Introduction
Health anxiety is characterised by a person’s pre-occupation with the belief that they have a serious illness due to an enduring tendency to misinterpret ambiguous bodily sensations as sinister symptoms (Abramowitz et al., Reference Abramowitz, Olatunji and Deacon2007b). Health anxiety can be defined in people with physical health conditions (PwPHC) as anxiety about a physical health condition or symptoms that exceeds what might be seen in others with the same symptoms or condition (Janzen Claude et al., Reference Janzen Claude, Hadjistavropoulos and Friesen2014), causing clinical levels of distress or impairment in social, occupational or other important areas of functioning (Salkovskis et al., Reference Salkovskis, Warwick and Deale2003). However, some argue elevated anxiety is a realistic and proportionate response to the presence of severe or life-threatening physical symptoms (Herschbach et al., Reference Herschbach, Berg, Waadt, Duran, Engst-Hastreiter, Henrich, Book and Dinkel2010; Melchior et al., Reference Melchior, Büscher, Thorenz, Grochocka, Koch and Watzke2013; Mehnert et al., Reference Mehnert, Berg, Henrich and Herschbach2009).
Health anxiety can be conceptualised to occur along a continuum (Asmundson et al., Reference Asmundson, Abramowitz, Richter and Whedon2010; Taylor and Asmundson, Reference Taylor and Asmundson2004) ranging from the absence of anxiety about health, to severe levels that present as clinical health anxiety (Asmundson and Fergus, Reference Asmundson, Fergus and Hedman-Lagerlöf2019). At the lower end of the scale, health related anxiety can also be protective and adaptive, helping to mobilise individuals to take action to improve health or avoid potential health stressors (Ștefan et al., Reference Ștefan, Fodor, Curt, Ionescu, Pantea, Jiboc and Tegzesiu2021); at the upper end, behaviours originally designed to prevent or protect from illness or illness progression become excessive, and thoughts of illness become a pre-occupation. These are core elements of the cognitive behavioural model of health anxiety (Salkovskis et al., Reference Salkovskis, Warwick and Deale2003), explaining the cycle of persistent health anxiety and health concerns that are not assuaged by medical reassurance (Hoffman et al., Reference Hoffmann, Rask, Frostholm and Hedman-Lagerlöf2019). The multi-centre study of Tyrer et al. (Reference Tyrer, Cooper, Crawford, Dupont, Green, Murphy, Salkovskis, Smith, Wang, Bhogal, Keeling, Loebenberg, Seivewright, Walker, Cooper, Evered, Kings, Kramo, McNulty, Nagar and Tyrer2011) reported the incidence of health anxiety in PwPHC to be higher than in the general population, ranging from 17.5% in endocrinology clinics to almost 25% in neurology clinics. Health anxiety appears more common when disease pathology is unknown, the nature of the condition is heterogeneous, or there is uncertainty around illness progression. Examples of this include multiple sclerosis (25%; Kehler and Hadjistavropoulos, Reference Kehler and Hadjistavropoulos2009), myalgic encephalomyelitis/chronic fatigue syndrome (42.4%; Daniels et al., Reference Daniels, Parker and Salkovskis2020) and chronic pain (51.5%; Rode et al., Reference Rode, Salkovskis, Dowd and Hanna2006) which exceed the range of incidence of Tyrer et al. (Reference Tyrer, Cooper, Crawford, Dupont, Green, Murphy, Salkovskis, Smith, Wang, Bhogal, Keeling, Loebenberg, Seivewright, Walker, Cooper, Evered, Kings, Kramo, McNulty, Nagar and Tyrer2011).
Current health anxiety screening questionnaires such as the Health Anxiety Inventory (HAI; Salkovskis et al., Reference Salkovskis, Rimes, Warwick and Clark2002) were developed based on the criteria for ‘hypochondriasis’ and use outdated terminology (American Psychiatric Association, 2000) with some reports that individuals with medical conditions find the language and inferences unacceptable (Daniels et al., Reference Daniels, Brigden and Kacorova2017; Fanous et al., Reference Fanous, Ryninks and Daniels2020). Given the controversy of the term ‘hypochondriasis’ and movement towards the new diagnostic criteria of illness anxiety and somatic symptom disorder (American Psychiatric Association, 2013) and more anxiety based conceptual models that are empirically well supported and evidence based (Cooper et al., Reference Cooper, Gregory, Walker, Lambe and Salkovskis2017), a review of the utility of current measurements of health anxiety is warranted.
Symptom-specific outcome measures are commonly used to screen for health anxiety in PwPHC to accommodate for ‘normal’ responses to living with physical symptoms; for example, the Falls Efficacy Scale in Parkinson’s disease (Yardley et al., Reference Yardley, Beyer, Hauer, Kempen, Piot-Ziegler and Todd2005), the Hypoglycaemia Fear Survey in diabetes (Gonder-Frederick et al., Reference Gonder-Frederick, Schmidt, Vajda, Greear, Singh, Shepard and Cox2011) and the Fear of Cancer Recurrence Inventory in cancer (Simard and Savard, Reference Simard and Savard2009). Such measures are limited to exploring anxiety related to existing specific physical symptoms and often lack normative data, making it difficult to distinguish when anxiety may be conceptualised as excessive or disproportionate. The HAI and Short Version HAI (SHAI; Salkovskis et al., Reference Salkovskis, Rimes, Warwick and Clark2002) and Health Anxiety Questionnaire (HAQ; Lucock and Morley, Reference Lucock and Morley1996) were developed specifically to screen for health anxiety. The SHAI and HAQ have not been directly compared; however, the SHAI has the strongest evidence base consistently demonstrating high convergent validity, construct validity, internal consistency and sensitivity to treatment across clinical and non-clinical samples (Abramowitz et al., Reference Abramowitz, Deacon and Valentiner2007a; Alberts et al., Reference Alberts, Sharpe, Kehler and Hadjistavropoulos2011; Salkovskis et al., Reference Salkovskis, Rimes, Warwick and Clark2002). When compared with the Illness Attitude Scale (IAS) and the Whiteley Index (WI), the HAI performed similarly, but was preferred for use in clinical settings. More recently, Alberts et al. (Reference Alberts, Hadjistavropoulos, Jones and Sharpe2013) conducted a systematic review and meta-analysis of the psychometric properties of the SHAI, providing further evidence that the SHAI demonstrated high internal consistency and strong validity across all core dimensions of measurement, in both clinical, non-clinical and medical examples.
