Frostholm, Lisbeth MSc; Fink, Per MD, PhD, DMSc; Christensen, Kaj S. MD, PhD; Toft, Tomas MD, PhD; Oernboel, Eva MSc; Olesen, Frede MD, PhD, DMSc; Weinman, John PhD
The experience of a physical symptom is the most important reason for seeking medical care. However, only a minority of those experiencing symptoms within a 14-day period consult their physician (1–3), meaning that the presence of a physical symptom does not alone explain help-seeking behavior.
Emotional distress has proved to be one of the strongest predictors of high use of nonpsychological health care (4–8). However, those findings do not provide us with much insight into the underlying processes involved in an individual’s response to the experience of a symptom. Research originating from Mechanic’s (9) work on illness behavior has focused on the individual’s help-seeking behavior from a biopsychosocial perspective, including aspects such as personal dispositions, attitudes and beliefs, and symptom appraisal (10,11).
Symptom appraisal involves defining and interpreting the symptoms, taking into account preexisting knowledge and the current context to make sense of the health threat. A number of studies have examined the differences between consulters and nonconsulters in primary care. Factors such as concern about the possible serious and severe nature of the symptoms, believing that the symptoms will interfere with social roles, engaging in more social communication concerning the symptoms, and generating fewer normalizing explanations for common bodily symptoms have been associated with consulting versus not consulting the physician for a given illness (12–18). Consequently, how an individual appraises a symptom seems to be crucial for the decision to consult.
One of the more recent measures of the individual’s understanding of his or her illness is the Illness Perception Questionnaire (19,20), which is based on the Leventhal (18) self-regulatory model proposing that the experience of a symptom is the starting signal of a cognitive search for understanding the health threat. This process results in the individual developing his or her own perception of the illness, which comprises a number of distinct but interrelated themes: identity (the illness label and number of symptoms), cause, timeline, consequences, perceived control, and emotional representations (the emotional reactions to the experience of a symptom).
Previous research on individuals with a diagnosed disease or condition found that a strong illness identity, believing that the illness would be long-lasting and have serious consequences, and low perceived control were associated with more negative outcomes (21–24). A recent randomized controlled trial reported promising results using the IPQ as a point of departure for a psychological intervention: Improvements on function and return to work were found among those myocardial infarction patients who received a 3-session intervention conducted by a psychologist with the purpose of changing their illness perceptions when compared with treatment as usual (25).
We wished to use an illness perception approach to examine the relationship between patients’ perceptions of a specific health problem presented to their physician at the day of inclusion in the study, and previous as well as future health care use.
According to the self-regulatory model, the understanding that a person develops about his or her health problem will guide the decision to consult or not to consult. Moreover, when consulting, the patient’s illness perceptions will have an effect on how the patient presents his or her health problem to the physician and hence the physician’s decision whether to refer the patient to a specialist. We hypothesized, in keeping with previous research, that patients with a strong illness identity (reporting many symptoms), a long timeline perspective, high illness worry, and low personal control would find it more difficult to cope with the health problem and thus have higher health care use in the 2-year follow-up period. We also assumed that those negative perceptions would reflect a disposition to develop negative illness perceptions and thus be associated with higher previous health care use. Regarding the causal attributions, previous research on persons with diagnosed diseases and conditions is less clear. We hypothesized that in the present study with primary care patients presenting a new health problem, the causal attributions would play a more important part. In particular, we expected patients reporting psychosocial, stress, and lifestyle attributions to experience more stress in their daily living, which could lead to more help-seeking behavior. Finally, we expected patients with negative emotional representations and high levels of anxiety and depression to be more prone to consult both before and after the index consultation.
The study took place in Aarhus County, which is a mixed rural and metropolitan area with 600,000 inhabitants, served by 431 physicians working in 271 practices. All physicians were invited to participate in an educational program and a subsequent randomized controlled intervention study on assessment and treatment of patients presenting with functional somatic symptoms and somatization (26).
