Genetic predisposition and antipsychotic treatment effect on metabolic syndrome in schizophrenia: a ten-year follow-up study using the Estonian Biobank

Summary Background Schizophrenia (SCZ) patients exhibit 30% higher prevalence of metabolic syndrome (MetS) compared to the general population with its suboptimal management contributing to increased mortality. Large-scale studies providing real-world evidence of the underlying causes remain limited. Methods To address this gap, we used real-world health data from the Estonian Biobank, spanning a median follow-up of ten years, to investigate the impact of genetic predisposition and antipsychotic treatment on the development of MetS in SCZ patients. Specifically, we set out to characterize antipsychotic treatment patterns, genetic predisposition of MetS traits, MetS prognosis, and body mass index (BMI) trajectories, comparing SCZ cases (n = 677) to age- and sex-matched controls (n = 2708). Findings SCZ cases exhibited higher genetic predisposition to SCZ (OR = 1.75, 95% CI 1.58–1.94), but lower polygenic burden for increased BMI (OR = 0.88, 95% CI 0.88–0.96) and C-reactive protein (OR = 0.88, 95% CI 0.81–0.97) compared to controls. While SCZ cases showed worse prognosis of MetS (HR 1.95, 95% CI 1.54–2.46), higher antipsychotic adherence within the first treatment year was associated with reduced long-term MetS incidence. Linear mixed modelling, incorporating multiple BMI timepoints, underscored the significant contribution of both, antipsychotic medication, and genetic predisposition to higher BMI, driving the substantially upward trajectory of BMI in SCZ cases. Interpretation These findings contribute to refining clinical risk prediction and prevention strategies for MetS among SCZ patients and emphasize the significance of incorporating genetic information, long-term patient tracking, and employing diverse perspectives when analyzing real-world health data. Funding EU Horizon 2020, 10.13039/501100004359Swedish Research Council, 10.13039/501100002301Estonian Research Council, Estonian Ministry of Education and Research, University of Tartu.


Study cohort
Estonian Biobank (EstBB)   Health records at EstBB were extracted from the data provided by regional hospitals, national registries, and the Estonian Health Insurance Fund, a national administrative database that pools detailed and individual-level billing data for all health care services as well as digital prescription information.Disease diagnoses are recorded based on the International Classification of Diseases, 10 th revision (ICD-10 codes) and prescribed medication according to the Anatomical Therapeutic Chemical (ATC) classification system.The activities of EstBB are regulated by the Human Genes Research Act, which was adopted in 2000 specifically for the operations of EstBB.All EstBB participants have signed a broad informed consent.The study was approved by the Estonian Committee on Bioethics and Human Research at the Estonian Ministry of Social Affairs (24 March 2020, nr 1.1-12/624) and carried out using data according to release S47 from EstBB.Data freeze 2023v1 with follow-up data until 31/03/2023 at EstBB was used for analyses 1,2 .
Selection of SCZ cases SCZ cases were defined as individuals with Schizophrenia Spectrum Disorder (SSD), according to ICD-10 3 codes F20-F29.These encompass several conditions: schizophrenia, the central diagnosis of the group, schizotypal disorders, persistent delusional disorders, and a larger group of acute and transient psychotic disorders.While there were 2,473 individuals with the SSD diagnosis in data freeze 2023v1, we excluded individuals i) with no available medication prescription data, ii) with a bipolar disorder diagnosis confirmed by a neurology or a psychiatry specialist, iii) whose earliest diagnosis age was <15 and >40, and iv) who had their first report of SCZ onset (i.e., either received their first antipsychotic prescription or their first SSD diagnosis) before year 2006.The latter filtering criterium was applied due to two observational characteristics of biobank data.First, some SCZ cases received their disease diagnosis after antipsychotic prescription (Supplementary Figure 17A).Since the primary objective was to track patients from treatment initiation, treatment trajectories had to be captured from treatment onset and not from disease diagnosis.Hence, the earliest follow-up start date was set at either the date of the first diagnosis or the date of the first prescription.Second, considering that the electronic filing of medical bills within the Health Insurance Fund started in 2004, a buffer period of two years was introduced.This was applied to ensure that individuals who received their first SSD diagnosis or began their antipsychotic treatment prior to the implementation of the electronic filing system were excluded (Supplementary Figure 17B).All defined SCZ cases (n = 677) were of European ancestry.In analyses using PRSs, we excluded i) samples who were included in the GWASs used for calculating PRSs considered in the current study, and ii) one member per related pairs of SCZ cases (PLINK PI_HAT >0.2), totalling 577 individuals.The flowchart of selecting SCZ cases is provided in Supplementary Figure 1.