It also has a growing evidence base for reliable use in medical populations (Alberts et al., Reference Alberts, Sharpe, Kehler and Hadjistavropoulos2011; Daniels et al., Reference Daniels, Brigden and Kacorova2017; Donkin et al., Reference Donkin, Ellis, Powell, Broadbent, Gamble and Petrie2006; Kehler and Hadjistavropoulos, Reference Kehler and Hadjistavropoulos2009; Rode et al., Reference Rode, Salkovskis, Dowd and Hanna2006), and therefore shows the most promise for assessing health anxiety in PwPHC.
However, distinguishing a ‘normal’ response from a ‘pathological’ response associated with distress remains controversial (Lebel et al., Reference Lebel, Mutsaers, Tomei, Leclair, Jones, Petricone-Westwood, Rutkowski, Ta, Trudel, Laflamme, Lavigne and Dinkel2020). Some studies have proposed higher cut-off scores are needed for PwPHC (Rode et al., Reference Rode, Salkovskis, Dowd and Hanna2006; Seivewright et al., Reference Seivewright, Salkovskis, Green, Mullan, Behr, Carlin, Young, Goldmeier and Tyrer2004), as those with medical complaints are likely to be more aware of aches/pain in their body all of the time (item 2) or feel afraid of having a serious illness (item 5), resulting in elevated scores. Evidence shows that individuals with physical health conditions report significantly higher levels of health anxiety compared with those without such conditions (Daniels et al., Reference Daniels, Parker and Salkovskis2020; Rode et al., Reference Rode, Salkovskis, Dowd and Hanna2006; Tyrer et al., Reference Tyrer, Cooper, Crawford, Dupont, Green, Murphy, Salkovskis, Smith, Wang, Bhogal, Keeling, Loebenberg, Seivewright, Walker, Cooper, Evered, Kings, Kramo, McNulty, Nagar and Tyrer2011), and so may be more likely to perceive somatic sensations and perceive them as ‘dangerous’ (i.e. signs of worsening/something new and problematic). This means that those with health conditions may be more prone to achieving elevated cut-off scores.
Different response patterns are likely to be elicited across PwPHC depending on the symptoms they experience, leading to differential item functioning and variable sample means across conditions (LeBouthillier et al., Reference LeBouthillier, Thibodeau, Alberts, Hadjistavropoulos and Asmundson2015). This makes it difficult to calculate appropriate cut-off scores generalisable to all medical populations. Kehler and Hadjistavropoulos (Reference Kehler and Hadjistavropoulos2009) found participants with multiple sclerosis (MS) scored approximately one standard deviation above age-matched controls when ‘other than Multiple Sclerosis (MS)’ was added to items of the SHAI, suggesting a higher incidence of health anxiety in MS as opposed to only elevated scores due to physical symptoms. Unlike symptom-specific measures, adaptation of the SHAI by adding ‘other than’ [the symptoms of my physical health condition] does not allow for exploring anxiety in relation to existing physical symptoms (in this case MS) and therefore excludes the possibility of screening for clinical levels of health anxiety related to MS, compared with others that may be managing the same symptoms. It is key that a health anxiety screening tool spans both illness anxiety and somatic symptom disorder, reflecting the discussed movement towards updated diagnostic classification systems, and more specifically, is able to capture health anxiety that is notwithstanding a medical condition. Crucially, Kehler and Hadjistavropoulos (Reference Kehler and Hadjistavropoulos2009) found over a third of those that scored within the clinical range for health anxiety did not score within this range on a measure of generalised anxiety, emphasising that health anxiety is a distinct phenomenon and should be screened for separately. This has also been found in myalgic encephalomyelitis/chronic fatigue syndrome (ME/CFS) research (Daniels et al., Reference Daniels, Brigden and Kacorova2017; Daniels et al., Reference Daniels, Parker and Salkovskis2020).
A recent study reported a number of barriers to clinicians using the SHAI with patients experiencing ME/CFS (Fanous et al., Reference Fanous, Ryninks and Daniels2020): firstly it unintentionally discredits concerns that patients have about their physical health and questions the authenticity of symptoms; secondly individuals felt judged as ‘hypochondriacs’; and thirdly the phrasing of the questionnaire was an obstacle to engagement. Similar barriers were found by studies exploring patient experiences of completing the SHAI in ME/CFS and chronic pain: patients reported items were inappropriate and did not give their symptoms credence (Parker et al., Reference Parker, Carlton, Harris and Daniels2023), items were associated with negative connotations, perceived stigma and evoked strong emotions (Fanous et al., Reference Fanous, Ryninks and Daniels2020), and some patients felt completely delegitimised (Daniels et al., Reference Daniels, Brigden and Kacorova2017). The ‘inappropriate’, ‘invalidating’ and ‘outdated’ language may be seen to locate the responsibility of distress within the individual or to categorise a patient’s effort to manage their symptoms as ‘weak’ (Fanous et al., Reference Fanous, Ryninks and Daniels2020); for example, participants commented ‘it’s saying that their illness might be in their mind’ (p. 6) and ‘it insinuates I am weak in character’ (p. 10).
The reported incidence of health anxiety in PwPHC is high, suggesting a pressing clinical need for a more appropriate health anxiety screening questionnaire, spanning both illness anxiety and somatic symptom disorder to be developed, and for individuals to be offered psychological intervention. Cognitive behavioural therapy for health anxiety (CBTHA) is effective in reducing health anxiety in PwPHC and can be superior to existing treatment protocols (Cooper et al., Reference Cooper, Gregory, Walker, Lambe and Salkovskis2017; Daniels and Loades, Reference Daniels and Loades2017; Daniels et al., Reference Daniels, Osborn and Davis2018; Tyrer et al., Reference Tyrer, Cooper, Salkovskis, Tyrer, Crawford, Byford, Dupont, Finnis, Green, McLaren, Murphy, Reid, Smith, Wang, Warwick, Petkova and Barrett2014; Tyrer et al., Reference Tyrer, Wang, Crawford, Dupont, Cooper and Nourmand2021). Measures of symptom-specific or generalised anxiety are not sufficient to screen for the multi-faceted aspects of health anxiety or to identify individuals who may benefit from intervention, as these have been established as separate, but related constructs (Daniels et al., Reference Daniels, Brigden and Kacorova2017).
