Five-year trajectories of symptom severity, physical and mental functioning in patients with persistent somatic symptoms: the PROSPECTS cohort study

Abstract

Objectives

Knowledge about the long-term course and prognosis of persistent somatic symptoms (PSS) is important to improve clinical decision-making and guidance for patients with PSS. Therefore, we aimed to: (1) identify distinct 5-year trajectories of symptom severity, physical and mental functioning in adult patients with PSS and (2) explore patient characteristics associated with these trajectories.

Design

We used longitudinal data (seven measurements over a 5-year period) of the PROSPECTS study: a prospective cohort of adult patients with PSS. We applied Latent Class Growth Mixture Modelling to identify distinct trajectories for the three outcomes.

Setting and participants

Patients were recruited in general practices and specialised treatment facilities for PSS throughout the Netherlands. The study population consisted of participants with three or more measurements available (n=297).

Primary outcome measures

Symptom severity (Patient Health Questionnaire 15), physical and mental functioning (RAND-36 Physical Component Summary and Mental Component Summary).

Results

For symptom severity, we identified two ‘stable’ trajectories: ‘severe symptoms, stable’ (15.8%) and ‘moderate symptoms, stable’ (84.2%). For physical functioning, we identified three trajectories: ‘poor physical functioning, marked improvement’ (8.5%); ‘poor physical functioning, stable’ (34.7%) and ‘moderate physical functioning, slight improvement’ (56.8%). For mental functioning, we identified three trajectories: ‘poor mental functioning, marked improvement’ (13.9%); ‘moderate mental functioning, deterioration’ (12.2%) and ‘moderate mental functioning, slight improvement’ (73.8%). Patients’ characteristics such as personal, social and environmental background, illness stressors, comorbid diseases, cognitive, emotional and behavioural responses varied for the distinct trajectories.

Conclusions

We identified distinct 5-year trajectories for the three outcomes. Our findings suggest a high prevalence of persistence of symptoms and limited improvement in physical and mental functioning in the majority of patients with PSS. In a small proportion of patients, we identified trajectories that showed considerable physical or mental improvement or deterioration. Patient characteristics differed for the identified trajectories and may guide early recognition, although predictive studies are warranted.

Correspondence to Dr Hieke Barends; [email protected]

STRENGTHS AND LIMITATIONS OF THIS STUDY

• Latent Class Growth Mixed Modelling (LCGMM)—a contemporary and flexible data-driven technique—was applied to identify distinct 5-year trajectories of symptom severity, physical and mental functioning in patients with persistent somatic symptoms (PSS).

• Strengths of this study include its multicentre prospective design, the long duration of follow-up using multiple measurements over time and that we assessed symptom severity (and persistence) as well as health-related functioning.

• The different trajectories found may adequately summarise general long-term trends, but we should keep in mind that LCGMM gives a simplified version of the complex heterogeneous reality.

• The majority of participants already experienced symptoms for an extensive period of time at inclusion (median duration 5 years; IQR 14 years), trajectories shortly after the onset of PSS thus may differ from the trajectories identified in this study.

Introduction

In primary care, up to one-third of patients present with somatic symptoms that cannot be (fully) explained by an organic disease.^{1 2} In specialist medical care, the numbers even run as high as 70%, depending on the specialty.^{3 4} When these symptoms persist, they are associated with substantial psychological distress and functional impairment.^5–7 They are also a burden for society because of high healthcare use and work-related disability.^{8 9} There is ongoing debate about terms and definitions in this field. Throughout this manuscript, we will use the umbrella term ‘persistent somatic symptoms’ (PSS), see box 1 on the used definition.

Box 1

Persistent somatic symptoms (PSS)

Patients can suffer from PSS in the context of well-understood (adequately treated) conditions or when no somatic explanation for symptoms is found.^{71 72} Biological, psychological and social factors may all play a role in the persistence of somatic symptoms. In this study, PSS were defined as somatic symptoms that last at least several weeks and for which no sufficient somatic explanation is found after proper medical examination by a physician. This definition is in line with the Dutch guidelines for medically unexplained symptoms.^{73 74}

The presentation and duration of PSS vary widely, from relatively mild and self-limiting symptoms at one end of the spectrum to persistent symptoms with severe impact on functioning at the other end. Knowledge about the long-term course and prognosis of PSS is scarce. Prior studies investigating the course of PSS have several shortcomings, for example, they are usually relatively limited in follow-up, use a single measurement to determine rates of improvement, worsening or chronicity and focus mainly on symptom persistence and not on physical and mental functioning, which are important aspects of prognosis for patients. A systematic review dating back to 2009 found that 50–75% of patients showed improvement in symptoms over time, while 10–30% worsened or became chronic.¹⁰ Most studies published after this review^11–18 showed higher rates of symptom persistence, even after longer periods. For example, 56.8% of the participants had persistent symptoms after 2 years of follow-up in a primary care population.¹⁵ In sum, there is currently no study available that assessed long-term trajectories of PSS and most studies ignore physical and mental functioning.

Various patient characteristics have previously been associated with the prognosis of PSS,^10–20 for example, sex^{16 17 19} and number of reported symptoms.^{10 11 13 15 18–20} There is also a range of psychological factors that may influence the course of PSS.²¹ Until now, no study examined these characteristics in relation to different long-term trajectories of symptoms and functioning in patients with PSS.

The PROSPECTS cohort study on symptoms and physical and mental functioning in patients with PSS started in 2013.²² We have previously reported on the 2-year course of the PROSPECTS study.²³ In the current publication, we focus on the 5-year course. We wanted to see if similar or new trajectories emerge over this longer period. In addition, we wanted to examine patient characteristics of the identified trajectories. To conclude, the aims of our current study are twofold: (1) to identify distinct trajectories of symptom severity, physical and mental functioning over a 5-year course in patients with PSS and (2) to explore patient characteristics associated with these trajectories.

Method

Context: the PROSPECTS study

The PROSPECTS study is a Dutch prospective cohort study on symptoms and physical and mental functioning in patients with PSS. A detailed research protocol and results of the recruitment process were published previously.^{22 24} In box 2 we present a summary of the patient selection procedure and in figure 1 a flow chart of the inclusion and follow-up of data collection of the PROSPECTS study. In total, 325 adult patients (18–70 years old) were recruited in general practices (n=218) and in specialised PSS programmes of secondary and tertiary care organisations (n=107) across the Netherlands in 2013–2015. Participants filled out questionnaires at predefined intervals. The questionnaires were sent by mail. The time intervals were kept constant. After the baseline measurement, follow-up measurements took place at 6 months, 1, 2, 3, 4 and 5 years. When participants missed one follow-up measurement, we offered them the option to resume participation at the next follow-up moment. As the study was extended after the first three follow-up years, written consent was again obtained before the 4-year measurement. Only participants who had not opted out of further participation were contacted.

Box 2

Patient selection and flow chart PROSPECTS study

Patient selection

In primary care, patients who visited their general practitioner (GP) ≥2 times with an unexplained physical symptom in the preceding 3 months were selected. An electronic database search was carried out, based on the ‘Robbins list’, a list of 23 frequently unexplained physical symptoms composed by Robbins et al,⁸² in combination with a lacking diagnosis International Classification of Primary Care (ICPC) code (meaning an ICPC code >70). Selected patients were checked for exclusion criteria by their own GP. In secondary and tertiary care, newly referred patients with medically unexplained symptoms as the reason for referral were screened for inclusion and exclusion criteria by the physician performing the intake consultation. The severity of symptoms was measured using the Patient Health Questionnaire 15 (PHQ-15).²⁵ The patient had to have a score of 2 on at least one symptom of the PHQ-15 questionnaire, indicating that the symptom was bothering a lot, in order to participate in the PROSPECTS study.

Exclusion criteria were: (1) a sufficient medical explanation for the symptoms (according to the physician), (2) incomplete diagnostic evaluation of the symptoms (according to the physician), (3) insufficient command of the Dutch language, (4) a cognitive or visual impairment that prohibited participating in a questionnaire survey, (5) severe psychopathology (eg, psychotic disorder, bipolar disorder), (6) pregnancy, (7) cancer diagnosed in the 5 years prior to inclusion or (8) another life-threatening condition or a short life expectancy.

Figure 1. Flow chart of inclusion and longitudinal data collection of the PROSPECTS study. GP, general practitioner; PHQ, Patient Health Questionnaire.

Patient and public involvement

Patients or the public were not actively involved in the design, conduct, reporting or dissemination plans of our research.

Study population

For the present study, we used seven measurements in time (baseline, 6 months, 1, 2, 3, 4 and 5 years data). For this analysis, we excluded participants (n=3) in whom their PSS turned out to be fully explained by a somatic disease that was diagnosed during follow-up (pancreas carcinoma, congenital hip dysplasia, Crohn’s disease). Participants with a minimum of three measurements available were selected as the study population for this analysis (n=297; n=196 recruited in primary care; n=101 recruited in specialised PSS care). Measurements of participants with a newly diagnosed serious disease during follow-up (such as a life-threatening cancer diagnosis) were censored from the time of disease onset.

Measures

The primary outcomes of the trajectories were symptom severity (Patient Health Questionnaire-15 (PHQ-15)), physical functioning (RAND-36 Physical Component Summary (PCS)) and mental functioning (RAND-36 Mental Component Summary (MCS)) collected at 0, 6 months, 1, 2, 3, 4 and 5 years of follow-up (see online supplemental appendix A).

