Cognition and direct costs in schizophrenia spectrum disorders. Preliminary results from a retrospective longitudinal study

Objectives. Schizophrenia spectrum disorders (SSD) accounts for a substantial economic burden on healthcare, with roughly half being direct costs. Hospitalization and use of services in general account for the majority (81%) of direct costs. Given the impact of cognitive impairment in SSD, a clearer understanding of cognition's impact on healthcare expenses for SSD is of clinical and scientific relevance.

Methods. A total of 70 subjects living with SSD were included in this preliminary ad-interim analysis: for each participant, cognitive, clinical, sociodemographic data, as well as information regarding the use of mental health services during the year 2023 was collected and analyzed. Predictors of total direct costs were assessed using multiple linear regression models.

Results. The average direct yearly healthcare expenditure per patient with SSD related to service utilization amounted to €58,932.44. Several variables were associated with increased costs (all p<0.05): poorer Brief Assessment of Cognition in Schizophrenia (BACS) verbal fluency performance, alcohol or substance abuse, higher Clinical Global Impression-Severity (CGI-S) scores, higher total and subscale scores on the Positive and Negative Syndrome Scale, higher Health of the Nation Outcome Scales (HoNOS) score, and lower Global Assessment of Functioning (GAF) scores. Higher CGI-S score (p<0.001, β=-0.581), lower BACS verbal fluency performance (p=0.003 β=−0.349), and the presence of substance use (p=0.014, β=0.293) emerged as individual predictors of increased costs.

Conclusion. The findings of this study highlight the importance of implementing tailored psychosocial interventions in individuals with SSD, with the additional goal of a reduction in overall healthcare expenses.

INTRODUCTION

Schizophrenia spectrum disorders (SSD) are complex mental illnesses that not only have a consistent negative impact on the quality of life and the overall well-being of those affected, but also place a considerable and often underestimated burden on healthcare systems internationally (Maj et al., 2020; Solmi et al., 2023).

Moreover, although the estimated prevalence of SSD is limited when compared to more frequently occurring medical conditions (Institute of health Metrics and Evaluation (IHME)., 2019), the far-reaching impact of these disorders is substantial, affecting both the personal lives of individuals and the financial sustainability of healthcare provision (Altamura et al., 2014). The Global Burden of Diseases study has provided a stark illustration of this impact, estimating that schizophrenia contributes to 0.6% of all Disability Adjusted Life Years (DALYs) worldwide. This seemingly modest figure equates to the staggering loss of approximately 15 million years of productive and healthy life annually, a consequence of premature mortality associated with the illness and years lived with significant disability.

Moreover, the DALYs proportion related to schizophrenia is in constant growth: in 1990, the estimated contribution was 0.35% of global DALYs, translating to roughly 9 million years lost (Institute of health Metrics and Evaluation (IHME)., 1990). In Italy, a 2018 analysis quantified the economic repercussions of schizophrenia at about 2.7 billion € (Marcellusi et al., 2018). Finally, across developed nations, schizophrenia absorbs a proportion ranging from 1.4 % to 3 % of the total healthcare expenditure and, specifically for Italy, 22% of the overall spending dedicated to mental health, reflecting the priority and resources required to manage this complex and costly disorder within the national healthcare landscape (Altamura et al., 2014; Knapp et al., 2004; Marcellusi et al., 2018). On a global level, a rapid progression in the cost attributable to SSD has been observed, and, in the Italian context, the estimated annual cost per patient living with schizophrenia grew from 25.000 € in 2000, to 41.290 € per year in 2022; more specifically, it was estimated to be further growing in 2024, as just direct costs, accounting for 50-70% of total costs, were estimated to be 16477.23 € per patient per year (Calzavara Pinton et al., 2024; Kadakia et al., 2022; Latorre et al., 2022; Marcellusi et al., 2018; Tarricone et al., 2000).