This study sought to adapt the SHAI for medical populations using a two-stage sequential design: firstly, eliciting empirical data on the utility of the measure and developing an adapted version of the SHAI for medical populations through an iterative process based on this empirical data; secondly, evaluating the psychometric properties of the adapted measure in conditions where a heightened incidence of health anxiety and clinical need has been identified: MS, ME/CFS and chronic pain.
Study 1: Adaptation of the Health Anxiety Inventory – Short Version (SHAI)
The aim of study 1 was to adapt the 18-item SHAI in order to improve its acceptability for administration in medical populations.
Method
Design
A multi-stage iterative Delphi approach was used in order to produce consensus agreement on the barriers to completion of the SHAI. The Delphi method is used to produce statistically reliable and valid responses (Fowles, Reference Fowles1978), and is appropriate when there is discrepancy within the literature or incomplete knowledge of a topic (Trevelyan and Robinson, Reference Trevelyan and Robinson2015), such as this area.
The Delphi study closely followed protocols described by Daniels (Reference Daniels2017) and Hsu and Sandford (Reference Hsu and Sandford2007). Three rounds were used to reach an evidence-based consensus for which adaptation of the SHAI could be based upon. This number of rounds is considered optimal (Trevelyan and Robinson, Reference Trevelyan and Robinson2015) and was decided a priori. One of the 12 participants was unable to attend round 1 due to technological difficulties (91.6% response rate) and all participants participated in rounds 2 and 3 (100% response rate).
Different methods have been used to measure consensus, with 67% agreement often used to indicate an evidence-based consensus on dichotomous scales (Heiko, Reference Heiko2012). Due to the lack of unified agreement, a more stringent value of 75% agreement was set for indicating consensus on dichotomous scales in this study. This value has been used in other studies using the Delphi method (Navarrete-Dechent et al., Reference Navarrete-Dechent, Liopyris, Molenda, Braun, Curiel-Lewandrowski, Dusza, Guitera, Hofmann-Wellenhof, Kittler, Lallas, Malvehy, Marchetti, Oliviero, Pellacani, Puig, Soyer, Tejasvi, Thomas, Tschandl, Scope, Marghoob and Halpern2020; Nieuwenhuys et al., Reference Nieuwenhuys, Õunpuu, Van Campenhout, Theologis, De Cat, Stout, Molenaers, De Laet and Desloovere2016). The criterion set for indicating consensus on Likert scales in Delphi studies has varied between 51 and 80% agreement (Green, Reference Green1982; Loughlin and Moore, Reference Loughlin and Moore1979; Putnam et al., Reference Putnam, Spiegel and Bruininks1995; Seagle and Iverson, Reference Seagle and Iverson2002). A rate of 70% expert agreement rating 3 or higher and the median of 3.25 or higher on a 4-point Likert scale was used in this study (Green, Reference Green1982).
Round 1
The first round consisted of two 1-hour online focus groups, facilitated on Microsoft Teams software. Each focus group included a mixture of experts by profession and by experience. The SHAI was sent to participants electronically a week prior to familiarise with the questionnaire. Experts were interviewed with cognitive interviewing scripted verbal probes (Willis, Reference Willis2005) as this method pays explicit attention to the cognitive processes respondents use to answer questionnaire items, with scripted probes particularly recommended for questionnaire development groups (Collins, Reference Collins2003; Willis, Reference Willis2005). This allows for the patient experience of completing the SHAI to be explored in a systematic and standardised way. The groups were audio recorded and data were transcribed verbatim for thematic analysis.
Thematic analysis followed the protocol described by Braun and Clarke (Reference Braun and Clarke2006). The researcher (J.C.) took a critical realist position exploring the reported experiences, meanings and reality of expert responses whilst acknowledging the broader social context of physical health conditions and health services. A deductive approach was used given the researchers’ knowledge of relevant theory. In order to account for bias towards pre-conceived themes, a researcher assistant (M.M.) also completed a thematic analysis to ensure qualitative trustworthiness (Elliot et al., Reference Elliot, Fischer and Rennie1999). Any differences were discussed before themes were agreed upon.
Round 2
A Qualtrics online survey was sent to the participants inviting them to comment if the themes accurately reflected their experience of the focus group discussions. Taking into account the themes identified, the experts were then invited to suggest adaptations, edits and deletions to the wording of each SHAI item, or to indicate no modifications were needed if the item was deemed appropriate.
The survey responses from round 2 were synthesised through quantitative content analysis, following Rourke and Anderson’s (Reference Rourke and Anderson2004) guidelines for scoring and interpreting coding schemes. Final codes were decided upon by majority consensus (75% of experts endorsing the code); this criterion for consensus was chosen based on endorsing or not endorsing the code being dichotomous. The researcher then used code frequencies to formulate item modifications through an iterative process in consultation with an expert working in the field and a person with lived experience.
Round 3
A version of the SHAI with the item modifications made in round 2 was sent to the experts via a second Qualtrics online survey and participants were invited to indicate their level of agreement with the adaptations made with regard to: how appropriate the modified item is for administration in medical populations (1=not appropriate to 4=completely appropriate), and to what extent the adaptations made have overcome the barriers to administration discussed in the focus groups (1=barriers have not been overcome to 4=barriers have been overcome). This scale was chosen following recommendations by Green (Reference Green1982) allowing for two set criteria to be used to indicate consensus on this 4-point Likert scale. A 4-point Likert scale is often chosen when measuring consensus as it excludes the possibility of experts providing neutral responses. Experts were invited to suggest further adaptations, edits and deletions for items where they lacked agreement. They were also asked to indicate whether the adapted measure was appropriate for administration in medical populations on a dichotomous scale (yes/no), with 75% agreement indicating consensus.