Symptom severity (primary outcome)

The 15-item PHQ-15²⁵ was used as an indicator of the level of somatic symptom severity. The PHQ-15 is a reliable measure of somatic symptom reporting in this field of research.^26–28 A higher score on the scale indicates more severe symptoms. We considered a change of at least five points a minimal important change (MIC) on the PHQ-15. Based on priory suggested cut-offs points, we considered a score below 5 as ‘minimal’, a score of 5–9 as ‘mild’, a score of 10–14 as ‘moderate’ and a score of 15 or higher as ‘severe’ somatic symptoms.²⁵

Physical and mental functioning (primary outcomes)

The PCS and MCS of the RAND-36 questionnaire (V.2.0) were used as indicators of physical and mental functioning.^{29 30} The RAND-36 questionnaire is a 36-item self-report survey of patient health in the preceding month. It was selected because of its validity³⁰ and previous use in this area of research. The RAND-36 PCS and RAND-36 MCS were calculated to summarise physical and mental functioning into a 0–100 scale score. A higher score indicates better functioning. A change of at least 0.5 times the SD (10 points on both scales in this study) for physical and mental functioning is considered an MIC.³¹ We considered a score above 70 as ‘average’, between 40–70 as ‘moderate’ and below 40 as ‘poor’ physical and/or mental functioning.

Patient characteristics associated with trajectories

We explored the association of several patient characteristics for the identified trajectories of PSS. The selection of patient characteristics and questionnaires (see online supplemental appendix A) was based on characteristics previously associated with the prognosis of PSS,^10–20 theoretical models³² and availability in our data set.²² Patient characteristics were measured partly at baseline (T0) and partly at the first follow-up measurement at 6 months (T1), in order to reduce participant burden at baseline; the characteristics measured at T1 are considered relatively stable over time in adults.³³ For clarity, patient characteristics were categorised into categories suggested in previous literature:³⁴ personal background, social and environmental background, illness stressors, comorbid diseases and cognitive, emotional and behavioural responses.

Statistical analysis

Data analyses were performed using IBM (New York, USA) SPSS Statistics (V.26) and Muthén & Muthén (Los Angeles, USA) Mplus (V.8.5). Descriptive statistics are presented as mean with SD for normally distributed continuous data; median, IQR for skewed continuous variables and as numbers and percentages for dichotomous and categorical variables.

Identifying distinct trajectories by Latent Class Growth Mixture Modelling

In Latent Class Growth Mixture Modelling (LCGMM), the assumption is that within a study sample, individuals may come from distinct underlying (or latent) subpopulations, that follow different and unique trajectories over time. The main aim is to identify the number and characteristics of these trajectories.

A ‘forward approach’ was applied,³⁵ similar to that in the analyses of the 2-year course of the PROSPECTS study,²³ starting with a model with one trajectory and adding more trajectories while evaluating the model fit. Models with one to six trajectories were run. We determined the optimal number of trajectories by using several statistical parameters combined with the clinical interpretability of the identified trajectories, in order to avoid clinically uninterpretable trajectories.³⁵ Statistical parameters included the Bayesian information criterion (BIC), the Akaike information criteria (AIC), posterior probabilities and entropy.³⁶ Both a lower BIC and AIC correspond to a better model fit.^{36 37} In addition, the higher the posterior probabilities and the higher the entropy (closer to 1), the fewer the classification errors and the less bias in the prediction of trajectory membership.³⁸ Posterior probabilities of at least 0.7 or 0.8 are advised.^39–41 Entropy indicates the ‘fuzziness’ of the model and is a common indicator of uncertainty in classification.^{41 42}

Unequally spaced time points of our data collection (ie, intervals of 6 months and 12 months) were modelled. As we anticipated within-class heterogeneity based on the 2-year analyses,²³ we used LCGMM instead of Latent Class Growth Analysis.^{43 44} LCGMM allows for variation in intercept and slope in one or more classes, leading to larger within-class heterogeneity.^{35 41 45} To accommodate anticipated fluctuations over time based on the 2-year analyses,²³ we estimated and compared models with linear and quadratic terms. In LCGMM, missing data are handled with the Expectation-Maximation Algorithm⁴⁶; therefore, there was no need to impute missing data. Trajectories resulting from the LCGMM were visualised alongside the mean trajectories of all participants. Subsequently, inter-individual and intra-individual variation within the different trajectories was visualised by producing graphs containing the individual courses of all participants per identified trajectory.

Differences in patient characteristics for the identified trajectories

We followed a standard three-step method as described by Andersen et al.⁴⁷ First, we determined the optimal number of latent trajectories as described above (step 1); subsequently, we selected the most likely trajectory for each participant (step 2) and finally, we explored differences in the selected patient characteristics between the identified trajectories (step 3).

For the selected patient characteristics, the questionnaires used^{25 29 30 48–68} can be found in online supplemental appendix A. Normally distributed data were tested with parametric tests; non-parametric tests were used for non-normally distributed data. For categorical variables, we used χ² tests. For continuous variables with a normal distribution we used either the (parametric) independent sample t-tests or analysis of variance and post hoc Bonferroni correction, depending on the number of trajectories identified. Post hoc Bonferroni correction was applied to control for the overall probability of type I error (false positive) for multiple hypothesis testing. For continuous variables with skewed distribution, we used the (non-parametric) Mann-Whitney test or Kruskal-Wallis test, depending on the number of trajectories.

Results

Study population

Characteristics of the study population at the baseline measurement are presented intable 1. At recruitment, participants showed on average moderate levels of somatic symptoms severity and moderate levels of physical and mental functioning. The three most frequently reported somatic comorbid diseases were ‘osteoarthritis’ (21.9%, n=65), ‘hypertension’ (19.5%, n=58) and ‘eczema or other skin disease’ (16.5%, n=49). The three most frequently mentioned psychiatric comorbid disorders were ‘burn-out syndrome’ (25.3%, n=75), ‘depressive disorder’ (18.2%, n=54) and ‘anxiety disorder’ (15.2%, n=45). The mean duration of symptoms at recruitment was 9.4 years (SD 11.1 years) and the median duration was 5 years (IQR 14 years), indicating that the distribution was skewed to the right. Based on cut-offs presented in prior literature, the study population scored on average mild on both anxiety and depression severity⁶⁹ and high on health anxiety⁷⁰ (table 1).

Table 1

Characteristics of the study population (n=297) at baseline

	Study population (N=297)	N missing (%)
Outcomes
Symptom severity (PHQ-15, scale 0–30), mean (SD)	12.3 (5.3)	0 (0)
Physical functioning (RAND-36 PCS, scale 0–100), mean (SD)	47.1 (19.5)	4 (1.3)
Mental functioning (RAND-36 MCS, scale 0–100), mean (SD)	53.0 (20.2)	4 (1.3)
Personal background
Women, n (%)	226 (76.1)	0 (0)
Education level, n (%)		0 (0)
Lower education	98 (33.0)
Intermediate education	117 (39.4)
Higher education	82 (27.6)
Age in years, range 19–70, mean (SD)	46.8 (12.4)	0 (0)
Potentially traumatic events in childhood (<16 years) (LEQ), median (IQR; total range)	0 (1; 0–7)	0 (0)
Perfectionism (MDPS, scale 29–145), mean (SD)*	70.1 (22.6)	30 (10.1)
Neuroticism (NEO-FFI, scale 12–60), mean (SD)*	32.6 (9.0)	16 (5.4)
Extraversion (NEO-FFI, scale 12–60), mean (SD)*	38.4 (6.5)	17 (5.7)
Social and environmental background
Marital status, n (%)		0 (0)
Married or cohabiting	181 (60.9)
Other (single, divorced, widow)	116 (39.1)
Social support (SoS, scale 12–60), median (IQR; total range)	56 (12; 15–60)	0 (0)
Illness stressors
Number of symptoms (PSQ), mean (SD)	22.5 (9.5)	2 (0.7)
Duration of symptoms in years, median (IQR; total range)	5 (14; 0.02–62)	8 (2.7)
Comorbid diseases
Number of somatic comorbid diseases (TiC-P), median (IQR; total range)	1 (2; 0–8)	2 (0.7)
Number of psychiatric comorbid disorders (TiC-P), median (IQR; total range)	0 (1; 0–3)	2 (0.7)
Cognitive responses
Fear-avoidance cognitions (CBRQ, scale 0–24), mean (SD)	9.9 (4.7)	3 (1.0)
Catastrophising cognitions (CBRQ, scale 0–16), mean (SD)	5.7 (3.5)	2 (0.7)
Damage cognitions (CBRQ, scale 0–20), mean (SD)	9.8 (3.6)	3 (1.0)
Embarrassment avoidance cognitions (CBRQ, scale 0–24), median (IQR; total range)	7.0 (8; 0–24)	3 (1.0)
Symptom focusing (CBRQ, scale 0–24), mean (SD)	10.4 (5.2)	5 (1.7)
Somatosensory amplification (SSAS, scale 10–50), mean (SD)	27.1 (6.3)	1 (0.3)
Identity (label of disease and associated symptoms, IPQ-brief, scale 0–10), mean (SD)	7.2 (1.9)	0 (0)
Consequences of illness (IPQ-brief, scale 0–10), median (IQR; total range)	8 (3.5; 0–10)	0 (0)
Timeline (expected duration of illness, IPQ-brief, scale 0–10), median (IQR; total range)	8 (4; 0–10)	1 (0.3)
Personal control (over illness, IPQ-brief, scale 0–10), mean (SD)	5.9 (2.8)	0 (0)
Treatment control (beliefs about treatment effect on illness, IPQ-brief, scale 0–10), mean (SD)	4.3 (2.6)	4 (1.3)
Comprehension of illness (extent to which patients believe to understand their illness, IPQ-brief, scale 0–10), mean (SD)	4.1 (2.7)	0 (0)
Emotional responses
Anxiety severity (BAI, scale 0–63), median (IQR; total range)	10 (11.3; 0–51)	19 (6.4)
Depression severity (QIDS-SR, scale 0–27), mean (SD)	9.2 (5.0)	3 (1.0)
Positive affect (PANAS, scale 10–50), mean (SD)	32.5 (8.2)	4 (1.3)
Health anxiety (WI, scale 0–14), mean (SD)	5.8 (2.6)	16 (5.4)
Concern (about illness, IPQ-brief, scale 0–10), mean (SD)	5.5 (2.8)	1 (0.3)
Emotions (extent to which illness affects mood, IPQ-brief, scale 0–10), mean (SD)	7 (2.6)	0 (0)
Behavioural responses
All-or-nothing behaviour (CBRQ, scale 0–20), mean (SD)	7.8 (4.5)	4 (1.3)
Avoidance/resting behaviour (CBRQ, scale 0–32), mean (SD)	8.2 (4.8)	11 (3.7)
Physical activity (IPAQ, total METs per week), median (IQR; total range)	2729 (5832; 0–63 312)	18 (6.1)

