would be stronger if there were more studies on each intervention.

Overall, the literature shows a trend towards faster treatments with fewer visits. For example, a 1973 study used 12 to 14 visits10 while a 2013 study required two 60-minute visits21 and a 2015 study required a single 15-minute visit.9 Appendix 2 (online at mja.com.au) shows that the median duration of therapy from 1980 to 1999 was 218 minutes and for 2000 to 2016 was 60 minutes. Another trend in therapy for acrophobia is the movement away from DS and IVE towards the use of virtual reality equipment. Such technology makes exposure therapy more accessible and easier for therapists to deliver. Finally, the use of medications to enhance the process of fear extinction is becoming more common, and has shown promising results when coupled with VRE.12,19,21 d-Cycloserine is a partial agonist of the N-methyl-d-aspartate glutamatergic receptor, and is thought to enhance the learning process underlying the extinction of fear, and corticosteroid hydrocortisone is also thought to speed the fear extinction process.12,19,21

Strengths and limitations

There are several strengths to our review, which we believe is the first to compare all treatments for acrophobia. We used broad inclusion criteria, and the studies included are representative of the existing body of literature. We also contacted all authors of included and ongoing studies in search of unpublished articles.

A limitation of our review is that only three studies had a power calculation.9,21,27 Power to detect differences was frequently undermined by dividing study participants into multiple small intervention groups (eg, 50 participants divided into five groups) and then reporting equivalence.11 A larger study may have shown differences in larger subgroups. There has been longstanding concern about this undermining of the power to detect differences in the psychological literature.30 Studies also reported numerous outcomes, and it was not clear which were primary and which were secondary. This, coupled with the lack of power calculation, made it difficult to determine the validity of the findings of each study.

Inadequate sample size was a common problem, as can be seen in the comparison in Box 2, B.9 The 1995 study by Menzies and Clarke had 49 participants, and a non-significant effect size of −0.28. To achieve a significant effect size of 0.42, with the same means and standard deviations, would have required 96 participants. In addition to the sources of potential biases (elaborated in Appendix 5, online at mja.com.au), intention-to-treat analyses were rarely reported, and only one study9 was designed according to a guideline for RCTs (the CONSORT statement).5 In addition, very few studies reported dichotomous outcomes, which are best understood by therapists.9,22,27 The proportion of participants who show improvement in their acrophobia is an important figure for clinicians to be able to judge effectiveness and to inform clients about potential benefits of therapy. Older studies also tended to report their results in the form of F tests, without publishing empirical data in any form (table, graph, or text), which limited our ability to compare studies and impairs the ability of readers to fully comprehend the results.

While there is a large array of studies and interventions in this area, our review provides a true overview of the literature. There are many sources of heterogeneity, including the use of different questionnaires for measuring outcomes, the presence or absence of a BAT, unclear sources of bias and small sample sizes. Our decision to exclude studies with fewer than 20 participants resulted in four papers being excluded (details of these studies and an assessment of their risk of bias are summarised in Appendix 6, online at mja.com.au). We made this decision when we were checking the abstracts and realised that some of the articles had very small sample sizes (19, 18, 12 and six participants, and there was no extractable data in any of these studies). The total number of participants in the included studies was 811 so the participants in the four excluded studies made up 7% of those in all the studies. Publication bias frequently occurs when small studies with positive results are published and small studies with negative results are not. Including the small studies would have not significantly increased the statistical power, but would have increased the risk of publication bias.

Comparison with existing literature

To our knowledge, there are no previously published reviews on acrophobia alone. There is a 2008 review of psychological therapies of a range of phobias, which found that treatments involving in vivo contact with the phobic target outperformed alternative modes of exposure (eg, imaginal exposure, virtual reality, etc.) at post-treatment but not at follow-up.31 This is broadly similar to our findings in this review focusing on acrophobia.

Implications for practice

There is a range of therapies that are effective for acrophobia, but the lack of adequately powered comparative studies makes it impossible to say which are the more effective treatments. IVE therapy and DS seem to be consistently effective. IVE therapy may be equivalent to VRE, so as the technology for VRE therapy becomes more available, it may become the first-line therapy because it can be performed in the clinician’s office and has high acceptability. NLP has also been shown to be a brief, effective treatment, easily delivered with minimal training required for practitioners, so this too may prove a popular choice for clients and clinicians alike.

Implications for research

The use of reporting guidelines such as the CONSORT statement would increase the quality of published papers.28 Such reporting would make clear what were primary outcomes and what were not. The inclusion of sample size calculations along with larger sample sizes would give more confidence in findings, especially when no difference is found.32 Larger studies comparing two treatments would help clinicians decide which treatment to employ first. This may require studies using an inferiority analysis.28,33 Dichotomous outcomes need to be reported to enable NNT to be calculated.

Box 2 – Participants treated with in vivo exposure (IVE) v controls

Box 3 – Participants treated with virtual reality exposure (VRE) v waitlist controls

Box 4 – Participants treated with desensitisation (DS) v waitlist controls