Items that failed to meet Green’s (Reference Green1982) two criteria for indicating consensus for both rating scales were deleted as the adaptations made were deemed insufficient to overcome the barriers to administration identified in the focus groups. Expert feedback was then used to modify the remaining items in consultation with an expert working in the field and a person with lived experience. These items were used to produce the Health Anxiety Inventory for Medical Populations (HAI-M) for pilot testing in study 2.
Participants
Twelve participants were recruited through the University of Bath People with Personal Experience Committee and through specialist health services. Research suggests 8–15 experts is a sufficient expert panel number and experts are required to have specific knowledge and experience of the issue being investigated (Keeney et al., Reference Keeney, Hasson and McKenna2011; Rowe and Wright, Reference Rowe and Wright2001; Skulmoski et al., Reference Skulmoski, Hartman and Krahn2007). This selection of both experts by profession and experience allows for a wider range of perspectives to be explored (Powell, Reference Powell2003).
All medical groups were represented by at least one ‘expert by experience’ who was living with one of the physical health conditions included within the scope of the study, and one expert by profession working within the respective physical health settings. Eight participants were experts by experience, defined as having a self-reported diagnosed physical health condition and having experienced symptoms for over 6 months. The final four participants were experts by profession, defined as professionals working clinically with PwPHC and health anxiety and who have demonstrated their expertise in peer reviewed publications in the clinical field studied in this research. The experts by experience worked across ME/CFS, chronic pain and neurorehabilitation services. Expert demographics can be seen in Table 1.
Table 1. Expert panel demographics
Results
Round 1
Thematic analysis led to identification of three over-arching themes: (1) overlaps between the symptoms of a physical health condition and health anxiety, (2) psychological adjustment to living with a physical health condition and (3) inappropriate wording. High overlap between the themes identified by the two researchers indicated high qualitative trustworthiness. Table 2 displays the subthemes, codes and example quotations associated with these themes.
Table 2. Themes, subthemes, codes and example quotations from focus group discussions
Round 2
On consultation, the participants who attended the focus groups indicated the themes accurately reflected their experience of the discussions (100% agreement). Consensus indicated by over 75% agreement suggested good face validity of the themes identified. Table 3 displays the codes and code frequencies derived from quantitative content analysis of survey responses and the item modifications made. The adapted questionnaire following round 2 can be seen in Appendix 1 of the Supplementary material.
Table 3. Codes, frequencies and item modifications made following content analysis
Round 3
Table 4 displays the percentage of ratings above three and median ratings for each rating scale to compare against Green’s two critera (Reference Green1982) for indicating consensus. Eleven (73.33%) of the 15 items met both criterion on ratings of how appropriate the modified items are for administration in medical populations and four items (26.67%) met neither criterion. The same eleven items (73.33%) met both criterion for overcoming barriers to administration and the same four (26.67%) met neither criterion.
Table 4. Ratings of appropriateness for administration in medical populations and overcoming barriers to administration
* Below criterion;
ᴿ item deleted following round 3;
ᴬ item adapted following round 3.
Three of the four items were removed for failing to meet consensus on the two criteria for both rating scales, as expert feedback suggested no further modifications would be sufficient to overcome the barriers identified. One item was further modified taking into account expert feedback. All other items were deemed appropriate for administration in medical populations. Nine of the 12 experts answered ‘yes’ (75% agreement) when asked if the adapted measure was appropriate for administration in medical populations, again indicating consensus on a dichotomous scale. Experts who answered ‘no’ commented the adapted measure has not overcome the issue of being ‘too lengthy’, ‘unnecessarily complicated’ and could model ‘more positive self-statements as opposed to endorsing the absence of negative ones’; however, these were individual reactions and therefore further modification of the measure was not deemed necessary.
The final study one pilot version of the Health Anxiety Inventory for Medical Populations (HAI-M) consisted of 12 items scoring from 0 to 3 (see Appendix 2 in the Supplementary material).
Study 2: Validation of the Health Anxiety Inventory for Medical Populations (HAI-M)
The aim of study 2 was to evaluate the psychometric properties of the HAI-M in samples with MS, ME/CFS and chronic pain.
Method
Design
An online cross-sectional mixed methods questionnaire design was used plus a follow-up one week later.
Participants and procedure
Snowball sampling was used to recruit participants to test the validity and reliability of the HAI-M between 10 September 2020 and 4 February 2021, inviting adults with a medically confirmed diagnosis of MS, ME/CFS or chronic pain to take part. Individuals meeting the inclusion criteria participated through clicking on a weblink distributed by social media. Participants provided informed consent and demographic information before completing a battery of online standardised self-report questionnaires. Participants were then invited to complete the HAI-M via a second weblink one week later. Debriefing information was provided to all participants. In addition to standard questionnaire completion, participants were asked to rate their experience of completing the SHAI and HAI-M on two Likert scales: how acceptable the wording is (1=not acceptable at all to 5=completely acceptable) and how relevant the wording is (1=not relevant at all to 5=completely relevant) for understanding health concerns.
Of 315 survey responses, 198 completed the whole battery of questionnaires (62.86%) and 78 completed the follow-up questionnaire one week later (24.76%). Chi squared analyses indicated no significant differences between demographics of people that did and did not complete the whole battery of questionnaires. Rates of fully completed survey responses were 72.7% in MS, 76.2% in ME/CFS and 45.7% in chronic pain. Imputation of missing data using person means has been demonstrated to be superior to analysing only complete data in previous studies involving the SHAI and HADS (Bell et al., Reference Bell, Fairclough, Fiero and Butow2016; Fergus and Valentiner, Reference Fergus and Valentiner2011), therefore this method of imputation was used for missing data on two SHAI (1%) and five HADS questionnaires (2%).
The final sample who completed the HAI-M at time 1 included 115 participants with MS (mean age=45.98, SD =12.86), 84 with ME/CFS (mean age=39.01years, SD=13.84) and 116 with chronic pain (mean age=47.77, SD=15.38). Table 5 displays descriptive statistics for the clinical populations and group differences.
Table 5. Descriptive statistics for individual clinical populations
* Significant difference between groups.
