Multidimensional social inclusion and its prediction in schizophrenia spectrum disorder

doi:10.21203/rs.3.rs-2608209/v1

Download PDF

Research Article

Multidimensional social inclusion and its prediction in schizophrenia spectrum disorder

https://doi.org/10.21203/rs.3.rs-2608209/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 07 Mar, 2024

Read the published version in Social Psychiatry and Psychiatric Epidemiology →

Version 1

posted

You are reading this latest preprint version

Social inclusion is poor among patients with chronic disorders such as schizophrenia spectrum disorder (SSD). It significantly impacts patient life, healthcare and society. We aimed to study multidimensional social inclusion (mSI) among patients diagnosed with SSD, and to test the prediction of mSI. We used the baseline and 3-year follow-up data of 1,119 patients from the Genetic Risk and Outcome in Psychosis (GROUP) cohort. The mSI was conceptualized by all subscales from social functioning (measured by Social Functioning Scale [SFS]) and quality of life (measured by the brief version of World Health Organization Quality of Life [WHOQOL-BREF]) questionnaires. K-means clustering was applied to identify mSI subgroups. Prediction models were built and internally validated via multinomial logistic regression (MLR) and random forest (RF) methods. Model fittings were compared by common factors, accuracy and the discriminability of mSI subgroups. We identified five mSI groups: “very low (social functioning)/very low (quality of life)”, “low/low”, “high/low”, “medium/high”, and “high/high”. The mSI was robustly predicted by genetic predisposition, premorbid social functioning, symptoms (i.e., positive, negative and depressive), number of met needs and baseline satisfaction with the environment and social life. The RF model was cautiously regarded to outperform the MLR model. We distinguished meaningful subgroups of mSI by combining rather than using two measurements standalone. The mSI subgroups were modestly predictable. The mSI has the potentials for personalized interventions to improve social recovery in patients. Different from conventional outcomes, we introduced mSI which has implications beyond clinics and could be applied to other disorders.

social inclusion

social exclusion

schizophrenia

psychosis

cluster

prediction

Five percent of the world’s population is suffering from severe psychiatric disorders, of which one out of five is affected by schizophrenia spectrum disorder (SSD) [1]. People with SSD, as a consequence of their disorder, face poor social inclusion which is often also a risk factor of SSD. This suggests a major challenge of a bidirectional relationship between SSD and poor social inclusion [2, 3]. Hence, social inclusion has become one of the ultimate goals in SSD recovery. Compared to physical diseases, the resources allocated to the management of psychiatric disorders including SSD are limited. For example, the Netherlands spent 0.47% of the total healthcare expenditures on psychiatric disorder management in 2017, when many countries do not have such category records of expenses [4–6]. Taking additional efforts to enhance social inclusion can improve clinical and social recovery, life controllability, and quality of life of the patients, and simultaneously bring tremendous benefits to stakeholders. Specifically, the healthcare organizations and governments can allocate fewer resources to the adverse impacts of poor social inclusion; and most importantly, patient families will bear less burden. Therefore, clearly defining social inclusion for an indicative purpose, and accurately identifying predictive factors of patients with SSD can be a crucial step to facilitate the recoveries and prevent the loss of social inclusion of the patients with SSD.

Generally, social inclusion is described as “a multidimensional state where prevailing conditions enable full and active participation in all aspects of daily life” [7–9]. By this definition, “multidimensional” is unrestricted to individual characteristics [10], relationships with other individuals and group environments [7, 10], subjective and objective living environments [11–14] and social-political rights [10]. Clearly, social inclusion is a broad holistic concept which is broader than conventional outcomes such as clinical recovery and quality of life (QoL) in medical field. However, in clinical research, social inclusion is mostly analyzed using only one of the indicators such as social functioning and employment, education, social recovery, and QoL as a proxy for social inclusion [15–17]. Moreover, with regard to “full and active participation”, the way patients with SSD subjectively view their functioning in society and feel about their surroundings should be a primary concern and of our interest.

In a previous study, we examined the factors associated with the self-reported social functioning which is an important element of social inclusion [18]. We found that ethnicity, age, education, genetic predisposition, illness duration, cannabis use, childhood trauma, marriage status, urbanicity and unmet needs of social support, on top of the symptomatic trajectories, are associated with social functioning. To date, quantitative studies have not yet captured the multidimensionality nature of the social inclusion of patients with SSD. This is probably caused by the ambiguous scope of social inclusion, conceptual complexity, and the lack of validated instruments to measure social inclusion [19]. Likewise, our earlier study still adopted a unidimensional construct of social inclusion. Since multidimensional social inclusion (mSI) is not only relevant in understanding the development of social recoveries over the years but also potentially informs clinicians at an early stage on the prognosis of their patients based upon the characteristics that are included in the mSI conceptualization, in the present study, we extend to mSI by using the subscales of social functioning and additional QoL. By doing so, we want to understand mSI that disparate from applying two measurements standalone, is likely to describe different subgroups demonstrating different social inclusion levels while accentuating subjective feelings and covering more extensive areas of living.

Furthermore, recently, data-driven approaches have gained popularity due to their methodological advantages and focus on prediction accuracy. In the context of SSD, using baseline clinical characteristics, machine learning models can reliably predict symptom remission, social recovery, vocational recovery and QoL at 1 year [16, 17], and can predict 3-year and 6-year symptomatic remission and dichotomous global functioning [20]. Particularly, data-driven approaches with relaxed assumptions such as linearity are anticipated to be more suitable for the intricate nature of mSI. Regardless of the advantages of data-driven approaches, they do not always outperform standard approaches such as regressions concerning the evaluation metrics, such as accuracy, relevant to the research question and/or the requirement in the clinical settings [21–24]. Therefore, to predict mSI most accurately, we are intrigued to employ and compare standard and data-driven approaches.

In this study, we aimed to study mSI clusters conceptualized by social functioning and QoL among the patients diagnosed with SSD; and to test mSI prediction through standard and data-driven models. We hypothesized that mSI clusters exist within the SSD cohort, which contribute to understanding of the needs of each patient subgroup. We also hypothesized that the data-driven approach might not perform worse and might identify a different set of predictors than the standard approach. The aforementioned hypotheses were corroborated by answering three research questions. Firstly, how many patient subgroups can be identified if social inclusion is defined as multidimensional? Secondly, what are the factors that predict 3-year mSI? In the pursuit of the most accurate factor identification, two methods are compared, which are multinomial logistic regression (MLR, standard approach) and random forest (RF, data-driven approach). Therefore, lastly, which model outperforms concerning accuracy and discriminability so that the patients with SSD can be correctly allocate into a mSI subgroup?

Study design and population

The current study is embedded within the framework of the Dutch-nationwide database Genetic Risk and Outcome in Psychosis (GROUP) project, using its data release 8.0 (Supplementary Fig. 1). Details on the GROUP project structure, participant recruitment, data collection and ethical approval have been published elsewhere [25]. In brief, patients were recruited across 36 partner mental health institutes including 4 university medical centers in Amsterdam, Groningen, Maastricht, and Utrecht that cover about 75% of the Dutch population. The inclusion criteria were as follows: 1) age of 16–50 years; 2) fluent Dutch speaking; 3) diagnosis of a non-affective psychotic disorder according to DSM-IV; 4) the first contact with mental health care service no longer than 10 years ago, and 5) being able and willing to give written consent. With 1,119 patients enrolled at baseline, the measurements taken at baseline and 3-year follow-up were used.

Outcome

Multidimensional social inclusion at 3-year follow-up was conceptualized from the 13 subscales of two validated self-reported surveys: Social Functioning Scale (SFS, including independence-performance, independence-competence, occupation, withdrawal, interpersonal behaviors, recreation and pro-sociality) [26] and World Health Organization Quality of Life (WHOQOL-BREF, including general quality of life, perceived health, and physical, psychological, social and environmental domains) [27].