Selection of controls
To select age-and sex-matched controls, the following individuals were considered: i) individuals without any behavioural or mental disorders (ICD-10 F* codes), ii) who had not been prescribed antipsychotic medications (ATC N05A* codes), iii) who were not related to any individuals with SSD in EstBB or other potential controls (PLINK PI_HAT <0.2), iv) had not been included in sample sets of GWAS used for calculating PRSs considered in the current study, and v) were of European ancestry, totalling 43,969 individuals.Four birth year-and sex-matched controls were defined for each SCZ case using R/MatchIt 4 with the method = nearest parameter.This approach yielded four controls for each SCZ case of the same sex and born in the same year.The flowchart of selecting controls is provided in Supplementary Figure 1.Data about sex was retrieved from the information reported at participant enrolment to EstBB.

Smoking and BMI data
For SCZ cases and controls, smoking status (ever/never) was derived based on data reported at participant enrolment: ever -in case former smoking or current smoking was marked, never -in case never smoking was reported.BMI information was extracted from all available electronic health records (including data extraction from free-text sections in medical case summaries) linked with EstBB.Measurements <15 and >50 BMI points were excluded (79 of 13,122 datapoints).Consecutive BMI measurements were required to be >2 months apart (943 datapoints omitted).We additionally excluded: i) datapoints of individuals whose difference in consecutive weight measurements was >15kg within a period of less than 2 years and not supported by later measurements, and ii) self-reported BMI measurements not supported by preceding and succeeding objective measurement data (17 datapoints omitted).Thus, after quality control, 2,163 and 9,920 BMI measurements were available for 664 SCZ cases and 2,691 controls, respectively.In survival analyses where BMI was considered as a covariate, the measurement closest to the entry date of the survival study was used.As the mean difference between the study entry date and the BMI measurement was 1,622 days (minimum 0, maximum 5,832 days, Supplementary Figure 3F), four BMI categories, i.e., (i) underweight (£18.5),ii) normal weight (>18.5 and £25), iii) overweight (>25 and £30), and iv) obese (>30)), were used instead of real values.