Instruments
The Hospital Anxiety and Depression Scale (HADS) is a standardised 14-item questionnaire for use in medical populations, made up of two subscales assessing generalised anxiety and depression (Snaith and Zigmund, Reference Snaith and Zigmund1986). Items are rated from 0 to 3, with a score of 8 indicating clinical symptomology with good sensitivity and specificity on either subscale (Bjelland et al., Reference Bjelland, Dahl, Haug and Neckelmann2002). Internal consistency for the anxiety (α=.93) and depression (α=.90) subscales is commendable, as is test–retest reliability, concurrent validity and discriminant validity (Bjelland et al., Reference Bjelland, Dahl, Haug and Neckelmann2002; Hermann, Reference Hermann1997). In the study sample, the HADS Cronbach’s alpha coefficient was high (α=.882) for the total scale, and for anxiety (α=.813) and depression (α=.848), respectively.
The SHAI is a standardised 18-item questionnaire used to screen for health anxiety. Items are scored from 0 to 3 and a cut-off score of 18 is used to indicate clinical caseness, with good internal consistency (α=.88), criterion validity and sensitivity to treatment reported (Salkovskis et al., Reference Salkovskis, Rimes, Warwick and Clark2002). The SHAI demonstrated excellent internal consistency in this sample, with a Cronbach’s alpha coefficient of .906. The items also have good convergent and divergent validity in non-clinical and health samples (Alberts et al., Reference Alberts, Sharpe, Kehler and Hadjistavropoulos2011; Abramowitz et al., Reference Abramowitz, Deacon and Valentiner2007a). Most studies exploring the factor structure of the 18-item SHAI provide support for a 2-factor model, illness likelihood and negative consequences. Often, the negative consequences scale is removed and a score of 20 is used as a conservative cut-off in medical populations (Seivewright et al., Reference Seivewright, Green, Salkovskis, Barrett, Nur and Tyrer2008; Tyrer et al., Reference Tyrer, Wang, Crawford, Dupont, Cooper and Nourmand2021). A 2-factor model is also reported for the 14-item version; however, there is a lack of consensus around how the two factors are labelled. Some studies describe labels of illness likelihood and body vigilance, and others suggest thought intrusion and fear of illness (LeBouthillier et al., Reference LeBouthillier, Thibodeau, Alberts, Hadjistavropoulos and Asmundson2015).
The Health Anxiety Inventory for Medical Populations (HAI-M) 12-item measure from study 1 was tested for its validity and reliability in study 2. The use of the HAI-M in this sample indicated good internal consistency for the total sample (α=.875) and for the respective clinical populations: MS (α=.855), ME/CFS (α=.877) and chronic pain (α=.887).
Planned analysis
Statistical analysis was performed using SPSS version 26, with a criterion of p=<.05 set for significance throughout. Effect sizes were chosen based on those given in the original SHAI validation study (Salkovskis et al., Reference Salkovskis, Rimes, Warwick and Clark2002). The total sample size required was 96 participants in order to complete correlational analyses and therefore a minimum of 48 participants per population was required. This was determined through power analysis using G*Power (effect size=.7, power=.95, α error probability=.05). The total sample size required was 107 participants in order to compare group differences and therefore a minimum of 54 participants per population was required. Again, this was determined through power analysis using G*Power (effect size=.5, power=.95, α error probability =.05). This study followed recommendations by Kline (Reference Kline1994) who suggests an absolute minimum of 100 participants per population for factor analyses is required.
Preliminary analyses
Preliminary analyses using Shapiro–Wilk tests revealed the HAI-M time 1, HAI-M time 2 and SHAI data were positively skewed (p=< .05), therefore statistical analysis used non-parametric tests where appropriate. Levene’s tests confirmed homogeneity of variance for all data. Spearman’s correlations identified no collinearity between the main study variables (r=< .90; see Table 6), and linear relationships were observed between all variables on scatterplots. Interquartile ranges were explored using box plots allowing for identification of outliers; however, they were not removed or transformed as this did not change the outcome of the result profile.
Table 6. Spearman’s correlations between main study variables
* p=<0.01.
Group differences
Descriptive statistics were used to summarise demographics and scores on the standardised questionnaires. Given F-statistics are robust against violations of normality (Glass et al., Reference Glass, Peckham and Sanders1972; Wilcox, Reference Wilcox2012), mixed ANOVAs (participant ratings: HAI-M time 1 and SHAI; Clinical population: MS, ME/CFS and chronic pain) were used to compare participant’s ratings of acceptability and relevance. One-way analyses of variance (ANOVA) were then used to test whether there were significant differences between (1) HAI-M time 1 scores and then (2) SHAI scores for populations with MS, ME/CFS and chronic pain. Post-hoc Hochberg’s GT2 was also used to account for unequal cell sizes when exploring group differences.
Reliability and validity
Cronbach’s alpha was used to evaluate internal consistency with a value of above .7 being deemed acceptable. Items were excluded if the measure was found to be more reliable following their removal. Spearman’s correlations were then used to explore test–retest reliability between scores on the HAI-M at time 1 and time 2, to explore convergent validity between scores on the HAI-M time 1 and SHAI, and to test the relationship between scores on the HADS anxiety scale and HAI-M at time 1 given the skewed data set. The SHAI was used as a comparison for convergent validity despite identified limitations in medical populations due to the measure’s demonstrated ability to differentiate health anxiety from other anxiety disorders with high specificity and sensitivity (Hedman et al., Reference Hedman, Lekander, Ljótsson, Lindefors, Rück, Andersson and Andersson2015), and in order to ensure the adaptations of the measure do not diminish the original measure’s demonstrated construct validity (Salkovskis et al., Reference Salkovskis, Rimes, Warwick and Clark2002). The HADS was used to explore divergent validity given its demonstrated sensitivity and specificity for differentiating generalised anxiety from other anxiety disorders (Bjelland et al., Reference Bjelland, Dahl, Haug and Neckelmann2002).