*As indicated in the method section, these characteristics were measured at T1.

BAI, Beck Anxiety Inventory; CBRQ, Cognitive Behavioural Response Questionnaire; IPAQ, International Physical Activity Questionnaire; IPQ-brief, brief illness perception questionnaire; LEQ, Life Events Questionnaire; MDPS, Multidimensional Perfectionism Scale; METs, Metabolic Equivalents; NEO-FFI, NEO Personality Questionnaire-Five Factor Inventory; PANAS, subscale of Positive and Negative Affect Schedule; PHQ-15, 15-item Patient Health Questionnaire; PSQ, Physical Symptom Questionnaire; QIDS-SR, Quick Inventory of Depressive Symptomatology; RAND-36, RAND 36-item Health Survey; RAND-36 MCS, RAND-36 Mental Component Summary; RAND-36 PCS, RAND-36 Physical Component Summary; SoS, Social Support Scale; SSAS, Somatosensory Amplification Scale; TiC-P, Treatment Inventory of Costs in patients with psychiatric disorders; WI, Whitely Index.

Results of LCGMM

Of the 297 participants, 217 (73%) completed all seven measurements. Overall, quadratic models with estimated within-class intercept variance performed best, these models are presented here (for model fit indices, see online supplemental appendix B).

Trajectories of symptom severity

We took into account BIC, posterior probabilities, entropy, class sample size as well as clinical interpretability. This resulted in the selection of the two-trajectory model (see figure 2), containing a ‘severe symptoms, stable’ (15.8%, n=47) and a ‘moderate symptoms, stable’ (84.2%, n=250) trajectory. Based on the MIC for the PHQ-15 (5-point change), both trajectories showed no clinically relevant improvement or deterioration over 5 years. Note that the posterior probability of the ‘stable moderate symptoms’ trajectory was suboptimal (0.618, ie, below the desired >0.7), indicating that for this trajectory there was more uncertainty in classification.

Figure 2. Identified 5-year trajectories. PHQ-15, Patient Health Questionnaire-15; RAND-36 MCS, RAND-36 Mental Component Summary; RAND-36 PCS, RAND-36 Physical Component Summary.

Trajectories of physical functioning

Model fit indices for the physical functioning trajectories are presented in online supplemental appendix B. Models with 3-trajectories and 4-trajectories were close in BIC, AIC was slightly better for the 4-trajectory model, however, adding an additional trajectory did not result in a clinically useful additional cluster. As posterior probabilities and entropy were acceptable for the 3-trajectory model, this model was selected (see figure 2), consisting of a ‘poor physical functioning, marked improvement’ (8.5%, n=25) trajectory, a ‘poor physical functioning, stable’ (34.7%, n=102) trajectory and a ‘moderate physical functioning, slight improvement’ (56.8%, n=167) trajectory. Based on the MIC for the RAND-36 PCS (10-point change), the ‘poor physical functioning, marked improvement’ trajectory showed clear clinically relevant improvement over 5 years (about four times the MIC). The ‘moderate physical functioning, slight improvement’ trajectory improved by about the MIC over the 5-year course.

Trajectories of mental functioning

Based on the combination of fit indices, class sample size as well and clinical interpretability we selected the 3-trajectory model (see figure 2), including a ‘poor mental functioning, marked improvement’ (13.9%, n=41) trajectory, a ‘moderate mental functioning, deterioration’ (12.2%, n=36) trajectory and a ‘moderate mental functioning, slight improvement’ (73.8%, n=217) trajectory. For the ‘poor mental functioning, marked improvement’ trajectory there was more uncertainty in classification (posterior probability below the desired >0.7, see online supplemental appendix B). Based on the MIC for the RAND-36 MCS (10-point change), the ‘poor mental functioning, marked improvement’ trajectory improved roughly 3.5 times the MIC over the course of 5 years. For the ‘moderate mental functioning, slight improvement’ trajectory, participants improved by about the MIC. Participants in the ‘moderate mental functioning, deterioration’ trajectory deteriorated by on average twice the MIC.

Visual evaluation of identified trajectories

We produced graphs for individual participants per trajectory (see online supplemental appendix C). Visual inspection of these graphs indicated considerable inter-individual and intra-individual variation in the identified trajectories.

Exploring patient characteristics for identified trajectories

Patient characteristics for trajectories of symptom severity

The results of the analyses are presented in table 2. Patients in the ‘severe symptoms, stable’ trajectory scored lower on extraversion, were less often married or cohabiting and reported lower social support. They reported a higher number of symptoms and more somatic comorbid diseases at recruitment. In addition, they reported more unfavourable cognitive, emotional and behavioural responses than patients in the ‘moderate symptoms, stable’ trajectory.

Table 2

Differences in patient characteristics for trajectories of symptom severity

Symptom severity (PHQ-15) trajectories
Trajectory	(A) Severe symptoms, stable (n=47)	(B) Moderate symptoms, stable (n=250)	P value between clusters
Personal background
Women, n (%)	41 (87.2)	185 (74.0)	p=0.051
Education level, n (%)			p=0.222
Lower education	20 (42.6)	78 (31.2)
Intermediate education	18 (38.3)	99 (39.6)
Higher education	9 (19.1)	73 (29.2)
Age in years, range 19–70, mean (SD)	47.1 (11.6)	46.7 (12.5)	p=0.830
Potentially traumatic events in childhood (<16 years) (LEQ), median (IQR; total range)	1 (2; 0–6)	0 (1; 0–7)	p=0.073
Perfectionism (MDPS), mean (SD)*	71.5 (24.5)	69.9 (22.3)	p=0.669
Neuroticism (NEO-FFI), mean (SD)*	35.0 (7.8)	32.2 (9.1)	p=0.056
Extraversion (NEO-FFI), mean (SD)*	36.6 (6.7)	38.7 (6.4)	p=0.044
Social and environmental background
Marital status, n (%)			p=0.030
Married or cohabiting	22 (46.8)	159 (63.6)
Other (single, divorced, widow)	25 (53.2)	91 (36.4)
Social support (SoS), median (IQR; total range)	53 (14; 22–60)	57 (11; 15–60)	p=0.004
Illness stressors
Number of symptoms (PSQ), mean (SD)	30.8 (7.6)	20.9 (9.0)	p<0.001
Duration of symptoms in years, median (IQR; total range)	10 (19; 0.06–45)	4 (13; 0.02–62)	p=0.096
Comorbid diseases
Number of somatic comorbid diseases (TiC-P), median (IQR; total range)	2 (2; 0–8)	1 (2; 0–7)	p<0.001
Number of psychiatric comorbid disorders (TiC-P), median (IQR; total range)	0 (1; 0–3)	0 (1; 0–3)	p=0.312
Cognitive responses
Fear-avoidance cognitions (CBRQ), mean (SD)	11.9 (4.7)	9.5 (4.7)	p=0.002
Catastrophising cognitions (CBRQ), mean (SD)	6.9 (3.2)	5.4 (3.5)	p=0.009
Damage cognitions (CBRQ), mean (SD)	10.7 (3.7)	9.6 (3.6)	p=0.069
Embarrassment avoidance cognitions (CBRQ), median (IQR; total range)	10 (7; 1–22)	6 (9; 0–24)	p<0.001
Symptom focusing (CBRQ), mean (SD)	10.8 (5.2)	10.3 (5.2)	p=0.62
Somatosensory amplification (SSAS), mean (SD)	29.8 (5.8)	26.6 (6.2)	p=0.001
Identity (label of disease and associated symptoms, IPQ-brief), mean (SD)	8.2 (1.6)	7.0 (1.9)	p<0.001
Consequences (beliefs about illness effects and outcomes, IPQ-brief), median (IQR; total range)	9 (3; 2–10)	7 (4; 0–10)	p<0.001
Timeline (expected duration of illness, IPQ-brief), median (IQR; total range)	9 (3; 0–10)	8 (4; 0–10)	p=0.257
Personal control (over illness, IPQ-brief), mean (SD)	6.0 (2.9)	5.9 (2.8)	p=0.700
Treatment control (beliefs about treatment effect on illness, IPQ-brief), mean (SD)	4.9 (2.9)	4.2 (2.5)	p=0.121
Comprehensibility (extent to which patients believe to understand their illness, IPQ-brief), mean (SD)	3.8 (2.4)	4.1 (2.8)	p=0.465
Emotional responses
Anxiety severity (BAI), median (IQR; total range)	17 (18; 6–44)	9 (10; 0–51)	p<0.001
Depression severity (QIDS-SR), mean (SD)	12.2 (4.6)	8.6 (4.6)	p<0.001
Positive affect (PANAS), mean (SD)	29.7 (7.1)	33.0 (8.3)	p=0.010
Health anxiety (WI), mean (SD)	7.3 (3.0)	5.6 (2.6)	p<0.001
Concern (about illness, IPQ-brief), mean (SD)	6.2 (2.9)	5.4 (2.8)	p=0.070
Emotions (extent to which illness affects mood, IPQ-brief), mean (SD)	7.4 (2.5)	6.0 (2.9)	p=0.003
Behavioural responses
All-or-nothing behaviour (CBRQ), mean (SD)	9.8 (4.0)	7.5 (4.5)	p=0.001
Avoidance/resting behaviour (CBRQ), mean (SD)	10.3 (4.6)	7.8 (4.8)	p=0.002
Physical activity (IPAQ, in METs), median (IQR; total range)	1916 (3142; 0–44 436)	2880 (6703; 0–63 312)	p=0.087