Median dose
Throughout the follow-up period, 20 different types of antipsychotic medications with different routes of administration (oral, short-acting and long-acting injections) and with patterns of concurrent use were purchased (Supplementary Table 2).To this end, the Delphi method [5][6][7] was applied to standardize the doses of purchased antipsychotics to an equivalent dose of 100 milligrams (mg) per day of chlorpromazine.Specifically, the conversion factors outlined in Gardner et al. 7 were preferred and if missing (e.g., for short-acting injectables and for cariprazine and melperone), the dose-adjustment factors from Leucht et al. 8,9 were applied (R/chlorpromazineR) 10 .Long-acting injectables were first converted to the daily dose based on the minimum duration recommendation (i.e., 1 ampoule of 50 mg/ml of haloperidol recommended for a minimum of 14 days were converted to the daily dose equivalent of 187 mg of chlorpromazine for two weeks).To derive the median antipsychotic dose for SCZ cases, digital drug dispensing data for oral and long-acting antipsychotics (ATC N05A* codes except lithium) from the Estonian National Health Insurance Fund were used.The medicines in the digital drug dispensing data belong to the list of medicines reimbursed by the Estonian Health Insurance Fund and can be purchased at the pharmacy.The data consist of information regarding the date of drug prescription and purchase, the dose and content specified on each package, and the number of packages bought.We additionally considered outpatient procedures for long-acting antipsychotics provided by a psychiatric nurse, either for i) risperidone, olanzapine, aripiprazole, paliperidone, or for ii) perphenazine, fluphenazine based on billing data stored from 2015 onwards.These outlined long-acting antipsychotics are reimbursed centrally by the Estonian Health Insurance Fund and are administered to patients as outpatient treatment services in psychiatric clinics.Such long-acting injections were converted to an equivalent dose of 300 mg per day of chlorpromazine for 4 weeks and were considered as purchases.Specifically, for each purchase, the package content (i.e., dose in mg and the number of pills in a package) was multiplied by the number of packages bought.Purchases within two weeks were consolidated, such that the purchase date for all consecutive purchases were assigned to the date of the first 14day purchase.Purchases separated by >365 days were considered as distinct purchase batches.Next, the calculated mg ´ pills in a package ´ number of packages variable for each purchase was divided by the number of days until the next purchase resulting in an estimate of a daily dose per purchase.Individual purchases or batch purchases consisting of a single purchase were excluded.Finally, the daily antipsychotic dose was derived by taking the median across the derived daily doses.In case different purchase batches were derived (>365 days between purchases), the median dose was calculated for each purchase batch, followed by a subsequent median dose calculation across all batches.Of note, the correlation with the widely used Defined Daily Dose method 11 was r = 0.98 (Supplementary Figure 10C).

Adherence
Four different adherence variables were derived using digital drug dispensing data for antipsychotics (ATC codes N05A* except lithium) and outpatient procedures for long-acting antipsychotics from the Estonian National Health Insurance Fund.Firstly, the proportion of days covered within the first treatment year (PDC1year) was calculated to provide a uniform estimate of adherence during the initial year of treatment.The PDC1year variable was derived for SCZ cases whose first antipsychotic prescription was dispensed at least one year before the end of follow-up (31/03/2023; n = 582; individuals who died within their first treatment year were excluded).It was calculated by counting the days with antipsychotic supply for all purchased antipsychotics within a year from the first purchase and divided by 365 days.Secondly, the proportion of days covered within the last available treatment year (PDClastyear) allowed to capture a cross-sectional estimate of adherence over a uniform period not affected by treatment initiation bias.The PDClastyear variable was derived for SCZ cases who had purchase information available for more than one year (n = 480).It was calculated by determining the days' supply for all purchased antipsychotics within their last available treatment year (starting 365 days before the last antipsychotic purchase) divided by 365 days.The median duration of treatment until the beginning of the last treatment year was 6.96 years (interquartile range 7.94).Thirdly, the proportion of days covered over the study period (PDCfollow-up) was calculated over follow-up.The PDCfollow-up variable was derived for all SCZ cases who had purchased antipsychotics at least once (n = 595) by calculating the days' supply for all purchased antipsychotics from the first purchase until the end of follow-up or death divided by the number of days between the first purchase and the end of follow-up or death.Lastly, the proportion of days covered over the dispensing period (PDCpurchase) captured the time from first to the last purchase.The PDCpurchase variable was derived for all SCZ cases who had purchased antipsychotics at least once (n = 595) by calculating days' supply for all purchased antipsychotics from the first purchase until the end of supply of the last purchase or death and divided by the number of days between the first purchase and the end of supply or death.If the end of supply of the last purchase extended beyond the follow-up end date, the end of supply was truncated to the end date of the follow-up period.For all adherence variables, the days' supply was calculated by multiplying the number of pills in a package by the number of packages bought per purchase, assuming one pill per day.Similarly to the median dose calculation, purchases within 14 days were consolidated, such that the purchase date of all consecutive purchases were assigned to the date of the first 14-day purchase.Purchased long-acting injectables were first converted to daily treatment based on the minimum duration recommendation.Longacting injections provided by a psychiatric nurse were converted to daily treatment for 4 weeks.