Predictors

Eighty-eight potential predictors were included at baseline and/or 3-year follow-up based on literature review, experts’ opinions and GROUP data availability. The inclusion of the 3-year follow-up was mainly attributed to the intention of inspecting the impacts of the change in a variable such as cannabis consumption between the baseline and 3-year on the outcome mSI, instead of the variable at baseline [28]. Predictors were categorized into five categories: 1) socio-demographic variables including medical centers, age, gender, ethnicity; 2) genetic and early-life factors including polygenic risk score for schizophrenia (PRS_SCZ), a mean score from the Premorbid Adjustment Scale (PAS-overall) [29], a mean score from the Childhood Trauma Questionnaire (CTQ-total, Dutch version: Jeugd Trauma Vragenlijst [JTV]) [30]; 3) disease profiles featured by IQ, age of onset, duration of psychosis, diagnosis of psychosis (i.e., affective versus non-affective), positive symptoms and core negative symptoms (i.e., Blunted affect, Emotional withdrawal, Poor rapport, Lack of spontaneity, Active social avoidance) from the Positive and Negative Syndrome Scale (PANSS) [31], frequency and distress level of depressive symptoms from the Community Assessment of Psychic Experience (CAPE) [32], disability and symptom subscales of the Global Assessment of Functioning (GAF) [33], remission; 4) work, social and overall environmental variables such as current urbanicity, current working status and payment, number of met needs from Camberwell Assessment of Need (CAN) [34]; and 5) medication and substance use covering duration of using antipsychotics and a change of antipsychotic dose (in chlorpromazine equivalents), awareness of antipsychotic use, changed uses of alcohol, cigarette and weeks of cannabis consumption. Covariables include domains of WHOQOL-BREF measured at baseline (i.e., physical, psychological, social and environmental domains) [27]. Overall, the variables conceptually overlapping with the outcome mSI were regarded as predictors to avoid complicating and invalidating the mSI outcome conceptualized on the two validated questionnaires. Detailed explanations of the predictors are in Table 1.

Table 1

Characteristics of study population and predictors
N = 1,119	Observed data		Complete data
	Missingness%	Median [IQR] / N (%)	Median [IQR] / N (%)
A. Model predictor ^d
Medical center	0.00
Amsterdam		283 (25.29)	283 (25.29)
Groningen		287 (25.65)	287 (25.65)
Maastricht		306 (27.35)	306 (27.35)
Utrecht		243 (21.72)	243 (21.72)
Age (year)	0.27	26.00 [22.00, 32.00]	26.00 [22.00, 32.00]
Gender: female	0.00	267 (23.86)	267 (23.86)
Ethnicity: Non-Caucasian	3.40	222 (20.54)	256 (22.88)
PRS_SCZ	36.64	-3.45 [-7.11, 0.97]	-3.16 [-5.92, 0.00]
PAS-overall	15.91	1.89 [1.37, 2.53]	1.95 [1.42, 2.47]
CTQ-total	32.53	1.48 [1.24, 1.84]	1.53 [1.32, 1.80]
IQ	6.88	94.00 [83.00, 105.00]	94.00 [83.50, 105.00]
Age of onset (year)	0.18	21.00 [18.00, 26.00]	21.00 [18.00, 26.00]
Duration of psychosis (year)	4.74	3.00 [1.00, 6.00]	3.00 [1.00, 6.00]
Diagnosis: affective	0.00	121 (10.81)	121 (10.81)
Positive symptopms	8.76	12.00 [9.00, 18.00]	13.00 [9.00, 17.50]
Core negative symptoms	8.13	9.00 [6.00, 14.00]	10.00 [6.00, 13.00]
Depressive symptom (frequency)	21.18	0.88 [0.62, 1.38]	1.00 [0.62, 1.29]
Depressive symptom (distress level)	23.77	1.43 [1.00, 2.00]	1.40 [1.00, 1.82]
GAF-disability	12.15	55.00 [41.00, 65.00]	54.00 [42.00, 65.00]
GAF-symptom	11.80	55.00 [45.00, 65.00]	55.00 [45.00, 65.00]
Remission (baseline to Year 3) ^a	32.62
No		309 (40.98)	365 (32.6)
Yes (≤ 6 months)		177 (23.47)	464 (41.5)
Yes (> 6 months)		268 (35.54)	290 (25.9)
Current urbanicity	33.24
No to little		309 (41.37)	422 (37.7)
Moderate		73 (9.77)	238 (21.3)
Strong to very strong		365 (48.86)	459 (41.0)
Work: Full time or part time	9.92	514 (51.00)	615 (55.0)
Work payment	9.92
None		414 (41.10)	429 (38.34)
Paid		329 (32.63)	404 (36.10)
Voluntary		112 (11.11)	132 (11.80)
Mixed		105 (10.42)	104 (9.29)
Unknown		50 (4.96)	50 (4.47)
Number of met needs	11.08	4.00 [2.00, 5.00]	4.00 [2.00, 5.00]
WHOQOL-BREF-environmental domain	11.62	3.57 [3.14, 4.00]	3.57 [3.14, 3.99]
WHOQOL-BREF-physical domain	11.62	3.43 [3.00, 3.86]	3.43 [3.00, 3.86]
WHOQOL-BREF-psychological domain	11.62	3.33 [2.83, 3.83]	3.33 [2.83, 3.83]
WHOQOL-BREF-social domain	11.62	3.33 [2.67, 4.00]	3.33 [2.67, 3.67]
Duration of using antipsychotics (year)	17.78	1.00 [0.00, 10.00]	2.00 [0.00, 12.00]
Antipsychotic dose (in Chlorpromazine equivalents) ^{b, c}	58.36
Drastic reduction		88 (13.48)	252 (22.52)
Moderate reduction		100 (15.31)	308 (27.52)
No change		114 (17.46)	114 (10.19)
Moderate increase		76 (11.64)	202 (18.05)
Drastic increase		88 (13.48)	243 (21.72)
Awareness of antipsychotic use: (Become) aware ^{b, d}	36.19	601 (84.17)	1,003 (89.64)
Change of alcohol units consumed per week ^b	69.35	0.00 [-2.00,2.00]	0.00 [-2.00, 3.00]
Change of cigarette units consumed ^b	69.26	0.00 [-1.00, 3.00]	0.00 [-1.00, 4.00]
Change of total week with cannabis consumption in past 12 months ^b	68.81	0.00 [0.00, 0.00]	0.00 [0.00, 2.00]
B. Outcome
Multidimensional social inclusion subgroup
Low/Low (LL)			240 (21.45)
High/Low (HL)			551 (49.24)
Medium/High (MH)			202 (18.05)
High/High (HH)			126 (11.26)
Note: ^a Remission is a longitudinal variable that recorded a patient’s remission status between the baseline and 3-year follow-up. ^b The variable indicates change from baseline to 3-year follow-up. For the continuous variables, a positive, negative and 0 value implies an increased, decreased and no change in use. ^c Specifically, the antipsychotic dose excluding no dose change (0 value) were divided by 25 percentiles into four groups which were denoted as drastic reduction, moderate reduction, moderate increase, and drastic increase. ^d Being (or becoming) aware of antipsychotic use refers to the patients who are aware of them using antipsychotic at baseline and 3-year follow-up (or being unaware at the baseline and be). ^e For explanation: a higher value of PRS_SCZ (polygenic risk score for schizophrenia) indicates a stronger genetic predisposition of developing SSD (schizophrenia spectrum disorder); PAS (premorbid adjustment score) has a 7-point scale for each item in different age periods (0 healthiest adjustment, 1 lowest adjustment) and PAS-overall is an average score retrospectively measured for childhood (before age 12), early adolescence (12–16) and late adolescence1(6–19) prior to the disease onset where a higher PAS-overall reflects a worse premorbid social functioning; CTQ (Childhood Trauma Questionnaire, Dutch Version) has a 5-point scale for each item (1 never true, 5 very often true) in each subscale (i.e., emotional abuse, physical abuse, sexual abuse, emotional neglect and physical neglect) and CTQ-total is an average score of all subscales where a higher CTQ-total reflect a severer childhood adversity; a higher IQ indicates a better cognitive ability; a higher value of positive symptoms or core negative symptoms from PANSS (positive and negative syndrome scale) indicates severer symptoms; depressive symptoms (frequency and distress level) from CAPE (community assessment of psychic experience) have a 4-point scale (1 never, 4 nearly always; 1 not distressed, 4 very distressed); a higher value of GAF (global assessment of functioning)-disability or -symptoms indicates severer disability or symptoms; a higher number of met needs suggests a better-fulfilled individual needs; four domains of WHOQOL-BREF (the abbreviated version of the World Health Organization Quality of Life) has a 5-point scale for each domain item, and each domain score ranges from 1 (very dissatisfied) to 5 (very satisfied).

