Latent profile analysis of psychiatric symptoms and the Ability to Participate in Social Roles and Activities

Abstract

Background

The objective of this study was to investigate latent response patterns in symptom severity, as measured by the Brief Symptom Inventory (BSI), and limitations in daily functioning, as assessed by the extended PROMIS item bank ‘Ability to Participate in Social Roles and Activities’ (APSRA), within a sample of psychiatric outpatients. In addition to identifying classes with converging test results, we hypothesized the existence of classes with divergent results: those exhibiting low symptom severity alongside severe functional limitations, and those demonstrating high symptom severity while maintaining high levels of functioning.

Methods

A sample of 1,010 psychiatric outpatients from the Netherlands completed the Dutch BSI and APSRA. Latent Profile Analysis (LPA) was employed to group patients with similar patterns in symptom severity and daily functioning limitations. After identifying the LPA classes and class membership of all patients, we analyzed the distribution of suicidal ideation, participation level (i.e., employment), diagnosis, age, sex, living situation, and education level, across the LPA classes.

Results

The correlation between APSRA and BSI scores (r = −.64) showed that higher APSRA scores were associated with lower psychopathology. LPA identified four distinct profiles of psychosocial dysfunction: minimal, mild, moderate, and severe. These profiles differed significantly in suicidal ideation and work participation but not in other demographic variables. While diagnosis had a statistically significant effect on class membership, the effect size was negligible. The hypothesized divergent classes were not observed.

Conclusion

The four profiles provide a clinically relevant framework for understanding self-reported psychosocial dysfunction, distinguishing patients on key outcomes such as suicidal ideation and work participation. This approach supports tailoring interventions, prioritizing treatment goals, and allocating resources based on shared patterns of characteristics. Future research should validate these profiles’ temporal stability and predictive value for treatment outcomes while exploring the benefits of combining symptom and functioning assessments for clinical decision-making.

Supplementary Information

The online version contains supplementary material available at 10.1186/s40359-025-02946-z.

Background

Heterogeneity is a hallmark of psychiatric disorders in adults. Individuals diagnosed with the same condition can exhibit a wide range of symptom profiles and limitations in daily functioning [1–4]. However, in clinical practice, the emphasis often lies on measuring symptom reduction [5]. While reducing symptoms is undoubtedly crucial, the focus on symptoms can leave a gap in understanding the effectiveness of established interventions and the development of therapies specifically addressing daily functioning [6, 7]. Moreover, returning to a usual or acceptable level of functioning is part of what psychiatric patients themselves consider an important aspect of recovery [8]. Data from the Netherlands Study of Depression and Anxiety (NESDA) shows that limitations in social functioning have a strong association with mood and anxiety disorders but that even after full symptomatic remission, impairments in social functioning remain [9]. Although some research suggests that both - symptoms and daily functioning - are interchangeable outcomes [10, 11], a positive relationship between symptom severity and limitations in daily functioning does not inevitably imply that changes in symptoms and functioning occur simultaneously or at the same rate or time and therefore should be treated as distinct outcomes [12, 13]. In conjunction with the heterogeneity among individuals with psychiatric disorders, this complex relationship between symptoms and functioning, directly relates to the importance of a person-centered approach that considers the possible existence of convergence and divergence of symptoms and functioning in individuals. Convergence describes a situation where the intensity of experienced symptoms relates to the level of difficulties people have in daily activities or vice versa. In other words, there’s a clear alignment between how bad someone feels and how well they can manage day-to-day tasks. For example, someone with severe depression might report symptoms such as intense sadness, fatigue, and difficulty concentrating. This would likely translate into limitations in daily functioning, such as struggling to get to work, neglecting personal hygiene, or withdrawing from social activities. Or, vice versa, conflicts with colleagues, inability to participate in leisure activities and limited social support, could trigger feelings of worthlessness, fatigue and a depressed mood. Divergence, on the other hand, indicates a mismatch between symptom severity and limitations in daily life in two ways. It can describe individuals who experience significant symptoms but still manage to maintain a relatively satisfactory level of daily functioning. For instance, someone with social anxiety might feel intense fear in social situations but still attends to work or social events. Conversely, this applies to individuals who experience mild symptoms yet face substantial limitations in daily functioning. For example, someone might report only occasional nervousness in social situations, but the fear of being judged or scrutinized leads to avoiding social interactions altogether. This can significantly impact work, education, and personal life, despite the anxiety itself being mild.

By examining convergence and divergence, healthcare professionals can develop more tailored interventions to meet individual needs, potentially improving treatment outcomes.

However, despite the appeal of these possible patterns, which seem logical at first glance, research typically examines symptomatic and functional outcomes separately. As a result, there is limited evidence regarding the existence of divergent profiles in individual psychiatric patients. The primary goal of this research is to address this gap.

This research is also relevant to the ongoing debate about the Dutch mental health funding system. Since 2022, the Netherlands has implemented the ‘care performance model,’ based on a model that was developed for the National Health Services (NHS), which links costs to care needs as assessed by the clinician-rated Health of the Nation Outcome Scale Plus (HoNOS+) [14]. However, this approach has been criticized for being conceptually weak, lacking scientific support, and relying on a poor predictor of healthcare costs [15, 16]. This study could offer insights that contribute to the discussion of whether self-reported assessments of symptom severity and functional limitations, as measured by the Brief Symptom Inventory (BSI) and the Ability to Participate in Social Roles and Activities (APSRA), might serve as a viable alternative.

Latent Profile Analysis (LPA) offers a structured approach to explore patterns of symptom severity and functioning within psychiatric populations and examine how those profiles are different in relations to other factors. By applying LPA, the likelihood of each patient belonging to a certain profile (class) can also be estimated. While the presence of divergent profiles remains uncertain, LPA provides an optimal framework to empirically test this possibility, notwithstanding that these aspects are correlated at group level [17]. Even if divergent profiles are not identified, LPA remains valuable for uncovering clinically meaningful subgroups based on shared patterns of symptoms and functioning. Variability within classes is an expected finding that merely reflects the inherent complexity of psychiatric presentations, where patients may differ individually while still sharing core characteristics. Identifying these latent profiles allows for a nuanced understanding of patient status, potentially informing treatment strategies and guiding future research into functional recovery and treatment outcomes.