Consecutive patients 18 to 65 years old, consulting their physician with a new health problem during a 3-week period, with scheduled successive inclusion of practices between March 3 and May 1, 2000, were invited by the physicians’ receptionists to participate in a study on improving the physicians’ understanding and treatment of their patients to avoid unnecessary examination, treatment, and suffering. The study excluded patients of non-Scandinavian descent, not speaking or reading Danish, and those whom the receptionists assessed as being too ill or too demented to participate. Moreover, administrative consultations (e.g., driver’s licenses and other certificates, pregnancy controls, and vaccinations) were excluded from the study.
Health Care Use
Denmark has a public health care system, which is tax-financed by the Danish Public Health Insurance. Within this system, 98% of the population is registered with a specific physician, who acts as a gatekeeper to the secondary health care system. Each person has a personal registration number used for all contacts with the health care system. Data were obtained from the Danish Public Health Register by using this registration number. Use of primary care such as general primary care physicians, private consultants, physiotherapists, psychologists (limited to emergency help), and out-of-hours services is reported to the regional Public Health Insurance Register and reimbursed on a monthly basis. Drugs were not included in this study.
Primary health care utilization cost in Danish Kroners (DKK) was obtained for the period of 3 years before and 2 years after inclusion in study. Patients included in March 2000 were defined to be at risk from March 1997 to February 2000 for previous use and from March 2000 to February 2002 for use after baseline (both months included). Patients included in April were defined to be at risk from April 1997 to March 2000 and from April 2000 to March 2002, respectively. Thus the baseline visit was included in the assessment of use after baseline. Decreased use of health care services due to time spent abroad or death was taken into account by subtracting those periods from the maximum time-at-risk (1095.75 and 730.5 days, respectively). Fifty-five periods (due to time spent abroad) for prior use and 26 periods (due to death or time spent abroad) during follow-up were identified. Mean use in DKK per year before and after inclusion respectively was computed taking into account number of days at risk.
The measure of illness perceptions was based on the IPQ (20). Each patient answered questions regarding their perception of the most important health problem, which they presented to their physician on the day of inclusion in the study. The IPQ was adapted and condensed for use in general practice and consisted of 20 illness perceptions and 20 causal attribution items. The response format was true/mainly correct/mainly false/false, and moreover, patients were given the option to tick off a “cannot answer” category. The emotional representation items were extended to include helplessness and hopelessness. Two questions regarding the patients’ certainty of the nature of their health problem were added. Principal components factor analysis showed a dimensional structure similar to the original IPQ, including the following dimensions: emotional representations, consequences, long timeline perspective, personal control, cyclical timeline perspective, and certainty (data available from authors). Principal components factor analysis of the causal attributions resulted in the following dimensions: psychosocial, stress, lifestyle, modern health hazards, infection/lowered immunity, and accident/change attributions. Sum scores of the different IPQ dimensions were calculated. The majority of the various components displayed from 10% to 15% missing. To increase the power of the analyses, we made imputations to replace missing values of the illness perception sum scores based on best subsets regression with the other illness perceptions, emotional distress scores, and sociodemographics of the patients as covariates. Due to the very skewed distribution of the sum scores, we chose to dichotomize the sum scores to identify the 20% with the most pessimistic illness perceptions as cases. We found almost no change in cut point when dichotomizing nonimputed and imputed data, respectively.
The Symptom Check List-Somatization Subscale (SCL-SOM) (27) was used as a measure of illness identity according to the self-regulatory model (18). Moreover, a 7-item measure of illness worry, the Whiteley-7 (28), was included.
To measure emotional distress, 3 variables were included: Measures of anxiety and depression respectively (subscales of the Symptom Check List–92 Items) (29), and emotional representations from the IPQ.