Data analysis and statistical modeling

Missing values and outliers. We inspected outliers and missingness of potential variables of use. Single-tone outliers that lie beyond 3 standard deviations (SD) from the variable means were identified and set to missing. All raw variables (Table-1A) have a missingness lower than 60% and were tested highly likely to be missing completely at random [35]. Expectation-maximization (EM) imputed the missingness and non-parametric paired tests between the complete and observed data inspected the potentially-introduced bias [36]. No median difference was found, indicating the validity of the complete dataset for further analyses.

Statistical modelling. To construct outcome mSI, all subscales of SFS and WHOQOL-BREF were standardized and used to cluster patients with K-means clustering analysis. The assumption of K, symbolizing the number of subgroups, was premised from statistical indexes (i.e., silhouette, duda, pseudot2, Hartigan and gap indexes with the Euclidean distance) [37] and supplemented with clinical knowledge. Subgroup (between clusters) difference was indicated by Kruskal-Wallis tests with \(p\) < 0.05 as statistically significant and two-group comparison was examined by Dunn’s Kruskal-Wallis Multiple Comparison tests and Bonferroni adjustment.

To build prediction models, in the MLR approach, potential predictors were selected by reaching a predictor \(p\)-value \(\le\) 0.2 in the univariate multinomial mixed-effect model with a random effect of four medical centers (only indexed variable included as predictor). The multivariate model used MLR with medical centers as one of the fixed covariates. The subgroup with the best mSI level was chosen as the reference group because of our particular interest in the groups with relatively worse mSI levels under the probable mimicry of the subgroup with the best mSI to healthy controls. The predictors were evaluated and chosen by backward selection and variable exclusion was assessed jointly by overall model performance statistics including Akaike Information Criterion (AIC), log-likelihood, and the analysis of variance (ANOVA test) and clinical knowledge. The absence of multicollinearity was investigated (correlation coefficient \(\ge\) 0.8) prior to the model finalization. Bootstrapping with 1,000 replications was used for internal validation. On top of the effect of predictors by the odds ratio (OR) and corresponding confidence interval (95%CI), we presented model performance metrics such as accuracy (i.e., 1-misclassification rate) and their 95%CI derived from bootstrapping. Additionally, considering the imbalanced outcome, \(p\)-value of the one-sided binomial test was reported to examine if the model accuracy is significantly better than no information rate (NIR), which suggests if the model can allocate a patient into the right outcome group significantly better than classification by chance.

In the RF approach, the dataset was randomly split into training and testing sets by an 80/20% ratio. Synthetic minority oversampling technique (SMOTE) was applied to balance the outcome classes. Then, a RF algorithm enhanced by inner repeated cross-validation (i.e., the process of training data split into 3 subsets for training and validating repeats 10 times) was developed. Feature selection in the initial runs was guided by feature importance indices (i.e., derived from a variable-specific out-of-bag decrease in accuracy averaged over all trees after permutation), where a higher value indicates higher usefulness of variable in prediction. Afterwards, variables were visualized in a descending order. To avoid excluding informative variables, backward selection was used. Subsequently, hyperparameter was tuned via grid search. The final RF model was next validated on the testing set. We reported the variable importance of the identified predictors. We also reported the model performance metric such as accuracy in both training and testing and the \(p\)-value of the one-sided binomial test.

Model comparison. MLR and RF models were compared by accuracy and mSI-cluster discriminability. We conducted simulations with 1,000 repeats using a random draw of 30%, 50%, 70%, 80% and 90% of the total sample. Furthermore, the individual-level prediction accuracy and mSI-cluster discriminability of the two models were also examined by scatterplots and confusion matrix.

The data analyses were conducted using R version 1.4.1103. Technical details are explained in Supplementary Methods.

Patient characteristics

In total, 1,119 patients were included. Patients were young adults with a median age of 26.00 years (IQR 22.00, 32.00). They were mostly male (N = 852, 76.14%) and mostly Caucasian (specifically, patients with Dutch and Belgian origin, N = 863, 77.12%). 10.81% of the patients had a diagnosis of schizoaffective disorder at baseline. Patients had an average IQ of 94.00 [83.50, 105.00]. They reported occasional depressive symptoms of 1.00 [0.62, 1.29] at a rather distressed level of 1.40 [1.00, 1.82] as measured with the CAPE. Patients were on average aged 21.00 [18.00, 26.00] years when experiencing first onset of psychosis, and until baseline, they on average had lived with psychosis for 3.00 [1.00, 6.00] years, had been using antipsychotics for 2.00 [0.00, 12.00] years and frequently were aware or became aware of antipsychotic use (N = 1,003, 89.64%). Table 1-A presents the characteristics of patients.

Clusters of multidimensional social inclusion

Figure 1 demonstrates the centroid features of five subgroups identified: 1) “very low (social functioning)/very low (QoL)” cluster (VLL, 8.58% of patients); 2) “low/low” cluster (LL, 12.87%); 3) “high/low” cluster (HL, 49.24%); 4) “medium/high” cluster (MH, 18.05%); and 5) “high/high” cluster (HH, 11.26%). Significant differences in 13 subscales among five groups were observed (Table 2). As VLL and LL do not demonstrate significant difference in independent living, interpersonal and withdrawal levels, to gain statistical power, we merged VLL and LL due to their similarities in low social functioning, ending up with four multinomial subgroups: “low/low” (LL, 21.45%), “high/low” (HL, 49.24%), “medium/high” (MH, 18.05%), and “high/high” (HH, 11.26%).

Table 2

Subgroup differences in multidimensional social inclusion by 13 subscales
SFS							WHOQOL-BREF
Subscale (7x)	mSI cluster						Subscale (6x)	mSI cluster
Independence performance		VLL	LL	HL	MH	HH	QoL		VLL	LL	HL	MH	HH
	VLL			3*	3*	3*		VLL		3*	3*	3*	3*
	LL			3*	3*	3*		LL				3*	3*
	HL				3*	2*		HL				3*	3*
	MH					3*		MH					3*
	HH							HH
Independence competence		VLL	LL	HL	MH	HH	Health		VLL	LL	HL	MH	HH
	VLL			3*	3*	3*		VLL		3*	2*	3*	3*
	LL			3*	3*	3*		LL			3*	3*	3*
	HL					3*		HL				3*	3*
	MH					3*		MH					3*
	HH							HH
Occupation		VLL	LL	HL	MH	HH	Environment domain		VLL	LL	HL	MH	HH
	VLL		3*	3*	3*	3*		VLL		2*		3*	3*
	LL					3*		LL				3*	3*
	HL					3*		HL				3*	3*
	MH					3*		MH					3*
	HH							HH
Withdrawal		VLL	LL	HL	MH	HH	Physical domain		VLL	LL	HL	MH	HH
	VLL			3*	3*	3*		VLL		3*	2*	3*	3*
	LL			3*	3*	3*		LL				3*	3*
	HL					3*		HL				3*	3*
	MH					3*		MH					3*
	HH							HH
Interperson		VLL	LL	HL	MH	HH	Psychological domain		VLL	LL	HL	MH	HH
	VLL			3*	3*	3*		VLL		3*	3*	3*	3*
	LL			3*	3*	3*		LL				3*	3*
	HL				3*	3*		HL				3*	3*
	MH					3*		MH					3*
	HH							HH
Recreation		VLL	LL	HL	MH	HH	Social domain		VLL	LL	HL	MH	HH
	VLL		1*	3*	3*	3*		VLL		3*	3*	3*	3*
	LL			3*	3*	3*		LL				3*	3*
	HL				3			HL				3*	3*
	MH					3		MH					3*
	HH							HH
Prosociality		VLL	LL	HL	MH	HH
	VLL		1*	3*	3*	3*
	LL			3*	3*	3*
	HL				3*	1*
	MH					3*
	HH
Note: Significance level of Dunn’s multiple comparison test is represented by \(1\text{}\text{ = p < 0.05; 2 = p < 0.01; 3* = p < 0.001}\). Following is a short description of each subscale for better understanding the difference among clusters: 1) Independence performance refers to the performance of using a set of skills for independent living; 2) Independence competence refers to the capability of using a set of skills for independent living; 3) Occupation refers to the engagement in the labor market or structured daily activities; 4) Withdraw refers to how socially engaged a person is such as time spent alone and initiation of conversations; 5) Interperson refers to interpersonal behaviors such as number of friends and communication quality; 6) Recreation refers to the engagement of leisure activities such as hobbies and interests; 7) Prosociality refers to the engagement of social activities; 8) QoL refers to perceived quality of life; 9) Health refers to perceived overall health, 10) Environmental domain includes facets such as freedom and safety, home environment, financial resources, accessibility and quality of health and social care, opportunities for acquiring new information and skills, transport and so forth; 11) Physical domain includes facets such as discomfort, rest, fatigue, mobility, work capacity, medication dependence and so forth; 12) Psychological domain includes facets such as positive and negative feelings, self-esteem, bodily image, personal beliefs and so forth; 13) Social domain includes facets such as personal relationships, social support and sexual activity. Abbreviations: VLL: the “very low/very low” mSI subgroup characterized by the lowest levels of social functioning and quality of life while the quality of life is even worse; LL: the “low/low” mSI subgroup featured by low levels of social functioning and quality of life but moderately better quality of life; HL: the “high/low” mSI subgroup with a high social functioning but low quality of life; MH: the “medium/high” mSI subgroup with a middle-level social functioning but a relatively high level of quality of life; HH: the “high/high” mSI subgroup featured by the highest level of both social functioning and quality of life.