Treatment years
Two different variables were derived to capture treatment years.Firstly, we counted the days from the first to the last purchase, added the supply of the last purchase and converted the summed days to a year (treatment_yearspurchase). Secondly, we added up the days' supply for each purchase and converted the number to years (treatment_yearssupply).These two variables offer different perspectives on treatment duration, the former focusing on the duration between the first and the last antipsychotic purchase, and the latter considering only the days with antipsychotic supply acquired during treatment.The proportion of metabolically more active antipsychotics taken over treatment years (i.e., treatment_yearssupply) was calculated by adding up the days' supply for clozapine, risperidone, quetiapine, and olanzapine and divided by treatment_yearssupply.Metabolically more active drugs were defined based on the British Association for Psychopharmacology (BAP) guidelines 12 .SCZ cases who received antipsychotics via outpatient procedures for long-acting antipsychotics (n=47) were excluded from this calculation.
Overview of the derivation of the treatment variables are depicted in Supplementary Figure 2 and the Spearman correlations of the calculated chlorpromazine-equivalent dose with adherence and treatment length variables in Supplementary Figure 18.

Genotype data and polygenic risk scores
All EstBB participants were genotyped using Illumina GSAv1.0,GSAv2.0, and GSAv2.0_ESTarrays with quality control conducted according to best practices (exclusion of individuals with call rate <95%, with mismatch between genotype and phenotype sex and who deviated ±3SD from the samples' heterozygosity rate mean; exclusion of SNVs with call rate <95%, HWE p<1e-4, MAF <1%) 13 .Pre-phasing was carried out with Eagle v2.3 14 and imputation with Beagle 15 v.28Sep18.79367using the population-specific imputation reference panel built from 2,297 whole genome sequencing samples 16 .One member per related individual pairs (PLINK PI_HAT >0.2) were excluded in association testing where PRSs were considered 13,17 .
For PRS calculation, GWAS summary statistics based on GRCh37 for SCZ 18 and 14 MetS-associated traits (coronary heart disease (CHD) 19 , total cholesterol (TC) 20 , LDL-cholesterol (LDL) 20 , HDL-cholesterol (HDL) 20 , nonHDL-cholesterol (nonHDL) 20 , triglycerides (TG) 20 , C-reactive protein (CRP) 21 , glycated haemoglobin (HbA1c) 22 , fasting glucose (FG) 22 , random glucose (RG) 23 , type II diabetes (T2D) 24 , body mass index (BMI) 25 , systolic blood pressure (SBP) 26 , and diastolic blood pressure (DBP) 26 ) were considered with European-specific results preferred when available.SNVs with imputation score <0.8, MAF <0.01, HWE P<1e-4, and ambiguous (A/T and C/G) SNPs in EstBB genotype data were excluded.PRSs were calculated with PRS-cs 27 , a Bayesian polygenic prediction method that places a continuous shrinkage prior on SNV effect sizes and infers posterior SNV weights using GWAS summary statistics restricted to ~1.1 million HapMap variants 28 and an external European sample-based LD reference from the 1000 Genomes Project 29 .The default parameters and the auto option were used, and the HLA region was excluded.For SCZ, however, the HLA region was considered and the global shrinkage parameter phi = 1 was used, given the significant association of the HLA region in SCZ GWAS and to capture the highly polygenic architecture of SCZ, respectively 18,[30][31][32][33] .The PRSs were normalized to follow a normal distribution with mean of 0 and SD of 1.
Correlation patterns among 15 PRSs were assessed with Pearson correlation using R/corrplot 34 .Significant associations were identified based on Bonferroni correction (0.05/15 = 0.0033).To validate the correlation structure of the PRSs, unrelated EstBB participants (PLINK PI_HAT <0.2), excluding SCZ cases and controls used in the current study were used, totalling 117,792 individuals.Genetic correlations for 15 traits were calculated with LDSC 35 , using summary statistics from GWAS restricted to SNVs in the HapMap 3 reference panel 28 and the LD scores from the European ancestry sample 29 .
For testing the association with the derived median chlorpromazine-equivalent dose, PRSs for clozapine drug metabolites, i.e., clozapine, norclozapine and their ratio 36 were additionally included.Given that the PRSs built only with genome-wide significant variants showed stronger association and greater explained variance for these traits compared to those built using less stringent pvalue thresholds 36 , we considered only independent variants (--clump-r2 0.1) below P<5×10 −8 and calculated the PRSs using PLINK (--score) 13,17 .