However, despite the abundance of studies investigating the clinical presentation of SSD, gaps persist in the understanding of exactly how clinical and socio-demographic factors influence the economic burden related to the disorder, and direct costs in particular. Currently, longer duration of illness, earlier age of onset, and higher clinical severity have been identified as predict higher direct costs in schizophrenia (Calzavara Pinton et al., 2024; Calzavara-Pinton et al., 2024).

Cognitive impairment associated with schizophrenia (CIAS) constitutes a cardinal feature of SSD and have a consistent negative impact on functional outcomes and on the rehabilitation in affected individuals (Vita et al., 2022a). In fact, CIAS can be observed in a large proportion of diagnosed individuals,, frequently preceding the manifestation of psychotic symptoms, and encompasses significant impairment in both neurocognitive and social cognition domains (Green et al., 2019; McCutcheon et al., 2023; Vita et al., 2025, 2021) and one with substantial implications for treatment and prognosis. Our understanding of the causes, consequences and interventions for cognitive impairment in schizophrenia has grown substantially in recent years. Here we review a range of topics, including: a. Considering that psychosocial functioning represents an individual predictor of direct costs, and considering the impact of CIAS on several aspects of the lives of individuals living with SSD, it could be hypothesized that CIAS also has an impact on the economic burden of SSD.

The aim of the present preliminary study is to estimate the yearly economic burden of use of services in people living with SSD, and to investigate whether cognitive performance, alongside other relevant clinical and socio-demographic characteristics, represents a significant predictor of increased direct costs.

MATERIALS AND METHODS

Sample

This preliminary study was carried out at the University Psychiatric Unit of the Department of Mental Health and Addictions of Brescia, Italy. The psychiatric unit’s service area covers approximately two-thirds of the population residing in the city and its surrounding areas, totaling 200,048 individuals. The unit comprises four Community Mental Health Centers (CMHCs), two acute inpatient wards, one rehabilitation center, medium-term care facilities and long-term care facilities. CMHCs are responsible for providing routine follow-up appointments and implementing comprehensive treatment strategies for all patients under the psychiatric unit’s care.

This retrospective observational study involved the collection of clinical data and information regarding service utilization in the time period starting January 1st, 2023, and ending December 31st, 2023. This report represents a preliminary presentation of the study, resulting from an ad-interim assessment of included data.

Inclusion criteria for participants were: a diagnosis of SSD as defined by DSM-5-TR or DSM-5 (American Psychiatric Association, 2022, 2013), being 18 years of age or older, and having had at least one contact with a CMHC during the specified timeframe. Subjects with an observation period shorter than the study’s duration were excluded from participation.

The Local Ethical Committee granted ethical approval for the study (registration code NP-2872), and the research was performed in accordance with the guidelines outlined in the Declaration of Helsinki. All necessary precautions were implemented to maintain the anonymity of patients and the confidentiality of their data.

Cost evaluation and use of services

Information pertaining to the utilization of psychiatric facilities and services within the observational timeframe was investigated using the regional mental health database, known as Sistema Informativo di Psichiatria di Regione Lombardia (SIPRL). This registry, compiled at the regional level, contains details regarding each interaction with mental health facilities and services across the Lombardy region and socio-demographic data. Moreover, hospitalizations’ data were accessed through the Milos database, which is an online platform that collects information on people admitted to the Spedali Civili di Brescia, Italy.

To determine individual direct costs, data on service utilization were translated via the regional tariff schedules for residential and semi-residential services, and outpatients’ activities (n. XI/7241 regional resolution of the year 2022), as well as the Diagnosis-Related Groups (DRG)-driven tariff schedules for hospitalization (regional law n. 17/2014 / regional law n. 27/2018).

Data collection and assessment

Various data were collected for the included subjects, with particular attention to: socio-demographic characteristics (including sex, age, marital status, educational attainment, employment status, duration of illness, and age at initial contact), data pertaining to the utilization of services (encompassing medication administration, clinical interventions, interviews with family members, psychosocial interventions for both patients and their families – group or individual -, interprofessional meetings on clinical cases involving CMHC personnel and other clinical or institutional colleagues, data concerning residential and semi-residential placements (such as the intensity of rehabilitation programs and the duration of stay) and related to hospitalizations (such as the length of stay and the assigned DRG upon discharge).

