Confirmatory factor analysis of the causal illness attribution scale in Chinese patients with multiple somatic symptoms

Abstract

Causal illness attributions influence how individuals cope with somatic symptoms and illnesses. Dimensions of causal symptom attributions have been examined in Western cultures with the subscale ‘causes’ of the revised Illness Perception Questionnaire (IPQ-R). Some previous studies have identified a stronger somatic attribution style in Asian patients. In this study it was examined if the factorial structure of causal attributions identified in Western populations can be identified in a large Chinese sample of patients presenting with somatic symptoms. We recruited 665 patients aged at least 18 who were visiting the hospital for reasons of treatment from departments of traditional Chinese medicine (TCM), neurology (Biomedicine), and psychosomatic medicine in six hospitals across China. All subjects completed the Patient Health Questionnaire (PHQ) and the causes subscale of the IPQ-R. We split the data-set by chance in two parts. On the first subsample, we conducted a confirmatory factor analysis (CFA) to check the fit of the originally proposed 4-factor structure and an exploratory factor analysis (EFA). The factor structure indentified in the EFA was rechecked with a CFA in the second subsample. The originally proposed 4-factor-model of the IPQ-R subscale causes showed no adequate fit in the first subsample. The EFA revealed two factors, psychological attributions and risk factors. The CFA in the second sample showed mediocre fit indices (RMSEA = .098, CFI = .923). For the Chinese sample we propose a two-factor structure for IPQ-R causes scale. As in other studies, we identified the relatively stable factor psychological attributions, indicating no fundamental differences in illness attributions between Western and Chinese samples.

Keywords:

• Illness perception questionnaire
• causal illness attribution
• China
• factor analysis

Introduction

Having an illness or experiencing chronic somatic symptoms has a significant impact on our psychological well-being. Especially excessive symptom-related thoughts play a crucial role since – regardless of a known or unknown etiology of the somatic symptoms – patients try to explain their symptoms (Groben & Hausteiner, 2011). Thus, illness attributions are an important psychological feature in the context of physical symptoms. One of the most impactful models of symptom attributions is Leventhal’s Self-Regulatory Model (Leventhal, Leventhal, & Contrada, 1998). It conceptualizes the cognitive representation of illness as a multidimensional concept with five dimensions: identity (the symptoms or labels that patients use to describe the illness), timeline (expected course of the illness), cause (the perceived causes of the illness), consequence (severity of the illness), and control (expected cure). Since these components of illness representations make up the patients perception of their illness, they are often referred to as illness perception (Weinman, Petrie, Moss-morris, & Horne, 1996).

Among these five dimensions, causal attributions have proven to be of special relevance because they influence internal processes associated with an illness such as cognitions and consequent coping behaviors in the context of illness (Roesch & Weiner, 2001) and an individual’s emotional response (Cameron, Petrie, Ellis, Buick, & Weinman, 2005). Causal attributions for example are highly correlated with symptoms of depression and anxiety (Rief, Nanke, Emmerich, Bender, & Zech, 2004; Robbins & Kirmayer, 1991). Studies in the context of health locus of control showed that believing in the controllability of a disease was associated with less depression in patients with end-stage renal disease (Christensen, Turner, Smith, Holman, & Gregory, 1991). Causal attributions influence external processes associated with an illness, such as preventive health behavior (Weinman, Petrie, Sharpe, & Walker, 2000), symptom report, illness consequences (Rief et al., 2004), and treatment outcomes (Henningsen, Jakobsen, Schiltenwolf, & Weiss, 2005; Kolk, Schagen, & Hanewald, 2004). These findings are in line with research in the context of health locus of control showing that people, who believed that their health status is determined by their own behavior are more likely to carry out healthy behavior than people who think that fate, luck or chance influence their health status (Wallston, 2005).