Factor analyses
Exploratory factor analyses were used as Salkovskis et al. (Reference Salkovskis, Rimes, Warwick and Clark2002) did not report eigenvalues, factor loadings or percentage of variance explained making it difficult to compare against using confirmatory factor analysis. Principal components analysis and an extraction method of ‘Eigenvalues above 1 retained’ was used to determine the factors extracted. Rotation using the oblique method (direct oblimin) was used due to Salkovskis et al. (Reference Salkovskis, Rimes, Warwick and Clark2002) reporting a moderate correlation between the factors of the SHAI, from which the HAI-M is adapted. Parallel analysis and factor interpretability was also used to determine the number of factors retained within each clinical population.
Results
Acceptability and relevance
Mean acceptability and relevance ratings are displayed in Table 7. No significant interaction effects between the clinical populations and participant ratings were found when exploring acceptability; however, a significant main effect of participant ratings was found (F 1,197 =9.555, p=<.01). Post-hoc analyses found that participants rated the HAI-M (mean=4.164, SE=.67) as significantly more acceptable for assessing their health concerns compared with the SHAI (mean=3.965, SE=.73). A significant main effect of clinical population was also found (F 2,197 =4.228, p=.016) and post-hoc analyses demonstrated that participants with MS (mean=4.295, SE=.095) rated the questionnaires as significantly more acceptable than participants with ME/CFS (mean=3.898, SE=.108). There were no significant differences between acceptability ratings for chronic pain (mean=4.000, SE=.118) and the two other clinical populations.
Table 7. Acceptability and relevance ratings
SE, standard error;
* statistically significant difference.
A significant main effect of clinical population was found when exploring relevance (F 2,197 =4.701, p=.01); however, there was no significant main effect of participant ratings or interaction effect between the clinical populations and participant ratings. Post-hoc analyses found that participants with MS (mean=4.084, SE=.101) rated the questionnaires as significantly more relevant for assessing their health concerns compared with ME/CFS (mean=3.641, SE=.115). There was no significant difference between relevance ratings for chronic pain (mean=3.745, SE=.126) and the two other clinical populations.
Group differences
No significant differences were found between the clinical populations for HAI-M scores at time 1 (F 2,240 =2.977, p=.053), confirmed by post-hoc Hochberg’s GT2 analyses (p=.117; see Table 8). Similarly, no significant differences were found between the clinical populations for SHAI scores (F 2,197 =2.228, p=.110), also confirmed by post-hoc Hochberg’s GT2 analyses (p=.138).
Table 8. Descriptive statistics for the clinical populations and total sample
CI, confidence interval; SD, standard deviation.
Reliability
The HAI-M alpha coefficient indicated high internal consistency for the total sample (α=.875) and clinical populations: MS (α=.855), ME/CFS (α=.877) and chronic pain (α=.887). All items were retained. Test–retest reliability was good for the total sample (r=.812, p=< .01) and for clinical populations with ME/CFS (r=.907, p=< .01) and chronic pain (r=.843, p=< .01). Test–retest reliability was acceptable (>.7) for participants with MS (r=.736, p=< .01).
Convergent validity
Strong (>.7) positive correlations were found between the HAI-M and SHAI for the total sample (r=.801, p=< .01) and for participants with MS (r=.796, p=< .01), ME/CFS (r=.775, p=< .01) and chronic pain (r=.852, p=< .01), supporting the convergent validity of the HAI-M.
Divergent validity
Moderate (<.6) positive correlations were found between scores on the HADS anxiety scale and the HAI-M at time 1 for the total sample (r=.515, p=< .01) and for the clinical populations: MS (r=.472, p=< .01), ME/CFS (r=.539, p=< .01) and chronic pain (r=.557, p=< .01). Moderate correlations support the HAI-M’s ability to differentiate health anxiety from generalised anxiety.
Factor analysis
Initial inspection of the correlation matrices confirmed the presence of many correlation coefficients above .3. The Kaiser–Meyer–Olkin value suggested good sampling adequacy (KMO=.841 to .851) and Bartlett’s test of sphericity indicated the correlations between items were sufficient to carry out a factor analysis for all clinical populations (p=< .01). Principal components analyses with direct oblimin rotation were conducted on the 12-item HAI-M to explore if items may load differently across the clinical populations: MS (n=103), ME/CFS (n=79) and chronic pain (n=61). All analyses revealed three factors with eigenvalues exceeding 1, and scree plot inflexion and parallel analysis justified these being retained. The three factors explained 57.98% of the variance in MS, 62.13% in ME/CFS and 65.17% in chronic pain. The correlations between the three factors can be found in the Supplementary material.
Eigenvalues, percentage of variance explained and factor loadings following rotation for the clinical populations can be found in the Supplementary material. Substantial factor loadings were set at greater than .4 as this value has been used in previous studies exploring the factor structure of the SHAI (Abramowitz et al., Reference Abramowitz, Deacon and Valentiner2007a; Alberts et al., Reference Alberts, Sharpe, Kehler and Hadjistavropoulos2011). Factor interpretability suggested items assessing pre-occupation with thoughts about health load onto factor 1 (items 1, 2, 4, 6, 7 and 12). Items assessing vigilance to bodily sensations loaded onto factor 2 (items 3, 9 and 11) and items assessing perceived illness likelihood loaded onto factor 3 (items 5, 8 and 10). Items 5 and 6 loaded onto two items in one clinical population, and so the highest factor loading was taken into account when interpreting factors.
The 3-factor model described differs from studies that report a 2-factor model for the 14- and 18-item SHAI (Alberts et al., Reference Alberts, Sharpe, Kehler and Hadjistavropoulos2011; Salkovskis et al., Reference Salkovskis, Rimes, Warwick and Clark2002; Wheaton et al., Reference Wheaton, Berman, Franklin and Abramowitz2010). These studies report either illness likelihood and negative consequences as the two factor labels or thought intrusion and fear of illness. Other studies have reported a 3-factor model labelled body vigilance, illness likelihood and illness severity (Abramowitz et al., Reference Abramowitz, Deacon and Valentiner2007a; Olatunji, Reference Olatunji2009), although the items load differently onto these factors across studies.
Although the same number of factors were retained across the clinical populations, there appear to be differences in how the items load onto these factors; for example, items appear to load most heavily onto factor 1 in MS and chronic pain and less heavily in the sample with ME/CFS. In contrast, items appear to load most heavily onto factor 3 in ME/CFS. This is consistent with previous studies that report differences in item scores across factors, for example Alberts et al. (Reference Alberts, Sharpe, Kehler and Hadjistavropoulos2011) found participants with MS scored higher on items relating to thought intrusion and fear of illness than participants without MS.