All subjects underwent assessment using the following instruments: the Global Assessment of Functioning (GAF) (Endicott et al., 1976), the Clinical Global Impression-Severity (CGI-S) scale (Busner and Targum, 2007), the Positive and Negative Syndrome Scale (PANSS) (Kay et al., 1987), the Health of the Nation Outcome Scales (HoNOS) (Wing et al., 1998), and the Brief Assessment Cognitive Schizophrenia (BACS) (Keefe et al., 2004).

The PANSS is a semi-structured interview designed to evaluate the severity of symptoms in individuals with SSDs; the CGI-S serves as a scale to assess the global severity of the disorder; the GAF is employed to evaluate an individual’s overall level of functioning in relation to their mental health; the HoNOS is a scale utilized to assess psychosocial functioning and the severity of symptoms in patients with psychiatric disorders. The BACS is a specialized assessment tool developed to measure the cognitive domains impaired in schizophrenia, focusing on areas such as working memory, attention, processing speed, verbal memory, and problem-solving abilities, and currently represents one of the most recommended instruments to assess cognitive performance and to evaluate the severity of CIAS due to its good coverage of different cognitive domains, its psychometric proprieties and its time of completion (Vita et al., 2022b).

Dedicated interviews were used to administer all assessments, performed by physicians specifically trained for this purpose and independent of the participants’ routine treatment. All patients with at least one contact with CMCHs during the year 2023 were contacted and evaluated during a dedicated visit that took place in CMHCs from January 2024 to June 2024. Data regarding the use of services between January 1st, 2023, and December 31st, 2023, were collected. The good inter-rater reliability of these assessments had been demonstrated in previous research conducted within the same Department. (Barlati et al., 2022b).

Statistical analysis

The normality of the data was verified with the Shapiro-Wilk test. Since the total direct cost value showed a non-normal distribution (p < 0.001), non-parametric tests were used for the analyses. To identify potential predictors of total direct costs, the Mann-Whitney test for categorical variables and Spearman’s correlation for continuous variables were used (Heinze et al., 2018). Multiple linear regression analyses were conducted to identify individual predictors of direct costs, with total cost as the dependent variable and variables emerging as significantly correlate with total cost in the correlation analyses as potential predictors. To distinctly investigate cognitive predictors, two separate regression models were performed, one including socio-demographic and clinical variables, and one including cognitive variables. The multiple linear regressions were completed using a stepwise regression process, which allows for the evaluation of the increase in variance explained by the model with each added predictor.

An appropriate number of potential predictors was introduced in each model considering conservative estimates (Austin and Steyerberg, 2015; Schmidt, 1971). The software SPSS version 29.0 (IMB, Armonk, NY, USA) was used for all statistical analyses. Statistical significance was considered for p-values < 0.05.

RESULTS

Sample characteristics

A total of 70 participants were included in the current analysis (which is a preliminary analysis including only a partial sample): the sample was composed by 27.1 % (n = 19) females and 72.9 % (n = 51) males; 38.6 % (n = 27) of participants were recruited in residential facilities while 61.4% (n = 43) were outpatients. The mean age was 43.91 years old (SD ± 17.86), the mean duration of illness was 15.81 years (SD ± 10.16). With regards to education, the mean education was 9.49 years (SD ± 2.71).

Regarding the clinical presentation, the mean PANSS positive score was 18.22 (SD ± 6.55), the mean PANSS negative score was 24.87 (SD ± 6.14), and the mean PANSS general score was 40.01 (SD ± 10.29), the mean BNRR score was 37.46 (SD ± 14.98) and the mean CGI score was 4.66 (SD ± 0.92). When analyzing cognition, the mean score of the BACS Global Cognition was -2.01 (SD ± 1.00). When analyzing the psychosocial functioning, the HoNOS mean score was 12.48 (SD ± 6.32) and the mean GAF score was 44,43 (SD ± 13,42). Concerning alcohol and/or substance abuse, it had been a diagnosis in 14 patients (24.1%).