Immediately after the consultation, the physicians completed a questionnaire for each patient. The questionnaire was tailored and initially tested in a pilot study by a number of experienced Danish physicians to acquire information on diagnostics, prognostics, estimate of the doctor-patient relationship, and functional somatic symptoms. Four items from this questionnaire regarding the presence/absence of chronic disorder, the course of the current health problem, estimated future duration of the patient’s health problem, and estimation of the consequences for the patient’s future functional level were included in the analysis (data available from authors).
Due to the skewed distribution of all the total sum scores of the dependent variables, associations between previous health care use and illness perceptions/emotional distress were examined through multivariate logistic regression models with illness perceptions/emotional distress as dependent variables (Table 1). The illness perception scores, including illness worry and emotional representations were dichotomized (high 20% to 25%/low 75% to 80%), identifying the 20% to 25% patients with negative illness perceptions as cases. The exact cut points were set as close to 20/80 as possible, dependent on the sum score. Illness identity was dichotomized (high 72%/low 28%) equal to cases being the 72% of patients reporting 1 or more symptoms on the SCL-SOM. The measures of anxiety and depression were dichotomized (high 30%/low 70% and high 23%/low 77%), respectively identifying patients with 1 or more symptoms as cases. Log-transformed previous health care use was included as continuous variable. All models were adjusted for age and gender. Furthermore, a variable indicating the slope of each patient’s health care use over the past 3 years (values indicating decreasing, stable or increasing use) was added to the analyses.
To assess severity of illness, we used 2 variables from the physician questionnaire: The time course of the current health problem (isolated occurrence, remittent, chronic) and the presence/absence of a chronic mental or physical disorder. We performed all analyses with an interaction term between chronic disorder and health care use in order to check for heterogeneity in the 2 groups’ chronic disorder and no chronic disorder, respectively. In cases of significant interaction, analyses are presented separately for the two groups. Nonsignificant interaction is presented in a single analysis adjusted for chronic disorder.
To validate the models, we estimated the Pearson χ2 statistic, the Hosmer-Lemeshow Ĉ fit statistic including the 2 × 10 table of observed and estimated expected frequencies, and we performed Stukel’s score test (30,31), and for discrimination the area under the receiver operating characteristic (ROC) curve (32,33).
Use During Follow-up
Associations between illness perceptions/emotional distress and future use of health care were examined through multivariate linear regression models with log-transformed health care as the dependent variable (Table 2). The illness perception sum scores were dichotomized as described above. Illness worry, the identity scale, and the anxiety and depression scales were added to the analyses as continuous variables. All models were adjusted for age and gender. The linear regressions with the illness perceptions were also adjusted for the physician variable “time course of the current health problem,” as the illness perception items were specifically targeted at this problem. An interaction term between illness perception and chronic disorder was included. Again, results were presented separately for patients with and without a chronic disorder, if the interaction was significant. Moreover, all regressions were performed with and without previous use to examine the impact of illness perceptions compared with previous use.
For the 2 illness perception variables long timeline perspective and serious consequences, we refined the analyses further by adding 2 physician variables to the analyses: “Future duration of the patient’s health problem according to your best judgment” (0–6 weeks/>6 weeks-chronic) and “Your estimation of the consequences for the patient’s future functional level” (none-minimal/moderate-severe), respectively. This was done to ascertain that associations between those variables and health care use did not merely reflect severity of disease from a professional judgment.
Finally, all illness perceptions/emotional distress variables were entered simultaneously into a linear regression model. Only variables with p values exceeding .8 were removed from this model according to the Steyerberg et al. (33) approach. Two full models are presented with and without previous health care use, respectively. We used regression diagnostics, comprising examination of residuals and influential points, to test all models. Furthermore, van Houwelingen and le Cessie’s shrinkage factor was estimated to investigate the amount of overfitting in the 2 full multivariate models (32). Statistical analysis was performed using STATA.