In spite of this research, most of the models concerning illness attributions are based on studies that have been conducted with Western populations. Investigating culturally specific characteristics of symptom attributions might require a modification of these models because studies have shown cultural differences in illness behavior and illness attributions (Gureje, Simon, Ustun, & Goldberg, 1997; Kleinman, 1986). Asian patients were often less likely to report psychological symptoms while emphasizing physical suffering in comparison to Western patients, who were more likely to attribute episodes of mixed symptoms as caused by psychological processes (Karasz, Dempsey, & Fallek, 2007). Additionally a variety of culture bound syndromes as listed in DSM-5 (American Psychiatric Association, 2013) shows the necessity of culture specific research. For example, the concept of shenjing shuairuo (SJSR) is very common in China and is often diagnosed in patients with sleep disorders, dizziness, headaches, exhaustion, and difficulties in concentrating. According to Traditional Chinese Medicine (TCM), symptoms of SJSR are caused by an imbalance of energy flow or ‘qi’ and a disrupted harmony of the ‘vital organs’ (Cheng, 1989). Nowadays most Chinese psychiatrist orient themselves toward the DSM and ICD classification systems and SJSR is only rarely diagnosed (Lee, 1999; Lee & Kleinman, 2007). However, there is a need for intercultural investigations of psychological variables associated with somatic symptoms, in particular illness attributions, which has already been addressed by a variety of Chinese researchers (Chen, Tsai, & Lee, 2008; Fritzsche et al., 2013; Mo et al., 2015; Yan et al., 2011; Yu, Tan, Song, & Yang, 2016). Yet, these studies focused only on patients diagnosed with a specific disorder or disease (hypertension, medically unexplained symptoms, substance dependence, acute myocardial infarction, androgenetic alopecia) and there has only been one study investigating illness attributions of experiences of everyday symptoms in general hospital outpatients (Zhang et al., 2014). Especially in China, there has been a variety of interesting validational studies including the causes subscale of the IPQ-R in China. In two studies including patients with myocardial infarction, the authors validated an adapted Chinese version of the IPQ-R (Song et al., 2007) that includes six additional items adjusted to this patient group. They identified six factors with exploratory factor analyses (Yan et al., 2014). A study including patients with hypertension revealed good model-fit of a four-factor structure (Chen et al., 2008). These examples show that the IPQ-R causes scale has only been validated in adapted forms for patients with specific medical or psychiatric conditions in China (hypertension and acute myocardial infarction). Thus, we wanted to examine if the factor structure suggested by the authors of the original IPQ-R causes scale (Moss-Morris et al., 2002) can be confirmed in a sample of general hospital outpatients in China presenting common symptoms or if the structure of illness attributions differs from the factorial structure identified in Western populations. Investigating different samples is important in order to detect differences that might be specific to subsamples. Illness attributions of patients with specific medical conditions might vary substantially from general hospital outpatients because both groups differ with regards to reported symptoms. As suggested by the authors of IPQ-R (Moss-Morris et al., 2002), we performed an exploratory factor analysis on a subsample of our total sample in order to identify the factor structure for our specific group of patients. Additionally we tried to confirm this factor structure with a confirmatory factor analysis on a second subsample.

In many studies investigating causal illness attributions the number and pattern of these attributions is being discussed. In Western countries the number of causal attributions seems to be positively related to the number of symptoms patients reported (Rief et al., 2004). One Chinese study showed no effect of number or severity of symptoms on illness attributions (Fritzsche et al., 2013) and another study found an effect of the number of symptoms on illness attributions (Chen et al., 2008). These findings show that high symptom reporters do not necessarily show a high number of causal illness attributions. It is possible, that they attribute a variety of symptoms to a few causal factors (for instance an undetected illness). Thus, the number of attributions can give us information about how complex patients’ explanatory model of their symptoms is. It can be assumed that partly patients would attribute a variety of symptoms to one cause whereas other patients take a variety of potential causes into account. Thus, as a second goal of the study was to find out whether symptom severity and the number of causal illness attributions are correlated.

Some Western studies showed that symptom report is influenced by personality variables (De Ridder, Fournier, & Bensing, 2004), mood (Bogaerts, Janssens, De Peuter, Van Diest, & Van den Bergh, 2010) and cognitive factors (Gunstad & Suhr, 2004). These findings suggest that individuals reporting a high level of symptom severity compared to those reporting a low level of symptom severity might differ in a variety of underlying psychological variables. Accordingly, some studies with high symptom reporters show that they show predominantly somatic attribution styles compared to healthy controls (Martin et al., 2007; Rief et al., 2004). However, other studies show that they report multiple causal attributions like healthy controls (Groben & Hausteiner, 2011). One study in China with high symptom reporters has shown that physicians and patients from all areas of medicine – biomedicine, Psychosomatic Medicine and Traditional Chinese Medicine (TCM) – most frequently reported psychological attributions (Fritzsche et al., 2013). These findings are in contrast to other studies reporting that Asian patients in general are likely to neglect psychological causes and emphasize somatic explanations of their illness (Karasz et al., 2007; Kleinman, 1986). However, no studies have been conducted so far investigating causal illness attributions in general hospital outpatients with high compared to low symptom reporting in China. Thus, as a third goal of this study was to compare low versus high symptom reporter in terms of their illness attributions.