The present study utilized LPA to investigate the number, nature (convergent vs. divergent) and prevalence (size) of latent profiles from data on psychiatric symptoms (Brief Symptom Inventory [18]) and daily functioning (Ability to Participate in Social Roles and Activities item bank [19–22]) in a Dutch adult psychiatric population. Given the expected correlation between symptoms and functioning, we hypothesized that convergent profiles were more likely to occur than divergent profiles and examined the prevalence of both. Furthermore, the relation between profiles and the external variables suicidal ideation, participation level (i.e., employment), diagnosis, age, sex, living situation, and education level, were examined. Investigating suicidal ideation is relevant as it reflects mental health severity and may align with profiles of greater symptom severity and functional impairments. Including it as an external variable enhances the LPA’s capacity to identify high-risk profiles, supporting more nuanced clinical understanding, risk assessment, and treatment planning. By exploring the characteristics of these profiles, this study can contribute to a deeper understanding of the relationship between psychiatric symptoms and daily functioning in adults.

Methods

Procedure

Participants in this study were patients from the Dimence Group (DG), a Dutch mental healthcare facility for psychiatric disorders. Upon referral, every patient underwent an initial interview with at least one healthcare professional. Following the interview, consultations with other professionals were conducted to reach a consensus on the most appropriate diagnosis. Each interview led to a diagnosis for the patient, and in cases of comorbidity, the primary diagnosis (considered the most relevant) was designated as the focus of treatment. While the Mini-International Neuropsychiatric Interview (MINI) [23, 24] was available as part of the interview process, its use was not mandatory, allowing the clinician to decide whether or not to utilize it. A digital version of the Brief Symptom Inventory (BSI) [18, 25, 26] was completed by the patients as a mandatory part of the standard intake process, either before or after the initial interview. Immediately after completing the BSI, they were given the option to voluntarily answer additional questions as part of a larger study aimed at improving the measurement of daily functioning [19]. When consent was given, the Ability to Participate in Social Roles and Activities (APSRA) questionnaire and the ‘participation ladder’ (see Measures section for more details) were presented immediately, ensuring that all instruments were completed at the same time. More details about the study were provided in a separate leaflet. Participants provided informed consent to the anonymous use of their data for research purposes.

Participants

Data of 1,010 Dutch adult psychiatric outpatients were collected for analysis. The patients were mainly diagnosed with a depressive disorder, an anxiety disorder, a personality disorder or a traumatic disorder. Demographics and clinical features of the sample are provided in Table 1.

Table 1.

Demographic profiles per class

Characteristic	Overall N = 1,010a	Class 1 Mild N = 344a	Class 2 Moderate N = 358a	Class 3 Minimal N = 232a	Class 4 Severe N = 76a
Sex
Male	339 (34%)	119 (35%)	106 (30%)	87 (38%)	27 (36%)
Female	671 (66%)	225 (65%)	252 (70%)	145 (63%)	49 (64%)
Age	36 (13)	36 (13)	35 (12)	36 (13)	36 (11)
BSI c	1.62 (0.70)	1.37 (0.28)	2.14 (0.30)	0.77 (0.30)	2.92 (0.29)
APSRA d	3.03 (0.80)	3.07 (0.51)	2.62 (0.53)	4.00 (0.48)	1.80 (0.36)
Diagnosis
ADHD b	10 (1.0%)	4 (1.2%)	3 (0.8%)	3 (1.3%)	0 (0%)
Trauma	127 (13%)	35 (10%)	54 (15%)	22 (9.5%)	16 (21%)
Personality Disorder	196 (19%)	72 (21%)	61 (17%)	53 (23%)	10 (13%)
Anxiety	214 (21%)	77 (22%)	63 (18%)	59 (25%)	15 (20%)
Depressive	396 (39%)	140 (41%)	157 (44%)	66 (28%)	33 (43%)
Other	10 (1.0%)	4 (1.2%)	2 (0.6%)	3 (1.3%)	1 (1.3%)
OCD b	46 (4.6%)	9 (2.6%)	14 (3.9%)	22 (9.5%)	1 (1.3%)
ASD b	3 (0.3%)	0 (0%)	2 (0.6%)	1 (0.4%)	0 (0%)
Somatic Symptoms	8 (0.8%)	3 (0.9%)	2 (0.6%)	3 (1.3%)	0 (0%)
Participation e
1	110 (11%)	28 (8.1%)	47 (13%)	7 (3.0%)	28 (37%)
2	185 (18%)	60 (17%)	82 (23%)	31 (13%)	12 (16%)
3	64 (6.3%)	28 (8.1%)	19 (5.3%)	13 (5.6%)	4 (5.3%)
4	66 (6.5%)	19 (5.5%)	20 (5.6%)	21 (9.1%)	6 (7.9%)
5	49 (4.9%)	15 (4.4%)	19 (5.3%)	12 (5.2%)	3 (3.9%)
6	536 (53%)	194 (56%)	171 (48%)	148 (64%)	23 (30%)
Suicidal Ideation f	568 (56%)	194 (56%)	247 (69%)	62 (27%)	65 (86%)
Suicide Thoughts
0 Not at all	442 (44%)	150 (44%)	111 (31%)	170 (73%)	11 (14%)
1 A little bit	301 (30%)	126 (37%)	103 (29%)	49 (21%)	23 (30%)
2 Moderately	119 (12%)	31 (9.0%)	62 (17%)	7 (3.0%)	19 (25%)
3 Quite a bit	104 (10%)	32 (9.3%)	55 (15%)	5 (2.2%)	12 (16%)
4 Extremely	44 (4.4%)	5 (1.5%)	27 (7.5%)	1 (0.4%)	11 (14%)
Education
High	288 (29%)	104 (30%)	93 (26%)	63 (27%)	28 (37%)
Middle	686 (68%)	226 (66%)	251 (70%)	164 (71%)	45 (59%)
Low	36 (3.6%)	14 (4.1%)	14 (3.9%)	5 (2.2%)	3 (3.9%)
Living Situation
Single	235 (23%)	82 (24%)	91 (25%)	45 (19%)	17 (22%)
Single Parent	68 (6.7%)	18 (5.2%)	30 (8.4%)	15 (6.5%)	5 (6.6%)
Child Single Parent	57 (5.6%)	26 (7.6%)	11 (3.1%)	14 (6.0%)	6 (7.9%)
Child Multiparent	141 (14%)	40 (12%)	55 (15%)	41 (18%)	5 (6.6%)
Partner And Child	299 (30%)	102 (30%)	98 (27%)	72 (31%)	27 (36%)
Partner No Child	179 (18%)	66 (19%)	57 (16%)	40 (17%)	16 (21%)
MH Institute	1 (< 0.1%)	1 (0.3%)	0 (0%)	0 (0%)	0 (0%)
Non-MH Institute	9 (0.9%)	3 (0.9%)	5 (1.4%)	1 (0.4%)	0 (0%)
Homeless	1 (< 0.1%)	0 (0%)	1 (0.3%)	0 (0%)	0 (0%)
Other	20 (2.0%)	6 (1.7%)	10 (2.8%)	4 (1.7%)	0 (0%)