Thirty-eight physicians working in 28 practices volunteered for the study (Figure 1). The participating physicians did not differ in age, gender, type of clinic, or postgraduate psychiatric training from nonparticipating physicians in the county. However, participating physicians had practiced family medicine for fewer years (mean 10.3 years versus 14.1 year, likelihood ratio test (LR-test) p = .005) than nonparticipating physicians and were more likely to have participated in longer (more than 3 days) courses in communication skills or psychological treatment (52.8% versus 39.5%, LR-test p = .05).
To evaluate the representativeness of the participating practices with respect to their patients’ primary health care use, 18 of the 28 practices (in which all physicians participated) were matched on physicians’ age, gender, and number of years working as a physician with 36 register control practices from the county. All patients aged 18 to 65 initiating consultations for a period of 3 weeks before to 3 weeks after the 3-week inclusion were selected (all in all, 9 weeks): 6952 patients from the 18 project practices and 14,325 patients from the 36 matched register control practices. We found no difference in previous primary health care use per year between the 2 groups; Mann Whitney U, Z = −1.627, p = .10: Patients from project practices had a mean health care use of 1248 DKK, SD = 1775, interquartile range (IR) = 386 to 1531, and patients from register control practices 1225 DKK, SD = 1829, IR = 372 to 1457.
Of the 2424 patients assessed for eligibility, 1785 patients joined the study (Figure 1). Declining patients had a mean age of 42.2 years compared with 38.8 years in the participating patients (p < .0001), whereas we found no significant gender differences. No significant differences were found in age or gender between participants and those not participating for other reasons.
Previous health care use was associated with a range of the illness perceptions (Table 1 and Figure 2). High users were more likely to express a belief that their current health problem would have serious consequences, odds ratio (95% confidence interval) (CI) = 1.26 (1.1–1.4) and would be long-lasting and to report general illness worry and a strong illness identity (SCL-SOM). For patients with a chronic disorder, there was a stronger association between number of symptoms reported and previous health care use and between illness worry and previous health care use compared with patients without a chronic disorder. A cyclical timeline perspective was associated with lower health care use (Table 1).
Regarding the causal attributions, higher use was associated with psychosocial, stress, and lifestyle attributions for all patients, whereas accident/chance attributions were associated with higher use for patients with a chronic disorder but with lower use for patients without a chronic disorder. Emotional representations, anxiety, and depression were all associated with higher use (Table 1).
The Hosmer-Lemeshow Ĉ fit statistic with 8 degrees of freedom displayed p values from .09 (depression) to .83 (illness identity for patients with chronic disorders). The Pearson χ2 statistic with 1757 degrees of freedom for all patients and 1172 and 579 for analyses performed with and without chronic disorder, respectively, had p values in the range of .09 (depression) to .59 (illness identity for patients with chronic disorders). p Values of Stukel’s test were all nonsignificant, except lifestyle attributions, p = .01. The area under the ROC curve varied from 0.61 (stress attributions) to 0.75 for long timeline perspective.
Use During Follow-up
A number of illness perception dimensions were also found to be associated with health care utilization in the 2 years following the consultation. Illness identity, believing that the health problem would have serious consequences and would be long-lasting, high illness worry, psychosocial attributions, and all the emotional distress variables were associated with higher use during follow-up, even when taking previous use into account (Table 2). Timeline cyclical was associated with higher use only when taking previous use into account.
Perceiving serious consequences was associated with increased use, β = 1.37; CI, 1.2–1.5, p < .001 in the analysis without previous use, β = 1.24; CI, 1.1–1.4, p < .001 including previous use. The physician variable “estimation of consequences for the patient’s future functional level” predicted later health care use β = 1.25; CI, 1.1–1.4, p < .001 in the analysis without previous use and β = 1.16; CI, 1–1.3, p = .003 in the analysis including previous use but did only have a minor impact on the association found between the patients’ perceived consequences and health care use. With respect to long-lasting timeline perspective (Table 2), the physician variable “future duration according to your best judgment” was not significantly associated with health care use during follow-up (β = 1.06; CI, 0.9–1.2; p = .32 in the analysis without previous use and β = 1.08; CI, 1.0–1.4; p = .15 in the analysis including previous use).