Method

Participants, study procedure, and setting

For the current study a sample of 689 participants was recruited. Of this total sample 491 participants were recruited for a multicenter cross-sectional study assessing the relationship between somatic symptom severity, quality of life and psychobehavioral characteristics in Chinese general hospital outpatients (Zhang et al., 2014). This multicenter trial comprised 9 outpatient departments from general hospitals specialized in Psychosomatic Medicine, Traditional Chinese Medicine (TCM), Gastroenterology, or Neurology in six hospitals across China: Chengdu (West China Hospital of Sichuan University), Beijing (Union Hospital), Kunming (Red Cross Hospital) and one specialized hospital in Shanghai (Tongji Hospital, Shanghai Mental Health Centre, Dong Fang Hospital) (see Figure 1). Additionally 198 subjects of the second sample were recruited from West China Hospital of Sichuan University in Chengdu (see Figure 1). Thus, all patients visited the hospital for specialized treatment and not because of acute illness such as influenza. The departments of psychosomatic medicine are comparable to Western standards and treat patients with psychiatric diseases such as depression, anxiety disorders, sleep disorders, somatization disorders etc.

Figure 1. Study flow chart.

Notes: ^aWe excluded all participants with more than 15% of missing items in PHQ-15 or IPQ-R causes scale from the analysis. TCM = Traditional Chinese Medicine. SOM+ = Participants with PHQ-15 score ≥ 10. SOM− = Participants with PHQ-15 score ˂10.

Participants were recruited in 2011 and 2012. All participants who entered one of the departments on a previously determined screening day were informed about the study and asked to participate. After providing written informed consent, research assistants assessed symptom severity using the somatic symptom scale of the Patient Health Questionnaire (PHQ-15) as a self-report measure (Kroenke, Spitzer, & Williams, 2002). Amongst others, participants completed questionnaires measuring causal illness attributions, demographic variables including basic information about the illness (single items measuring the number of doctor visits during the past 12 months and whether the complaint has caused an impairment in daily life or not) alone in the waiting room before another interview was conducted which is reported elsewhere (Zhang et al., 2014). Subjects were assigned to one of two groups according to their PHQ-15 score (group ‘SOM+ ‘ with PHQ-15 ≥ 10, group ‘SOM−’ with PHQ-15 ˂10).

Participants were included when they were aged at least 18, visited the hospital for reasons of treatment (i.e. not only picking up a prescription or visiting a patient) and had sufficient reading and writing skills. We excluded participants from the study if they, were cognitively impaired (e.g. organic brain damage, dementia), had a current psychosis or acute suicidal tendencies. Postgraduate medical students in psychiatry served as research assistants and were supervised by the heads of each department. Additionally a German research team supervised data management. The study was approved by the ethics committees of Shanghai Dong Fang Hospital (China) and the University Medical Centre Freiburg (Germany) and has therefore been performed in accordance with the ethical standards laid down in the 1964 Declaration of Helsinki and its later amendments.

Measures

Somatic symptom scale of the patient health questionnaire (PHQ-15)

The PHQ-15 (Kroenke et al., 2002) assesses the severity of 15 somatic symptoms such as stomach or back pain, dizziness, nausea, constipation or trouble sleeping. Participants have to answer on a 3-point scale how much they were bothered by each symptom during the last 4 weeks (0 = ‘not bothered at all’, 1 = ‘bothered a little’, 2 = ‘bothered a lot’). The answers add up to a total score with a range from 0 to 30 serving as an indicator for overall symptom severity. For detecting clinically relevant somatic symptom distress, a cutoff score of 10 points has been shown to have the highest predictive validity resulting in a sensitivity of 80.2% and specificity of 58.5% (Körber, Frieser, Steinbrecher, & Hiller, 2011). In the current study this cutoff score was used to create a subgroup of high symptom reporters and a subgroup of low symptom reporters. Studies on Chinese samples have shown, that the Chinese version of the PHQ-15 exhibits satisfactory reliability and validity in the general population (Lee, Ma, & Tsang, 2011; Qian, Dehua, Xiaoyan, & Chunbo, 2014). The reliability of the PHQ-15 total score was Cronbachs’s α = .782 in our sample (N = 665).