Multinomial logistic regression model

There were 22 factors included in the MLR model after selection. To highlight, Table 3-B displays that one additional unit of PAS increases the 66% and 43% likelihoods of people respectively with LL (1.66 [1.22, 2.26]) and HL (1.43 [1.07, 1.92]) relative to HH. One additional unit of PRS_SCZ has 1.05 and 1.10 times of decreased chance respectively of people in favor of HL (0.95 [0.91, 0.99]) and MH (0.93 [0.88, 0.98]) relative to HH. Table 3-C exhibits that one count more presence of positive symptoms has 1.10 times decreased chance of people with MH relative to HH (0.91 [0.86, 0.96]), while one count more presence of core negative symptoms increases the 8.00% and 7.00% likelihood correspondingly of people being favor of LL (1.08 [1.02,1.15]) and HL (1.07 [1.01, 1.13]) relative to HH. One additional unit in frequency sub-score of depressive symptoms deceases 48% of the likelihood of people in favor of MH relative to HH (0.58 [0.37, 0.91]). During a period of 3 years from the baseline, compared to no remission, people once in short-time remission have threefold odds of being in MH relative to HH (3.19 [1.90, 5.36]), and meanwhile, people who ever experienced longer-time remission have 2.7 times decreased chance of being in HL relative to HH (0.37 [0.26, 0.54]). Table 3-D demonstrates that one more count of met needs decreases the likelihood of people in MH relative to HH by 14% (0.86 [0.78, 0.95]), while no significant effect was disclosed in other mSI subgroups. Contrarily, one unit increase in the score of WHOQOL-BREF environment domain has 3 times the odds of people in favor of MH relative to HH (3.09 [1.79, 5.35]). One unit increase in the score of WHOQOL-BREF social domain has 1.55 times the odds of people in favor of MH relative to HH (1.55 [1.06,2.27]). Overall, the MLR model accuracy is 59.16% (95%CI: [55.75%, 62.58%], \(p\) = 0.994) (Table 3-F).

Table 3

Effects (odds ratios) of predictors estimated by the MLR model
Predictor	Outcome				Predictor	Outcome
	LL	HL	MH	HH (ref)		LL	HL	MH	HH (ref)
	N = 240	N = 551	N = 202	N = 126		N = 240	N = 551	N = 202	N = 126
	OR [95%CI]	OR [95%CI]	OR [95%CI]			OR [95%CI]	OR [95%CI]	OR [95%CI]
A. Socio-demography					D. Work, social and overall environment
Medical center					Current urbanicity
Amsterdam (ref)				1.00	No to little (ref)				1.00
Groningen	1.47 [0.73, 2.95]	1.72 [0.90, 3.32]	1.56 [0.73, 3.34]		Moderate	0.79 [0.54, 1.17]	1.09 [0.76, 1.57]	1.00 [0.65, 1.54]
Maastricht	0.83 [0.41, 1.64]	1.49 [0.80, 2.77]	1.05 [0.50, 2.18]		Strong to very strong	1.71 [1.07, 2.76]*	1.09 [0.72, 1.65]	1.25 [0.74, 2.09]
Utrecht	0.85 [0.43, 1.68]	1.15 [0.61, 2.17]	1.06 [0.48, 2.35]		Work
Age	1.01 [0.98, 1.05]	1.02 [0.98, 1.05]	0.98 [0.94, 1.02]		Not working or unknown (ref)				1.00
Gender					Full time or part time	1.55 [0.56, 4.35]	3.00 [1.13, 7.99]*	1.34 [0.40, 4.55]
Male (ref)				1.00	Work payment
Female	0.53 [0.29, 0.97]**	0.74 [0.44, 1.25]	1.00 [0.54, 1.85]		None (ref)				1.00
Ethnicity				1.00	Paid	0.63 [0.20, 1.97]	0.33 [0.11, 0.98]*	1.35 [0.36, 5.10]
Caucasian					Voluntary	0.61 [0.27, 1.39]	0.40 [0.19, 0.88]*	0.98 [0.38, 2.54]
Non-Caucasian	1.56 [0.80, 3.06]	2.22 [1.20, 4.12]*	1.32 [0.61, 2.86]		Mixed	0.40 [0.12, 1.36]	0.24 [0.08, 0.77]*	0.57 [0.13, 2.47]
B. Gene and early-life					Unknown	0.55 [0.11, 2.65]	0.41 [0.10, 1.77]	2.19 [0.38, 12.57]
PAS-overall	1.66 [1.22, 2.26]**	1.43 [1.07, 1.92]*	1.31 [0.92, 1.85]		Number of met needs	1.02 [0.93, 1.10]	0.94 [0.87, 1.02]	0.86 [0.78, 0.95]**
PRS_SCZ	0.98 [0.93, 1.03]	0.95 [0.91, 0.99]**	0.93 [0.88, 0.98]**		WHOQOL-BREF-environment	1.03 [0.65, 1.65]	1.35 [0.87, 2.09]	3.09 [1.79, 5.35]**
C. Disease profile					WHOQOL-BREF-social	1.12 [0.80, 1.57]	1.32 [0.97, 1.80]	1.55 [1.06, 2.27]*
Diagnosis					E. Medication use
Non-affective (ref)				1.00	Duration of using antipsychotics	1.01 [1.00, 1.02]*	1.01 [1.00, 1.02]	1.00 [0.99, 1.01]
Affective	0.80 [0.40, 1.58]	0.62 [0.33, 1.14]	1.26 [0.62, 2.54]		Antipsychotic dose (in Chlorpromazine equivalents)
Duration of psychosis	0.95 [0.88, 1.03]	0.94 [0.87, 1.00]	0.97 [0.89, 1.06]		Drastic reduction (ref)				1.00
Positive symptoms	0.98 [0.94, 1.02]	0.98 [0.94, 1.02]	0.91 [0.86, 0.96]**		Moderate reduction	1.47 [0.75, 2.88]	1.87 [1.03, 3.41]*	2.21 [1.08, 4.50]*
Core negative symptoms	1.08 [1.02,1.15]**	1.07 [1.01, 1.13]*	0.97 [0.91, 1.05]		No change	1.34 [0.61, 2.97]	0.49 [0.23, 1.05]	1.58 [0.68, 3.66]
Depressive symptom (frequency)	1.14 [0.71, 1.83]	0.58 [0.37, 0.91]**	0.65 [0.37, 1.13]		Moderate increase	1.47 [0.67, 3.21]	2.57 [1.27, 5.17]**	2.29 [0.99, 5.30]
GAF-symptom	0.99 [0.97, 1.01]	0.98 [0.96, 1.00]**	0.96 [0.94, 0.99]**		Drastic increase	1.42 [0.71, 2.84]	1.81 [0.96, 3.41]	0.98 [0.43, 2.23]
Remission (baseline to Year 3)					Awareness of antipsychotic use
No (ref)				1.00	No (ref)				1.00
Yes (≤ 6 months)	0.68 [0.44, 1.06]	1.11 [0.73, 1.68]	3.19 [1.91, 5.36]**		(Become) aware	0.69 [0.33, 1.45]	2.89 [1.38, 6.06]**	0.65 [0.30, 1.41]
Yes (> 6 months)	1.04 [0.67, 1.61]	0.37 [0.26, 0.54]**	0.66 [0.42, 1.05]		Constant	0.31 [0.02, 5.57]	0.19 [0.01, 2.79]	0.23 [0.01, 6.53]
F. Model Performance
AIC	2336.84 [2232.61, 2441.08]				Accuracy	59.16% [55.75%, 62.58%]
Log Likelihood	-1069.42 [-1121.54, -1017.30]				\(p\)-value [Accuracy > NIR]	0.994
McFadden pseudo R²	0.23 [0.19, 0.26]				Kappa	0.34 [0.28, 0.40]
Adjusted McFadden pseudo R²	0.21 [0.18, 0.24]
Note: The significance level is represented by \(\text{}\text{ = p < 0.05; * = p < 0.01; *** = p < 0.001}\). 95%CI of model performance were derived from bootstrapping. Abbreviations: LL: mSI subgroup with both low social functioning and quality of life; HL: mSI subgroup with high social functioning and low quality of life; MH: mSI subgroup with medium social functioning and high quality of life; HH: mSI subgroup with both high social functioning and quality of life; PAS: premorbid adjustment score; PRS_SCZ: polygenic risk score for schizophrenia; GAF: global assessment of functioning; WHOQOL-BREF: the abbreviated version of World Health Organization Quality of Life; NIR: no information rate.

