The association between medically unexplained physical symptoms and health care use over two years and the influence of depressive and anxiety disorders and personality traits: a longitudinal study

The Netherlands Study of Depression and Anxiety (NESDA) aims to describe the long-term course and consequences of depressive and anxiety disorders and to examine its predictors. A detailed description of its rationale and design has been published elsewhere [25]. In summary, the study sample consisted of 2981 participants (age 18–65) with current depressive and/or anxiety disorders, participants with a lifetime risk or subtreshold depressive and/or anxiety symptoms and healthy controls. Recruitment took place across primary care practices (n = 1610), outpatient secondary mental health care institutions (n = 807) and the general population (n = 564). Exclusion criteria were not being fluent in the Dutch language and a primary diagnosis of a psychotic, obsessive compulsive, bipolar or severe substance abuse disorder. Baseline data were collected between 2004 and 2007. Assessments, including written questionnaires and interviews, were repeated after 1, 2, 4 and 6 years. Non-response among participants was not significantly related to mental health status, but slightly higher among younger and male respondents. The research protocol was approved centrally by the ethical review board of VU University medical center. Subsequently it was approved by the local ethical review boards of Leiden University Medical Center and University Medical Center Groningen. The study was performed in accordance with the ethical standards of the Declaration of Helsinki. All participants provided written informed consent.

For the present study, we used data of all participants who completed the questionnaire used for our study, and used the measurements at baseline (T0), one (T1) and 2 years (T2) of follow-up. Baseline measurements were obtained from 2981 participants. At T1 and T2, 2045 (68.6 %) and 2395 (80.3 %) participants had a follow-up assessment, respectively.

HCU was measured with the Trimbos and iMTA questionnaire on costs associated with psychiatric illness (TIC-P) [26]. The TIC-P is a widely used, feasible and reliable questionnaire on health care consumption and productivity losses for patients with mental health disorders. For this study we focused on the first part of the TIC-P, consisting of dichotomous questions on relevant medical services, followed by a question on the consumption volume (number of contacts) in the past 6 months; e.g. ‘did you consult with a family physician? No/Yes, namely … times’. We counted the number of medical services used (range 0–14) and additionally categorized these into three subgroups: mental health care services (primary care psychologists, social workers/social psychiatric nurses, secondary mental health care institutions, centers for drugs or alcohol, self-help groups and private psychiatrists/psychotherapists); somatic health care services (family physicians, medical specialists and hospital admissions); and miscellaneous health care services (homecare, complementary alternative professionals, occupational health physicians, physiotherapists). Participants completed the TIC-P at T0, T1 and T2.

MUPS were measured with the somatisation scale of the validated Four Dimensional Symptoms Questionnaire (4DSQ) [27]. The self-report 4DSQ has been developed to measure distress, depression, anxiety and somatisation as separate dimensions. The somatisation scale comprises 16 items including physical symptoms that often remain medically unexplained (e.g. dizziness and abdominal pain). The scale highly correlates with instruments used in other countries measuring MUPS; 0.82 in case of the SCL-90 [27, 28] and 0.84 in case of the PHQ-15 [29, 30]. In the present sample Cronbach’s alpha of the 4DSQ somatisation scale was 0.92, 0.89 and 0.97 at the three measurements, respectively. The items on the somatisation scale are scored on a 5-point Likert scale: “no”, “sometimes”, “regularly”, “often”, and “very often or constantly”. In order to arrive at scale scores, the responses were recoded as 0 for “no”, 1 for “sometimes” and 2 for “regularly”, “often” and “very often or constant” and summated, resulting in a score ranging from 0 to 32. Additionally, in order to facilitate clinical use and to overcome the fact that there is no linear relation between MUPS and HCU, we repeated the analyses with a dichotomized scale using 11 points as a cut-off score, since a score of 11 or higher is considered to indicate MUPS [27]. Participants completed the 4DSQ at T0, T1 and T2.

The presence of depressive and anxiety disorders was assessed with the validated Composite International Diagnostic Interview (CIDI, WHO 2.1) at T0 and T2. Trained research staff interviewed all participants. Depressive disorders included major depressive disorder and dysthymia. Anxiety disorders included generalized anxiety disorder, panic disorder with or without agoraphobia, social phobia and/or agoraphobia without panic disorder. We only took into account diagnoses established during the previous six months at both assessments.

Personality traits were measured with the NEO Five Factor Inventory (NEO-FFI) at T0 and T2. The NEO-FFI measures the five most important personality domains in adults: neuroticism, extraversion, openness, agreeableness and conscientiousness. Each domain is measured with 12 items, using a five-point Likert response format (sum score: range 12–60). More detailed information about the contents, validity and reliability of the NEO FFI has been published elsewhere [31‐33].