The observed mean annual cost related to the use of services per patient was of 50’236,83 € (SD ± 52’816.68 €).

Socio-demographic variables are shown in Table I, data regarding the clinical presentation are shown in Table II.

Costs, socio-demographic and clinical variables: correlations

From the correlation analyses, few the analyzed socio-demographic variables: a significant difference (p < 0.001) in total costs was found between participants with or without a history of alcohol/substance abuse, with a mean difference between the two groups of 58’932.44 €/year (± 15’806.30 €/year). No significant correlations were observed between the cost and sex (p = 0.208), age (p = 0.522), level of education (p = 0.970), or the duration of their illness (p = 0.222). Regarding clinical variables, a significant correlation was identified between greater disorder severity, considering different parameters, and increased costs (CGI-S p < 0.001, PANSS total p < 0.001, PANSS positive p < 0.001, PANSS negative p = 0.002, and PANSS general p < 0.001). Also, the increased cost was significantly correlated with reduced performance in the verbal fluency domain (p = 0.003) and worse psychosocial functioning as measured by the GAF score (p = 0.001) and the HoNOS score (p = 0.001).

Potential predictors are summarized in Tables III (socio-demographic variables – categorical variables) and IV (socio-demographic variables – continuous variables), and Table V (clinical variables).

Individual predictors of total cost

Considering the model investigating socio-demographic and clinical predictors, the main individual predictor of costs was disorder’s severity expressed through the CGI-S score (β = 0.497, p < 0.001), followed by substance/alcohol (β = 0.293, p = 0.014); the model explained a considerable proportion of the observed variance [F = 18.530, R² = 0.416, Adj. R² = 0.394, p < 0.001].

Considering cognitive predictors, verbal fluency emerged as a significant individual predictor of costs (β = − 0.349, p = 0.003); this model explained a smaller proportion of the total variance [F = 9.446, R² = 0.122, Adj. R² = 0.109, p = 0.003]. Individual predictors of costs are summarized in Table VI.

DISCUSSION

This preliminary study aimed to identify the annual cost of service use for patients with SSD and its individual predictors: the resulting mean annual cost was 50.236,83 € (SD ± 52.816,68 €) per patient; this data highlights a significant increase compared to historical estimates and reflects a global trend of increasing costs associated with the management of patients with severe psychiatric disorders (Marcellusi et al., 2018; Tarricone et al., 2000). A recent literature review reported that the annual direct cost per patient with schizophrenia varies between € 4,394 and € 31,798 (Kotzeva et al., 2023). However, data reported in this review does not coincide with historical estimates in Italy: in 2001, direct costs were estimated between € 798.35 and € 1,433.51 per two-month period, including pharmacotherapy costs (Garattini et al., 2001) and the same research group identified in 2004 an average annual cost of € 7,025.09 (Garattini et al., 2004). More recent studies have estimated the proportion of direct cost related to service use per patient with schizophrenia at €7,338 annually (Mennini et al., 2021). Moreover, another recent study identified a total annual cost of 16,477.23 (± 32,856.47) €, although the sample was larger and included a more diverse patient population (Calzavara Pinton et al., 2024; Calzavara-Pinton et al., 2024). The analyses included in this report are performed on a preliminary sample and a large proportion of participants included in the present work were recruited in residential facilities, hence increasing the total direct costs.

Also, it can be hypothesized that part of the reason why the average cost identified in this study deviates from that reported in previous literature is also due to the progressive increase in global health costs, a well-documented phenomenon in recent literature: in this context, the cost related to treatment of SSD has been estimated to be constantly growing globally (Kadakia et al., 2022). Finally, differences in socio-economic contexts, organizational models of health services, and reimbursement policies may contribute to discrepancies between costs reported in different countries and different settings.

