Dimensionality of the Whiteley Index: Assessment of hypochondriasis in an Australian sample of primary care patients

Abstract

Objective

The Whiteley Index (WI) is a widely used instrument for measuring hypochondriacal worries and beliefs. Several studies explored the structural validity of the WI obtaining contrary results concerning the number of factors as well as the item composition. The main aim of this study is to compare factor solutions from previous studies to draw conclusions about the most valid scale model of the WI for administration in primary care.

Methods

Weighted least squares (WLS) confirmatory factor analyses of the WI were conducted. The sample in study consisted of 1800 patients from primary care practices. Seven different models were compared, including single- and three-factor conceptualisations.

Results

A seven-item, single-factor model best described the data, while three-factor models were clearly inadequate.

Conclusions

Results support a one-dimensional conceptualisation of the WI and suggest a certain subscale of the WI, the WI-7, to constitute the most psychometrically sound scale for use as a screening instrument for hypochondriasis in primary care. In addition to psychometric considerations, the brevity and simplicity of the WI-7 also make it attractive as a screening tool in the context of primary care. A cutoff score of 2/3, calculated on the basis of general practitioners' diagnoses, yielded the best balance of sensitivity and specificity in the present study.

Introduction

Hypochondriasis is a relatively prevalent disorder in primary care, with prevalence estimates of between 3% and 6.3% [1], [2], [3]. The economic effects of hypochondriasis are tremendous [4]. Even at subclinical levels of severity, hypochondriacal worries significantly increase health care utilization at the level of primary care services, associated pathology testing, and secondary care services [5], [6]. Alongside this substantial community burden, the disorder has a major impact on interpersonal relationships, in general [7], [8], and on the doctor–patient relationship in particular. Despite the patient's experienced symptoms, often, no physical abnormalities can be found, resulting in consultations that are experienced by both doctor and patient as frustrating and unsatisfactory [4], [9].

Given the substantial negative impact of hypochondriasis for patients, doctors, and the community, it is surprising that this disorder remains extremely underdiagnosed in primary care [10], [11]. There appears to be need for a short screening instrument able to validly and reliably identify potential hypochondriacal patients in primary care.

Pilowsky [12] attempted to develop such a measure, the Whiteley Index (WI), by comparing responses of hypochondriacal and nonhypochondriacal patients on a range of items designed to tap general health worries. Those items discriminating significantly between groups were chosen for further analyses. The final scale comprised 14 items said to measure hypochondriacal fears and beliefs. Exploratory principal component analysis (EPCA) has yielded three factors, interpreted as ‘disease phobia, ‘bodily preoccupation’, and ‘disease conviction’ [12]. The factor disease phobia refers to the fear of having or developing a serious illness. The second factor, bodily preoccupation, describes the subjective impression of the patient as suffering from multiple bodily symptoms and pains. The third factor, disease conviction, is said to represent the strength of the patients' belief that they suffer from a serious illness.

The structural validity of the WI has been explored and challenged in several studies. A three-factor solution was found by Rief et al. [13], Hiller et al. [14], and Hinz et al. [15]. However these three factors did not correspond with the original factor solution of Pilowsky. Furthermore, a one-factor solution was proposed by Speckens et al. [16] based on investigations using three different samples. The latter four studies used EPCA. In another study, Fink et al. [17] identified a seven-item scale with good psychometric properties. The Whiteley-7 scale fitted into a modified Rasch model, which means that the slopes of the item–characteristic curve are equal and the pattern of responses to the items supports an additive scale. Table 1 gives a detailed overview of the range of differing factor solutions discussed.

Several reasons can be hypothesized for the contrary results. First, the study samples stem from dissimilar populations. Originally, Pilowsky [12] validated the WI in a sample of 200 inpatients of a psychiatric hospital, so that the three-factor solution might be a specific result of a tertiary care sample. Indeed, two of the studies that could replicate the three-factor solution were run in tertiary care [13], [14], although a third confirmation came from a general population survey [15]. Thus, the number of factors might be dependent upon the diagnostic compositions of the samples in study.

Second, different methods were employed. Whereas Fink et al. [17] used exploratory and latent trait methods to remove weak items and to confirm a short scale, all other studies made use of EPCA. Despite the frequent use of EPCA, it is assumed to be an inappropriate tool for the handling of dichotomous item responses, resulting in an underestimation of item intercorrelations, inconsistent estimates of parameters, and standard errors [18], [19], [20]. Weighted least squares (WLS) factor analysis employing the asymptotic covariance matrix of the tetrachoric correlations is recommended instead. Third, criteria for assigning items to factors differ between studies. For example, Speckens et al. [16] accepted factor loadings >.15 in their proposed one-factor model, although the authors flagged three items that contributed little to the common factor. Rief et al. [13] and Hiller et al. [14] accepted factor loadings >.40. Fink et al. [17] and Hinz et al. [15] employed higher standards, with loadings >.50.

The main aim of this study is to compare differing factor solutions to draw conclusions about the most valid scale model of the WI for the administration in primary care. Therefore, we firstly present endorsement frequencies, item–total correlations, and internal consistency for the WI on data from a primary care sample. Second, we compared seven different factor models of the WI by means of confirmatory factor analysis (CFA). Third, psychometric properties and a proposed cutoff score for the most valid scale are presented.