Revised illness perception questionnaire (IPQ-R)

The IPQ-R (Moss-Morris et al., 2002) assesses cognitive representations of illness. The original version consists of three sections: identity, illness representation, and causes. Since we wanted to focus on the patients’ causal illness attributions, we only used the causes subscale (IPQ-R) in our study. This scale consists of 18 items explaining different possible factors influencing illness such as stress or worry, family problems, germs, pollution, ageing or accident (see also Table 2). Participants were asked to consider what may have been the cause of their illness and rate a list of potential causes on a five-point Likert type scale: 1 = strongly disagree to 5 = strongly agree. The authors encourage other researchers to modify this scale in order to suit specific illnesses, cultural settings or populations. They also suggest that factor analysis can be conducted in studies with sufficient sample size to identify groups of causal beliefs that can be used as subscales (Moss-Morris et al., 2002). The original scale consists of four subscales: psychological attribution, risk factors, immune system, and accident or chance that accounted for 57% of the total variance (Moss-Morris et al., 2002). A high score on these dimensions of the causes scale indicates that participants consider these dimensions as a relevant cause for their illness. However, comparisons are only possible on a descriptive level since the number of items is highly variable between the scales. The IPQ-R does not provide a total score defining the total number of illness attributions. Mean scores of the identified subscales can be computed to compare their relevance in the sample. A psychometric validation of the Chinese version of the IPQ-R shows that it exhibits satisfactory reliability and validity (Chen et al., 2008).

Statistical analyses

We excluded 24 cases from the original data-set with 689 participants because they had more than 15% of missing values in PHQ-15 or IPQ-R causes scale. For all other cases missing values were imputed using the expectation maximization technique. The final data-set included 665 participants. A probability value of 5% was used as the limit for Type I error. Data analyses were conducted using SPSS 22.0 and M-Plus Version 7. We performed post hoc independent t-tests to identify differences between male and female participants in terms of their illness attributions concerning smoking and alcohol.

To analyze the factor structure of the causes subscale of the IPQ-R in our sample, we first split the data-set by chance in two parts. On subsample 1 (n = 332), we tested with a confirmatory factor analysis (CFA) if all 18 items load on one general factor. Second, we tested the original 4-factor structure suggested by the authors of the questionnaire (Moss-Morris et al., 2002, p. 6) with another CFA on subsample 1 (n = 332).

In the case that the originally postulated 4-factor structure does not fit sufficiently our data we decided to perform an exploratory factor analysis (EFA) on subsample 1 in order to identify latent constructs underlying our data. To test for the assumptions needed to compute an EFA, we used an anti-image correlation matrix. According to a suggestion by Kaiser and Rice (1974), items with a measure of sampling adequacy less than .70 were excluded from further analyses. Additionally we calculated the Kaiser-Meyer-Olkin measure of sampling adequacy and Bartlett’s test of sphericity to assure that the data-set was suitable for factor analyses. Since the Kaiser criterion is an error-prone extraction method, in the sense of overestimating the number of extracted factors (Costello & Osborne, 2005), we used Horn’s parallel analysis (Horn, 1965) to determine the number of extracted factors. We used the method of common factor analysis as a standard procedure in Mplus. We applied oblique promax rotation of the resulting components of the EFA because we expected correlations between the latent factors (Costello & Osborne, 2005).

Then we performed another CFA on subsample 2 to confirm the factor structure we identified with the EFA on subsample 1. As measures for the goodness of fit of the CFAs we report the following fit indices that are often reported in structural equation modeling literature (Kline, 2010): the root mean square error of approximation (RMSEA), the weighted root mean square residual (WRMR), the comparative fit index (CFI), and the Tucker-Lewis Index (TLI). Different cut-off levels are suggested for these fit indices. For RMSEA values less than .06 indicate good fit, values less than .08 reflect adequate fit, and values between .08 and .10 demonstrate mediocre fit while values higher than .11 demonstrate inadequate fit (Hu & Bentler, 1999). Regarding WRMR values less than 1.0 indicate good fit (C. Yu, 2002). For the CFI values above .90 represent an acceptable fit while values close to .95 are indicative for a good fit (Hu & Bentler, 1999). Regarding the TLI values higher than .95 are representative for a well-fitting model (Hu & Bentler, 1999). Since we used non-nested models it was difficult to compare the fit indices statistically.