Random forest model

Totally, 22 factors were identified in the RF model. Among all, WHOQOL-BREF domain scores and CTQ-total contribute the most to predicting the mSI group (Fig. 2). Observing the feature importance indices, other important factors are age, PRS_SCZ, PAS-overall, symptom severity (i.e., positive, core negative and depressive) and the number of met needs. The accuracy on training and testing sets, respectively, is 70.46% ± 2.03% and 61.61% (95%CI: [54.90%, 68.01%], \(p\) = 0.013).

Model comparison

For the simulation of 30% (up to 90%) of the observed patients, the mean accuracy of the MLR model was 59.12% ± 2.50% (59.13% ± 1.20%) and the mean of the RF model was 92.29% ±1.34% (92.26% ± 0.67%) (Supplementary Table 1). In the comparison of observed and predicted mSI subgroups among 1,119 patients, 662 patients (59.16%) and 1,033 patients (92.31%) were correctly predicted by the MLR and RF models correspondingly. We saw a similar pattern and percentage of mSI subgroup partition between the observations and RF-predictions, and contrarily, differences occurred between the observations and MLR-predictions (Supplementary Fig. 2-A). For example, a distinctly higher HL percentage with a (62.73% vs 49.24% observed, 48.88% RF-predicted, Supplementary Fig. 2-B).

In this study, we aimed to study the clusters of multidimensional social inclusion (mSI) conceptualized by social functioning and QoL among the patients diagnosed with SSD; and to test mSI prediction using MLR and RF models. Through K-means clustering analyses, five mSI subgroups were identified, including VLL, LL, HL, MH and HH. The MLR and RF models consistently regard genetic predisposition for schizophrenia, premorbid social functioning, symptoms, baseline environment and social domains and numbers of met needs as paramount. Comparatively, the RF model is cautiously considered better because of its capability of discriminating all mSI subgroups. The mSI clusters intend to preliminarily define and bring awareness to social inclusion, an understudied but crucial outcome. The mSI prediction models have clinical and policy implications.

Clusters of multidimensional social inclusion among patients with schizophrenia spectrum disorder

We identified five mSI subgroups among the patients with SSD. The “very low (social functioning)/very low (QoL)” (VLL) and “low/low” (LL) both indicated a low mSI level. VLL was featured by the lowest social functioning and QoL in the spectrum with an even worse QoL, while LL was characterized by relatively low levels of social functioning and QoL with a slightly elevated QoL. These two clusters were merged for the analysis; however, disparities exist when the patients in VLL compared to those in LL might have restricted access to the labor market, recreational activities, and social engagement with simultaneously affected QoL and mSI. Clinicians are possibly not concerned with VLL and LL when advising interventions for similarly low social functioning among VLL and LL. As clinical interventions would probably be a priority, the two clusters could be regarded as one.

The “high/low” (HL) and “medium/high” (MH) clusters implied a medium mSI level. Compared to LL, HL had a better level of social functioning, particularly in the subscales of independence performance, interperson, recreation and prosociality, but shared a nearly similar level of QoL (i.e., moderately lower than the average QoL of all patients with SSD), pointedly with a distinctively lower satisfaction towards general health, environment, and physical conditions. Differently, MH showed an average social functioning level and a mildly higher QoL than the average QoL of all patients with SSD. It is noteworthy that about 50% of patients with SSD are in HL. The feature disparity across LL, HL and MH might point out that the patients in HL had better performance in independent living and (possibly resulted) increasing social exposure through recreational and social activities. Such an involvement could cause patients to compare themselves to the healthy population. This could shift the perception and stimulate other psychological impacts such as internalized stigma, which consequently may lead to a systematically perceived low QoL [38]. Moreover, a significant difference in occupation was absent. A likely elucidation for these patients ending up in different mSI levels is that patients in LL might join a simple task from a program or struggle to continue at work due to low functioning and a lack of support from workplaces [39, 40]. Comparatively, patients in HL and MH who function better are likely to have and stay at a job. However, with increasing functioning, patients in HL might be not satisfied with work suitability, financial earnings, skills required and meaning [40, 41], which results in a lower QoL and mSI.

Eventually, the “high/high” (HH) showed a high mSI level with the highest levels of social functioning and QoL, suggesting that this patient subgroup possibly mimics healthy controls, albeit on an overall lower level in comparison to healthy controls presumably due to their psychiatric conditions.

The characteristics displayed in the mSI subgroups implicated that good functioning or satisfaction in one aspect does not necessarily signify a better perceived mSI. For instance, patients doing equivalently well in occupation do not equate to showing similarly good mSI. Therefore, mSI is essential, as a holistic approach, to provide a comprehensive overview of social inclusion of an individual. The subgroup characteristics could also guide the intervention focus. The patients in VLL and LL may require more psychosocial interventions to manage symptoms and improve social functioning. They should also be prioritized to the eligibility for protected living when independent living is not achievable. The patients in HL and MH could target current programs designed for stigma reduction. The patients in HH could be offered training for advanced skills and opportunities for more challenging working position to further grow self-esteem and self-actualization that are beneficial to mSI enhancement.

Common factors predictive of multidimensional social inclusion

PRS_SCZ, symptoms (i.e., positive, negative, depressive), premorbid social functioning, baseline environment and social domain satisfactions and number of met needs were found to be predictors of mSI.

Congruent with previous studies [18, 42–45], in the MLR model, more severe core negative symptoms increase the risk of having low-to-medium mSI (i.e., LL and HL relative to HH). Surprisingly, worse positive symptoms significantly predict good mSI (i.e., MH relative to HH) with a mild protective effect. Other studies have found that positive symptoms do not contribute much to QoL or social cognition [46–48], although the cross-sectional symptomatic remission [49] can significantly improve social functioning [50]. Our result might be the consequence of heterogeneous effects from various positive symptoms at baseline on a prospective mSI [42]. A higher genetic vulnerability towards SSD displayed a significant protective effect on good mSI (HL and MH relative to HH). This is possibly due to the single comparison in a relatively low sample size of HH in a multivariate model, which may occupy the variability of mSI concerning PRS_SCZ, and consequently yield a dubious finding. Therefore, the relationship between the genetic predisposition and mSI should be independently investigated in well-powered research. Counterintuitively, we found that the more often an SSD patient has depressive symptoms, the more likely the patient is to be in high mSI (i.e., HH) than in medium mSI (i.e., HL) in the MLR model, which is discordant with previous studies [51–53]. One possible explanation is that in the diagnostic categories of SSD, patients with a higher level of depressive symptoms are more likely to have affective symptoms and to be associated with affective dysregulation, which results in a better outcome (i.e., mSI in our case) than the one of the patients with non-affective symptoms such as withdrawal in HL [54]. Specifically, the early detection and interventions of depressive symptoms could be essential to help patients with SSD, which might further impact on their lives and subsequent mSI. All the aforesaid factors were confirmed informative in the RF model as well.