^a n (%); Mean (SD)

^b ADHD = Attention Deficit Hyperactivity Disorder; OCD = Obsessive Compulsive Disorder; ASD = Autism Spectrum Disorder

^c BSI = Brief Symptom Inventory

^d APSRA = Ability to Participate in Social Roles and Activities

^e 1 = a withdrawn life, 2 = social contacts outside the home, 3 = participation in organized activities, 4 = unpaid work, 5 = paid work with support, 6 = paid work without support

^f Suicide Thoughts BSI item 9 > 0

Measures

The brief symptom inventory

The Brief Symptom Inventory (BSI) is a widely used tool to assess general psychological distress [18]. For our study we used the Dutch version of the BSI [26]. The BSI is a self-report measure with 53 items that are scored on a 5-point Likert scale ranging from 0 (not at all) to 4 (very much). Higher scores indicate more severe problems. As many researchers favor a single general factor, the present study used the Global Severity Index (GSI, i.e., the average of all items) to reflect overall psychopathology severity [26].

The ability to participate in social roles and activities item bank

The Ability to Participate in Social Roles and Activities item bank (APSRA) is part of the innovative Patient-Reported Outcomes Information System (PROMIS) implemented by the ‘HealthMeasures’ initiative1 [20, 27]. PROMIS develops item banks by using Item Response Theory (IRT) making them suitable for Computerized Adaptive Tests and for creating short forms. For this study we used the (experimental) extended Dutch version of APSRA item bank which is part of a larger study to improve the measurement of daily functioning. According to a recent psychometric evaluation, the extended APSRA item bank provided more accurate assessments in clinical populations [19, 21]. The extended APSRA item bank is a unidimensional scale that consists of 52 negatively worded items (e.g. “I have trouble doing my regular daily work around the house”) that are scored on a 5-point Likert scale (1 = always, 2 = usually, 3 = sometimes, 4 = rarely, 5 = never), with higher scores indicating a greater ability to participate. For this study we use the total score (i.e., the average score of all items).

The participation ladder

As estimate of the level of actual real-life participation of patients, the participation ladder was used [28]. The participation ladder is a freely available Dutch adaptation of the ladder of citizen participation developed by a joint initiative from the twelve municipalities in the Netherlands [29] to measure social and economic participation in society. The participation ladder is a 6-point ordinal scale, with a potential emphasis on paid labor relative to other forms of societal contribution. Participants self-rated their level of participation by selecting the category that most accurately reflected their current situation: (1) a withdrawn life; (2) social contacts outside the home; (3) participation in organized activities; (4) unpaid work; (5) paid work with support; (6) paid work without support. Higher scores indicate greater participation.

Statistical analysis

Analyses were conducted using the R statistical language [30]. The main package used for the latent profile analysis was tidyLPA [31].

Latent profile analysis

To identify unobserved subgroups (classes) within the data, we conducted Latent Profile Analysis (LPA) using the total scores from the BSI and APSRA item bank. LPA and Latent Class Analysis (LCA) are comparable modeling techniques for detecting unobserved groups within data and are largely equivalent. As such, the terms “profile” and “class” are used interchangeably [32].

Given the theoretical framework emphasizing the identification of convergent and divergent profiles (defined by combinations of high/low symptoms and high/low functioning) a range of 1 to 6 latent classes was deemed appropriate. A two-class solution typically captures broad convergent patterns (e.g., high symptoms with high impairment, low symptoms with low impairment), while models with three to six classes allow for the identification of more nuanced divergent profiles (e.g., high symptoms with low impairment or vice versa). Limiting the analysis to 6 classes ensures a balance between theoretical interpretability and statistical robustness, minimizing the risk of overfitting while adequately representing the complexity of the data. Furthermore, although traditional LPA models often assume equal variances and covariances fixed to zero, our approach compared models under varying assumptions of variance-covariance structures. Model 1 is characterized by equal variances across the classes and covariances fixed to 0. Model 2 features varying variances across the classes but also has covariances fixed to 0. Model 3 assumes equal variances across the classes and equal covariances between the variables within each class. Model 4 allows for both varying variances across the classes and varying covariances between the variables within each class. The Tidy LPA package is an interface for the mclust package that uses model-based clustering based on parameterized finite Gaussian mixture models. Specifically, it uses hierarchical agglomerative clustering to group the data before the Expectation-Maximization (EM) algorithm is applied to estimate the model parameters. This hierarchical clustering approach starts by individually assigning each data point to its own cluster and then iteratively merges the closest clusters based on similarity. The clusters identified in this initial step serve as a more informed starting point for the EM algorithm, reducing the risk of the model converging on a local sub-optimal solution [33–35].

The Analytic Hierarchy Process (AHP) [36, 37], based on the fit indices Aikake Information Criterion (AIC), Approximate Weight of Evidence (AWE), Bayesian Information Criterion (BIC), Classification Likelihood Criterion (CLC), and Kullback Information Criterion (KIC) [38–42], was used to determine the best model with lower values indicating a better fit, alongside with the Bootstrapped Likelihood Ratio Test (BLRT) [17, 43–45].

The BLRT evaluates the improvement in fit between a model with k-classes and a simpler model with k − 1 classes. A significant p-value < 0.05 indicates that the k-class model provides a statistically better fit than the k − 1 class model. The BLRT helps avoid overfitting by identifying when adding a new class does not significantly improve the model.