Association testing between median chlorpromazine-equivalent dose and treatment variables
To test the association of the median chlorpromazine-equivalent dose with adherence (PDCpurchase specifically, given its high correlation with other adherence variables) and treatment years over supply (treatment_yearssupply), linear regression was used with sex and birth year as covariates.Individuals whose median chlorpromazine-equivalent dose was greater than ±3SD from the mean on log10 scale were excluded (n = 6; Supplementary Figure 10B).

Association testing between PRSs with disease status and the median chlorpromazine-equivalent dose
For association testing with disease status, multivariate logistic regression analysis was performed in a forward stepwise manner.Namely, each PRS of interest was first modelled independently with the SCZ/control status as a dependent variable, and baseline characteristics, i.e., sex, birth year, and 10 genotype principal components (PCs), as covariates.Next, the PRSs were incrementally added to the main model, starting with the one with the lowest p-value and scanning through all other PRSs.The effective number of tests (meff; R/poolR 37 ) was applied to account for the PRS correlation structure (0.05/13 = 0.0038).
To test whether the median chlorpromazine-equivalent dose was associated with the genetic liability to MetS, multivariate linear regression analysis was applied similarly to the multivariate logistic regression model.Median dose on log10 scale as a dependent variable was regressed on 18 PRSs (including the PRSs for clozapine, norclozapine and their ratio) in a forward stepwise manner with PDCpurchase, treatment_yearssupply, sex, birth year and ten genotype PCs as covariates.Individuals whose median chlorpromazine-equivalent dose on log10 scale was greater than ±3SD from the mean were excluded (n = 6; Supplementary Figure 10B).Normality assumption for BMI was assessed with diagnostic plots of model residuals.Since indications of nonnormality of the residuals were observed, Box-Cox transformation was used to determine the best possible transformation of the response variable.However, as the Box-Cox transformation did not result in perfect normality and the analysis with transformed response variable showed basically the same ANOVA p-values for effects of interest (<2e-16 for both models), we opted for using the model with untransformed BMI values for easier interpretability.

Survival analyses
To assess the difference in the incidence of MetS in SCZ cases compared to controls, multivariate Cox proportional hazards modelling was applied using R/survival 38 .The following disease endpoints were considered: type II diabetes (T2D, ICD-10 E11), hypercholesterolemia (ICD-10 E78), essential hypertension (ICD-10 I10), coronary heart disease (CHD, ICD-10 I20-I25), cardiovascular diseases (CVD, ICD-10 I20-I25, I61, I63, I64), and any metabolic disorder (the first occurrence of any of the outlined MetS endpoints).The study entry date for SCZ cases was the onset of SCZ, either first diagnosis or prescription, and the study entry date for controls was the date of SCZ onset for the matched SCZ case.The study end date was either the first occurrence of the diagnosis of the considered MetS endpoint, death, or the end of follow-up.All prevalent cases were excluded (i.e., SCZ cases and controls with the endpoint of interest before the study start date).Sex, BMI group, smoking status, PRSs, and 10 genotype PCs were considered as covariates.Bonferroni correction (0.05/6 = 0.0083) was applied to account for six endpointspecific analyses.In each of the six endpoint-specific model, the PRSs were modelled in a forward stepwise manner and the effective number of tests was applied to account for the PRS correlation structure as described in the previous paragraph.
To assess the association between MetS and adherence captured within the first treatment year, multivariate Cox proportional hazards modelling was applied as described previously.Only SCZ cases who had dispensed antipsychotics at least a year (n = 480) were considered.The study entry date was one year after the first antipsychotic purchase.