The three-factor model here also demonstrates construct and content validity as evidenced by tests of convergent validity and associated strong significant relationship (>.8) between the HAI-M and the SHAI measure, tests of divergent validity with the HADS and the mapping of the three factors onto the theoretical construct of the HA model (i.e. pre-occupation, hypervigilance and perceived illness likelihood) for which the HAI has been designed to measure against.
Discussion
Study 1 describes the systematic development of the HAI-M, a health anxiety screening tool for use in medical populations which was produced with a high level of agreement. The limitations of administering the SHAI in medical populations, identified in the first round of the Delphi study, are similar to those identified in a previous study by Fanous et al. (Reference Fanous, Ryninks and Daniels2020): the wording of the SHAI is perceived to be inappropriate, lacking in sensitivity and questions the authenticity of symptoms associated with physical health conditions. The experts in study 1 recommended changes to phrases such as ‘serious illness’, ‘resist’, ‘lastingly relieved’ and ‘hypochondriac’ which were also identified to be problematic phrases in the studies of Fanous et al. (Reference Fanous, Ryninks and Daniels2020) and Daniels et al. (Reference Daniels, Brigden and Kacorova2017). Changes made in relation to phrasing were considered acceptable as reflected in the final round of the Delphi study, with psychometric properties reflecting internal consistency of the measure.
The adapted screening tool was rated to be more acceptable compared with the SHAI, suggesting this study met its primary aim of developing a more acceptable measure for use in medical populations whilst also remaining both reliable and valid. Correspondence from participants raised further concerns around the wording of the SHAI during study 2, with participants stating the SHAI wording is ‘problematic’ and ‘dismissive’. Further comments stated ‘to assume that having images of being ill is a negative thing is really ableist’, that it is ‘concerning to see common physical symptoms being asked about to ascertain mental health difficulties’ and that the wording suggests a conceived assumption that these conditions are not ‘serious illnesses’. These words and phrases were adapted or removed in the development of the HAI-M and therefore such concerns have been addressed. This is reflected in the statistically significant difference in ratings of acceptability between the HAI-M and SHAI, although an even greater difference may have been expected given this qualitative information and a rating scale allowing for a greater spread of data may be useful in future research.
In contrast to studies that report highly variable scores across medical populations using the SHAI (Alberts et al., Reference Alberts, Hadjistavropoulos, Jones and Sharpe2013; Lebel et al., Reference Lebel, Mutsaers, Tomei, Leclair, Jones, Petricone-Westwood, Rutkowski, Ta, Trudel, Laflamme, Lavigne and Dinkel2020), the findings of study 2 suggest HAI-M items are not differentially endorsed across three physical health conditions where this was previously documented as an issue.
These findings suggest that a pooled cut-off for the HAI-M could be used to indicate ‘clinical’ levels of health anxiety in MS, ME/CFS and chronic pain, signifying a threshold at which psychological distress associated with the medical problem is likely to be affecting social and occupational functioning beyond what might be usually be expected in the circumstances. However, while clinical cut-offs in clinical settings can be used to signal that a fuller assessment of health concerns may be warranted, it is imperative that the HAI-M, and indeed all screening tools, are interpreted with caution and in clinical context of the presenting illness, taking into account relevant factors such as stage and prognosis of illness, and personal and familial history of the same or similar illnesses. The HAI-M has been adapted to aid the recognition of the added burden of health anxiety in those with medical problems, aiming to understand but not pathologise normal human responses to the personal catastrophe that illness can often represent.
This study differed from previous studies in sampling the general population using an online questionnaire, as opposed to sampling clinical populations attending medical settings, and it should be acknowledged that health anxiety is known to be elevated during the process of medical consultations and investigations, which may explain this difference in findings (Hadjistavropoulos et al., Reference Hadjistavropoulos, Craig and Hadjistavropoulos1998). Additionally, the global pandemic has had a significant impact on the incidence of health anxiety (Rettie and Daniels, Reference Rettie and Daniels2021), with emerging evidence that health anxiety may be a particular difficulty for PwPHC who have been advised to shield, which could have led to diminishing group differences across conditions (Chaplin, Reference Chaplin2021; Sloan et al., Reference Sloan, Gordon, Harwood, Lever, Wincup, Bosley, Brimicombe, Pilling, Sutton, Holloway and D’Cruz2020). That being so, further validation is needed to be sure a pooled cut-off is generalisable across physical health conditions.
The HAI-M was found to be psychometrically acceptable, and this study reports comparable statistics to those reported in the SHAI validation study of Salkovskis et al. (Reference Salkovskis, Rimes, Warwick and Clark2002). The good convergent validity of the HAI-M conveys that the measure is able to screen for the multi-faceted phenomenology of health anxiety drawing on cognitive behavioural theory. Good divergent validity also suggests the HAI-M and HADS are measuring distinct but related entities. This is consistent with previous research documenting a moderate correlation (.57) between the SHAI and HADS anxiety scale (Tang et al., Reference Tang, Salkovskis, Poplavskaya, Wright, Hanna and Hester2007a; Tang et al., Reference Tang, Wright and Salkovskis2007b). Weaker correlations (.29–.42) tend to be observed between the SHAI and other measures of generalised anxiety (Abramowitz et al., Reference Abramowitz, Deacon and Valentiner2007a; Olatunji et al., Reference Olatunji, Deacon, Abramowitz and Valentiner2007), which may be attributable to the HADS being developed for medical populations and therefore capturing some aspects of anxiety related to health.