The AHP is an automated, systematic decision-making method employed in the tidyLPA package in R to determine the optimal number of latent profiles (classes) in LPA. This approach facilitates model selection by hierarchically weighing multiple criteria (e.g., AIC, BIC) and synthesizing the results to identify the most suitable solution. This approach ensures that no single index dominates the decision, and that the final recommendation balances multiple statistical considerations. However, since AHP does not incorporate clinical interpretability or domain knowledge into the ranking, we carefully assessed whether additional classes provided meaningful distinctions beyond statistical improvements. Rather than relying solely on automated selection, we evaluated whether the identified classes reflected qualitatively distinct subgroups. Specifically, we aimed to determine whether solutions aligned with our hypotheses regarding the existence of convergent classes (high symptoms with high functional limitations, or low symptoms with low functional limitations) and divergent classes (high symptoms with good functioning, or low symptoms with poor functioning). Our final model selection was based on a combination of statistical fit indices and substantive interpretability, while also considering AHP’s automated ranking as a supportive tool. Additionally, we examined entropy. Entropy measures classification uncertainty, meaning higher values indicate greater uncertainty. To align with interpretability, tidyLPA automatically reverse-codes entropy, so that 1 represents complete classification certainty and 0 indicates complete uncertainty [39]. We also evaluated the minimum and maximum probability for most likely class membership by assigned class, as well as the proportion of the sample assigned to the smallest and largest classes. These values should be as high as possible, reflecting greater classification certainty [46]. In LPA, average latent posterior probabilities are used to assess the reliability and precision of class assignments. Each individual is assigned to a latent profile based on their highest posterior probability, which reflects the likelihood of class membership given their data and the model. The average probabilities across individuals within each class are then calculated. Values equal or above 0.7 typically indicate well-defined, distinct classes, with high confidence in class assignments [47]. Lower values may suggest greater imprecision in class assignments or potential overlap between classes.

LPA classes and external variables

After identifying the LPA classes and determining class membership for all participants, we examined differences in suicidal ideation, participation level, diagnosis, age, sex, living situation, and education level across these classes.

Suicidal ideation was defined as being at least somewhat bothered by the symptom, based on item 9 of the BSI (“Thoughts of ending your life”).

For diagnosis, DSM classifications were grouped into the following main categories: Attention Deficit Hyperactivity Disorder (ADHD), Obsessive-Compulsive Disorder (OCD), Autism Spectrum Disorder (ASD), Trauma-Related Disorders, Personality Disorders, Anxiety Disorders, Depressive Disorders, Somatic Symptom Disorders, and Other.

Participation level was measured by the participation ladder as described earlier. Age was recorded in years at the time of assessment, and sex was categorized as male or female.

Living situation was classified into several groups: individuals living alone were categorized as “Single,” while those who were single parents were placed in the “Single Parent” category. Children living with a single parent were classified as “Child of Single Parent,” whereas those living with more than one parent were categorized as “Child of Multiple Parents.” Participants living with a partner and children were classified as “Partner and Child,” while those living with a partner but without children fell under “Partner, No Child.” Those residing in a mental health institution were placed in the “Mental Health Institution” category, whereas those in assisted living facilities were classified as “Non-Mental Health Institution.” Participants who were homeless were categorized as “Homeless,” and any other living situations were grouped under “Other.”

Education level was classified into three categories based on the Dutch Standard Education Classification and aligned with U.S. equivalents. The low category included elementary and early secondary education, corresponding to middle school and early high school. The middle category encompassed high school, including both college preparatory and vocational tracks, as well as community college. The high category included university education at the bachelor’s, master’s, and doctoral levels.

To determine if there were significant differences between the means of age between the LPA classes, we used a one-way Analysis of Variance (ANOVA). To test the association between suicidal ideation, participation, diagnosis, sex, living situation, and education level, with LPA classes we used Chi Square tests. Effect sizes of 0.2, 0.5, and 0.8 represent small, medium, and large effects for where Cohen’s d was used and 0.3, 0.5 and > 0.5 where Cramer’s V was used [48]. To examine the effect sizes of the mean differences on the BSI and APSRA between all latent classes, we conducted pairwise comparisons using the Tukey’s Honestly Significant Difference (HSD) test, with adjusted p-values to control for inflated Type I error due to multiple comparisons. To explore the degree of association between APSRA scores (reflecting potential and ability in daily functioning) and the participation ladder (reflecting actual societal participation), we calculated Kendall’s rank correlation coefficient. Although this correlation does not establish whether the instruments measure distinct constructs, it offers insight into their relationship and is consistent with our conceptual view that these tools likely assess related but not identical aspects of functioning.

Results

Latent profiles

Drawing upon the overall evidence presented in the analysis of latent profile models, it can be concluded that a 4-class solution appears to be generally preferable for understanding the latent structure within the data, irrespective of the specific model type (see Table 2). Despite the differences in their underlying assumptions about the (co)variance structure, a consistent pattern emerges from the analysis regarding the optimal number of latent classes. Across the majority of the models examined (Model 1, Model 2, and Model 3), the Bootstrapped Likelihood Ratio Test (BLRT) consistently indicated statistically significant improvements in model fit when transitioning from a 3-class to a 4-class solution. Examining the AIC and BIC values across all models also generally suggests that the 4-class model represents a good balance between model fit and parsimony, as the decrease in these information criteria tends to level off or even increase beyond four classes. This suggests that the addition of a fourth latent profile provides a meaningful and statistically justifiable enhancement in the ability of these models to capture the underlying heterogeneity within the sample. Conversely, a common trend observed across the model specifications was the lack of statistically significant improvements in fit when moving beyond a 4-class solution to either a 5 or 6-class model. This is evidenced by the non-significant BLRT p-values for these subsequent increases in the number of classes, and decreasing entropy with additional classes, indicating greater overlap between them. While Model 4 exhibited a non-significant BLRT p-value for the 3 to 4 class transition (p =.26), the consistent pattern of significant improvement up to 4 classes observed in the other model types, coupled with the general lack of justification for adding more classes beyond this point across the board, continues to provide a compelling argument for the overall preference of a 4-class solution. Adding further complexity to the models beyond four classes may not be warranted from a statistical perspective and could potentially lead to overfitting without providing a substantial gain in explanatory power.

Table 2.