In bold p<0.05.

*As indicated in the method section, these characteristics were measured at T1 (6 months follow-up); differences between trajectories were tested with independent sample t-test for mean values, with χ² tests for proportions and Mann-Whitney U test for median values.

BAI, Beck Anxiety Inventory; CBRQ, Cognitive Behavioural Response Questionnaire; IPAQ, International Physical Activity Questionnaire; IPQ-brief, Brief Illness Perception Questionnaire (higher scores indicating more negative illness perceptions); LEQ, Life Events Questionnaire; MDPS, Multidimensional Perfectionism Scale; METs, Metabolic Equivalents; NEO-FFI, NEO Personality Questionnaire-Five Factor Inventory; PHQ-15, Patient Health Questionnaire-15; PSQ, Physical Symptom Questionnaire; QIDS-SR, Quick Inventory of Depressive Symptomatology; SoS, Social Support Scale; TiC-P, Treatment Inventory of Costs in patients with psychiatric disorders; WI, Whitely Index.

Patient characteristics for trajectories of physical functioning

Between-trajectories differences in patient characteristics are presented in table 3. Participants in the different trajectories of physical functioning varied in all categories studied. Participants who improved markedly (‘poor physical functioning, marked improvement’) reported the lowest physical activity and highest personal control over illness at recruitment. In addition, they reported fewer symptoms and somatic comorbid diseases and higher social support than participants who remained ‘stable’ (‘poor physical functioning, stable’ trajectory). Compared with participants in the ‘moderate physical functioning, slight improvement’ trajectory, participants in both ‘poor’ trajectories (‘poor physical functioning, marked improvement’ and ‘poor physical functioning, stable’) reported more unfavourable cognitive, emotional and behavioural responses at recruitment. Participants in the least favourable trajectory (‘poor physical functioning, stable’) were older and suffered from more symptoms and comorbid diseases than participants in the ‘moderate physical functioning, slight improvement’ trajectory.

Table 3

Differences in patient characteristics for trajectories of physical functioning

Physical functioning (PCS) trajectories
Trajectory	(A) Poor physical functioning, marked improvement (n=25)	(B) Poor physical functioning, stable (n=102)	(C) Moderate physical functioning, slight improvement (n=167)	P value between clusters
Personal background
Women, n (%)	18 (72.0)	82 (80.4)	125 (74.9)	p=0.498
Education level, n (%)				p=0.003
Lower education	9 (36.0)	46 (45.1)	41 (24.6
Intermediate education	11 (44.0)	38 (37.3)	67 (40.1)
Higher education	5 (20.0)	18 (17.6)	59 (35.3)
Age in years, range 19–70, mean (SD)	47.2 (12.9)	49.9 (11.4)	44.8 (12.6)	B>C (p=0.003)
Potentially traumatic events in childhood (<16 years) (LEQ), median (IQR; total range)	0 (1; 0–4)	1 (2; 0–7)	0 (1; 0–6)	p=0.151
Perfectionism (MDPS), mean (SD)*	71.0 (21.9)	71.7 (22.1)	68.7 (23.1)	p=0.956
Neuroticism (NEO-FFI), mean (SD)*	33.7 (9.6)	34.8 (8.3)	31.0 (8.9)	B>C(p=0.003)
Extraversion (NEO-FFI), mean (SD)*	37.6 (6.2)	36.6 (6.6)	39.6 (6.3)	B<C (p<0.001)
Social and environmental background
Marital status, n (%)				p=0.507
Married or cohabiting	17 (68.0)	58 (56.9)	104 (62.3)
Other (single, divorced, widow)	8 (32.0)	44 (43.1)	63 (37.7)
Social support (SoS), median (IQR; total range)	59 (17; 22–60)	54 (16; 15–60)	57 (9; 33–60)	p=0.047
Illness stressors
Number of symptoms (PSQ), mean (SD)	22.4 (6.7)	28.0 (8.8)	19.0 (8.7)	B>A (p=0.012); B>C (p<0.001)
Duration of symptoms in years, median (IQR; total range)	6 (12; 0.17–40)	8 (19; 0.06–62)	3 (11; 0.02–50)	p=0.018
Comorbid diseases
Number of somatic comorbid diseases (TiC-P), median (IQR; total range)	1 (2; 0–4)	2 (2.5; 0–8)	1 (2; 0–6)	p<0.001
Number of psychiatric comorbid disorders (TiC-P), median (IQR; total range)	0 (1.5; 0–3)	0 (1; 0–3)	0 (1; 0–3)	p=0.529
Cognitive responses
Fear-avoidance cognitions (CBRQ), mean (SD)	9.9 (5.1)	12.01 (4.6)	8.5 (4.2)	B>C (p<0.001)
Catastrophising cognitions (CBRQ), mean (SD)	7.5 (3.6)	7.1 (3.0)	4.4 (3.2)	A>C (p<0.001); B>C (p<0.001)
Damage cognitions (CBRQ), mean (SD)	9.8 (3.4)	11.0 (3.2)	9.0 (3.6)	B>C (p<0.001)
Embarrassment avoidance cognitions (CBRQ), median (IQR; total range)	7 (10.5; 0–18)	8 (6; 0–22)	6 (8; 0–24)	p<0.001
Symptom focusing (CBRQ), mean (SD)	10.4 (5.3)	11.0 (4.6)	10.0 (5.5)	p=0.325
Somatosensory amplification (SSAS), mean (SD)	28.6 (6.3)	27.6 (6.4)	26.6 (6.0)	p=0.398
Identity (label of disease and associated symptoms, IPQ-brief), mean (SD)	8.6 (1.2)	8.0 (1.6)	6.4 (1.9)	A>C (p<0.001); B>C (p<0.001)
Consequences (beliefs about illness effects and outcomes, IPQ-brief), median (IQR; total range)	9 (3; 3–10)	8 (3; 3–10)	7 (4; 0–10)	p<0.001
Timeline (expected duration of illness, IPQ-brief), median (IQR; total range)	10 (3.5; 4–10)	9.5 (2; 3–10)	7 (4; 0–10)	p<0.001
Personal control (over illness, IPQ-brief), mean (SD)	7.7 (2.3)	5.8 (3.0)	5.6 (2.7)	B<A (p=0.008); C<A (p=0.002)
Treatment control (beliefs about treatment effect on illness, IPQ-brief), mean (SD)	4.4 (2.9)	4.4 (2.5)	4.3 (2.6)	p=1.00
Comprehensibility (extent to which patients believe to understand their illness, IPQ-brief), mean (SD)	3.9 (3.3)	3.9 (2.4)	4.2 (2.8)	p=0.912
Emotional responses
Anxiety severity (BAI), median (IQR; total range)	11 (11; 1–35)	12.5 (12.75; 1–51)	8 (10, 0–38)	p<0.001
Depression severity (QIDS-SR), mean (SD)	10.4 (5.4)	11.1 (4.9)	7.8 (4.6)	A>C (p=0.032); B>C (p<0.001)
Positive affect (PANAS), mean (SD)	30.6 (10.0)	31.0 (7.6)	33.8 (8.1)	B<C (p=0.019)
Health anxiety (WI), mean (SD)	6.0 (2.5)	6.8 (2.6)	5.1 (2.7)	B>C (p<0.001)
Concern (about illness, IPQ-brief), mean (SD)	7.3 (2.6)	6.1 (2.8)	4.8 (2.7)	A>C (p<0.001); B>C (p<0.001)
Emotions (extent to which illness affects mood, IPQ-brief), mean (SD)	7.2 (2.9)	6.8 (2.6)	5.7 (2.9)	B>C (p=0.009); A>C (p=0.046)
Behavioural responses
All-or-nothing behaviour (CBRQ), mean (SD)	8.7 (4.4)	9.1 (4.2)	7.0 (4.5)	B>C (p>0.001)
Avoidance/resting behaviour (CBRQ), mean (SD)	10.0 (5.7)	9.8 (4.6)	7.0 (4.5)	A>C (p=0.012); B>C (p<0.001)
Physical activity (IPAQ, in METs), median (IQR; total range)	1097 (3139; 0–35 523)	1875 (4327; 0–63 312)	3534 (6923; 0–42 846)	p<0.001

In bold p<0.05.