Based on previous studies, we considered the following sociodemographic variables as possible confounders: gender, age, level of education, marital status and the number of chronic diseases [34, 35]. The level of education was derived from the standard classification of education from Statistics Netherlands [36] and categorized into three groups (basic, intermediate, high). Marital status divided participants in those being married/living with a partner and those living alone. Participants were asked if they had one or more diseases from the following chronic diseases categories: respiratory, cardiometabolic, musculoskeletal, digestive, neurological, endocrine and cancer. We only considered and summated the diseases if participants were currently treated with medication and/or under specialist control. All variables were assessed at T0.

Descriptive statistics are presented as mean with standard deviation for normally distributed continuous data, median and inter-quartile range for skewed continuous variables and as numbers and percentages for dichotomous and categorical variables.

Generalized estimating equations (GEE) with an exchangeable correlation structure were used to assess the relationship between MUPS and HCU longitudinally (Fig. 1). We used GEE because it takes into account the dependency of repeated observations within the participants and because it is capable of analysing non-complete longitudinal data. As the total number of consulted medical services showed a Poisson distribution, we used Poisson GEE analysis to assess its association with MUPS. For the total number of contacts, we used negative binomial GEE analysis because the Poisson distribution was skewed to the right (a Poisson distribution with overdispersion). The effect sizes of both the Poisson and the negative binomial GEE analyses are expressed as rate ratios (RRs). This RR represents the association between MUPS and HCU on average over time and reflects both a within and between subjects interpretation [37]. Besides crude analyses, we adjusted the relationships for the sociodemographic variables, and additionally we examined the influence of depressive and anxiety disorders and personality traits on the association between MUPS and HCU. The influence is expressed as the percentage decrease in the regression coefficient as a result of including each separate variable. As depressive and anxiety disorders and personality traits were only measured at T0 and T2, these last analyses were based on these two measurements only. We repeated the crude and adjusted analyses for each of the three medical resource subgroups. Furthermore, as we performed our analyses in a population with an oversampling of depressive and anxiety disorders, we analysed whether the effect between MUPS and HCU was modified by depressive and/or disorders by adding an interaction term to the model. Finally, we carried out the same set of analyses with a time lag model in order to assess if MUPS at a certain point in time was related to HCU one year later (Fig. 1). We used all observations in our analyses.

The reporting of this manucripted adhered to the STROBE guidelines (Additional file 1).

MUPS were significantly associated with the total number of consulted medical professionals and the total number of associated contacts, respectively, on average over time in both the crude and adjusted GEE analyses (Table 2). To illustrate the interpretation of the results: the estimated adjusted RR of 1.02 found for MUPS in relation to the total number of medical resources can be interpreted as follows: for every unit increase in the 4DSQ, a 2 % increase in the number of medical services is observed, within and between participants. The small difference between the crude and adjusted analyses was mainly driven by the number of chronic diseases. MUPS defined with the dichotomized 4DSQ showed results in the same direction (services: RR 1.35; 95 % CI 1.31–1.40; contacts: RR 1.41; 95 % CI 1.28–1.55, not in table). Outcome defined as the three categories of medical services also showed comparable results (Table 2). The strongest association was found for both the number of mental health care services used and the number of contacts with these services. Table 3 shows the influence of depressive and anxiety disorders and personality traits on the association of HCU with MUPS. For both HCU outcomes, neuroticism had the strongest influence, followed by depressive disorders. When taking neuroticism and depressive disorders together, the magnitude of the regression coefficient decreased by 48 % (services) and 44 % (contacts). Despite the contribution of these mental health characteristics, HCU remained significantly associated with MUPS. Adding anxiety disorders and other personality traits did not further affect the association. Also for the dichotomized 4DSQ score, neuroticism and depressive disorders together showed the strongest influence (decrease in regression coefficients of 56 and 59 %, respectively) and HCU remained significantly associated with MUPS. When we examined whether the association between MUPS and HCU was modified by depressive and/or anxiety disorders, we found a significant inverse interaction effect (p < 0.001), meaning that the association between MUPS and HCU (both services and contacts) was weaker for patients with depressive and anxiety disorders (data not shown).

Table 4 shows the results of the time-lag analyses. When comparing the results of the time-lag analyses with the standard analyses, we found comparable RRs for the number of medical services, but slightly higher RRs for the number of contacts in the time-lag analyses. For the dichotomized 4DSQ score, results were in the same direction (services: RR 1.19; 95 % CI 1.15–1.22; contacts: RR 1.64; 95 % CI 1.48–1.83). For the influence of depressive and anxiety disorders and personality traits on the association between MUPS and HCU, we found the same pattern with neuroticism as the strongest influencing variable, followed by depressive disorder and again even stronger when taken together (data not shown). MUPS was still associated with HCU over a longer period of time, as reflected in the time lag analyses (data not shown).