BMI trajectory analyses
To test the difference of BMI trajectories between SCZ cases and controls, we employed linear mixed modelling (R/lme4 39 , R/lmerTest 40 ) considering that individuals had multiple BMI measurements, that BMI trajectory over time could be individual-specific and accounting for non-linear age effects (as quartic polynomial) and the polygenic predisposition to BMI (BMI PRS).Population parameters (sex, smoking status (ever vs never), BMI PRS, 10 genotype PCs) were modelled as fixed effects and subject-specific effects (subject-specific random intercept and subject-specific age effect) were considered as random effects.The interaction term between BMI PRS and age at BMI measurement was additionally included to capture the age-dependent effect of BMI PRS.Firstly, we considered two models, a baseline model and a model that additionally included an interaction term between disease status and the age at BMI measurement (models outlined in Supplementary Table 5).Next, to account for the treatment effect, linear mixed modelling was similarly applied as in the first approach.We considered the model with the better fit identified in the first approach and compared it with a model where the quartic polynomial of treatment years was additionally included (models outlined in Supplementary Table 6).Model fit was assessed with ANOVA.In both analyses, 1,785 and 9,446 BMI measurements with on average 3.34 and 3.69 BMI measurements per individual were available for 532 SCZ cases and 2,519 controls, respectively.
To assess the impact of treatment factors, we focused on SCZ cases.We considered sex, age at BMI measurement, the PRS for BMI, smoking status (ever vs never), and 10 genotype PCs as fixed effects.Additionally, we used the median chlorpromazineequivalent dose and adherence calculated based on purchase information two years prior to each BMI measurement and treatment years based on all supply information up to each BMI timepoint.Subject-specific effects (subject-specific random intercept and subject-specific age effect) were considered as random effects.The median chlorpromazine-equivalent dose was transformed to log10 scale.Timepoints with a single antipsychotic purchase preceding the BMI measurement within the two-year timeframe and individuals whose median dose on log10 scale deviated more than >3SD from the mean were excluded.In total, 745 BMI measurements were available for 282 SCZ cases with an average of 2.61 BMI measurements per individual.In all analyses, the BOBYQA method 41 for optimization was applied.BMI trajectories were derived using the predict function.
Normality assumption for BMI was assessed with diagnostic plots of the model residuals.As there were indications of nonnormality of the residuals, Box-Cox transformation was used to determine the best possible transformation of the response variable.However, as the Box-Cox transformation did not achieve perfect normality and the analysis with transformed response variable showed basically the same ANOVA p-values for effects of interests (<2e-16 for both models), we opted for using the model using untransformed BMI values due to easier interpretability.
We have provided 83% confidence intervals for Figure 3.This confidence level was chosen to make comparisons between group means easier.Namely, Goldstein and Healy (1995) highlight that to indicate statistically significant differences (alpha = 0.05) between two means with non-overlapping confidence intervals, one should use 83% confidence intervals in graphs 42 .
Statistical analyses were conducted with R software version 4.2.2. 43Supplementary Figure 2. Overview of the derivation of treatment variables from digital drug dispensing data.In each panel, the top part displays four purchases (1 package of 30 tablets of 1mg, 1 package of 60 tablets of 1mg, 2 packages of 30 tablets of 2mg, and 1 package of 30 tablets of 1mg) for an individual over a 270-day period.A shorter time period (<1 year) is considered for simplicity.The half balloons below each purchase in each section indicate the values used for the derivation of the treatment variables.For each treatment variable, one pill per day is considered.Of note, the second purchase of 1 package of 60 tablets of 1mg is depicted to occur 5 days before the end of the supply of the first purchase.Median dose: For each purchase, the package content (i.e., dose in mg and the number of pills in a package) was multiplied by the number of packages bought.The calculated mg ´ pills in a package ´ number of packages variable for each purchase was divided by the number of days until the next purchase resulting in an estimate of a daily dose per purchase.The median daily antipsychotic dose was derived by taking the median across the derived daily doses.Adherence: The proportion of days covered within the first treatment year (PDC1year) was calculated by counting the days with antipsychotic supply for all purchased antipsychotics within a year from the first purchase and divided by 365 days.The proportion of days covered within the last available treatment year (PDClastyear) was calculated by determining the days' supply for all purchased antipsychotics within their last available treatment year divided by 365 days.The proportion of days covered over the dispensing period (PDCpurchase) was derived by calculating days' supply for all purchased antipsychotics from the first purchase until the end of supply of the last purchase and divided by the number of days between the first purchase and the end of supply.The proportion of days covered over the study period (PDCfollow-up) was derived by calculating the days' supply for all purchased antipsychotics from the first purchase until the end of follow-up divided by the number of days between the first purchase and the end of follow-up.For all adherence variables, the days' supply was calculated by multiplying the number of pills in a package by the number of packages bought per purchase, assuming one pill per day.Overlapping days with supply were counted only once as indicated by an asterisk.Treatment years: To derive treatment_yearspurchase, we counted the days from the first to the last purchase, added the supply of the last purchase and converted the summed days to a year.For treatment_yearssupply, we added up the days' supply for each purchase and converted the number to a year.Overlapping days with supply were counted only once as indicated by an asterisk.