*As indicated in the method section, these characteristics were measured at T1 (6 months follow-up); differences between trajectories were tested with ANOVA and post hoc Bonferroni correction for mean values to control for multiple testing, with χ² tests for proportions and Kruskal-Wallis test for median values. For ANOVA and post hoc Bonferroni correction, we report only the results for trajectories that showed significant differences.

ANOVA, analysis of variance; BAI, Beck Anxiety Inventory; CBRQ, Cognitive Behavioural Response Questionnaire (higher scores indicating more negative cognitive and behavioural responses); IPAQ, International Physical Activity Questionnaire; IPQ-brief, Brief illness Perception Questionnaire (higher scores indicating more negative illness perceptions); LEQ, Life Events Questionnaire; MDPS, Multidimensional Perfectionism Scale; METs, Metabolic Equivalents; NEO-FFI, NEO personality questionnaire- Five Factor Inventory; PANAS, subscale of Positive and Negative Affect Schedule; PCS, Physical Component Summary; PSQ, Physical Symptom Questionnaire; QIDS-SR, Quick Inventory of Depressive Symptomatology; SoS, Social Support Scale; SSAS, Somatosensory Amplification Scale; TiC-P, Treatment Inventory of Costs in patients with psychiatric disorders; WI, Whitely Index.

Patient characteristics for trajectories of mental functioning

The results of the analyses are shown in table 4. Patients with different trajectories of mental functioning also varied in all studied categories. In the ‘poor mental functioning, marked improvement’ trajectory, participants scored lowest on physical activity and reported more psychiatric comorbid disorders and a higher anxiety level at recruitment. Participants in the ‘deterioration’ trajectory scored lowest on social support. Participants in both the ‘marked improvement’ and ‘deterioration’ trajectories reported more symptoms and scored less favourable in terms of personality traits and cognitive, emotional and behavioural responses than participants in the ‘moderate mental functioning, slight improvement’ trajectory.

Table 4

Differences in patient characteristics for trajectories of mental functioning

Mental functioning (MCS) trajectories
Trajectory	(A) Poor mental functioning marked improvement (n=41)	(B) Moderate mental functioning, deterioration (n=36)	(C) Moderate mental functioning, slight improvement (n=217)	P value between clusters
Personal background
Women, n (%)	32 (78.0)	27 (75.0)	166 (76.5)	p=0.951
Education level, n (%)				p=0.577
Lower education	15 (36.6)	14 (38.9)	67 (30.9)
Intermediate education	17 (41.5)	15 (41.7)	84 (38.7)
Higher education	9 (22.0)	7 (19.4)	66 (30.4)
Age in years, range 19–70, mean (SD)	47.0 (11.8)	46.0 (11.1)	46.8 (12.7)	p=1.00
Potentially traumatic events in childhood (<16 years)(LEQ), median (IQR; total range)	0 (2; 0–5)	0 (1; 0–7)	0 (1; 0–6)	p=0.824
Perfectionism (MDPS), mean (SD)*	81.1 (23.9)	81.7 (22.1)	66.0 (21.1)	B>C (p<0.001); A>C (p<0.001)
Neuroticism (NEO-FFI), mean (SD)*	40.0 (7.8)	37.3 (6.4)	30.5 (8.5)	B>C (p<0.001); A>C (p<0.001)
Extraversion (NEO-FFI), mean (SD)*	35.9 (6.9)	35.4 (6.4)	39.4 (6.2)	B<C (p=0.004); A<C (p=0.006)
Social and environmental background
Marital status, n (%)				p=0.397
Married or cohabiting	23 (56.1)	19 (52.8)	137 (63.1)
Other (single, divorced, widow)	18 (43.9)	17 (47.2)	80 (36.9)
Social support (SoS), median (IQR; total range)	55 (18; 22–60)	52.5 (15.5; 34–60)	57 (11; 15–60)	p=0.027
Illness stressors
Number of symptoms (PSQ), mean (SD)	25.9 (8.8)	26.3 (9.5)	21.1 (9.4)	B>C (p=0.008); A>C (p=0.010)
Duration of symptoms in years, median (IQR; total range)	6 (13; 0.42–30)	7 (16; 0.25–45)	4 (14.5; 0.02–62)	p=0.235
Comorbid diseases
Number of somatic comorbid diseases (TiC-P), median (IQR; total range)	1 (2; 0–7)	1 (2.75; 0–6)	1 (2; 0–8)	p=0.799
Number of psychiatric comorbid disorders (TiC-P), median (IQR; total range)	1 (2; 0–3)	0 (1; 0–3)	0 (1; 0–3)	p<0.001
Cognitive responses
Fear-avoidance cognitions (CBRQ), mean (SD)	10.2 (5.8)	11.9 (4.5)	9.5 (4.4)	B>C (p=0.014)
Catastrophising cognitions (CBRQ), mean (SD)	8.0 (3.6)	7.0 (3.4)	4.9 (3.1)	B>C (0.002); A>C (<0.001)
Damage cognitions (CBRQ), mean (SD)	10.3 (3.6)	10.9 (3.9)	9.4 (3.4)	p=0.086
Embarrassment avoidance cognitions (CBRQ), median (IQR; total range)	11 (7.5; 0–24)	10 (5.5; 0–22)	6 (7.25; 0–18)	p<0.001
Symptom focusing (CBRQ), mean (SD)	13.1 (6.0)	11.5 (4.5)	9.6 (4.9)	A>C (p<0.001)
Somatosensory amplification (SSAS), mean (SD)	29.5 (6.9)	28.4 (4.8)	26.5 (6.1)	A>C (p=0.013)
Identity (label of disease and associated symptoms, IPQ-brief), mean (SD)	8.2 (1.7)	7.5 (1.7)	6.9 (2.0)	A>C (p<0.001)
Consequences (beliefs about illness effects and outcomes, IPQ-brief), median (IQR; total range)	9 (2; 5–10)	8 (3.75; 2–10)	7 (4; 0–10)	p<0.001
Timeline (expected duration of illness, IPQ-brief), median (IQR; total range)	9 (3; 3–10)	9 (3.75; 2–10)	8 (4; 0–10)	p=0.111
Personal control (over illness, IPQ-brief), mean (SD)	6.8 (3.3)	6.1 (2.7)	5.6 (2.7)	A>C (p=0.040)
Treatment control (beliefs about treatment effect on illness, IPQ-brief), mean (SD)	4.1 (2.9)	4.4 (2.8)	4.3 (2.5)	p=1.00
Comprehensibility (extent to which patients believe to understand their illness, IPQ-brief), mean (SD)	4.6 (3.3)	3.7 (2.5)	4.0 (2.6)	p=1.00
Emotional responses
Anxiety severity (BAI), median (IQR; total range)	15 (14.5; 1–51)	12 (12.5; 1–44)_	9 (10; 0–42)	p<0.001
Depression severity (QIDS-SR), mean (SD)	13.2 (5.0)	10.9 (4.5)	8.1 (4.6)	B>C (p=0.003); A>C (p<0.001)
Positive affect (PANAS), mean (SD)	26.3 (9.1)	30.2 (6.7)	34.1 (7.6)	B<C (p=0.017); A<C (p<0.001)
Health anxiety (WI), mean (SD)	7.4 (2.8)	6.9 (2.4)	5.3 (2.6)	B>C (p=0.002); A>C (p<0.001)
Concern (about illness, IPQ-brief), mean (SD)	7.0 (3.0)	6.2 (2.9)	5.1 (2.7)	A>C (p<0.001)
Emotions (extent to which illness affects mood, IPQ-brief), median (IQR; total range)	7.7 (2.5)	7.4 (1.9)	5.8 (2.9)	B>C (p=0.003); A>C (p<0.001)
Behavioural responses
All-or-nothing behaviour (CBRQ), mean (SD)	9.8 (4.3)	8.5 (4.4)	7.4 (4.4)	A>C (p=0.004)
Avoidance/resting behaviour (CBRQ), mean (SD)	11.7 (6.0)	9.9 (4.4)	7.3 (4.3)	B>C (p=0.009); A>C (p<0.001)
Physical activity (IPAQ, in METs), median (IQR; total range)	1166 (4442; 0–63 312)	2696 (5627; 132–28 668)	2984 (6455; 0–44 436)	p=0.046

In bold p<0.05.

*As indicated in the method section, these characteristics were measured at T1 (6 months follow-up); differences between trajectories were tested with ANOVA and post hoc Bonferroni correction for mean values to control for multiple testing, with χ² tests for proportions and Kruskal-Wallis test for median values. For ANOVA and post hoc Bonferroni correction, we report only the results for trajectories that showed significant differences.

ANOVA, analysis of variance; BAI, Beck Anxiety Inventory; CBRQ, Cognitive Behavioural Response Questionnaire (higher scores indicating more negative cognitive and behavioural responses); IPAQ, International Physical Activity Questionnaire; IPQ-brief, Brief Illness Perception Questionnaire (higher scores indicating more negative illness perceptions); LEQ, Life Events Questionnaire; MCS, Mental Component Summary; MDPS, Multidimensional Perfectionism Scale; METs, Metabolic Equivalents; NEO-FFI, NEO Personality Questionnaire-Five Factor Inventory; PANAS, subscale of Positive and Negative Affect Schedule; PSQ, Physical Symptom Questionnaire; QIDS-SR, Quick Inventory of Depressive Symptomatology; SoS, Social Support Scale; SSAS, Somatosensory Amplification Scale; TiC-P, Treatment Inventory of Costs in patients with psychiatric disorders; WI, Whitely Index.