Fit indices LPA models

Model	Classes	AIC	AWE	BIC	CLC	KIC	Entropy	prob_min	prob_max	n_min	n_max	BLRT_val	BLRT_p
1	1	5738.51	5795.85	5758.18	5732.51	5745.51	1.00	1.00	1.00	1.00	1.00	-	-
1	2	5331.76	5434.27	5366.19	5319.10	5341.76	0.67	0.88	0.91	0.46	0.54	412.75	< 0.01
1	3	5214.56	5361.55	5263.74	5195.93	5227.56	0.68	0.77	0.88	0.17	0.52	123.20	< 0.01
1	4	5183.19	5374.72	5247.12	5158.52	5199.19	0.67	0.75	0.84	0.08	0.35	37.37	< 0.01
1	5	5188.11	5424.33	5266.80	5157.26	5207.11	0.58	0.49	0.80	0.10	0.33	1.08	0.50
1	6	5193.40	5474.18	5286.83	5156.48	5215.40	0.54	0.06	0.85	0.01	0.30	0.72	0.47
2	1	5738.51	5795.85	5758.18	5732.51	5745.51	1.00	1.00	1.00	1.00	1.00	-	-
2	2	5312.39	5444.51	5356.65	5295.80	5324.39	0.70	0.88	0.93	0.39	0.61	436.12	0.01
2	3	5197.59	5403.84	5266.44	5171.04	5214.59	0.72	0.78	0.90	0.14	0.58	124.80	0.01
2	4	5164.36	5444.78	5257.79	5127.81	5186.36	0.73	0.78	0.86	0.08	0.45	43.24	0.01
2	5	5169.33	5524.16	5287.36	5122.55	5196.33	0.61	0.54	0.85	0.07	0.30	5.03	0.37
2	6	5174.02	5602.95	5316.63	5117.32	5206.02	0.65	0.51	0.89	0.01	0.32	5.31	0.63
3	1	5211.49	5283.67	5236.08	5203.49	5219.49	1.00	1.00	1.00	1.00	1.00	-	-
3	2	5207.89	5325.74	5247.23	5192.73	5218.89	0.42	0.77	0.85	0.42	0.58	9.60	0.03
3	3	5185.53	5347.86	5239.63	5164.39	5199.53	0.43	0.50	0.83	0.21	0.40	28.36	0.01
3	4	5179.69	5386.35	5248.53	5152.72	5196.69	0.52	0.64	0.85	0.11	0.34	11.85	0.01
3	5	5179.00	5430.19	5262.60	5146.02	5199.00	0.51	0.56	0.77	0.11	0.25	6.69	0.07
3	6	5183.91	5479.61	5282.27	5144.93	5206.91	0.51	0.12	0.83	0.03	0.24	1.09	0.27
4	1	5211.49	5283.67	5236.08	5203.49	5219.49	1.00	1.00	1.00	1.00	1.00	-	-
4	2	5176.82	5338.91	5230.92	5155.93	5190.82	0.55	0.79	0.88	0.32	0.68	46.67	< 0.01
4	3	5177.42	5428.67	5261.02	5144.37	5197.42	0.47	0.57	0.85	0.24	0.43	11.40	0.07
4	4	5182.58	5522.79	5295.69	5137.58	5208.58	0.50	0.53	0.87	0.13	0.39	6.84	0.26
4	5	5175.89	5605.14	5318.50	5118.86	5207.89	0.49	0.46	0.87	0.13	0.26	18.69	0.07
4	6	5190.00	5708.15	5362.12	5121.08	5228.00	0.54	0.51	0.79	0.09	0.24	-2.11	0.88

Note AIC = Aikake Information Criterion; AWE = Approximate Weight of Evidence; BIC = Bayesian Information Criterion; CLC = Classification Likelihood Criterion; KIC = Kullback Information Criterion; Entropy = Classification uncertainty reversed coded; prob_min = the lowest probability that a patient truly belongs to the class they were assigned to; prob_max the highest probability that a patient belongs to the class they were assigned to; n_min = Proportion of the sample assigned to the smallest class; n_max = Proportion of the sample assigned to the largest class; BLRT = Bootstrapped Likelihood Ratio Test; BLRT_p = p-value for BLRT; Model 1 = equal variances and covariances fixed to 0; Model 2 = Varying variances and covariances fixed to 0; model 3 = Equal variances and equal covariances; model 4 = Varying variances and varying covariances; The 1-class model serves as a baseline assuming a single homogeneous group, its fit indices do not reflect latent structure. BLRT is not reported for the 1-class model because the test compares models with k and k–1 classes, and a 0-class model is not statistically defined

Interestingly, while our theoretical expectations anticipated the identification of two classes with convergent patterns (high symptoms/low functioning and low symptoms/high functioning) and two with divergent patterns (high symptoms/high functioning and low symptoms/low functioning), the analysis revealed that all identified profiles were convergent. This means that our sample did not exhibit the expected divergent patterns of symptom severity and functional limitations. Despite this finding of only convergent classes, the statistical evidence from the BLRT across most model types still supports the preference for a 4-class solution. Furthermore, given that our analysis did not reveal divergent classes, changing the number of classes would not meaningfully alter our conclusions. This suggests that while the specific nature of the classes might differ from initial hypotheses, the overall preference for a 4-class representation of the data remains robust based on the statistical fit indices. Although within-class heterogeneity (i.e., divergence) was present, this was expected and an interpretable feature of LPA. LPA identifies latent classes as overarching prototypes, grouping individuals with shared core symptom and functioning profiles, despite some variability within each class. Individuals assigned to the same class share similar response patterns based on the model’s estimates, which may highlight clinically relevant groupings.

In sum, the balance between achieving statistically significant improvements in model fit and maintaining parsimony strongly suggests that a 4-class model offers a robust and generally applicable representation of the latent profiles in this analysis, regardless of the nuances in specific model parameterizations and despite the unexpected absence of divergent classes.

LPA class characteristics

On average, APSRA scores were M = 3.03 (SD = 0.79, min = 1.02, max = 5), while the BSI psychopathology scores were M = 1.62 (SD = 0.70, min = 0.11, max = 3.79).

The Pearson’s product-moment correlation between APSRA and BSI scores was negative, statistically significant, and large (r = -.64, 95% CI [-0.67, -0.60], t(1008) = -26.34, p <.001), indicating that higher APSRA scores were generally associated with lower levels of psychopathology, and vice versa. The mean BSI score for the current sample falls within the expected range for outpatient populations, exhibiting a level of symptom distress that is above the normative average as defined by the Dutch BSI manual [25].

ANOVA suggested that the main effect of class for the BSI was statistically significant and large (F(3, 1006) = 1627.66, p <.001; Eta² = 0.83, 95% CI [0.82, 1.00]). The main effect of class for the APSRA was also statistically significant and large (F(3, 1006) = 526.95, p <.001; Eta² = 0.61, 95% CI [0.58, 1.00]). For both, the BSI and the APSRA, the effect sizes for all class comparisons were very large (see Table 3).

Table 3.

