نگاهی به ایجاد اضطراب سلامتی با استفاده از مدل سازی مخلوط عاملی در یک نمونه غیربالینی

Cognitive-behavioral models conceptualize health anxiety as a construct that varies in degree along a continuum rather than existing as nonpathological versus pathological classes or taxons. Only two studies have empirically evaluated the latent structure of health anxiety, both using taxometric statistical methods and both supporting its conceptualization as continuous (Ferguson, 2009 and Longley et al., 2010). We sought to further evaluate the latent structure of health anxiety using factor mixture modeling (FMM), which involved a combination of exploratory factor analysis (EFA) and mixture modeling that allowed comparison of models comprising one or more latent classes. Health anxiety symptom data were obtained from the Illness Attitude Scales (IAS) administered to 1768 university undergraduate students. Indicators of health anxiety, derived from EFA of IAS item data, included disease worry, disease conviction, health-related safety behaviors, fear of death, somatic focus, interference due to symptoms, and treatment seeking. FMM of these indicators suggested that health anxiety consists of two classes: (a) an “anxious” class comprising 81.4% of the sample and characterized primarily by somatic focus and interference due to symptoms, and (b) a “nonanxious” class comprising 18.6% of the sample with low scores on all indicators. Contrary to current conceptualizations and taxometric findings, the FMM results indicate the latent structure of health anxiety to be taxonic rather than continuous. Implications for the measurement and conceptualization of health anxiety are discussed and future research directions are highlighted.

Health anxiety often arises when bodily sensations or changes are believed to be indicative of serious disease. The magnitude of health anxiety can vary over time for a given individual, and levels can vary across people; consequently, contemporary cognitive-behavioral models conceptualize it along a continuum ranging from minimal to severe (Abramowitz et al., 2002, Salkovskis and Warwick, 1986 and Taylor and Asmundson, 2004). Mild to moderate degrees of health anxiety can be adaptive, motivating one to seek clinical care in cases in which clinical care is warranted (Asmundson, Abramowitz, Richter, & Whedon, 2010); however, both minimal and severe health anxiety can be maladaptive. Minimal health anxiety is associated with ignoring or minimizing the potential importance of bodily sensations and changes that may be indicative of disease, not seeking medical attention, and sometimes leading to disease progression or death (Taylor & Asmundson, 2004). Conversely, high levels of heath anxiety are characterized by preoccupation and worry that often lead to undue personal suffering, impaired social and occupational functioning, as well as over utilization of general and specialty health care services (Asmundson & Taylor, 2007). Health anxiety is characterized by several core cognitive, somatic, and behavioral features that typically manifest following periods during which one is stressed, seriously ill, or has suffered the loss of a family member (Barsky & Klerman, 1983) or after exposure to disease-related popular media (Taylor & Asmundson, 2004). The core cognitive feature is disease conviction; that is, the belief that bodily sensations and changes are due to disease processes rather than benign bodily perturbations, symptoms of minor ailments, or autonomic nervous system arousal. Other dysfunctional beliefs (e.g., the doctor has missed something critical) may accompany disease conviction and, together with disease-related preoccupation and worry, motivate maladaptive coping behaviors. Reassurance seeking and recurrent checking behaviors, while providing transient relief to distress (Haenen, de Jong, Schmidt, Steven, & Visser, 2000), perpetuate dysfunctional beliefs and are detrimental in the long term (Warwick & Salkovskis, 1990). People with high levels of health anxiety are often diagnosed with hypochondriasis, a DSM-IV somatoform disorder characterized by presentations of core features of health anxiety (Asmundson et al., 2010). Whether health anxiety is a construct that varies between people in degree along a continuum, as opposed to existing as nonpathological versus pathological classes or taxons, has important implications for theorists, researchers, and clinicians (Ruscio, Haslam, & Ruscio, 2006). Only two studies have empirically evaluated the latent structure of the health anxiety construct (Ferguson, 2009 and Longley et al., 2010). Ferguson (2009) used three procedures belonging to the applied mathematical approach of taxometrics (i.e., MAXimum EIGenvalue [MAXEIG; Meehl & Yonce, 1994], mean above minus below cut [MAMBAC; Meehl & Yonce, 1994], and latent mode [L-Mode; Waller & Meehl, 1998]) to examine this issue in a sample of 501 healthy individuals who completed the Whiteley Index (WI; Pilowsky, 1967). Results suggested that health anxiety is a dimensional construct. Longley et al. (2010) applied a similar selection of taxometric analyses to four symptom indicators derived from exploratory factor analysis (EFA) of several health anxiety measures (i.e., Illness Attitude Scales [IAS; Kellner, 1986 and Kellner et al., 1987]; Multidimensional Inventory of Hypochondriacal Traits [MIHT; Longley, Watson, & Noyes, 2005]; WI), completed by a sample of 1083 undergraduate students. As with Ferguson (2009), results suggested a dimensional construct. The findings of Ferguson (2009) and Longley et al. (2010) are consistent with contemporary cognitive-behavioral models, suggesting that health anxiety varies along a continuum. Conceptualizing health anxiety as dimensional diverges from the categorical approach used in numerous studies of health anxiety (e.g., Hadjistavropoulos et al., 1998, Hitchcock and Mathews, 1992 and Owens et al., 2004) and espoused by the DSM-IV (American Psychiatric Association [APA], 2000). Accordingly, additional empirical scrutiny of the latent structure of health anxiety using alternative statistical methods is warranted. Taxometric statistical methods are useful for understanding the latent structure of psychological phenomena, but have disadvantages relative to the alternate method of factor mixture modeling (FMM). First, because taxometric statistical methods were principally designed to estimate whether a model consisting of two latent classes (i.e., a taxonic model) has a better goodness-of-fit than a model consisting of a single latent class (i.e., a dimensional model), they are generally less informative with regard to whether a two-class model has a better or worse fit to the data than models consisting of three or more classes (e.g., a model of health anxiety comprising classes characterized by health anxiety that is either pathologically low, adaptive, or pathologically elevated). Second, taxometric statistical methods assume that indicators are uncorrelated within classes, whereas empirical investigations (using FFM) have shown that a better fitting model can be obtained when indicators are allowed to correlate within classes (e.g., Bernstein et al., 2010). Thus, compared to taxometric statistical methods, FFM has advantages for investigating the latent structure of health anxiety. The present study was conducted to further evaluate the latent structure of health anxiety using FMM, which entailed a combination of EFA and mixture modeling that allowed the comparison of models comprising one or more latent classes. The IAS – a trait measure of the construct (Hadjistavropoulos et al., 2004 and Stewart and Watt, 2000) considered by some to be the gold-standard (Sirri, Grandi, & Fava, 2008) – was used to measure health anxiety in a large sample of university students who completed the measure as part of participating in several other studies. Despite some debate (cf. Otto et al., 1998, Otto et al., 1992, Stewart and Watt, 2000, Taylor, 1992 and Taylor, 1995), there is evidence indicating that the construct assessed by the IAS and measures of arousal-reactive constructs (e.g., anxiety sensitivity) are distinct (Sirri et al., 2008 and Watt and Stewart, 2000). The base rate of clinical forms of health anxiety is not known; however, epidemiological studies suggest community past 12-month prevalence rates of between 0.4% (Bleichhardt and Hiller, 2007 and Looper and Kirmayer, 2001) and 4.5% (Faravelli et al., 1997) and a lifetime prevalence rate of approximately 1–5% (APA, 2000). Since one of the two prior taxometric studies of health anxiety (Ferguson, 2009) utilized a sample that may have been too small to detect a health anxiety taxon class had it existed, the empirical basis for concluding that health anxiety is either continuous or taxonic remains insufficient. Accordingly, the current study can be considered exploratory in nature.