Regarding the primary aim of the study, in this preliminary report we identified worse verbal fluency as a predictor of higher healthcare costs: verbal fluency is a fundamental cognitive ability that is severely compromised in patients with SSD, even when other cognitive capacities remain relatively preserved (Bokat and Goldberg, 2003; Juhasz et al., 2012). Cognitive Remediation (CR) is proposed as a fundamental strategy to improve impaired cognitive processes in patients with SSD and scientific evidence is clear regarding its effectiveness: numerous studies and meta-analyses have confirmed that CR interventions lead to moderate improvements at both the cognitive and functional levels (Lejeune et al., 2021; Vita et al., 2021; Wykes et al., 2011). Therefore, a more widespread implementation of targeted CR interventions could not only improve CIAS and psychosocial functioning, but could also contribute to reduce healthcare costs associated with the management of SSD, as promoting better social and work functioning could reduce the need for intensive health interventions.

Another aim of the study was to identify both the clinical and psychosocial parameters more strongly correlated with the total annual expenditure, and the main cost predictors related to use of service, thus providing a more comprehensive picture of the economic determinants of SSD.

Firstly, the direct correlation between the total cost and the severity of the illness (measured both with the CGI-S and with the PANSS), and the fact that greater clinical severity (measured with the CGI-S) emerged as a predictor of higher direct costs, confirms what has been reported by previous studies (Calzavara Pinton et al., 2024; Knapp et al., 2004). These studies show that costs increase proportionally to the complexity and severity of the clinical picture, as more severe patients require more intensive interventions both from a pharmacological and a psychosocial care perspective. Moreover, the analysis of specific PANSS domains further clarified that positive, negative, and general psychopathology symptoms are all significantly correlated with increased costs. This data reflects what was reported by Kennedy et al. (2014), according to which not only the most evident symptoms, such as delusions and hallucinations, but also the less conspicuous ones, such as apathy and social withdrawal, significantly contribute to the increase in healthcare costs. These results highlight the need to adopt integrated therapeutic approaches, capable of intervening on all dimensions of psychopathology to improve treatment effectiveness and reduce costs (Galderisi et al., 2021; Maj et al., 2020; Vita et al., 2022a).

Substance abuse has emerged as an individual predictor of total cost, highlighting the significant economic impact of this clinical condition: this increase in costs is often attributable to the need for targeted interventions, which tend to be more complex and expensive compared to standard treatments (Bartels et al., 1993; Hoff and Rosenheck, 1999; Whiteford et al., 2013). Moreover, the phenomenon of stigmatization still represents a significant barrier to access, and the quality of health services aimed at this population: several studies have shown that people with substance abuse are frequently victims of prejudice, which compromises both the success of treatment and the continuity of care (Corrigan et al., 2009; Schomerus et al., 2011) in turn, are viewed more harshly than those with physical disabilities. Endorsement of such stereotypes often lead to less helping behavior and more avoidance of people with drug addiction compared to those with mental illness. In this study, attribution and dangerousness models are tested on a stratified random sample of the US population. The sample was recruited from a national online research panel (N = 815. Stigma not only reduces the likelihood of adherence to therapies but also intensifies psychological distress and hinders effective management of related pathologies (Livingston and Boyd, 2010) psychosocial, and psychiatric variables for people who live with mental illness. An exhaustive review of the research literature was performed on all articles published in English that assessed a statistical relationship between internalized stigma and at least one other variable for adults who live with mental illness. In total, 127 articles met the inclusion criteria for systematic review, of which, data from 45 articles were extracted for meta-analyses. None of the sociodemographic variables that were included in the study were consistently or strongly correlated with levels of internalized stigma. The review uncovered a striking and robust negative relationship between internalized stigma and a range of psychosocial variables (e.g., hope, self-esteem, and empowerment.