Our findings on the positive association between MUPS and HCU are in accordance with previous studies, irrespective of the differences in methodology [8, 10, 13, 17, 18, 22]. However, when we examined the three categories of medical resources, we found the strongest association between MUPS and mental HCU, in contrast to some other studies [8, 13, 18]. Fink et al. concluded in their study among patients with somatoform disorders that these patients used more non-psychiatric health care facilities than patients without somatoform disorders [13]. Also Barsky et al. found large amounts of medical, but not mental, health care use among their somatising patients [8]. Their findings support the assumption that patients with MUPS attribute their complaints to their physical symptoms, thereby seeking somatic health care instead of mental health care. The difference with our findings could be explained by the difference in setting (primary care only versus primary care, mental health care and general population).

We found that depressive disorders and neuroticism had the strongest influence on the association between MUPS and HCU over time. As far as we know, no research has been published on the comorbidity of MUPS with psychiatric disorders and personality traits with regard to HCU, especially not in a longitudinal design. De Waal et al. found that somatoform disorders and depressive disorders were almost equally associated with HCU, but that the undifferentiated somatoform disorder had an independent effect after adjusting for psychiatric disorders [22]. Noyes et al. found that MUPS were associated with specific personality traits as neuroticism and that this led to increased care seeking behaviour, which is in accordance with our findings [24]. However, in contrast with our findings, Carlier et al. found no association between somatoform disorders and specific personality traits in their cross-sectional study [38]. As an explanation for this result they argued that their somatoform disorders patients were mostly highly educated and married, indicating a stable personal life.

One should realize that we examined the relation between MUPS and HCU in a sample with predominantly depressive and/or anxiety disorders. This may impede the interpretation and generalizability of the results. Therefore, we investigated possible effect modification between MUPS and depressive disorders and between MUPS and anxiety disorders. These analyses showed that within the population without these disorders the relationship between MUPS and HCU was stronger than in the population with these disorders. Based on those analyses, we believe that the observed relations may also hold for the general population.

A main strength of our study was that we used the NESDA cohort to answer our research questions. By using this cohort, longitudinal data over 2 years from a large sample of participants were available, recruited from both primary care, secondary mental health care and the general population. Also, we adjusted all analyses for chronic somatic diseases. Furthermore, we used structured diagnostic interviews and not only self-report questionnaires.

However, our findings should be interpreted in the light of several limitations. First, the NESDA cohort study used the 4DSQ somatisation scale to measure MUPS. As with all existing MUPS questionnaires, it lacks judgement of a clinician to verify that symptoms are really unexplained [39]. However, the 4DSQ highly correlates with the PHQ-15 and SCL-90, questionnaires that are widely used to measure MUPS [27], and may be considered as an adequate proxy measure for MUPS. Second, HCU was measured over the past 6 months, while MUPS were measured over the past week, leading to incongruence. However, we do not believe that the results were affected much by this. We found a correlation coefficient of 0.72 of MUPS over the 2 years (data not shown), indicating a quite stable pattern of MUPS over time. Also, our time-lag analyses showed results that were similar to those of the standard analyses, with even a slightly higher effect of MUPS on HCU for the number of contacts. Third, as HCU was asked over the past 6 months, the risk of recall bias exists. However, the direction of this bias is unclear. Other studies have used electronic medical records to assess HCU, but these can also be incomplete [40]. Fourth, we have no information on the actual reasons for health care use and have only assessed the quantity and not the appropriateness of provided health care. Also we did not control for the use of psychopharmacological therapy which could have influenced the results, nor can we determine whether MUPS were the primary problems or symptoms from a psychiatric disorder. Fifth, only the baseline sociodemographic variables and number of chronic diseases were included as covariates in the analyses, although several of these covariates may have changed over the course of 2 years. However, we do not believe that the results of our adjusted analyses would be different because the influence of these covariates was only marginal. Finally, based on our study we cannot infer causality. Therefore it might at least theoretically be possible that an increase in HCU is leading to more MUPS.

As we found that MUPS are independently associated with HCU, attention should be paid to early identification and adequate treatment of MUPS in clinical practice. Also, physicians should be aware of signs of depression and anxiety and personality traits as they have an influence on the association between MUPS and HCU.

A possible explanation for prolonged high HCU is that care for patients with MUPS is often fragmented, as diagnostics and treatments are carried out by different health care providers. To reduce high and possibly inadequate health care for these patients, the issue of adequacy and fragmentation of health care patterns should be further examined, also in relation to the patient’s quality of life. Guidelines for MUPS across disciplines and different health care settings may be instrumental in this examination [41, 42].

Also, it would be interesting to examine the role of health anxiety as a predictor of HCU in a longitudinal design. As we found neuroticism to be a predictor, it is possible that health anxiety could play an important role as well as these are often related.