Compared to the previously mentioned factors, premorbid social functioning has the highest negative effect on low-to-medium mSI in the MLR model (i.e., LL and HL relative to HH). Thus, although ranked after symptoms and next to genetic predisposition in the RF model, premorbid social functioning shows moderate predictability of the 3-year mSI. Earlier studies have shown that worse premorbid social functioning, resulting from environmental (socio-economic status and cultural specificity) and genetic factors [55, 56], may lead to severer negative symptoms and poorer social outcomes later in the course of SSD [57, 58]. Therefore, the premorbid social functioning is undoubtfully vital [18], and can be potentially used for screening of low mSI. Yet, it has not caught enough attention in the field given the limited literature.

As expected, the baseline environment and social domain satisfactions are consistently crucial. Abundant studies have emphasized the importance of occupation (and thus financial income and opportunities for acquiring new skills) and social relationships [2, 59], a safe stable housing, family support [60] and inclusive and accessible support systems across sectors such as transportation [60, 61] for social inclusion. On the other hand, a higher level of fulfilled needs significantly distinguishes MH and HH only in the MLR model. It is conceivable that increased met needs might indicate better social functioning and thus improve mSI among the patients with medium-to-high mSI. In other mSI clusters where needs were met equivalently to the HH subgroup, the heterogeneity within and between subgroups (i.e., LL, HL and HH) and subjective criteria and expectations on met needs could attenuate the significant effects. Particularly in the RF model, the number of met needs is regarded important. Therefore, with a growing emphasis on extramural care [62], the local communities and mental health organizations need to incorporate the heterogeneous environmental and social needs, beyond the medical needs, of patients with SSD at different mSI levels. More generally, given the possibility of a blend of a healthy population and a population with varying physical and mental conditions along severity scales, the mixed demands of such a population should be well considered. Stakeholders with different responsibilities should unitedly, collaboratively and actively participate in the course and co-design the community with local residents. Thus, the right infrastructures and additional human touches might boost perceived satisfaction and elevate mSI.

Model performance in prediction

By the goal of prediction, the predictivities of MLR and RF models are fair and comparable. We infer the RF model outperforms as it supposedly allocates people into the right mSI subgroup significantly better than solely by chance. It means that the MLR model should be applied with caution as some predictors in the MLR model are indiscriminative to mSI subgroups except for the HL cluster due to the imbalanced mSI outcome. Therefore, cautiously, the RF model outperforms regarding discriminability. The RF model successfully passing the generalizability test can be readily implemented in clinical practices through the electronic health record system (see Supplementary Clinical Illustration) [63]. Such prediction could serve as an indicator on how much external assistance an SSD patient may require in the upcoming future to be prevented from and/or helped get through deteriorating mSI. Furthermore, exemplified by our study (i.e., different approaches with different modelling procedures), when constructing prediction models, one should seek candidate strategies in a wide range of methodologies, try and compare different combinations, and choose the most appropriate approach to achieve an optimized balance among predictivity, discriminability, interpretability and applicability of a model, depending on the research goal and population at stake.

Future perspectives

Future studies should continue working on the conceptualization of mSI and study the generalizability of mSI across diagnoses. Following this study, the clinical differences of the identified mSI clusters and comparability to healthy controls should be investigated to derive more clinical meanings and understanding of the mSI subgroups in the practices. Meanwhile, developing a validated composite score might bring more implications due to a higher level of data granularity and a resultant possibility of monitoring over time. Methodologically, future studies should take extra steps in modeling procedures such as outer cross-validation and different feature-selection algorithms and give opportunities to the latest interpretable machine learning algorithms for exertion to pick up next-level clinical utility. Clinically, to build prognostic models, future studies should test the utilities of the potential factors that measure the same clinical outcomes but with slight differences in submodels and to understand the multidimensional mechanisms hidden under the effect sizes of such as premorbid social functioning. Furthermore, the investigations on the genetic effects on behaviors and mSI along the course of SSD could be necessary for early screening and interventions. Statistically, we underscored the importance of satisfaction towards environment and social life. This can possibly be modified by other observable and non-observable contributive factors, such as personality traits, coping strategies, diversity of community residents, community social-economic status, relationship with caregivers, and so forth. Therefore, future studies might consider extending coverage of the data collection from clinical practices in a balance of feasibility and costs. With growing awareness of social inclusion and more efforts in building increasingly sophisticated prediction models on mSI, and in the implementation of personalized interventions and supporting policies, patients with SSD would be able to harness the necessary skills and receive indispensable resources, which consequently assist them in managing their conditions better and be better included in society, eventually.

Study strengths and limitations

We quantified the multidimensional nature of social inclusion by combining the subscales of the self-reported SFS and WHOQOL-BREF. Considering the situation that the questionnaires previously developed regarding social inclusion and its similar terms for other target populations are not well validated, our conceptualization emphasizes the multidimensionality that provides a more comprehensive overview of an individual’s status of social inclusion through a broad range of activities, the perception of an individual, and the exploitation of the existing large cohort and standardly collected data, which can be of convenience for the use beyond clinics. Furthermore, the longitudinal measurements in the cohort were utilized. We also built and compared standard and data-driven models to examine the robustness and enhance the credibility of the factors and the mSI predictability. However, the models were not externally validated.

On the other hand, despite earlier efforts in developing conceptual frameworks of social inclusion, one cannot ignore that the constitution of social inclusion might be meant to be nebulous, implying a variable boundary of this multifaceted construct. The readily accessible data do not provide many flexibilities or balance the eligibility and validity of the elements used for the construct. No interview was conducted to preliminary determine the most applicable and relevant scope of social inclusion. Regardless of the limitations, our attempt to pick up the measurement development of multidimensional social inclusion is vitally implicative to clinical practices and societies.

In our study, we conceptualized mSI from social functioning and quality of life, resulting to five identified subgroups (i.e., “very low/very low”, “low/low”, “high/low”, “medium/high”, and “high/high”). We found that genetic predisposition, premorbid social functioning, symptoms (i.e., positive, negative and depressive), number of met needs and baseline satisfaction with the environment and social life are robust factors predictive of mSI among the patients with SSD. We cautiously conclude that the RF model offers a better prediction, compared to the MLR model, of the 3-year mSI among patients with SSD in the Netherlands due to its better discriminability. Yet, external validation is still required.

Our findings indicate that mSI is applicable and offers possibilities of personalized treatment strategies and policymaking tailored for patients with SSD at different mSI levels. Our study sheds light on the special proposition of mSI that should be an imperative goal for patients with SSD and a solution to expensive healthcare and harmony for society.

COMPETING INTERESTS

Author Jiasi Hao, Natalia Tiles-Sar, Edith J. Liemburg, Tesfa Dejenie Habtewold, GROUP Investigators, Richard Bruggeman, Lisette van der Meer and Behrooz Z. Alizadeh declare that they have no conflict of interest.

FUNDING

This work was supported by Geestkracht programme of the Dutch Health Research Council (Zon-Mw) (10-000-1001), and matching funds from participating pharmaceutical companies (Lundbeck; AstraZeneca; Eli Lilly and Janssen Cilag) and Universities and Mental Health Care organizations (Amsterdam: Academic Psychiatric Centre of the Academic Medical Center and the mental health institutions: GGZ Ingeest; Arkin, Dijk en Duin; GGZ Rivierduinen; Erasmus Medical Centre and GGZ Noord Holland Noord. Groningen: University Medical Center Groningen and the mental health institutions: Lentis, GGZ Friesland; GGZ Drenthe; Dimence; Mediant; GGNet Warnsveld; Yulius Dordrecht and Parnassia psycho-medical center The Hague. Maastricht: Maastricht University Medical Centre and the mental health institutions: GGZ Eindhoven en De Kempen; GGZ Breburg; GGZ Oost-Brabant; Vincent van Gogh voor Geestelijke Gezondheid; Mondriaan; Virenze riagg; Zuyderland GGZ; MET ggz; Universitair Centrum Sint-Jozef Kortenberg; CAPRI University of Antwerp; PC Ziekeren Sint-Truiden; PZ Sancta Maria Sint-Truiden; GGZ Overpelt and OPZ Rekem. Utrecht: University Medical Center Utrecht and the mental health institutions: Altrecht; GGZ Centraal and Delta). The sponsors have no role in designing the study, in the collection, analysis, and interpretation of data, in the writing of the report and in the decision to submit the paper for publication.