In order to compare SOM+ and SOM− participants in terms of their illness attributions, we assigned participants with a PHQ-15 score of at least 10 to the SOM+ group and participants with a score less than ten to the SOM− group. Since the IPQ-R does not provide a total score of illness attributions, we computed the total numbers of agreed illness attributions (4 ‘agree’ or 5 ‘strongly agree’) and disagreed illness attributions (1 ‘strongly disagree’ or 2 ‘disagree’) for every participant leaving out those causes that participants neither ‘agree(d) nor disagree(d)’ with (3). To assess correlations between the PHQ-15 mean score and the number of illness attributions (agreed and disagreed) we computed the Pearson product moment correlations. Finally a mean score for each of the two factors of the IPQ-R subscale ‘causes’ identified in the first EFA was computed. We conducted multivariate analyses of covariance (MANCOVA) in order to analyze difference between SOM+ and SOM− controlled for age and sex as covariates. Additionally we calculated partial eta squared (η ² _p ) as an indicator for the effect size.

Results

Participants characteristics

A total of 665 participants were included in the analysis. Participants of the SOM+ group and the SOM− group did not differ in terms of health insurance, residence, education and age. The level of formal education was comparable to the general Chinese population (Li, Whalley, & Xing, 2014). The SOM+ group showed significantly higher scores than the SOM− group regarding the number of female patients, number of doctor visits during the past 12 months, impairment in daily life, symptom duration, somatic symptom intensity and illness perceptions. The results can be derived from Table 1. Results of an independent t-test show that males considered alcohol, t(662)= 7.67, p < .001, or smoking, t(662)= 9.48, p < .001, more often as a cause for their illness than females.

Table 1. Characteristics of the SOM+ group and the SOM− group.

	SOM+, n = 334	SOM−, n = 331	χ^2-value (df)	p-values^a
	n (%)	n (%)
Gender (female)	247 (73.7%)	184 (55.8%)	25.17(1)	<.001
Health insurance (yes)	277 (82.9%)	269 (81.3%)	.163 (1)	.686
Residence (urban)	246 (73.7%)	246 (74.3%)	.004 (1)	.947
Educational level			6.42(3)	.093
Elementary school	56 (16.8%)	47 (14.2%)
Middle school	83 (24.9%)	71 (21.5%)
High school	83 (24.9%)	72 (21.8%)
University degree or higher	107 (32.0%)	138 (41.7%)
Number of dr. visits			21.94 (4)	<.001
0	21 (6.3%)	49 (14.8%)
1–2 times	61 (18.3%)	69 (20.8%)
3–10 times	116 (34.7%)	123 (37.2%)
11–20 times	56 (16.8%)	38 (11.5%)
>20 times	78 (23.4%)	49 (14.8%)
Impairment daily life (yes)	250 (74.9%)	164 (49.5%)	40.52 (1)	<.001
Symptom duration			31.99 (4)	<.001
< 4 weeks	19 (5.7%)	65 (19.6%)
4 weeks–6 months	66 (19.8%)	56 (16.9%)
6 months–1 year	43 (12.9%)	49 (14.9%)
1–2 years	55 (16.5%)	44 (13.3%)
> 2 years	149 (44.6%)	115 (34.7%)
	M (SD)	M (SD)	t-value (df)	p-values^b
Age	44.60(15.71)	45.25 (15.93)	.53 (663)	.595
Somatic symptom intensity (PHQ-15 Total score)	13.85 (3.45)	5.29 (2.48)	−36.73 (663)	<.001
Illness perceptions (IPQ-R causes total score)	47.72 (9.55)	42.67 (11.02)	−6.31 (663)	<.001

Notes: M = Mean; SD = Standard Deviation. PHQ-15 = Somatic symptom scale of the Patient health questionnaire. IPQ-R = Revised Illness Perception Questionnaire. SOM+ = participants with a PHQ-15 score ≥ 10. SOM− = participants with a PHQ-15 score < 10.

^a p-value based on Pearson Chi-square test (χ²).

^b p-value based on t-test.

The factor structure of the IPQ-R causes scale

In a first step, a general model including all 18 items was analyzed in subsample 1 (n = 332). The χ²-test was highly significant, χ²(df = 135, N = 332) = 1817.265, p < .001, and none of the goodness-of-fit indices indicated a good model-fit, RMSEA = .194, WRMR = 3.058, CFI = .646 and the TLI = .599.