Effect sizes class mean differences

Class Comparisons	BSI		APSRA
	Cohen’s d	95% CI	Cohen’s d	95% CI
mild vs. moderate	-2.70	-2.89– -2.54	0.87	0.72–1.04
mild vs. minimal	2.04	1.84–2.28	-1.88	-2.08– -1.70
mild vs. severe	-5.42	-5.94– -5.01	2.87	2.63–3.18
moderate vs. minimal	4.60	4.32–4.91	-2.75	-2.98– -2.54
moderate vs. severe	-2.62	-2.94– -2.37	1.81	1.63–2.06
minimal vs. severe	-7.22	-7.96– -6.63	5.21	4.83–5.67

Note The adjusted p-values for multiple comparisons were < 0.05 for all comparisons; BSI = Brief Symptom Inventory; APSRA = Ability to Participate in Social Roles and Activities

The four distinct classes of patients identified by LPA, likely represented a spectrum of increasing severity of symptoms and impairments, i.e., a gradient of self-reported psychosocial dysfunction (see Fig. 1 and supplement Figure S2). Consequently, classes 3, 1, 2, and 4 (LPA assigns class labels arbitrarily and does not determine them based on an intrinsic order) were characterized as representing minimal, mild, moderate and severe cases (see Table 4). These severity levels align with the standardized categories outlined in the BSI manual for individuals with comparable scores within the outpatient population, which include below average, above average, high, and very high [18]. The Dutch BSI further categorizes severity levels based on cumulative percentages: ≤ 5% (very low), ≤ 20% (low), ≤ 40% (below average), ≤ 60% (above average), ≤ 80% (high), ≤ 95% (very high), and > 95% (extremely high) [25].

Fig. 1.

Profile plot 4 class model. Note: From top left to bottom right: Class 4 Severe (plusses), Class 2 Moderate (triangles), Class 1 Mild (dots), and Class 3 Minimal (squares). Scores were standardized into standardized z-scores (mean = 0; SD = 1) to ensure equal contribution and improve interpretability. APSRA_Z = Ability to Participate in Social Roles and Activities z-score; BSI_Z = Brief Symptom Inventory z-score. The rounded lines in the plot enhance visualization by highlighting data concentration. The outermost lines provide a smooth, general overview, while the inner lines follow the data more closely, revealing finer clustering patterns for better interpretation

Table 4.

Descriptive statistics 4-class model

Class	BSIa	APSRAa	BSI_Za, b	95% CIc	APSRA_Za, b	95% CIc
Mild	1.37 (0.28)	3.07 (0.51)	-0.32 (0.49)	[-0.64, − 0.017]	0.06 (0.68)	[-0.10, 0.22]
Moderate	2.14 (0.30)	2.62 (0.53)	0.68 (0.49)	[0.56, 0.81	-0.50 (0.68)	[-0.62, -0.38]
Minimal	0.77 (0.30)	4.00 (0.48)	-1.16 (0.49)	[-1.28, -1.03]	1.13 (0.68)	[0.96, 1.29]
Severe	2.92 (0.29)	1.80 (0.36)	1.73 (0.49)	[1.55, 1.91]	-1.38 (0.68)	[-1.60, -1.15]

^a Mean (SD); ^b standardized z-scores (Mean = 0, SD = 1) were computed using the sample mean and standard deviation for each variable; ^c 95% confidence interval standardized z-scores; BSI = Brief Symptom Inventory; APSRA = Ability to Participate in Social Roles and Activities

The distribution of patients across LPA classes was relatively balanced: minimal (n = 232; 23%), mild (n = 344; 34%), moderate (n = 358; 35%), and severe (n = 76; 8%). The smaller proportion of cases in the severe class aligns with expectations for this outpatient sample. (see online supplement Figure S1 for density plots of the latent classes). The LPA model demonstrated an acceptable level of accuracy in predicting individual class membership, as indicated by probabilities of 0.8 or higher in Table 5 [49].

Table 5.

Average latent posterior probabilities

Class	cprob1	cprob2	cprob3	cprob4
Mild	0.76	0.13	0.11	< 0.0001
Moderate	0.14	0.80	< 0.0001	0.06
Minimal	0.15	< 0.0001	0.85	< 0.0001
Severe	< 0.0001	0.17	< 0.0001	0.83

Note The table represent the average probability that individuals, based on the model’s output, belong to each latent class. These values are the averages of the conditional probabilities (cprob), which indicate the likelihood that an individual in a given observed class belongs to each latent class. For example, cprob1 of 0.76 for Class Mild in the first column, means that individuals assigned to Class 1 have an 76% probability of actually belonging to latent Class 1 based on the model. Higher diagonal values indicate higher accuracy in classifying individuals into their correct latent class, as they reflect the probability of correctly predicting class membership. Off-diagonal values represent the probability of misclassification into other latent classes

We analyzed whether the external variables suicidal ideation (SI), participation level, diagnosis, age, sex, living situation, and education level, showed significant variations across the identified LPA classes (see also Table 1).

Figure 2 shows that class membership was significantly associated with SI and that this effect was moderate (𝜒²_pearson (3) = 132.28, p <.001, V_Cramer = 0.36, 95% CI [0.29, 0.42]). Within class minimal, 27% were at risk for SI, while in class mild, 56% were at risk. For the moderate and severe classes, the proportion of individuals at risk for SI was substantially higher (69% and 86% respectively). This illustrates an increasing risk of SI as severity of self-reported psychosocial dysfunction increases. The prevalence of SI in our sample was consistent with expectations, given that our participants were psychiatric outpatients referred for specialized psychological treatment, with the majority experiencing depressive or anxiety disorders. The variation in prevalence of SI between classes reflects the expected association between higher symptom severity, greater functional limitations, and increased SI prevalence, consistent with clinical observations in similar populations.

Fig. 2.

Class membership and suicidal ideation

The majority of patients (53%) rated themselves at category 6 of the participation ladder, indicating that they were engaged in paid work without support. The Kendall’s tau correlation between APSRA scores and participation ladder scores was positive, statistically significant, but small (tau = 0.19, z = 7.80, p <.001), indicating only a modest association between individuals’ self-reported functional abilities and their actual level of societal participation. While this does not conclusively demonstrate that the two instruments measure distinct constructs, the modest correlation is consistent with the interpretation that they reflect related but potentially different aspects of functioning (see Table 6).

Table 6.