ACKNOWLEDGEMENTS

We sincerely thank the GROUP recruited participants for the time spared, efforts made and generosity of sharing data for research. Furthermore, we would like to thank all research personnel involved in the GROUP project, in particular: Joyce van Baaren, Erwin Veermans, Ger Driessen, Truda Driesen, Erna van ’t Hag and Jessica de Nijs.

AUTHOR CONTRIBUTIONS

Jiasi Hao. and Behrooz Z. Alizadeh formulated the research questions and designed analysis plan. Richard Bruggeman contributed to the formulation of the research questions. Jiasi Hao conducted data analysis and checked for reproducibility. Jiasi Hao drafted and revised the manuscript. Natalia Tiles-Sar, Edith J. Liemburg, Tesfa Dejenie Habtewold, GROUP Investigators, Richard Bruggeman, Lisette van der Meer and Behrooz Z. Alizadeh assisted with results interpretation and provided constructive comments on the manuscript. All authors read and approved the final version.

ETHICS APPROVAL

The GROUP protocol was approved by the Medical Ethics Review Committee (Medisch Ethische Toetsingscommissie [METC]) of the University Medical Center Utrecht. The protocol was conducted in accordance with the Helsinki Declaration that was revised in 2008. Thereof, no further ethical approval is required for this study.

CONSENT TO PARTICIPATE

All GROUP participants have given written informed consent.