Psychosocial and lifestyle attributions were associated with health care use (Table 2); however, lifestyle attributions turned nonsignificant when previous use was included. Infection/lowered immunity attributions were significant for patients with chronic disorders only (Table 2). For all models, previous health care use was a very strong predictor of later use, for which reason the explanatory value of the illness perceptions varied considerably from models with and without previous use.
For the full model, including all variables, 5 illness perceptions predicted higher use during follow-up over and above chronic disorder, gender, and age: illness identity, serious consequences, and a long timeline perspective predicted higher use, whereas stress and accident/chance attributions predicted lower use (Table 3). After adding previous use to the model, only illness identity, stress and accident/chance attributions, chronic disorder, and gender still contributed to explaining the variance (Table 3).
To check the assumptions of linear regression modeling, we made diagnostic plots and found the assumptions to be fulfilled. The shrinkage factor was 0.95 and 0.97 for the full model, excluding and including previous use, respectively.
Missing Values and Imputations
For the vast majority of the analyses performed, estimates changed only slightly when performing the analyses on nonimputed data. For example, in the full models predicting health care use during follow-up, results were very similar, except for long timeline perspective being near significant in the model including previous use when using nonimputed data β (95% CI) = 1.14 (1.0–1.3), p = .059 as opposed to 1.11 (1.0–1.3), p = .11 for imputed data.
Adjustment for Multiple Analyses Using the Bonferroni Method
To take into account the problem of type 1 error due to multiple analyses, we performed Bonferroni’s correction of the final p value. Including all 3 sets of analyses (previous use, use during follow-up without previous use, and use during follow-up including previous use), the p value was corrected to .05/(3 × 17) = .001. Thus, a large number of the analyses were robust to the correction of the p value (Table 1 and Table 2).
In this study, we examined the association between patients’ perceptions of a current health problem at a self-initiated consultation in general practice and their primary health care utilization 3 years before and 2 years after the index consultation. We found that a range of the illness perceptions was associated with previous, as well as later, health care use. Perceiving the health problem to be highly symptomatic and believing that the health problem would have long-lasting and serious consequences were key predictors of health care use. Emotional distress reactions to the current health problem and anxiety and depression were, as expected, associated with use.
The strong association between mental disorders and use of health care has been confirmed in numerous studies. This study contributes with a health psychological approach to health care use. To our knowledge, no other study has examined the associations between illness perceptions and health care use in such a large sample or used the IPQ in a general practice population. The sample size and several data sources (register, physician, and patient) support the value of the study. Our findings are strengthened by the fact that our health care data cover a total time period of 5 years and were obtained from registers. Therefore, the data are not subject to self-report bias, which may be particularly prominent with regard to ambulatory physician visits (34). Moreover, most of the analyses are robust to the Bonferroni method, which leads to very conservative estimates of the associations between illness perception and health care use.
The IPQ assessed patients’ perceptions of the main health problem that they presented to their physician. Some patients may have presented more than 1 complaint at a time, and we can make no conclusions regarding the potential effect of that. Furthermore, we did not have access to a precise diagnosis of the health problem causing the patient to consult, which has probably been as distinct as initial symptoms of cardiovascular disease to warts on the foot. Thus, one of the main objections to this study may be that expecting negative consequences could merely reflect severity of biomedical disease. To counter this argument, we took into account the course of the current illness and the effect of having a chronic disorder. Furthermore, regarding 2 of the stronger associations between health care use and illness perceptions (serious consequences and long timeline perspective), we included the physicians’ estimation of those same parameters in the follow-up analyses. As this did not impair our findings, the patients’ perceptions must reflect something more than severity of disease from a biomedical point of view. However, given the heterogeneity of the presented health problems, we should take particular caution in deducing from the overall results to the meaning of a specific belief in the individual case.