The positive correlation between the total cost and the HoNOS scale score, indicative of a higher level of psychosocial impairment, aligns with what is described in other studies, according to which patients with high psychosocial disability tend to require greater use of care resources (Parker et al., 2016; Franklin et al., 2019). This finding further highlights the importance of including targeted psychosocial interventions, which could improve the patient’s overall functioning and, at the same time, contain healthcare costs in the long term.

On the other hand, the inverse correlation between the total cost and the GAF score, as well as with verbal fluency, suggests that better global and cognitive functioning is associated with a reduction in costs. This finding is supported by some evidence (Breitborde et al., 2021; Reeder et al., 2014; S. Yamaguchi, 2016; Sevy and Davidson, 1995) provision of such treatments within Coordinated Specialty Care (CSC, which argue that interventions aimed at improving cognitive skills and global functioning can bring benefits not only in terms of quality of life, but also in terms of economic sustainability for health systems.

Strengths and limitations

The present study has some notable points of strength. It was conducted in a real-world setting, with participant recruitment from CMHCs during routine clinical practice, which strengthens the generalizability of its findings. The use of well-validated instruments for the assessment of multiple cognitive domains and clinical correlates represents another point of strength. Finally, another significant methodological strength is the direct measurement of direct costs through data collection, unlike many studies that estimate expenses probabilistically. This direct approach allows for a more accurate calculation of the average annual cost per patient for service use and enables the examination of associated clinical factors. However, this direct measurement is also a weakness, as probabilistic methods allow for a broader analysis of cost types, including indirect costs.

The main drawback of this study is its exclusive focus on direct costs related to service utilization, omitting the costs of pharmacological treatments, which will be addressed in future research. Another key limitation is the preliminary nature of the sample, which predominantly consists of patients from residential settings, potentially skewing the cost analysis and creating a possible sampling bias.

Data regarding psychiatric and medical comorbidity of participants and regarding non-pharmacological treatments, including psychotherapy and other evidence-based psychosocial interventions, were not retrieved for the present preliminary study and could not be included in the analyses. Finally, no dedicated measure of treatment adherence, which has a considerable impact on relapse and hospitalization risk and therefore on direct treatment costs, was included in the assessment of participants.

Future perspectives

Future research could enhance our findings by incorporating more data on the costs of pharmacological treatments, as well as examining additional potential predictors of costs, such as social cognition performance domains. These domains are often impaired in these patients and significantly influence psychosocial functioning and clinical severity. Moreover, the sample is currently being expanded with the goal of including a more diverse clinical population and in particular more outpatients, to make the sample more representative of the clinical population.

CONCLUSIONS

The results of this study show how, in SSD, the mean direct cost associated to use of service is 50236.83 € per patient per year. Clinical severity, measured through both the CGI-S and the PANSS, has a direct impact on healthcare expenses, contributing to incremental costs. Conversely, a better cognitive profile and better psychosocial functioning, measured through verbal fluency and the GAF, was associated to lower costs, suggesting efficacy of specific rehabilitation programs and prompting cost-effectiveness studies to better define this matter. Moreover, it is important to underline how substance abuse is a strong contributor to higher healthcare costs, pointing out how patients with substance abuse comorbidity present more complex therapeutic needs, reflecting higher healthcare costs. Therapeutic intervention specifically designed for these patients could contribute to a long-term decrease in healthcare costs, improving the quality of the provided treatments and the overall efficiency of the healthcare system.

A better organization of mental health services, supported but an integrated net capable of identifying early patients with vulnerabilities, including substance abuse, represents a crucial factor to optimize the use of healthcare resources. A multidisciplinary approach, integrating psychopharmacological treatments, psychotherapy and psychosocial interventions, could efficiently respond to the complex needs of these people. Such a multidisciplinary approach could optimize healthcare resources allocation, as well as improving clinical and psychosocial results. In particular, the growing knowledge that the importance of a targeted and integrated treatment represents a key factor in improving clinical outcomes and supports the economical sustainability of healthcare systems. Targeted psychosocial interventions, combined with personalized psychopharmacological treatments, could contribute not only to decrease long term costs, but also to improve patients’ quality of life, with the goal of a complete functional recovery.