3.1. Exploratory factor analysis Parallel analysis for both mean and 95th percentile eigenvalues indicated 7-factors, which accounted for 48% of the variance. The first 10 obtained eigenvalues were 7.03, 2.20, 1.81, 1.50, 1.43, 1.27, 1.17, 0.95, 0.86, and 0.80. Factor loadings and the interpretation of each factor (i.e., factor label) appear in Table 1. The interpretation of each factor was based on its salient loadings, particularly the highest loadings of each factor. The factor labels are largely self-explanatory, although some elaboration is warranted for two factors. The factor concerning health-related safety behaviors pertains to avoidance and checking behaviors (e.g., avoidance of harmful health habits, checking body), which are performed in order to prevent or limit disease. The factor concerning somatic focus pertains to awareness of, or attention to, bodily sensations (e.g., worry that pain indicates serious illness; worry about bodily sensations). Table 1. Factor loadings for 7-factor solution. Item 1. Illness worry 2. Disease conviction 3. Health-related safety behaviors 4. Fear of death 5. Somatic focus 6. Interference due to symptoms 7. Treatment seeking 1. Worry about health .56 .04 .14 −.05 .05 .02 .06 2. Worry about future illness .80 .15 −.04 .02 −.05 .05 −.02 3. Scared by thought of illness .57 −.07 .02 .29 .10 −.03 .02 4. Worry that pain indicates illness .40 .24 −.02 .00 .30 .02 .00 5. If pain persists, visits a physician .00 −.10 .12 .06 .21 .01 .38 6. If pain persists, worries about serious illness .12 .20 .05 .00 .39 .02 .05 7. Avoids harmful health habits (e.g., smoking) .02 −.08 .46 .05 .01 .05 −.09 8. Avoids unhealthy foods −.05 .02 .77 −.03 −.05 .00 .00 9. Checks body .16 .05 .33 .00 .13 −.06 .17 10. Believes he/she has an undiagnosed serious illness .04 .52 −.03 −.15 .11 .05 .06 11. Refuses to believe doctor when told he/she is healthy −.03 .46 .04 −.14 .06 .05 .04 12. Interprets doctor's report as evidence of illness −.22 .34 .07 .03 .03 .00 .01 13. Afraid of reminders of death −.07 .23 .04 .64 .05 .03 −.02 14. Fear of death .03 .02 −.03 .88 −.02 .02 .04 15. Afraid will die soon .10 .43 −.02 .39 .01 .00 .02 16. Afraid has cancer .22 .56 .03 .12 −.02 −.01 −.04 17. Afraid has heart disease .13 .55 .01 .07 −.01 .01 −.04 18. Afraid has some other serious illness .04 .59 .00 .11 .01 −.02 .03 19. Learns of an illness, then develops symptoms −.05 .41 −.06 −.02 .36 .00 .02 20. Unable to distract self from bodily sensations −.03 .04 −.06 .01 .72 .11 −.04 21. Worries about bodily sensations .05 −.01 .05 .03 .81 .00 .00 23. Frequency of medical appointments −.01 .10 −.01 .03 −.01 −.04 .66 24. Number of different doctors consulted .07 −.11 −.01 .00 .04 .02 .62 25. Frequency of medical treatment −.02 .04 −.02 −.01 −.08 .10 .68 27. Symptoms interfere with work functioning −.03 .00 .01 −.01 −.03 .73 .07 28. Symptoms interfere with concentration .02 .04 −.01 −.02 .01 .87 .01 29. Symptoms impair quality of life .04 −.03 .01 .07 .06 .80 −.02 Bold: salient (≥.30) loading. Table options Factor inter-correlations ranged from .01 to .48 (M = .23). When the factor scores were factor analyzed, using RML and parallel analysis, a single factor was obtained, accounting for 27% of the variance: eigenvalues: 2.56, 1.04, 0.99, 0.76, 0.66, 0.51, and 0.48. The loadings of the factors (F) on the higher-order factor were as follows: F1 = .63, F2 = .68, F3 = .11, F4 = .43, F5 = .68, F6 = .51, F7 = .38. Note that the higher-order factor accounted for only about a quarter of the variance. The majority of variance (73%) was due to the unique (nonshared) aspects of each factor. Factor 3 (health-related safety behaviors) had the lowest loading on the higher-order factor, with only 1% of its variance due to the higher-order factor. Given that the higher-order factor does not capture the bulk of the variance in the seven factors, the latter rather than the higher-order factor were used in the analyses described in the following section, which address the question of whether there are distinct classes of health anxiety. The seven factors were defined as the essential building blocks for determining whether health anxiety consists of discrete classes. 