Discussion

We identified distinct 5-year trajectories for symptom severity, physical and mental functioning in adult patients with PSS. For symptom severity, we identified two ‘stable’ trajectories, indicating a high prevalence of persisting symptoms in this population. Physical and mental functioning over the 5-year period improved slightly for the large majority of the participants. In a minority of participants, we identified trajectories that showed considerable improvement in physical or mental functioning (8.5%, respectively, 13.9% of participants), as well as a trajectory for deteriorating mental functioning (12.2% of participants). As anticipated, we found high inter-individual and intra-individual variation within the identified trajectories. A wide range of patient characteristics—such as personal, social and environmental background, illness stressors, comorbid diseases and cognitive, emotional and behavioural responses—varied between the distinct trajectories for symptom severity, physical and mental functioning.

This is the first study that applied LCGMM to study 5-year trajectories of symptom severity, physical and mental functioning in patients with PSS. Strengths of this study include its multicentre prospective design, the long duration of follow-up using multiple measurements over time, the low loss to follow-up and the use of LCGMM—a contemporary and flexible data-driven technique to distinguish distinct trajectories over time in a study population. We did not only focus on symptom persistence, but also on trajectories of physical and mental functioning: important aspects of prognosis in patients with PSS. An important limitation is the fact that trajectories identified by LCGMM represent a simplified version of the complex heterogeneous reality. Though the different trajectories found may be indicative of long-term general trends and summarise these adequately, individual trajectories can be more variable and diverse, with exacerbations and remissions that remain undetected by the LCGMM. Another limitation concerns the possibility of selection and attrition bias. The majority of participants in the PROSPECTS study already experienced symptoms for an extensive period of time when recruited (median duration 5 years; IQR 14 years), indicating that we may have selected more patients with long-lasting PSS by our inclusion procedure. Therefore, we should keep in mind that trajectories identified in this study may differ from trajectories shortly after onset of PSS. An additional limitation concerns the small number of participants in some of the trajectories; due to these small numbers, prediction modelling was not possible within our study, and examining differences in patient characteristics between trajectories was also hampered by the small subgroups. Some attrition bias may have occurred as well, though the study sample for the LCGMM consisted of over 90% of the initially recruited participants in the PROSPECTS study. Furthermore, the moment at which participants completed the questionnaire may have been influenced by the perceived severity of symptoms and their functional abilities, which in turn may affect our findings.

In most previous studies on the course of PSS, the majority of patients showed symptom improvement over time, and in a minority of patients, symptoms worsened or became chronic.^{9–13 71–74} Our findings show a different picture: when using multiple measurements and applying LCGMM over a 5-year period, participants in the PROSPECTS cohort followed ‘stable’ trajectories for symptom severity, indicating a higher prevalence of symptom persistence than expected based on these previous studies. Our findings were largely consistent with our previous results on the 2-year course of PSS.²³ The frequent occurrence of ‘slight improvement’ rather than ‘stable’ trajectories for physical and mental functioning in the current 5-year analyses indicates either a genuine adjustment leading to better functioning or a response shift leading to different reporting of functioning. A response shift can occur when individuals learn to adapt to their changed functioning and tend to report scores differently over time.⁷⁵ The inter-individual and intra-individual heterogeneity within trajectories, was also present in our current analyses. Patterns of remissions and exacerbations of symptoms were confirmed in a qualitative study conducted among participants of the PROSPECTS study and were an important element of their symptom experience.⁷⁶ Patterns of fluctuating symptoms have been described in patients with various long-term or chronic conditions, such as osteoarthritis,^{77 78} fibromyalgia^{79 80} and long COVID-19 symptoms.⁸¹

Patients in the distinct trajectories differed with regard to a wide range of patient characteristics previously associated with an (un)favourable prognosis of PSS. In the interpretation of our findings, we have to be aware that these analyses were primarily explorative in nature. However, our findings are largely consistent with previous studies and the theoretical background of (un)favourable prognostic factors in PSS.^10–21 Of note is that participants who improved markedly in mental functioning, reported more unfavourable characteristics at baseline (eg, higher anxiety levels, more psychiatric comorbid disorders) than those in the other mental functioning trajectories. This finding may underline the importance of timely diagnosing psychiatric comorbid disorders, although it may also mean that this subgroup had more room for improvement (‘regression to the mean’) or was more receptive to treatment.

As indicated in the introduction, patients can suffer from PSS in the context of well-understood (adequately treated) conditions or when no somatic explanation for symptoms is found. Though explorative in nature, our findings on patient characteristics for the distinct trajectories may add to this current concept of PSS. For example, participants in the ‘severe symptom’ and ‘poor physical functioning’ trajectory reported more somatic comorbid diseases and lower social support. In addition, we found differences in cognitive, emotional and behavioural responses to symptoms for a range of trajectories, whereas the number of psychiatric comorbid diseases did not differ. These findings suggest that biological, psychological and social factors may all play a role in the persistence of somatic symptoms, irrespective of aetiology and question the outdated term and definition of ‘medically unexplained symptoms’.

The identified trajectories help to better understand long-term trajectories and prognosis of symptom severity, physical and mental functioning in adult patients who repeatedly visit their general practitioner or are referred to a specialised treatment facility for their PSS. Clinicians should be more aware of the high prevalence of persisting symptoms and limited improvement in physical and mental functioning with frequent exacerbations and remissions over time. We also identified trajectories of marked improvement in physical or mental functioning, as well as a trajectory of mental deterioration. Gaining more knowledge about these potentially clinically relevant subgroups may further assist clinical decision-making and evaluation. Future studies that follow patients shortly after the onset of their PSS and map their long-term course of PSS with multiple measurements over time and multiple outcomes are warranted. As are further predictive studies and studies on the identification of clinically relevant subgroups (as in the current study potentially relevant subgroups were small, for this purpose large study samples may be necessary). This may ultimately enable a more patient-centred approach to providing care for patients with PSS.

Conclusion

We identified distinct 5-year trajectories for symptom severity, physical and mental functioning in adult patients with PSS. Our findings suggest, partly contrary to previous research, a high prevalence of persistence of symptoms and limited improvement in physical and mental functioning over the longer-term in a majority of adult patients with PSS. In a small proportion of patients, we identified trajectories that showed considerable improvement in physical or mental functioning, or deterioration in mental functioning. These may represent potentially clinically relevant subgroups of patients with PSS. A wide range of patient characteristics differed between the distinct trajectories, and these characteristics may help in early recognition, although predictive studies are warranted.

We want to thank our research assistants for their help and assistance in gathering all necessary data. We are particularly grateful to all the participants who took part in the PROSPECTS study for their time and efforts.

Data availability statement

No data are available.

Ethics statements

Patient consent for publication

Not applicable.

Ethics approval

The institutional review board of the VU University Medical Center (IRB00002991) approved the research protocol (WC2013.003). Participants gave informed consent to participate in the study before taking part.

Footnote

X @TrynkeHoekstra

Contributors HB and TH developed the study protocol. HEvdH, JCvdW, NC and JD provided feedback on the study protocol. HB and TH conducted the analyses. HB drafted the manuscript. All authors read, provided critical revisions and approved the manuscript. HB acted as guarantor.

Funding This work was financially supported by ZonMw (grant number 839110018) and the Stoffels-Hornstra Foundation (grant number NA).

Competing interests None declared.

Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.

Provenance and peer review Not commissioned; externally peer reviewed.

Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.

References

¹ Peveler R, Kilkenny L, Kinmonth AL. Medically unexplained physical symptoms in primary care: a comparison of self-report screening questionnaires and clinical opinion. J Psychosom Res 1997; 42: 245–52. doi:10.1016/s0022-3999(96)00292-9

² Rosendal M, Carlsen AH, Rask MT, et al. Symptoms as the main problem in primary care: A cross-sectional study of frequency and characteristics. Scand J Prim Health Care 2015; 33: 91–9. doi:10.3109/02813432.2015.1030166

³ Nimnuan C, Hotopf M, Wessely S. Medically unexplained symptoms: an epidemiological study in seven specialities. J Psychosom Res 2001; 51: 361–7. doi:10.1016/s0022-3999(01)00223-9

⁴ Reid S, Wessely S, Crayford T, et al. Frequent attenders with medically unexplained symptoms: service use and costs in secondary care. Br J Psychiatry 2002; 180: 248–53. doi:10.1192/bjp.180.3.248

⁵ Dirkzwager AJE, Verhaak PFM. Patients with persistent medically unexplained symptoms in general practice: characteristics and quality of care. BMC Fam Pract 2007; 8: 33. doi:10.1186/1471-2296-8-33

⁶ Hanssen DJC, Lucassen P, Hilderink PH, et al. Health-Related Quality of Life in Older Persons with Medically Unexplained Symptoms. Am J Geriatr Psychiatry 2016; 24: 1117–27. doi:10.1016/j.jagp.2016.07.013

⁷ Rasmussen S, Jensen CT, Rosendal M, et al. Multiple physical symptoms and individual characteristics - A cross-sectional study of the general population. J Psychosom Res 2020; 131: 109941. doi:10.1016/j.jpsychores.2020.109941