In a second step the first-order, 4-factor model suggested by Moss-Morris and colleagues (2002) was tested on subsample 1. None of the fit indices showed adequate fit of the model, RMSEA = .124, WRMR = 1.945, CFI = .861 and the TLI = .835.

Consequently, we conducted an EFA on subsample 1. Bartlett’s test (χ² = 1664.70, df = 153, p < .001) and the KMO criterion (KMO = .809) showed excellent suitability of the data for this analysis. However, item statistics for item 14 (alcohol consumption) and item 15 (tobacco use) indicated low means and standard deviations (see Table 2). They also showed measures of sampling adequacy in the anti-image correlation matrices below .70 (MSA = .637; MSA = .660) indicating that they are not suitable for factorial analyses (Kaiser & Rice, 1974). Therefore, we decided to eliminate these two items from the analysis and evaluated a 2-factor solution with 16 items on subsample 1. The EFA extracted two components that were retained by Horn’s parallel analysis. After oblique promax rotation the items stress or worry, own behavior, mental attitude, family worries, overwork, emotional state and personality loaded on one factor that could be termed psychological attributions, the items hereditary, germ, diet, chance, poor past medical care, pollution, ageing, accident and immunity loaded on a factor that could be characterized as risk factors. Table 2 shows the factor loadings for the EFA with two components. Correlation of the factors was r = .333.

Table 2. Results of the common factor analysis (subsample 1, n = 332): factor loadings with promax rotation.

Item	Description	Factor 1	Factor 2	M (SD)
	What do you consider may have been the cause of your illness?	Psychological attributions	Risk factors
Item 1	Stress or worry	.751	−.147	3.18 (1.30)
Item 8	My own behavior	.503	.248	2.81(1.22)
Item 9	My mental attitude – e.g. Thinking about life negatively	.804	.030	2.67 (1.27)
Item 10	Family problems or worries caused my illness	.719	.055	2.79 (1.28)
Item 11	Overwork	.497	.088	2.86 (1.28)
Item 12	My emotional state – e.g. Feeling down, lonely, anxious, empty	.891	−.117	3.03 (1.28)
Item 17	My personality	.749	−.021	2.91 (1.25)
Item 2	Hereditary – it runs in my family	.108	.431	2.14 (1.10)
Item 3	A germ or virus	−.078	.758	2.08 (1.02)
Item 4	Diet or eating habits	−.021	.700	2.44 (1.19)
Item 5	Chance or bad luck	.003	.414	2.30 (1.10)
Item 6	Poor medical care in my past	.027	.600	2.23 (1.10)
Item 7	Pollution in the environment	−.132	.703	2.46 (1.20)
Item 13	Ageing	−.042	.506	2.50 (1.20)
Item 16	Accident or injury	.036	.577	1.91 (.98)
Item 18	Altered immunity	.050	.456	3.15 (1.25)
Item 14	Alcohol	^1	^1	1.87 (1.01)
Item 15	Smoking			1.86 (1.02)
M (SD)		2.90 (.88)	2.36 (.62)

Bold figures indicate each item’s highest factor loading.

Notes: M = Mean; SD = Standard Deviation.

¹ Item 14 and 15 had to be excluded from the factorial analysis due to poor item characteristics.

Additionally, we tested the fit of the 2-factor model on subsample 2 after eliminating item 14 and 15. Of the fit indices RMSEA = .098 and WRMR = 1.501 reflected mediocre fit while CFI = .923 and TLI = .910 reflected adequate fit.

Because the model-fit was not satisfactory for all fit-indices we lastly decided to exclude all items with a low factor loading (<.40) from the analysis: item 2, 5, 11 and 18. However, the fit indices of the 2-factor model with 12 items did not change substantially compared to the previous 2-factor solution with 16 items: RMSEA = .093, WRMR = 1.223, CFI = .957 and TLI = .947. Thus, we rejected this model and would recommend using a 2-factor model with 16 items and analyze items 14 and 15 separately. The correlation of the subscales was r = .320.

Differences between SOM+ and SOM− in illness attributions

SOM+ participants had higher scores than SOM− participants on the illness attribution factors psychological attributions, F(1,660)=42.64; p < .001; ${\eta }_p^2=.061$, and risk factors, F(1,660)=29.90; p < .001; ${\eta }_p^2=.043$, (F scores are controlled for age and sex as covariates). Symptom severity (assessed with the PHQ-15) was significantly positively correlated with the number of agreed illness attributions (r = .30; p < .001) and negatively correlated with the number of rejected illness attributions (r = −.28; p < .001). This result indicates that patients reporting more symptom severity considered more causal illness attributions. In both groups, SOM+ and SOM−, participants mainly chose items of the scale psychological factors as the cause for their illness (see Table 3).