Acknowledgments

All authors who contributed to this paper are listed as authors. No professional medical writer was involved in any portion of the preparation of the manuscript.

Financial support

The Authors received no specific funding for this work.

Statement of interest

The authors declare no conflict of interest in the design, execution, interpretation, or writing of the study.

Ethical statement

The study was approved by the Local Ethical Committee (registration code NP20909) and conducted according to the procedures described in the Declaration of Helsinki. Patients agreeing to participate in the study signed a dedicated written informed consent form. All necessary precautions were adopted to maintain patients’ anonymity and data confidentiality.

Figures and tables

TABLE I. Sociodemographic data.

Total sample - 70 individuals
Gender [N (%)]
Male	Famale
51 (72.9 %)	19 (27.1 %)
Recruitment setting
Inpatient (Residential facilities)	Outpatient
27 (38.6%)	43 (61.4%)
Age (years) [mean (SD)]
43.91 (± 17.86)
Illness duration (years) [mean (SD)]
15.81 (± 10.16)
Education (years) [mean (SD)]
9.49 (± 2.71)

TABLE II. Clinical characteristics.

Total sample - 70 individuals
PANSS score [mean (SD)]
PANSS positive	PANSS negative	PANSS general	PANSS total
18.22 (± 6.55)	24.87 (± 6.14)	40.01 (± 10.29)	83.10 (± 19.95)
BNSS score [mean (SD)]
37.46 (± 14.98)
BACS Global Cognition score [mean (SD)]
-2.01 (± 1.00)
CGI-S score [mean (SD)]
4.66 (SD ± 0.92)
GAF score [mean (SD)]
44,43 (± 13,42)
HONOS score [mean (SD)]
12.48 (± 6.32)
Substance and/or alcohol abuse [N (%)]
Yes		No
14 (24.1 %)		44 (75.9 %)

TABLE III. Correlations – Socio-demographic variables – categorical variables.

Gender	N	Mean costs in € (± SD)	p-value
M	51	45363.35 (±50272.85)	0.208
F	19	63318.28 (±58524.40)

TABLE IV. Correlations – Socio-demographic variables – continuous variables.

Substances and/or alcohol abuse	N	Mean costs in € (± SD)	p-value
Yes	14	97559.83 (± 51319.89)	< 0.0001
No	44	38627.38 (± 51569.63)

TABLE V. Correlations – Clinical variables.

Variable		Direct cost
	Spearman’s rho	p-value
PANSS positive	+ 0.410	p = 0.001
PANSS negative	+ 0.369	p = 0.002
PANSS general	+ 0.411	p < 0.001
PANSS total	+ 0.460	p < 0.001
CGI-S	+ 0.571	p < 0.001
GAF	- 0.391	p = 0.001
HONOS	+ 0.380	p = 0.001
Verbal fluency	- 0.349	p = 0.003

TABLE VI. Individual predictors of direct costs.

DEPENDENT VARIABLE: TOTAL COST
	Standardized Beta (β)	t	p		MODEL DATA
MODEL 1 (GAF score, the CGI-S score, the HONOS total score and the PANSS total score, and the scores of each subscale)	CGI-S	+ 0.581	5.194	< 0.001	F = 26.981, R2 = 0.337, Adj. R2 = 0.325, p < 0.001
MODEL 2 (Model 1 + substance and/or alcohol abuse)	CGI-S	+ 0.497	4.494	< 0.001	F = 18.530, R2 = 0.416, Adj. R2 = 0.394, p < 0.001
	Substance and/or alcohol abuse	+ 0.293	+ 2.540	0.014
MODEL 3 (Cognitive variables)	Verbal fluency	- 0.349	- 3.073	0.003	F = 9.446, R2 = 0.122, Adj. R2 = 0.109, p = 0.003