⁸ Konnopka A, Schaefert R, Heinrich S, et al. Economics of medically unexplained symptoms: a systematic review of the literature. Psychother Psychosom 2012; 81: 265–75. doi:10.1159/000337349

⁹ Rask MT, Ørnbøl E, Rosendal M, et al. Long-Term Outcome of Bodily Distress Syndrome in Primary Care: A Follow-Up Study on Health Care Costs, Work Disability, and Self-Rated Health. Psychosom Med 2017; 79: 345–57. doi:10.1097/PSY.0000000000000405

¹⁰ olde Hartman TC, Borghuis MS, Lucassen PLBJ, et al. Medically unexplained symptoms, somatisation disorder and hypochondriasis: course and prognosis. A systematic review. J Psychosom Res 2009; 66: 363–77. doi:10.1016/j.jpsychores.2008.09.018

¹¹ Jackson JL, Kroenke K. Prevalence, Impact, and Prognosis of Multisomatoform Disorder in Primary Care: A 5-Year Follow-up Study. Psychosom Med 2008; 70: 430–4. doi:10.1097/PSY.0b013e31816aa0ee

¹² Koch H, van Bokhoven MA, Bindels PJE, et al. The course of newly presented unexplained complaints in general practice patients: a prospective cohort study. Fam Pract 2009; 26: 455–65. doi:10.1093/fampra/cmp067

¹³ Creed FH, Davies I, Jackson J, et al. The epidemiology of multiple somatic symptoms. J Psychosom Res 2012; 72: 311–7. doi:10.1016/j.jpsychores.2012.01.009

¹⁴ Steinbrecher N, Hiller W. Course and prediction of somatoform disorder and medically unexplained symptoms in primary care. Gen Hosp Psychiatry 2011; 33: 318–26. doi:10.1016/j.genhosppsych.2011.05.002

¹⁵ Budtz-Lilly A, Vestergaard M, Fink P, et al. The prognosis of bodily distress syndrome: a cohort study in primary care. Gen Hosp Psychiatry 2015; 37: 560–6. doi:10.1016/j.genhosppsych.2015.08.002

¹⁶ van Eck van der Sluijs JF, ten Have M, de Graaf R, et al. Predictors of Persistent Medically Unexplained Physical Symptoms: Findings From a General Population Study. Front Psychiatry 2018; 9. doi:10.3389/fpsyt.2018.00613

¹⁷ Lamahewa K, Buszewicz M, Walters K, et al. Persistent unexplained physical symptoms: a prospective longitudinal cohort study in UK primary care. Br J Gen Pract 2019; 69: e246–53. doi:10.3399/bjgp19X701249

¹⁸ Kooiman CG, Bolk JH, Rooijmans HGM, et al. Alexithymia does not predict the persistence of medically unexplained physical symptoms. Psychosom Med 2004; 66: 224–32. doi:10.1097/01.psy.0000116714.38868.06

¹⁹ Speckens AEM, Van Hemert AM, Bolk JH, et al. Unexplained physical symptoms: outcome, utilization of medical care and associated factors. Psychol Med 1996; 26: 745–52. doi:10.1017/S0033291700037764

²⁰ Claassen-van Dessel N, van der Wouden JC, Twisk JWR, et al. Predicting the course of persistent physical symptoms: Development and internal validation of prediction models for symptom severity and functional status during 2 years of follow-up. J Psychosom Res 2018; 108: 1–13. doi:10.1016/j.jpsychores.2018.02.009

²¹ Mewes R. Recent developments on psychological factors in medically unexplained symptoms and somatoform disorders. Front Public Health 2022; 10: 1033203. doi:10.3389/fpubh.2022.1033203

²² van Dessel N, Leone SS, van der Wouden JC, et al. The PROSPECTS study: design of a prospective cohort study on prognosis and perpetuating factors of medically unexplained physical symptoms (MUPS). J Psychosom Res 2014; 76: 200–6. doi:10.1016/j.jpsychores.2013.12.011

²³ Claassen-van Dessel N, van der Wouden JC, Hoekstra T, et al. The 2-year course of Medically Unexplained Physical Symptoms (MUPS) in terms of symptom severity and functional status: results of the PROSPECTS cohort study. J Psychosom Res 2018; 104: 76–87. doi:10.1016/j.jpsychores.2017.11.012

²⁴ Claassen-van Dessel N, van der Wouden JC, Dekker J, et al. Clinical value of DSM IV and DSM 5 criteria for diagnosing the most prevalent somatoform disorders in patients with medically unexplained physical symptoms (MUPS). J Psychosom Res 2016; 82: 4–10. doi:10.1016/j.jpsychores.2016.01.004

²⁵ Kroenke K, Spitzer RL, Williams JBW. The PHQ-15: validity of a new measure for evaluating the severity of somatic symptoms. Psychosom Med 2002; 64: 258–66. doi:10.1097/00006842-200203000-00008

²⁶ Sitnikova K, Dijkstra-Kersten SMA, Mokkink LB, et al. Systematic review of measurement properties of questionnaires measuring somatization in primary care patients. J Psychosom Res 2017; 103: 42–62. doi:10.1016/j.jpsychores.2017.10.005

²⁷ Zijlema WL, Stolk RP, Löwe B, et al. How to assess common somatic symptoms in large-scale studies: a systematic review of questionnaires. J Psychosom Res 2013; 74: 459–68. doi:10.1016/j.jpsychores.2013.03.093

²⁸ Terluin B, Barends H, van der Horst HE, et al. Head-to-head comparison of somatic symptom scales: The Patient Health Questionnaire (PHQ-15) and the somatization scale of the Four-Dimensional Symptom Questionnaire (4DSQ-S). J Psychosom Res 2022; 162: 111031. doi:10.1016/j.jpsychores.2022.111031

²⁹ van der Zee KI, Sanderman R, Heyink JW, et al. Psychometric qualities of the rand 36-item health survey 1.0: A multidimensional measure of general health status. Int J Behav Med 1996; 3: 104–22. doi:10.1207/s15327558ijbm0302_2

³⁰ van der Zee KI, Sanderman R. Measuring general health with the RAND-36, a manual (in Dutch). Groningen: Noordelijk Centrum voor Gezondheidsvraagstukken Research Institute SHARE; 2012.

³¹ Norman GR, Sloan JA, Wyrwich KW. Interpretation of changes in health-related quality of life: the remarkable universality of half a standard deviation. Med Care 2003; 41: 582–92. doi:10.1097/01.MLR.0000062554.74615.4C

³² Deary V, Chalder T, Sharpe M. The cognitive behavioural model of medically unexplained symptoms: a theoretical and empirical review. Clin Psychol Rev 2007; 27: 781–97. doi:10.1016/j.cpr.2007.07.002

³³ McCrae RR, Costa PT, Ostendorf F, et al. Nature over nurture: temperament, personality, and life span development. J Pers Soc Psychol 2000; 78: 173–86. doi:10.1037//0022-3514.78.1.173

³⁴ Dekker J, de Groot V. Psychological adjustment to chronic disease and rehabilitation - an exploration. Disabil Rehabil 2018; 40: 116–20. doi:10.1080/09638288.2016.1247469

³⁵ Jung T, Wickrama KAS. An Introduction to Latent Class Growth Analysis and Growth Mixture Modeling. Soc Pers Psych 2008; 2: 302–17. doi:10.1111/j.1751-9004.2007.00054.x

³⁶ van de Schoot R, Sijbrandij M, Winter SD, et al. The GRoLTS-Checklist: Guidelines for Reporting on Latent Trajectory Studies. Struct Equ Modeling 2017; 24: 451–67. doi:10.1080/10705511.2016.1247646

³⁷ Schwarz G. Estimating the Dimension of a Model. Ann Statist 1978; 6: 461–4. doi:10.1214/aos/1176344136

³⁸ Celeux G, Soromenho G. An entropy criterion for assessing the number of clusters in a mixture model. J Classif 1996; 13: 195–212. doi:10.1007/BF01246098

³⁹ Musliner KL, Munk-Olsen T, Eaton WW, et al. Heterogeneity in long-term trajectories of depressive symptoms: Patterns, predictors and outcomes. J Affect Disord 2016; 192: 199–211. doi:10.1016/j.jad.2015.12.030

⁴⁰ Nagin DS. Group-based modeling of development. Cambridge, Mass: Harvard University Press, 2005.

⁴¹ Hoekstra T. Applied latent class models for epidemiology. Amsterdam, 2013

⁴² Nylund KL, Asparouhov T, Muthén BO. Deciding on the Number of Classes in Latent Class Analysis and Growth Mixture Modeling: A Monte Carlo Simulation Study. Struct Equ Modeling 2007; 14: 535–69. doi:10.1080/10705510701575396

⁴³ Muthén B. The potential of growth mixture modelling. Infant Child Dev 2006; 15: 623–5. doi:10.1002/icd.482

⁴⁴ Muthén B, Muthén LK. Integrating person-centered and variable-centered analyses: growth mixture modeling with latent trajectory classes. Alcohol Clin Exp Res 2000; 24: 882–91.