One of the main weaknesses of the study was the relatively large number of missing values. We made imputations of the missing values based on the responses on the other illness perceptions, which are all correlated, and we found almost no change in mean values and standard deviations of imputed and nonimputed data. Moreover, results from analyses conducted on imputed and nonimputed data were very similar. Older patients were less likely to complete the questionnaire, which may have been due to a cohort effect in that they have a more traditional perception of their role as patients and are less accustomed to making self-evaluations about their illness. It is likely that some patients have not developed a full model including all the dimensions from the self-regulatory model when consulting with a new health problem, which would explain the high rate of missing data on particularly the timeline dimension. A qualitative approach would probably have resulted in better data capture, especially regarding the older patient group. Finally, the revised version of the IPQ used in this study may be in need of further adjustment to optimize its usability in primary care. The revised version of the IPQ also warrants some caution with respect to the generalizability to other studies using the original IPQ.
The participation in the educational program on assessment and treatment of patients presenting with functional somatic symptoms did not have any impact on the results reported here. Regarding the follow-up analyses, we found significant associations between illness perceptions and health care use even when previous use was accounted for in the analyses. However, the results from the analyses not including previous use are probably a more accurate reflection of actual effect sizes on the theory that previous use is inseparable from previous illness perception. To illustrate by example (applying Cox’s method to calculate mean health care use (35) and the results from Table 3, full multivariate model without previous use), a 40-year-old female with 2 symptoms has a mean primary health care use per year during follow-up amounting to 938 DKK CI (890–989). However, if she perceives the health problem to have serious consequences, her mean health care use per year rises to 1142 DKK CI (1082–1204); if she furthermore has a long timeline perspective, her mean health care use will increase further to 1311 CI (1243–1382), differences equal to 2 and 4 primary care consultations, respectively (priced year 2000, no additional health care services), indeed a clinically significant difference. However, it is important to take caution in generalizing results from the full model (Table 3) due to colinearity of the independent variables. In particular, a number of the illness perceptions are highly correlated, which is why they may not come out significant in the 2 joint analyses. In particular, serious consequences, long timeline perspective, and emotional representations were highly correlated.
A general finding, supported by our data, is that emotional distress and illness worry are associated with higher health care use (4,8). In line with a number of other studies on primary care patients (12,14–17), we found that illness identity (reporting many symptoms) and believing that the symptoms were of a severe nature and would have serious consequences were associated with higher health care use, even when taking the physicians’ estimation into account. Other studies of patients with diagnosed illness have also confirmed the impact of those specific perceptions on treatment outcomes (22,23, 36). Thus, a strong illness identity, a long timeline perspective, and a belief in serious consequences may be core illness perceptions.
Perceived personal control was not significantly associated with health care use. This may be due to the fact that this study investigated new health problems, and therefore control was not yet a prominent factor compared with more long-lasting conditions, for which management and control of persistent symptoms are essential.
In general, causal attributions have been less successful in predicting health outcomes than the other illness perceptions. This may be due to the fact that the majority of studies have been conducted on patients with established illnesses (37). In this study, we focused on new symptoms, which had not yet been professionally diagnosed and found significant associations. The main finding was that psychosocial and lifestyle attributions were significantly associated with both previous as well as succeeding health care use. Stress attributions were associated with higher previous use but with lower prospective use in the multivariate model. Thus, attributing symptoms to stress may be an expression of transient strain, whereas psychosocial and lifestyle attributions may reflect dispositional patterns of reacting.
We found differences between patients with and without a chronic disorder. There was a stronger association between previous use and illness worry and illness identity for patients with a chronic disorder. Moreover, accident/chance attributions were associated with higher previous use and infection/lowered immunity attributions with higher prospective use for patients with a chronic disorder only. The heterogeneity of the chronic disorders category in the present study, including all physical as well as mental disorders, renders difficult an understanding of those findings. However, attributing symptoms to accident/chance may be a subsequent rationalization of a complicated previous course of illness. Regarding the virus/lowered immunity attributions, the burden of having a chronic disorder may make those patients both more susceptible to as well as more prone to consult with symptoms of infectious disease.