Table 3. Results of the ANCOVAs in order to compare the SOM+ and SOM− group in terms of aspects of illness attributions (N = 665).

	SOM+, n = 334	SOM−, n = 331
	M (SD)	M (SD)	F-values	p-values	${\eta }_p^2$	Cronbach’s α
IPQ-R component 1 (7 items) psychological attributions	3.11 (.85)	2.68 (.92)	42.6	<.001	.061	.844
IPQ-R component 2 (9 items) risk factors	2.48 (.60)	2.23 (.65)	29.9	<.001	.043	.732

Notes: M = Mean (all items of the scale were summed up for each participant and divided by the number of items, then we computed the group mean); SD = Standard Deviation; IPQ-R = Revised Illness Perception Questionnaire causes subscale. ${\eta }_p^2$= partial eta squared. SOM+ = participants with a PHQ-15 score ≥ 10. SOM− = participants with a PHQ-15 score < 10, α = Cronbach’s alpha.

Discussion

The first aim of this study was to identify the factor structure of the causes subscale of the Illness Perception Questionnaire (IPQ-R) in a Chinese sample of general hospital outpatients.

Considering the results of our exploratory and confirmatory factor analysis we propose a 2-factor structure. Other validational studies with the original 18 items of the IPQ-R causes scale (Abubakari et al., 2012; Chilcot, Norton, Wellsted, & Farrington, 2012) or with modified versions (Hagger & Orbell, 2005; Noureddine & Froelicher, 2013; Rief et al., 2004) have often identified 3 or 4 factor-solutions for the causal attributions scale. Meanwhile, other studies could not find interpretable solutions with 3 or 4 factors (Nicholls, Hill, & Foster, 2012; Snell, Siegert, Hay-Smith, & Surgenor, 2010; Wittkowski, Richards, Williams, & Main, 2008). Only one study investigating the factor structure of a slightly modified IPQ-R causes scale (they added one item) in a sample of 1113 healthy Portuguese proposed a 2-factors structure with the factors psychological attributions and general risk factors/lifestyle (Figueiras & Alves, 2007) which show similar item loadings as our analysis. As in our sample, they had to exclude among others the items smoking and alcohol from their calculations. Our results in a Chinese sample with general hospital outpatients increase the validity of the finding. The form and number of symptoms in general hospital outpatients are highly variable. Figueiras and Alves (2007) asked healthy controls about their causal illness attributions concerning a variety of illnesses (AIDS, tuberculosis and skin cancer) and performed the factorial analysis in the total sample including answers about all illnesses. Most other studies have been conducted with samples of patients with a specific medical condition or with modified version of the IPQ-R causes scale (Chen et al., 2008; Chilcot et al., 2012; Rief et al., 2004). Thus, the consistency of the findings might be due to the variability of investigated symptoms and the use of the original causes scale in the study of Figueiras and Alves (2007) and our study.

In spite of obvious differences in the factor structure between different samples, the loadings of the factor psychological attributions has proven to be relatively stable in all conducted validational studies examining the causes subscale (Abubakari et al., 2012 [.56–.80]; Chilcot et al., 2012 [.39–.85]; Figueiras & Alves, 2007 [.62–.86]; Hagger & Orbell, 2005 [.61–.85]; Yan et al., 2011 [.70–.84]) and has been replicated in our study. In all of these studies, the items family problems, emotional state, mental attitude, stress and overwork loaded on a factor called psychological causes. In some studies, factors such as personality (Figueiras & Alves, 2007; Yan et al., 2011), or diet and eating behavior (Hagger & Orbell, 2005) also loaded o this factor. Thus, we can rule out fundamental differences in psychological illness attributions between samples of Western countries and Chinese participants. We could categorize illness attributions broadly into psychological attributions and general risk factors, the latter seeming to vary between different patient and cultural samples. Even though the fit indices RMSEA and WLMR did only show mediocre fit of the model, the indices CFI and TLI showed excellent suitability of our model.