⁴⁵ Twisk J, Hoekstra T. Classifying developmental trajectories over time should be done with great caution: a comparison between methods. J Clin Epidemiol 2012; 65: 1078–87. doi:10.1016/j.jclinepi.2012.04.010

⁴⁶ Dempster AP, Laird NM, Rubin DB. Maximum Likelihood from Incomplete Data Via the EM Algorithm. J R Stat Soc Ser B 1977; 39: 1–22. doi:10.1111/j.2517-6161.1977.tb01600.x

⁴⁷ Andersen SB, Karstoft K-I, Bertelsen M, et al. Latent trajectories of trauma symptoms and resilience: the 3-year longitudinal prospective USPER study of Danish veterans deployed in Afghanistan. J Clin Psychiatry 2014; 75: 1001–8. doi:10.4088/JCP.13m08914

⁴⁸ Garnefski N, Kraaij V. Levensgebeurtenissen Vragenlijst [life event questionnaire]. Leiden University; 2001. Available: http://www.cerq.leidenuniv.nl

⁴⁹ Frost RO, Marten P, Lahart C, et al. The dimensions of perfectionism. Cogn Ther Res 1990; 14: 449–68. doi:10.1007/BF01172967

⁵⁰ Flos CHGM, Peeters F, van Eiisden M. The “multidimensional perfectionism scale”; reliability and validity of the Dutch translation. Maastricht, the Netherlands Maastricht University; 2000.

⁵¹ Costa PT, McCrae R. Revised NEO personality inventory (NEO-PI-R) and the five factor inventory (NEO-FFI): professional manual. Odessa, Florida: Psychological Assessment Resources Inc, 1992.

⁵² Hoekstra H, Ormel J, de Fruyt F. NEO-PI-R en NEO-FFI: big five persoonlijkheidsvragenlijsten: handleiding [NEO-PI-R and NEO-FFI: big five personality questionnaires: a manual]. Amsterdam, the Netherlands: Hogrefe Uitgevers, 1996.

⁵³ Feij DC, van Kampen D, van den Berg G. Sensation seeking and social support as moderators of the relationship between life events and physical illness/psychological distress. In: Maes JAMWS, ed. Lifestyles, stress, and health. New developments in health psychology. Leiden: DSWO Press, 1992: 285–302.

⁵⁴ van Hemert AM. Lichamelijke Klachten Vragenlijst [physical symptoms questionnaire]. Leiden, the Netherlands Leiden University; 2003.

⁵⁵ Hakkaart-van Roijen L. Trimbos / iMTA questionnaire for costs associated with psychiatric illness (TiC-P). Rotterdam, the Netherlands: Institute for Medical Technology Assesment, 2002.

⁵⁶ Ryan EG, Vitoratou S, Goldsmith KA, et al. Psychometric Properties and Factor Structure of a Long and Shortened Version of the Cognitive and Behavioural Responses Questionnaire. Psychosom Med 2018; 80: 230–7. doi:10.1097/PSY.0000000000000536

⁵⁷ Picariello F, Chilcot J, Chalder T, et al. The Cognitive and Behavioural Responses to Symptoms Questionnaire (CBRQ): Development, reliability and validity across several long-term conditions. Br J Health Psychol 2023; 28: 619–38. doi:10.1111/bjhp.12644

⁵⁸ Barsky AJ, Wyshak G. Hypochondriasis and somatosensory amplification. Br J Psychiatry 1990; 157: 404–9. doi:10.1192/bjp.157.3.404

⁵⁹ Speckens AE, Spinhoven P, Sloekers PP, et al. A validation study of the Whitely Index, the Illness Attitude Scales, and the Somatosensory Amplification Scale in general medical and general practice patients. J Psychosom Res 1996; 40: 95–104. doi:10.1016/0022-3999(95)00561-7

⁶⁰ Broadbent E, Petrie KJ, Main J, et al. The brief illness perception questionnaire. J Psychosom Res 2006; 60: 631–7. doi:10.1016/j.jpsychores.2005.10.020

⁶¹ de Raaij EJ, Schröder C, Maissan FJ, et al. Cross-cultural adaptation and measurement properties of the Brief Illness Perception Questionnaire-Dutch Language Version. Man Ther 2012; 17: 330–5. doi:10.1016/j.math.2012.03.001

⁶² Beck AT, Epstein N, Brown G, et al. An inventory for measuring clinical anxiety: psychometric properties. J Consult Clin Psychol 1988; 56: 893–7. doi:10.1037//0022-006x.56.6.893

⁶³ Muntingh ADT, van der Feltz-Cornelis CM, van Marwijk HWJ, et al. Is the Beck Anxiety Inventory a good tool to assess the severity of anxiety? A primary care study in the Netherlands Study of Depression and Anxiety (NESDA). BMC Fam Pract 2011; 12: 66. doi:10.1186/1471-2296-12-66

⁶⁴ Rush AJ, Trivedi MH, Ibrahim HM, et al. The 16-Item Quick Inventory of Depressive Symptomatology (QIDS), clinician rating (QIDS-C), and self-report (QIDS-SR): a psychometric evaluation in patients with chronic major depression. Biol Psychiatry 2003; 54: 573–83. doi:10.1016/s0006-3223(02)01866-8

⁶⁵ Watson D, Clark LA, Carey G. Positive and negative affectivity and their relation to anxiety and depressive disorders. J Abnorm Psychol 1988; 97: 346–53. doi:10.1037//0021-843x.97.3.346

⁶⁶ Pilowsky I. Dimensions of hypochondriasis. Br J Psychiatry 1967; 113: 89–93. doi:10.1192/bjp.113.494.89

⁶⁷ Ainsworth BE, Bassett DR, Strath SJ, et al. Comparison of three methods for measuring the time spent in physical activity. Med Sci Sports Exerc 2000; 32: S457–64. doi:10.1097/00005768-200009001-00004

⁶⁸ Vandelanotte C, De Bourdeaudhuij I, Philippaerts R, et al. Reliability and Validity of a Computerized and Dutch Version of the International Physical Activity Questionnaire (IPAQ). J Phys Act Health 2005; 2: 63–75. doi:10.1123/jpah.2.1.63

⁶⁹ Maust D, Cristancho M, Gray L, et al. Psychiatric rating scales. Handb Clin Neurol 2012; 106: 227–37. doi:10.1016/B978-0-444-52002-9.00013-9

⁷⁰ Hedman E, Lekander M, Ljótsson B, et al. Optimal cut-off points on the health anxiety inventory, illness attitude scales and whiteley index to identify severe health anxiety. PLoS One 2015; 10: e0123412. doi:10.1371/journal.pone.0123412

⁷¹ Burton C, Fink P, Henningsen P, et al. Functional somatic disorders: discussion paper for a new common classification for research and clinical use. BMC Med 2020; 18: 34. doi:10.1186/s12916-020-1505-4

⁷² Löwe B, Andresen V, Van den Bergh O, et al. Persistent SOMAtic symptoms ACROSS diseases - from risk factors to modification: scientific framework and overarching protocol of the interdisciplinary SOMACROSS research unit (RU 5211). BMJ Open 2022; 12: e057596. doi:10.1136/bmjopen-2021-057596

⁷³ Olde Hartman T, Blankenstein A, Molenaar B, et al. NHG guideline on medically unexplained symptoms. NHG Huisarts Wetenschap 2013; 56: 222–30.

⁷⁴ van der Feltz-Cornelis CM, Hoedeman R, Keuter EJW, et al. Presentation of the Multidisciplinary Guideline Medically Unexplained Physical Symptoms (MUPS) and Somatoform Disorder in the Netherlands: disease management according to risk profiles. J Psychosom Res 2012; 72: 168–9. doi:10.1016/j.jpsychores.2011.11.007

⁷⁵ Schwartz CE, Sprangers MAG. Methodological approaches for assessing response shift in longitudinal health-related quality-of-life research. Soc Sci Med 1999; 48: 1531–48. doi:10.1016/S0277-9536(99)00047-7

⁷⁶ Barends H, Walstock E, Botman F, et al. Patients’ experiences with fluctuations in persistent physical symptoms: a qualitative study. BMJ Open 2020; 10: e035833. doi:10.1136/bmjopen-2019-035833

⁷⁷ Parry E, Ogollah R, Peat G. Significant pain variability in persons with, or at high risk of, knee osteoarthritis: preliminary investigation based on secondary analysis of cohort data. BMC Musculoskelet Disord 2017; 18: 80. doi:10.1186/s12891-017-1434-3

⁷⁸ van Berkel AC, Schiphof D, Waarsing JH, et al. Course of pain and fluctuations in pain related to suspected early hip osteoarthritis: the CHECK study. Fam Pract 2022; 39: 1041–8. doi:10.1093/fampra/cmac030

⁷⁹ Zautra AJ, Fasman R, Parish BP, et al. Daily fatigue in women with osteoarthritis, rheumatoid arthritis, and fibromyalgia. Pain 2007; 128: 128–35. doi:10.1016/j.pain.2006.09.004

⁸⁰ Bartley EJ, Robinson ME, Staud R. Pain and Fatigue Variability Patterns Distinguish Subgroups of Fibromyalgia Patients. J Pain 2018; 19: 372–81. doi:10.1016/j.jpain.2017.11.014

⁸¹ Wynberg E, Verveen A, van Willigen HDG, et al. Two-year trajectories of COVID-19 symptoms and their association with illness perception: A prospective cohort study in Amsterdam, the Netherlands. Influenza Other Respir Viruses 2023; 17: e13190. doi:10.1111/irv.13190

⁸² Robbins JM, Kirmayer LJ, Hemami S. Latent variable models of functional somatic distress. J Nerv Ment Dis 1997; 185: 606–15. doi:10.1097/00005053-199710000-00003