This study has confirmed our hypotheses based on the self-regulatory model; illness perceptions guide decisions to consult when faced with a new health problem and influences the physician’s decision to refer a patient to a specialist. In addition, our study points to the central role of a dispositional basis of symptom perception in explaining this relationship (38). “Here-and-now” illness perceptions are probably associated with previous use through dispositional or learned ways of reacting to symptoms as many of the associations are significant both back and forth in time. If phasic, genetic, and dispositional factors all contribute to illness behavior, illness perceptions may be a complex mix of them all (38). Thus, it should be possible to optimize treatment outcomes by taking illness perceptions as a starting point for future interventions (25). Asking patients open questions about their thoughts and ideas about their illness could be a straightforward way of bringing the biopsychosocial model into play in clinical practice. Minor misconceptions could be brought up for discussion, and hopefully this could be of particular benefit to patients at risk of developing chronic health problems, which could have otherwise been avoided or alleviated.
1. Kellner R, Sheffield BF. The one-week prevalence of symptoms in neurotic patients and normals. Am J Psychiatry 1973;130:102–5.
2. Pennebaker JW, Burnam MA, Schaeffer MA, Harper DC. Lack of control as a determinant of perceived physical symptoms. J Pers Soc Psychol 1977;35:167–74.
3. Kjøller M, Rasmussen NK, Keiding L, Petersen HC, Nielsen GA. Sundhed og Sygelighed i Danmark 1994-og Udviklingen Siden 1987. København: DIKE; 1995.
4. Bellon JA, Delgado A, Luna JD, Lardelli P. Psychosocial and health belief variables associated with frequent attendance in primary care. Psychol Med 1999;29:1347–57.
5. Andersen R, Francies A, Lion J, Daughtey VS. Psychologically related illness and health services utilization. Med Care 1977;XV:59–73.
6. Vedsted P, Fink P, Olesen F, Munk-Jørgensen P. Psychological distress as a predictor of frequent attendance in family practice: a cohort study. Psychosomatics 2001;42:416–22.
7. Fink P. The use of hospitalizations by persistent somatizing patients. Psychol Med 1992;22:173–80.
8. Hansen MS, Fink P, Frydenberg M, Oxhoj ML. Use of health services, mental illness, and self-rated disability and health in medical inpatients. Psychosom Med 2002;64:668–75.
9. Mechanic D. The concept of illness behavior. J Chronic Dis 1961;15:189–94.
10. Skelton JA, Croyle RT. Mental Representation in Health and Illness. New York: Springer-Verlag; 1991.
11. Petrie K, Weinman J. Perceptions of Health and Illness. Amsterdam: Harwood Academic Publishers; 1997.
12. Kettell J, Jones R, Lydeard S. Reasons for consultation in irritable bowel syndrome: symptoms and patient characteristics. Br J Gen Pract 1992;42:459–61.
13. Kessler D, Lloyd K, Lewis G, Gray DP. Cross-sectional study of symptom attribution and recognition of depression and anxiety in primary care. BMJ 1999;318:436–9.
14. Cornford CS. Why patients consult when they cough: a comparison of consulting and non-consulting patients. Br J Gen Pract 1998;48:1751–4.
15. MacLeod AK, Haynes C, Sensky T. Attributions about common bodily sensations: their associations with hypochondriasis and anxiety. Psychol Med 1998;28:225–8.
16. Sensky T, MacLeod AK, Rigby MF. Causal attributions about common somatic sensations among frequent general practice attenders. Psychol Med 1996;26:641–6.
17. Lydeard S, Jones R. Factors affecting the decision to consult with dyspepsia: comparison of consulters and non-consulters. J R Coll Gen Pract 1989;39:495–8.
18. Cameron L, Leventhal EA, Leventhal H. Symptom representations and affect as determinants of care seeking in a community-dwelling, adult sample population. Health Psychol 1993;12:171–9.