3.2. Factor mixture modeling Table 2 shows the FMM results. The best-fitting model consisted of two classes in which indicators were correlated within each class. This model had the best fit on four of five fit indices. If FMM had been conducted with uncorrelated within-class correlations in which only two classes were compared, the results would similarly have supported a two class model (see Table 2). Table 2. Fit indices for factor mixture modeling. Classes Factors correlated within classes Akaike information criterion Bayesian information criterion (BIC) Sample size adjusted BIC Bootstrap likelihood ratio test (p-value) Entropy 1 Yes 32,501.92 32,693.55 32,582.36 – – 1 No 35,063.10 35,139.76 35,095.28 – – 2 Yes 31,258.91 31,647.66 31,422.10 <.001 0.95 2 No 32,672.04 32,830.82 32,738.69 <.001 0.75 3 Yes 31,663.09 32,248.96 31,909.02 <.001 0.61 3 No 31,735.52 31,976.44 31,836.66 <.001 0.82 Bold = best-fitting model; (–) = not applicable. Table options As mentioned, FMM provides estimates of the probability that a person would be classified in one class or another. These estimates were used to classify respondents into classes. For the best-fitting two class model, n = 324 for class 1 and n = 1440 for class 2. Table 3 shows the comparison of classes on the seven factors, as well as on the total score on the IAS (defined as the unit-weighted sum of the 27 items). The latter is presented because an IAS score, unlike the factor scores, has a clearly anchored scale; that is, low IAS scores indicate low health anxiety and high scores indicate high health anxiety. Factor scores do not have this inherent scaling. Low factor scores could indicate either low or high health anxiety, depending on the factor scaling. Therefore, the IAS total score provides a useful reference point for interpreting the factor scores. Table 3. Comparison of the 2 classes on factor scores and IAS total score. Factor Class 1 (n = 324): M (SD) Class 2 (n = 1440): M (SD) t(1762) η2 1. Illness worry −0.45 (0.90) 0.10 (0.99) 9.17**** .046 2. Disease conviction −0.42 (0.57) 0.09 (1.05) 8.51**** .040 3. Health-related safety behaviors −0.11 (1.14) 0.03 (0.96) 2.29* .003 4. Fear of death −0.32 (0.85) 0.07 (1.02) 6.54**** .024 5. Somatic focus −0.71 (0.80) 0.16 (0.97) 14.97**** .113 6. Interference due to symptoms −1.34 (0.06) 0.30 (0.85) 34.67**** .406 7. Treatment seeking −0.54 (0.83) 0.12 (0.99) 11.19**** .066 IAS total score (sum of items) 24.20 (9.28) 38.70 (12.78) 19.30**** .175 * p<.05. **p<.01. ***p<.005. **** p<.001. Table options Table 3 shows that class 1 (“nonanxious”) had significantly lower scores on all factors relative to class 2 (“anxious”); however, the effect sizes (η2) indicate that some of the significant between-class differences were insubstantial. The η2 statistic is the proportion of variance in scores that is due to class membership. Table 3 shows that for most factors, class membership accounted for less than 7% of variance for most factors (i.e., factors 1, 2, 3, 4, and 7). Higher proportions of variance due to class membership were found for Factor 5 (somatic focus; 11%) and particularly Factor 6 (interference due to symptoms; 41%). In other words, the classes were primarily distinguished in terms of the degree of bodily preoccupation and functional impairment due to distressing bodily sensations. There were significant between-class differences in the age of participants, t(1762) = 2.11, p < .05; however, the magnitude of the difference was insubstantial, as indicated by the effect size, η2 = .003. That is, class membership (class 1 vs. class 2) accounted for only 0.3% of variance in age. The mean (SDs) ages in years were as follows: class 1 = 20.8 (3.8), class 2 = 20.4 (3.6). There were larger differences in gender distribution, in which the proportion of women was 63% in class 1 and 73% in class 2; χ2(1) = 13.14, p < .001, η2 = .086.