Charlson F, Van Ommeren M, Flaxman A, Cornett J, Whiteford H, Saxena S. New WHO prevalence estimates of mental disorders in conflict settings: a systematic review and meta-analysis. The Lancet. 2019;394(10194):240-8.
Killaspy H, White S, Lalvani N, Berg R, Thachil A, Kallumpuram S, et al. The impact of psychosis on social inclusion and associated factors. International Journal of Social Psychiatry. 2014;60(2):148 − 54.
Morgan C, Burns T, Fitzpatrick R, Pinfold V, Priebe S. Social exclusion and mental health: conceptual and methodological review. British Journal of Psychiatry. 2007;191:477 − 83.
Kovács G, Almási T, Millier A, Toumi M, Horváth M, Kóczián K, et al. Direct healthcare cost of schizophrenia – European overview. European Psychiatry. 2018;48(1):79–92.
Schizofrenie Kosten Zorguitgaven: Volksgezondheidenzorg.info; 2017 [Available from: https://www.volksgezondheidenzorg.info/onderwerp/schizofrenie/kosten/zorguitgaven.
Charlson FJ, Ferrari AJ, Santomauro DF, Diminic S, Stockings E, Scott JG, et al. Global Epidemiology and Burden of Schizophrenia: Findings From the Global Burden of Disease Study 2016. Schizophrenia Bulletin. 2018;44(6):1195 − 203.
Nhunzvi C, Langhaug L, Mavindidze E, Harding R, Galvaan R. Occupational justice and social inclusion in mental illness and HIV: a scoping review protocol. BMJ Open. 2019;9(3):e024049.
Pereira RB, Whiteford GE. Understanding Social Inclusion as an International Discourse: Implications for Enabling Participation. British Journal of Occupational Therapy. 2013;76(2):112-5.
Cobigo V, Stuart H. Social inclusion and mental health. Current Opinion in Psychiatry. 2010;23(5):453-7.
Simplican SC, Leader G, Kosciulek J, Leahy M. Defining social inclusion of people with intellectual and developmental disabilities: An ecological model of social networks and community participation. Research in Developmental Disabilities. 2015;38:18–29.
Izquierdo A, Cabello M, Leal I, Ayora M, Rodriguez-Jimenez R, Ibáñez Á, et al. How does neighbourhood socio-economic status affect the interrelationships between functioning dimensions in first episode of psychosis? A network analysis approach. Health & Place. 2021;69:102555.
Izquierdo A, Cabello M, Leal I, Mellor-Marsá B, Ayora M, Bravo-Ortiz M-F, et al. The interplay between functioning problems and symptoms in first episode of psychosis: An approach from network analysis. Journal of Psychiatric Research. 2021;136:265 − 73.
Fong P, Cruwys T, Haslam C, Haslam SA. Neighbourhood identification and mental health: How social identification moderates the relationship between socioeconomic disadvantage and health. Journal of Environmental Psychology. 2019;61:101 − 14.
Eaton S, Harrap B, Downey L, Thien K, Bowtell M, Bardell-Williams M, et al. Incidence of treated first episode psychosis from an Australian early intervention service and its association with neighbourhood characteristics. Schizophrenia Research. 2019;209:206 − 11.
Menezes NM, Arenovich T, Zipursky RB. A systematic review of longitudinal outcome studies of first-episode psychosis. Psychological Medicine. 2006;36(10):1349-62.
Leighton SP, Krishnadas R, Chung K, Blair A, Brown S, Clark S, et al. Predicting one-year outcome in first episode psychosis using machine learning. PLOS ONE. 2019;14(3):e0212846.
Leighton SP, Upthegrove R, Krishnadas R, Benros ME, Broome MR, Gkoutos GV, et al. Development and validation of multivariable prediction models of remission, recovery, and quality of life outcomes in people with first episode psychosis: a machine learning approach. The Lancet Digital Health. 2019;1(6):e261-e70.
Tiles-Sar N, Habtewold T, Liemburg E, Småstuen M, GROUP investigators, Alizadeh BZ, et al. Prediction of social functioning in patients with schizophrenia: three- and six-years follow-up in Dutch cohort. 2022.
Cordier R, Milbourn B, Martin R, Buchanan A, Chung D, Speyer R. A systematic review evaluating the psychometric properties of measures of social inclusion. PLOS ONE. 2017;12(6):e0179109.
De Nijs J, Burger TJ, Janssen RJ, Kia SM, Van Opstal DPJ, De Koning MB, et al. Individualized prediction of three- and six-year outcomes of psychosis in a longitudinal multicenter study: a machine learning approach. npj Schizophrenia. 2021;7(1).
Lynam AL, Dennis JM, Owen KR, Oram RA, Jones AG, Shields BM, et al. Logistic regression has similar performance to optimised machine learning algorithms in a clinical setting: application to the discrimination between type 1 and type 2 diabetes in young adults. Diagnostic and Prognostic Research. 2020;4(1).
Weng SF, Vaz L, Qureshi N, Kai J. Prediction of premature all-cause mortality: A prospective general population cohort study comparing machine-learning and standard epidemiological approaches. PLOS ONE. 2019;14(3):e0214365.
Li J, Zhou Z, Dong J, Fu Y, Li Y, Luan Z, et al. Predicting breast cancer 5-year survival using machine learning: A systematic review. PLOS ONE. 2021;16(4):e0250370.
Christodoulou E, Ma J, Collins GS, Steyerberg EW, Verbakel JY, Van Calster B. A systematic review shows no performance benefit of machine learning over logistic regression for clinical prediction models. Journal of Clinical Epidemiology. 2019;110:12–22.
Korver N, Quee PJ, Boos HB, Simons CJ, de Haan L, GROUP investigators, a multi site longitudinal cohort study focused on gene–environment interaction: objectives, sample characteristics, recruitment and assessment methods. International journal of methods in psychiatric research. 2012;21(3):205 − 21.
Birchwood M, Smith J, Cochrane R, Wetton S, Copestake S. The Social Functioning Scale the Development and Validation of a New Scale of Social Adjustment for use in Family Intervention Programmes with Schizophrenic Patients. British Journal of Psychiatry. 1990;157(6):853-9.
The WHOQOL Group. Development of the World Health Organization WHOQOL-BREF Quality of Life Assessment. Psychological Medicine. 1998;28(3):551-8.
Bruins J, Pijnenborg GHM, Visser E, Castelein S. The association of cannabis use with quality of life and psychosocial functioning in psychosis. Schizophrenia Research. 2021;228:229 − 34.
Cannon-Spoor HE, Potkin SG, Wyatt RJ. Measurement of Premorbid Adjustment in Chronic Schizophrenia. Schizophrenia Bulletin. 1982;8(3):470 − 84.
Arntz A, Wessel I. Jeugd trauma vragenlijst [Dutch version of the childhood trauma questionnaire]. The Netherlands. 1996.
Galderisi S, Mucci A, Dollfus S, Nordentoft M, Falkai P, Kaiser S, et al. EPA guidance on assessment of negative symptoms in schizophrenia. European Psychiatry. 2021;64(1):1–91.
van Os J, Verdoux H, Hanssen M. CAPE (Community Assessment of Psychic Experiences). University of Maastricht, Maastricht, the Netherlands. 1999.
Endicott J. The Global Assessment Scale. Archives of General Psychiatry. 1976;33(6):766.
Phelan M, Slade M, Thornicroft G, Dunn G, Holloway F, Wykes T, et al. The Camberwell Assessment of Need: The Validity and Reliability of an Instrument to Assess the Needs of People with Severe Mental Illness. British Journal of Psychiatry. 1995;167(5):589 − 95.
Little RJA. A Test of Missing Completely at Random for Multivariate Data with Missing Values. Journal of the American Statistical Association. 1988;83(404):1198 − 202.
Dong Y, Peng C-YJ. Principled missing data methods for researchers. SpringerPlus. 2013;2(1):222.
Islam MA, Alizadeh BZ, van den Heuvel ER, Bruggeman R, Cahn W, de Haan L, et al. A comparison of indices for identifying the number of clusters in hierarchical clustering: A study on cognition in schizophrenia patients. Communications in Statistics: Case Studies, Data Analysis and Applications. 2015;1(2):98–113.
Dong M, Lu L, Zhang L, Zhang Y-S, Ng CH, Ungvari GS, et al. Quality of Life in Schizophrenia: A Meta-Analysis of Comparative Studies. Psychiatric Quarterly. 2019;90(3):519 − 32.
Henderson M, Harvey SB, Overland S, Mykletun A, Hotopf M. Work and common psychiatric disorders. J R Soc Med. 2011;104(5):198–207.
Greenberg N, Tarn M, Tracy D. Lessons from the pandemic: why having a good understanding of occupational psychiatry is more important now than ever before. The British Journal of Psychiatry. 2022;221(4):589 − 90.
Lexén A, Hofgren C, Bejerholm U. Reclaiming the worker role: Perceptions of people with mental illness participating in IPS. Scandinavian Journal of Occupational Therapy. 2013;20(1):54–63.
Pazoki R, Lin BD, Van Eijk KR, Schijven D, De Zwarte S, Guloksuz S, et al. Phenome-wide and genome-wide analyses of quality of life in schizophrenia. BJPsych Open. 2021;7(1).
Maat A, Fett A-K, Derks E. Social cognition and quality of life in schizophrenia. Schizophrenia Research. 2012;137(1):212-8.
Brill N, Levine SZ, Reichenberg A, Lubin G, Weiser M, Rabinowitz J. Pathways to functional outcomes in schizophrenia: The role of premorbid functioning, negative symptoms and intelligence. Schizophrenia Research. 2009;110(1–3):40 − 6.
Desalegn D, Girma S, Abdeta T. Quality of life and its association with psychiatric symptoms and socio-demographic characteristics among people with schizophrenia: A hospital-based cross-sectional study. PLOS ONE. 2020;15(2):e0229514.
Galuppi A, Turola M, Nanni M, Mazzoni P, Grassi L. Schizophrenia and quality of life: how important are symptoms and functioning? International Journal of Mental Health Systems. 2010;4(1):31.
Landeen J, Pawlick J, Woodside H, Kirkpatrick H, Byrne C. Hope, quality of life, and symptom severity in individuals with schizophrenia. Psychiatric Rehabilitation Journal. 2000;23:364-9.
Svirskis T, Korkeila J, Heinimaa M, Huttunen J, Ilonen T, Ristkari T, et al. Quality of life and functioning ability in subjects vulnerable to psychosis. Comprehensive Psychiatry. 2007;48(2):155 − 60.
Helldin L, Kane JM, Hjärthag F, Norlander T. The importance of cross-sectional remission in schizophrenia for long-term outcome: A clinical prospective study. Schizophrenia Research. 2009;115(1):67–73.
Brissos S, Dias VV, Balanzá-Martinez V, Carita AI, Figueira ML. Symptomatic remission in schizophrenia patients: Relationship with social functioning, quality of life, and neurocognitive performance. Schizophrenia Research. 2011;129(2):133-6.
Van Rooijen G, Van Rooijen M, Maat A, Vermeulen JM, Meijer CJ, Ruhé HG, et al. Longitudinal evidence for a relation between depressive symptoms and quality of life in schizophrenia using structural equation modeling. Schizophrenia Research. 2019;208:82 − 9.
B. N. S, Grover S. Depression in schizophrenia: Prevalence and its impact on quality of life, disability, and functioning. Asian Journal of Psychiatry. 2020;54:102425.
Reininghaus U, Priebe S, Bentall RP. Testing the Psychopathology of Psychosis: Evidence for a General Psychosis Dimension. Schizophrenia Bulletin. 2012;39(4):884 − 95.
van Os J, Kapur S. Schizophrenia. Lancet. 2009;374(9690):635 − 45.
Klorman R, Strauss JS, Kokes RF. Premorbid adjustment in schizophrenia: III. The relationship of demographic and diagnostic factors to measures of premorbid adjustment in schizophrenia. Schizophrenia Bulletin. 1977;3(2):214 − 25.
Shapiro DI, Marenco S, Spoor EH, Egan MF, Weinberger DR, Gold JM. The Premorbid Adjustment Scale as a measure of developmental compromise in patients with schizophrenia and their healthy siblings. Schizophrenia Research. 2009;112(1):136 − 42.
Jeppesen P, Petersen L, Thorup A, Abel M-B, Øhlenschlæger J, Christensen Ø, T., et al. The association between pre-morbid adjustment, duration of untreated psychosis and outcome in first-episode psychosis. Psychological Medicine. 2008;38(8):1157-66.
Bailer J, Bräuer W, Rey E-R. Premorbid adjustment as predictor of outcome in schizophrenia: results of a prospective study. Acta Psychiatrica Scandinavica. 1996;93(5):368 − 77.
Turner N, Ferguson L, Hill M, Nesbitt T, O’Callaghan E, O’Mahony P, et al. An exploratory study of the extent of social inclusion among people with psychosis and psychotic-related conditions. International Journal of Social Psychiatry. 2017;63(3):195–202.
Filia K, Jackson H, Cotton S, Killackey E. Understanding what it means to be socially included for people with a lived experience of mental illness. International Journal of Social Psychiatry. 2019;65(5):413 − 24.
Hall T, Kakuma R, Palmer L, Minas H, Martins J, Kermode M. Social inclusion and exclusion of people with mental illness in Timor-Leste: a qualitative investigation with multiple stakeholders. BMC Public Health. 2019;19(1).
(TNO) NOfASR. Vision Of Extramural Care in 2030: The Right Care With The Right Logistics: TNO; 2020 [Available from: https://www.tno.nl/nl/aandachtsgebieden/mobiliteit-logistiek/roadmaps/smart-and-safe-traffic-and-transport/smart-mobility-and-logistics/visie-extramurale-zorg-2030/.
Lee TC, Shah NU, Haack A, Baxter SL. Clinical Implementation of Predictive Models Embedded within Electronic Health Record Systems: A Systematic Review. Informatics. 2020;7(3):25.

Download PDF

Journal Publication

published 07 Mar, 2024

Read the published version in Social Psychiatry and Psychiatric Epidemiology →

Version 1

posted

You are reading this latest preprint version

Multidimensional social inclusion and its prediction in schizophrenia spectrum disorder

Status:

Journal Publication

Version 1

Abstract

Figures

Introduction

Methods

Results

Discussion

Conclusion

Declarations

COMPETING INTERESTS

FUNDING

ACKNOWLEDGEMENTS

AUTHOR CONTRIBUTIONS

ETHICS APPROVAL

CONSENT TO PARTICIPATE

References

Supplementary Files

Status:

Journal Publication

Version 1