Supplementary Figure 3 .Supplementary Figure 11 . 1 −
bw/ first and last purchase + last supply 365 days = 2/3 of a year 175 days with supply* 365 days = 1/2 of a year Smoking and BMI data.(A) Overview of the distribution of birth years for SCZ cases and controls coloured for disease status.(B) Proportion of individuals who reported ever/never smoking.(C) For SCZ cases, the difference between the age when smoking was reported in EstBB and the age at SCZ onset (either first prescription or first diagnosis) based on electronic health records.The age at SCZ onset is subtracted from the age when smoking was reported in EstBB on x-axis.(D) Distribution of the median BMI measurement values per individual.(E) Count of SCZ cases and controls by the number of BMI measurements over follow-up.(F) Difference in days between the study entry date in survival analyses (i.e., date of SCZ onset for SCZ cases and matched controls).PRSs of MetS traits and SCZ.(A) Correlation matrix of the PRSs for 14 MetS traits and SCZ using unrelated EstBB participants (n = 117,792).(B) Genetic correlations retrieved with LDSC based on published GWAS for 14 MetS traits and SCZ.(C) Boxplots of BMI PRS distribution for SCZ cases and controls.(D) Boxplots of CRP PRS distribution for SCZ cases and controls.

Table 5 . Overview of the considered linear mixed models and the results for assessing BMI trajectories over time between SCZ cases and controls
. PRS for BMI and PCs are scaled to follow a normal distribution with mean of 0 and SD of 1. Coefficients for BMI age correspond to orthogonal polynomial coefficients.Model 1: BMI ~ poly(BMI age, 4) + sex + smoking + BMI PRS + PC1 … PC10 + BMI PRS : BMI age + (1 + BMI age | subject) Model 2: BMI ~ disease status * poly(BMI age, 4) + smoking + sex + BMI PRS + PC1 … PC10 + BMI PRS : BMI age + (1 + BMI age | subject)

Table 6 . Overview of the considered linear mixed models and the results for assessing BMI trajectories over time between SCZ cases and controls while accounting for treatment length
. PRS for BMI, and PCs are scaled to follow a normal distribution with mean of 0 and SD of 1. Coefficients for BMI age and treatment years correspond to orthogonal polynomial coefficients.Model 1: BMI ~ disease status * poly(BMI age, 4) + sex + smoking + BMI PRS + PC1 … PC10 + BMI PRS : BMI age + (1 + BMI age | subject) Model 2: BMI ~ disease status * poly(BMI age, 4) + poly(treatment_yearssupply, 4) + sex + smoking + BMI PRS + PC1 … PC10 + BMI PRS : BMI age + (1 + BMI age | subject)