Items 14 (alcohol) and 15 (smoking) could not be assigned to any of the factors due to statistical reasons. We recommend analyzing them separately because they are well-known causes of illness and should not be disregarded. However, previous validational studies with the IPQ-R causes scale in Western countries have also deleted the Items smoking and alcohol due to bad measures of sampling adequacy (Chilcot et al., 2012) or poor factor loadings (Figueiras & Alves, 2007). In other studies including samples with a non-Western cultural background smoking (Abubakari et al., 2012) or alcohol (Yan et al., 2011) were included but showed very low mean values. This is in line with the expectations of the authors of IPQ-R who have emphasized that the factorial structure and the relevance of the causal items can be variable in different cultural settings (Moss-Morris et al., 2002).

Investigations including a comparison of illness attributions between general hospital outpatients with high (SOM+) and low (SOM−) symptom severity revealed that they show substantial differences with respect to all extracted factors of the IPQ-R causes scale. SOM+ patients showed significantly higher values for psychological attributions and risk factors. Additionally we found a positive relationship between symptom severity (measured with the PHQ-15) and the number of agreed illness attributions. Hence, our findings are in line with various studies that have shown a higher number of illness attributions in other groups of high symptom reporters and a correlation between the number of attributions and the number of symptoms (Chen et al., 2008; Rief et al., 2004). Descriptively we found that high symptom reporters in China (SOM+) showed multiple causal attributions like high symptom reporters in other studies (Groben & Hausteiner, 2011; Henningsen et al., 2005). We found psychological attributions to be most common in both groups indicating that our patients did not show predominantly somatic attribution styles as Asian patients in other studies (Karasz et al., 2007; Kleinman, 1986).

To our knowledge, this is the first study investigating the factor structure of the original IPQ-R causes scale in general hospital outpatients in China. A major strength of the study lies in the large sample size and the carefully selected participants: We balanced the amount of patients recruited for the SOM+ group and the SOM− group as well as the type of department – Traditional Chinese Medicine, Biomedicine and Psychosomatic Medicine. We also recruited in different districts of China – Beijing, Kunming, Shanghai and Chengdu – to enhance external validity by including habitants of urban as well as rural areas to the sample. Nevertheless, patients with a low educational level often had problems to meet the study requirements and were difficult to recruit. Thus, there might be a selection bias and generalizability to other – especially less educated – populations is limited. However, our results are comparable with those of other studies (Figueiras & Alves, 2007; Rief et al., 2004). It should be noted that by using the original items of the IPQ-R causes scale we have used a Western biopsychosocial model of illness. Thus, we might not have identified possible culture-specific characteristics. However, to date there are no investigations that account for fundamental differences in illness attributions between Asian and Western populations (Yan et al., 2011; Zhang et al., 2014). Unfortunately, we did not acquire data concerning the form and number of comorbid psychiatric disorders and were thus not able to control for them in our analysis. However, a further strength of this study is that we compared values of the IPQ-R causes scale in general hospital outpatients with high (SOM+) and low (SOM−) symptom intensity.

For future research in the field of illness attributions in China it might be interesting to follow recommendations of the authors of the IPQ-R (Moss-Morris et al., 2002) and several research colleagues (Chen et al., 2008; Yan et al., 2011) by developing a culture specific modification of the IPQ-R causes scale. This new version should include the possibility to choose between more culture specific causes such as feng shui (predicted luck in a given year), blood blockage, and temperature as in other studies (Chen et al., 2008) and the original items. Applying such a culturally adapted version of the IPQ-R subscale could help ruling out the possibility, that recurring similarities between Western and Asian samples are due to the use of Western measurements. Further research in the field of illness attributions including any version of the IPQ-R should include data concerning the comorbidities of patients since they seem to play an important role in illness attributions (Rief et al., 2004). Additionally, longitudinal studies should investigate the influence of illness attributions on aspects like treatment outcome and health behavior that have been the focus of many studies in Western countries (Frostholm et al., 2007; Henningsen et al., 2005; Rief et al., 2004).

In conclusion, the factor structure of the Chinese version of the causes subscale of the IPQ-R was identified and confirmed. No substantial differences in causal illness attributions between samples of Western countries and Chinese participants have been found. As in other studies, a relatively stable factor psychological attributions has been identified which showed the highest means among SOM+ and SOM− patients. Our findings enable researchers to investigate the relationship between causal illness attributions and other aspects of somatic and mental health such as health behavior, treatment outcome, and functional impairment in Chinese patients.

Disclosure statement

No potential conflict of interest was reported by the authors.