Mean APSRA scores per participation ladder category

	Overall N = 1,0101	1 N = 1101	2 N = 1851	3 N = 641	4 N = 661	5 N = 491	6 N = 5361
APSRA	3.03 (0.80)	2.45 (0.78)	2.92 (0.72)	3.20 (0.80)	3.10 (0.81)	2.86 (0.73)	3.18 (0.77)

¹Mean (SD)

Our Chi-square analysis (χ²_pearson (15) = 93.66, p <.001, V_Cramer = 0.16, 95% CI [0.10, 0.19]) indicates that participation level is not independent of class membership (i.e., the distribution of participation ladder categories differs significantly between latent classes). Specifically, we observe that as class severity increases, the proportion of patients engaged in paid work decreases significantly. Patients classified as minimal exhibited the highest prevalence of having paid work (64%), followed by the mild class (56%), moderate class (48%), and the severe class with the lowest prevalence of having paid work (30%). This trend suggests an association between increasing psychosocial dysfunction and a lower likelihood of having paid work (see Fig. 3).

Fig. 3.

Class Membership and Participation. Note: Levels of participation: 1 = a withdrawn life, 2 = social contacts outside the home, 3 = participation in organized activities, 4 = unpaid work, 5 = paid work with support, 6 = paid work without support

The Pearson’s Chi-squared test of independence between the classes and diagnoses suggests that the effect is statistically significant, but very small (𝜒²_pearson (24) = 51.02, p <.001, V_Cramer = 0.09, 95% CI [0.00, 0.10]). To evaluate the robustness of the result given concerns about small or empty cells, we reran the analysis after excluding diagnostic categories with fewer than five observations (e.g., ASD). The results remained essentially the same, suggesting that the association was not driven by sparse data. Overall, these findings indicate that diagnostic category has limited value in distinguishing between latent profiles. Similarly, the remaining variables - age, sex, living situation, and education level - did not exhibit statistically significant differences across profiles. For living situation, a Chi-squared test was conducted after removing levels with fewer than five observations (Homeless, MHInstitute), and results remained unchanged; these details were omitted for brevity.

Discussion

The main purpose of the present study was to identify unique latent response patterns in psychiatric symptoms and daily functioning by conducting Latent Profile Analysis (LPA) on the Brief Symptom Inventory [18, 25, 26] and the Ability to Participate in Social Roles and Activities item bank [19–22]. In addition to identifying classes with convergent (i.e., aligned) patterns of symptom severity and functional limitations, we hypothesized the presence of divergent profiles: one characterized by low symptom severity paired with significant functional limitations, and another marked by high symptom severity alongside relatively preserved functional capacities.

A four-class model emerged as the optimal solution. Notably, all four profiles exhibited convergent patterns, with no evidence of divergent classes in our sample. Although the specific characteristics of the classes differed from our initial hypotheses, the preference for a four-class representation of the data is supported by robust statistical fit indices. In general, individuals with higher symptom severity reported greater functional limitations and vice versa. This finding is consistent with other evidence of a bidirectional relationship between symptoms and functioning in patients with internalizing symptoms [10, 50]. The four patient profiles could be organized along a continuum of increasing symptom severity and functional impairment, representing a gradient of self-reported psychosocial dysfunction categorized as minimal, mild, moderate, and severe. Significant differences were observed between the classes in terms of suicidal ideation (SI) and work participation. The likelihood of SI increased progressively with class severity, while the proportion of patients reporting paid work decreased. These patterns underscore the interplay between psychiatric symptomatology, functional outcomes, and critical life domains, highlighting the need for interventions tailored to address these interconnected challenges. Additionally, the analysis of diagnostic categories revealed a statistically significant, yet negligible, association with the identified profiles. This suggests that, while diagnosis remains important for clinical decision making, it has limited utility in differentiating patient profiles within the context of LPA. Moreover, demographic variables such as age, sex, living situation, and education level did not show statistically significant differences across the profiles.

These findings emphasize the importance of focusing on symptom severity and functional limitations as primary determinants in understanding patient heterogeneity, rather than relying solely on demographic or diagnostic characteristics. Our study shows that it is feasible to identify patient subgroups based on symptom severity and daily functioning, with these subgroups differing meaningfully in suicidal ideation and participation in work - both key treatment targets. While there is an ongoing discussion regarding the prevailing symptom-reduction model in mental healthcare, in favor of an alternative approach that prioritizes existential recovery and social participation [51], the reciprocal relationship between symptomatic and functional dimensions of mental disorders does not necessarily imply the exclusion of either in clinical assessment or treatment. The connection between symptoms and functioning could offer complementary advantages, for instance in cases where symptom reporting may be biased due to factors like stigma, cultural beliefs, or limited self-awareness. In these instances, asking about daily functioning can offer additional context on the overall impact of mental health issues [11]. Daily functioning, being based on observable behaviors, allows for a more objective assessment, reducing the influence of personal interpretation. Furthermore, questions about functioning are often perceived as less stigmatizing than those focused on symptoms like anxiety or depression, which may encourage individuals to respond more honestly. Culturally, daily functioning is more universally understood, making it easier for individuals to report without the same biases that might affect symptom reporting. Additionally, people often have a clearer sense of their daily functioning than of their symptom levels, leading to more accurate assessments. Finally, difficulties in functioning can serve as a reference point for symptom severity, even when individuals are reluctant to directly label or discuss their symptoms.

At the same time, assessing symptoms alongside functional impairments has its own set of advantages. Symptom measures can provide detailed insights into specific mental health conditions, capturing nuances like the intensity, duration, and frequency of experiences such as anxiety and depression. These details can guide targeted interventions by identifying precise areas requiring attention. Symptoms may precede changes in functioning, serving as early indicators of emerging functional impairments or potential recovery as symptom severity changes [12, 13]. Moreover, self-reported symptoms allow individuals to communicate their subjective experiences, providing a valuable perspective that functioning assessments alone might overlook. Together, these complementary approaches may lead to a more nuanced understanding of an individual’s experience while potentially minimizing the biases associated with self-reporting [52, 53].

Furthermore, these results are relevant to the ongoing political debate regarding the current Dutch funding system in mental health care. Since 2022, mental health care costs in the Netherlands have been charged by means of the ‘care performance model’, a derivative of the model that was developed for the National Health Services (NHS) in the United Kingdom. This system requires that the healthcare provider records the type of care that is needed according to the severity of the mental health problems and social status, as indicated by the clinician rated Health of the Nation Outcome Scale Plus (HoNOS+) [14]. This method has been criticized because this clustering system is conceptually weak, lacks sufficient scientific evidence, is complex, and the HoNOS + is not a good predictor of health care costs [15, 16]. This raises the question of whether a more efficient approach, through self-reported assessments of symptom severity and functional limitations, might represent a more suitable alternative. Profiling patients within the identified LPA classes could further enhance this efficiency by uncovering distinct patterns of symptoms and functional impairments. In addition, using self-report measures could also reduce potential bias from clinicians and contribute to a transparent, person-centered approach considering what is important to patients themselves.