There is disagreement within the literature of the factor structure of the SHAI (Alberts et al., Reference Alberts, Hadjistavropoulos, Jones and Sharpe2013). Study 2 suggests a 3-factor model for the HAI-M which bears good face and content validity when comparing with the main tenets of the health anxiety model (pre-occupation, symptom hypervigilance, perceived illness likelihood). The item modifications and factor interpretation take into account changes in diagnostic terminology and are likely to explain some disparities with previous literature. The factor of vigilance to sensations acknowledges PwPHC are understandably more likely to be aware or to notice physical symptoms, and items were adapted with the aim to distinguish ‘normal’ reactions from pathological responses associated with dysfunctional behaviours and distress as discussed by Lebel et al. (Reference Lebel, Mutsaers, Tomei, Leclair, Jones, Petricone-Westwood, Rutkowski, Ta, Trudel, Laflamme, Lavigne and Dinkel2020). The identified factors share characteristics with those reported in previous studies exploring the SHAI factor structure; for example, factors labelled illness likelihood, thought intrusion, fear of illness and body vigilance (Alberts et al., Reference Alberts, Hadjistavropoulos, Jones and Sharpe2013; Abramowitz et al., Reference Abramowitz, Deacon and Valentiner2007a; Salkovskis et al., Reference Salkovskis, Rimes, Warwick and Clark2002). The factors were also inter-correlated as found in previous studies (Abramowitz et al., Reference Abramowitz, Deacon and Valentiner2007a; Fergus and Valentiner, Reference Fergus and Valentiner2011). In accordance with the findings of Alberts et al. (Reference Alberts, Sharpe, Kehler and Hadjistavropoulos2011), items do appear to load differently onto these factors across clinical populations, which may reflect heterogeneous expressions of health anxiety in different physical health conditions. This is an unsurprising finding given the literature documenting disease-specific cognitions and different physical symptoms of varying degrees and pervasiveness, and therefore conveys an important rationale for cognitive behavioural intervention to be informed by item responses on the HAI-M.
The Collaborative Outcomes Resource Network (2007) states that a score of one standard deviation above the mean can be used to identify clinical significance. Given no significant differences were found between HAI-M scores across clinical populations, one standard deviation above the total sample mean in this study would be ≥24. It is currently unclear whether a pooled cut-off would be clinically meaningful for use across physical health conditions. Further work is needed to establish this. The use of ‘range’ categories as seen in the original SHAI, HADS and other questionnaires might prove more useful than a definitive clinical threshold.
Limitations and future research
The two-stage adaptation of the SHAI was based largely on the opinion of white British participants, therefore the relevance and acceptability of the HAI-M for other cultures is unclear. Study 2 is also limited in generalisability due to the under-representation of males which may be important due to differences in behavioural responses between men and women, for example engaging in more reassurance-seeking and worry (MacSwain et al., Reference MacSwain, Sherry, Stewart, Watt, Hadjistavropoulos and Graham2009). Results should additionally be considered in light of an additional limitation of study 2; participants were recruited through social media, which meant that confirmation of their medical diagnoses was not achieved. The study also took place during the COVID-19 pandemic where studies have reported a higher incidence of health anxiety in the general population particularly for those who may be vulnerable due to health conditions (Rettie and Daniels, Reference Rettie and Daniels2021; Jungmann and Witthöft, Reference Jungmann and Witthöft2020). It is unclear if the scores on the HAI-M could be influenced by this context, and future research may seek to clarify this and to validate this measure in other physical health conditions. We also acknowledge that data pertaining to the acceptability of the HAI measure were limited to those with ME/CFS; however, the issues reported by those with ME/CFS are unlikely to be specific to this clinical group. When given the opportunity to refine, adapt and edit the HAI for medical conditions, there was consensus agreement across all clinical groups and professionals that for purposes of acceptability, the measure should be adapted, where there was an option to retain in original form. This is also in context of existing criticisms regarding the suitability of the HAI for this group (Daniels et al., Reference Daniels, Brigden and Kacorova2017; Daniels et al., Reference Daniels, Parker and Salkovskis2020; Parker et al., Reference Parker, Carlton, Harris and Daniels2023). Finally, there is a need for further research to provide further validation data and to evaluate the sensitivity and specificity of a cut-off score when compared with practitioner report using the Structured Clinical Interview for the DSM (First, Reference First2014).
Conclusion
This study provides preliminary evidence for a more acceptable and a psychometrically robust health anxiety screening tool for use in medical populations. The HAI-M appears to consist of more appropriate items for assessing health anxiety in medical populations when compared with the SHAI, and to demonstrate good reliability and validity. It therefore provides a solution to the current unmet clinical need and may seek to improve access to therapy for those with PwPHC.
Key practice points
-
(1) Rates of health anxiety are reported to be high in medical conditions. Practitioners should consider assessing for health anxiety in order to gain a more thorough understanding of presenting difficulties and psychological distress.
-
(2) When using standardised measures such as the HAI, do consider where questions may be artificially inflated by virtue of the nature of having a medical problem.
-
(3) The health anxiety model by Salkovskis and colleagues can be used as a transdiagnostic model for health anxiety, and can easily be used to work with health anxiety occurring alongside medical problems. See ‘Further reading’ list below for more details.
-
(4) DO take account of how illness is considered within a person’s culture and familial systems; this is an important part of a holistic assessment. See ‘Further reading’ list for key papers that examine anxiety cross-cultural and cross cultural models of CBT.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S1754470X2400045X
Data availability statement
The data that support the findings of this study are available from the corresponding author, J.C., upon reasonable request. The data are not publicly available due to them containing information that could compromise research participant privacy and consent.
Acknowledgements
The study authors would like to extend thanks to all who contributed to this project including participants. The authors would also like to thank Mary-Jane Marffy for help with data analysis and Rita De Nicola and Chloe Lee for help in preparing the manuscript.
Author contributions
Jessica Colenutt: Data curation (lead), Formal analysis (lead), Investigation (lead), Methodology (lead), Resources (lead), Software (lead), Writing - original draft (lead), Writing - review & editing (lead); Jo Daniels: Conceptualization (lead), Supervision (lead), Writing - review & editing (supporting).
Financial support
This research received no specific grant from any funding agency, commercial or not-for-profit sectors.
Competing interests
The authors declare none.
Ethical standards
Ethical approval was given by the Health Research Authority (reference: 20/YH/0085) and University of Bath Psychology Research Ethics Committee (reference: 20-036). Any necessary informed consent to participate and for the results to be published has been obtained. The authors have abided by the Ethical Principles of Psychologists and Code of Conduct as set out by the BABCP and BPS.
Open access
Comments
No Comments have been published for this article.