19. Weinman J, Petrie KJ. Illness perceptions: a new paradigm for psychosomatics? J Psychosom Res 1997;42:113–6.
20. Moss-Morris R, Weinman J, Petrie KJ, Horne R, Cameron LD, Buick D. The Revised Illness Perception Questionnaire (IPQ-R). Psychol Health 2002;17:1–16.
21. Hagger MS, Orbell S. A meta-analytic review of the common-sense model of illness representations. Psychol Health 2003;18:141–84.
22. Scharloo M, Kaptein AA, Weinman J, Bergman W, Vermeer BJ, Rooijmans HG. Patients’ illness perceptions and coping as predictors of functional status in psoriasis: a 1-year follow-up. Br J Dermatol 2000;142:899–907.
23. Scharloo M, Kaptein AA, Weinman J, Hazes JM, Willems LN, Bergman W, Rooijmans HG. Illness perceptions, coping and functioning in patients with rheumatoid arthritis, chronic obstructive pulmonary disease and psoriasis. J Psychosom Res 1998;44:573–85.
24. Moss-Morris R, Petrie KJ. Discriminating between chronic fatigue syndrome and depression: a cognitive analysis. Psychol Med 2001;31:469–79.
25. Petrie KJ, Cameron LD, Ellis CJ, Buick D, Weinman J. Changing illness perceptions after myocardial infarction: an early intervention randomized controlled trial. Psychosom Med 2002;64:580–6.
26. Fink P, Rosendal M, Toft T. Assessment and treatment of functional disorders in general practice: the extended reattribution and management model: an advanced educational program for nonpsychiatric doctors. Psychosomatics 2002;43:93–131.
27. Derogatis LR, Cleary PA. Confirmation of the dimensional structure of the SCL-90: a study in construct validation. J Clin Psychol 1977;33:981–9.
28. Fink P, Ewald H, Jensen J, Sorensen L, Engberg M, Holm M, Munk-Jorgensen P. Screening for somatization and hypochondriasis in primary care and neurological in-patients: a seven-item scale for hypochondriasis and somatization. J Psychosom Res 1999;46:261–73.
29. Christensen KS, Fink P, Toft T, Frostholm L, Ørnbøl E, Olesen F. A brief case-finding questionnaire for common mental disorders: the CMD-SQ. 2003.
30. Hosmer DW, Lemeshow S. Applied Logistic Regression. 2nd ed. New York: John Wiley & Sons; 2001.
31. Hosmer DW, Hosmer T, Le Cessie S, Lemeshow S. A comparison of goodness-of-fit tests for the logistic regression model. Stat Med 1997;16:965–80.
32. Harrell FE Jr. Regression Modeling Strategies: With Applications to Linear Models, Logistic Regression, and Survival Analysis. New York: Springer; 2001.
33. Steyerberg EW, Eijkemans MJ, Harrell FE Jr, Habbema JD. Prognostic modelling with logistic regression analysis: a comparison of selection and estimation methods in small data sets. Stat Med 2000;19:1059–79.
34. Roberts RO, Bergstralh EJ, Schmidt L, Jacobsen SJ. Comparison of self-reported and medical record health care utilization measures. J Clin Epidemiol 1996;49:989–95.
35. Zhou XH, Gao S. Confidence intervals for the log-normal mean. Stat Med 1997;16:783–90.
36. Petrie KJ, Weinman J, Sharpe N, Buckley J. Role of patients’ view of their illness in predicting return to work and functioning after myocardial infarction: longitudinal study. BMJ 1996;312:1191–4.
37. Robbins JM, Kirmayer LJ. Attributions of common somatic symptoms. Psychol Med 1991;21:1029–45.
38. Watson D, Pennebaker JW. Situational, dispositional, and genetic bases of symptom reporting. In: Skelton JA, Croyle RT, editors. Mental Representations in Health and Illness. New York: Springer-Verlag; 1991:60–84.