Strengths, limitations, and future research

This study has various strengths and limitations. An important strength is the large sample size of Dutch outpatients which allows for reliably exploring the heterogeneity and analysis of subgroups. Another strength of this study is the use of well-known, validated measuring instruments. However, it is important to consider the following limitations. First, the profiles are based on cross-sectional data and therefore it is uncertain how stable these profiles are over time. We recognize that patients may shift between classes over time, which is both expected and clinically meaningful, especially as symptoms, functioning, or treatment status change. However, we consider the reliance on cross-sectional data a limitation, as it prevents us from assessing the stability of the class structure over time. While it is possible that the latent profiles identified in cross-sectional data are stable, with only individual patients transitioning between classes, it is also likely that profiles derived from longitudinal data could differ. As patients progress through various stages of treatment or illness, the relationships between symptoms and functioning may change, potentially leading to the emergence of new or modified profiles that cannot be captured by a single time point. Second, the absence of specific data on the number of clients who chose not to answer the APSRA questions after the mandatory BSI (since the APSRA was not mandatory) may have introduced potential bias into the sample. However, based on our experience with the healthcare system, the sample seems to represent the targeted population well. Third, a potential limitation of this study is that, while the MINI was available as part of the interview process, its use was not mandatory, allowing the psychologist to decide whether or not to administer it. Unfortunately, we do not have case-wise data on whether the MINI was used. However, we emphasize that the diagnostic procedure was conducted carefully, with diagnoses made through consultation and consensus among professionals. Nonetheless, we acknowledge that the variability in MINI use could pose a limitation.

An important direction for future research is to assess the clinical utility of profile-based classifications compared to models that rely on continuous measures such as the BSI and APSRA scores. While the current study focused on identifying patient profiles that primarily differed in overall severity, future studies could evaluate whether these latent profiles or individual-level continuous scores are more effective in predicting relevant clinical outcomes, such as suicidality or treatment response. This comparison could also extend to practical considerations, such as clinician satisfaction and ease of use in routine care. Such research would help determine whether categorical classifications or dimensional approaches offer greater value in clinical decision-making and personalized treatment planning. Investigating the impact of dual assessment on treatment outcomes through randomized controlled trials (RCTs) is also warranted. For example, patients could be assessed using both symptom and functioning measures, or only one dimension, with comparisons of treatment outcomes. Such studies could determine if dual assessment enhances predictions of treatment response, recovery trajectories, or goal prioritization. Future research should also focus on analyzing latent profiles in longitudinal data across diverse clinical populations. The longitudinal stability of the classes is relevant because it helps confirm whether the identified classes remain meaningful and consistent over time, rather than being arbitrary subdivisions of a continuous spectrum. Our findings suggest that the convergent classes consistently represent this gradient of severity, and thus, the stability of class membership over time enhances the interpretability of the model. It could further explore whether these profiles can predict treatment needs, healthcare utilization, recovery trajectories, and minimize biases related to self-reporting of symptoms, thereby refining integrated care approaches. Additionally, machine learning could be instrumental in advancing this area by analyzing large patient datasets that include both symptom and functioning measures. These models could identify complex patterns and interactions between symptoms, functioning, and outcomes, helping predict which combinations of symptom severity and functioning levels are most associated with treatment needs or outcomes. Machine learning could also aid real-time decision-making, offering clinicians personalized treatment recommendations based on dual assessment data. This approach would validate the added value of dual assessments, improving their practical application in clinical settings and supporting clinicians in optimizing care pathways and resource allocation.

Conclusions

The four latent profiles represent a clinically meaningful spectrum of psychosocial dysfunction, distinguishing patients based on key outcomes such as suicidal ideation and work participation. These profiles provide a potentially valuable framework for personalizing interventions, prioritizing treatment goals, and allocating resources according to shared patterns of characteristics. Future research should focus on validating the temporal stability of these profiles, assessing their predictive value for treatment outcomes, and investigating whether dual assessment improves clinical decision-making. Nevertheless, the current findings indicate that integrating symptom and functioning assessments offers a more comprehensive understanding of psychiatric presentations, promoting a holistic, person-centered approach to mental healthcare.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Abbreviations

ADHD

Attention Deficit Hyperactivity Disorder

ASD

Autism Spectrum Disorder

AIC

Aikake Information Criterion

ANOVA

Analysis of Variance

APSRA

Ability to Participate in Social Roles and Activities item bank

AWE

Approximate Weight of Evidence

BIC

Bayesian Information Criterion

BLRT

Bootstrapped Likelihood Ratio Test

BSI

Brief Symptom Inventory

CLC

Classification Likelihood Criterion

DG

Dimence Group

FWA

Federal Wide Assurance

GSI

Global Severity Index

HoNOS+

Health of the Nation Outcome Scale Plus

IRT

Item Response Theory

KIC

Kullback Information Criterion

LCA

Latent Class Analysis

LPA

Latent Profile Analysis

MERC

Medical Ethics Review Committee

NESDA

Netherlands Study of Depression and Anxiety

OCD

Obsessive Compulsive Disorder

OHRP

Office for Human Research Protections

PROMIS

Patient-Reported Outcomes Information System

SI

Suicidal Ideation

VUmc

VU University medical center

WMO

Medical Research Involving Humans Subjects Act

Author contributions

GLW, EdB, and PS collaboratively developed the project’s core concepts and theoretical framework. GLW took the lead on data analysis, manuscript preparation, and visualization. EdB and PS contributed to project planning, oversight, research design, result interpretation, and manuscript writing. All authors reviewed and endorsed the final manuscript.

Funding

The first author received financial support from the Dimence Group for the initial data collection.

Data availability

The data and the R codes used for analysis are available upon reasonable request from the corresponding author.

Declarations

Ethics approval and consent to participate

This study is part of a larger project concerning the development of Computerized Adaptive Tests for measuring the level of functioning in psychiatric patients. The Medical Ethics Review Committee (MERC) of VU University Medical Center (VUmc) confirmed that the Medical Research Involving Humans Subjects Act (WMO) does not apply to this study, and an official approval by the MERC was not required. The MERC of VUmc is registered with the US Office for Human Research Protections (OHRP) as IRB00002991. The Federal Wide Assurance (FWA) number assigned to the VUmc is FWA00017598. Informed consent was obtained from all individual participants included in the study. This study was conducted in accordance with the ethical standards of the institutional research committee and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.