The Impact of Selection Bias on Genetic Prediction Using the Bipolar Polygenic Risk Score in First-Admission Psychosis

The Impact of Selection Bias on Genetic Prediction Using the Bipolar Polygenic Risk Score in First-Admission Psychosis | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article The Impact of Selection Bias on Genetic Prediction Using the Bipolar Polygenic Risk Score in First-Admission Psychosis Katherine Jonas, Amna Asim, Yuan Yang, Urs Heilbronner, Thomas Schulze, and 4 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4536236/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Polygenic risk scores (PRS) have potential utility as biomarkers of psychiatric disorders. However, while the schizophrenia (SZ) PRS has been consistently associated with case-control status and a more severe course of illness, the associations between the bipolar (BP) PRS and markers of bipolar disorder vary considerably between studies, with studies of population and case-control samples identifying many effects that cannot be replicated in case-only analyses. These analyses demonstrate that the heterogeneity in studies of the BP PRS is driven by selection bias. Specifically, selecting samples on the basis of diagnostic status or other phenotypes associated with genetic risk attenuates the correlation between the BP and SZ PRS. In such high-severity samples, while the SZ PRS predicts poor outcomes, the BP PRS predicts better outcomes. These findings highlight the importance of understanding the impact of selection bias in translational research evaluating PRS as biomarkers of psychiatric disorders, particularly when the intended application is populations enriched for high levels of genetic risk. Biological sciences/Genetics Biological sciences/Psychology Figures Figure 2 Introduction Polygenic risk scores (PRS) have promise as biomarkers for psychiatric disorders. PRS are the weighted sum of single nucleotide polymorphisms (SNPs) carried by an individual, quantifying an individual’s predisposition to a given phenotype explained by common genetic variants ( 1 – 3 ). PRS have proved to be a clinically useful biomarker of inflammatory bowel disease, coronary artery disease, breast cancer, and type II diabetes ( 4 – 6 ). Women of European descent with PRS in the top percentile have 3-fold increased risk of breast cancer compared to those 40-60th percentile ( 7 ). PRS for coronary artery disease provide incremental predictive utility relative to well-validated clinical prediction models ( 8 , 9 ). With the advancement of new low-cost and time-efficient genotyping techniques, practical barriers to the deployment of PRS are eroding, and there is hope that PRS may eventually improve the prediction diagnosis and treatment of psychiatric illness ( 10 ). In population and case-control studies, the bipolar disorder polygenic risk score (BP PRS) has been consistently associated with diagnostic status and its correlates (see Table 1 for a review of this literature). The BP PRS discriminates between controls and both individuals at risk for psychosis and with psychotic disorders ( 11 ). Multiple studies identify an association between the BP PRS and intelligence in population samples ( 12 – 14 ), with a similar trend-level effect observed in older adults ( 15 ). Associations with clinical characteristics of bipolar disorder are less consistent ( 14 , 16 ), likely due to low base-rates of these phenotypes in the population. Overall, of 25 population-based or case control studies in Table 1, 16 (64%) reported significant results in the anticipated direction. In contrast to population and case-control samples, in clinical cohorts, associations between the BP PRS and correlates of the disorder have been inconsistent. In clinical cohorts, the BP PRS differentiates diagnostic categories ( 17 ), and number of hospitalizations ( 18 ). The BP PRS has also been associated with positive formal thought disorder and mania ( 19 , 20 ). However, a majority of studies in cases have observed no effect of the BP PRS on clinical outcomes (6 of 22 studies listed in Table 1 report significant effects, 27%). For example, studies of clinical samples have detected no association between the BP PRS and general intelligence ( 21 ), age of onset ( 22 ), response to lithium ( 23 ), suicide attempts ( 24 ), or clinical subgroups derived from symptom and cognitive profiles ( 25 ). A number of studies report effects in the direction opposite of what clinical research would hypothesize. In one analysis, the BP PRS was inversely associated with rapid cycling ( 26 ), such that those with more genetic risk were less likely to experience rapid cycling, and in another sample the BP PRS was associated with less severe depression symptoms ( 20 ), despite the common comorbidity of mania and depression. In another large study of cases, a higher BP PRS was associated with higher odds of remission between episodes, and better functioning ( 27 ). In general, results linking the BP PRS to phenotypes associated with bipolar disorder are more robust in population and case-control samples than among cases. This trend is highlighted in two case-control studies in which analyses were conducted in both the full sample and in case-only analyses. The BP PRS predicted both affective and non-affective diagnoses in the full sample of both cases and controls, but among cases was associated with lower odds of bipolar disorder relative to major depression with psychotic features ( 28 ). Similarly, the BP PRS was sensitive to differences in cognitive trajectories in a case-control sample, but not among cases ( 29 ). Both studies had a large number of cases (N > 800), making it unlikely the lack of significant effects is an issue of statistical power. Notably, in many of these and other case-only samples, the schizophrenia (SZ) PRS remained significantly associated with poor clinical ( 28 , 30 – 32 ) outcomes. A lack of statistical power also fails to account for reverse effects observed in a number of studies. In general, performance of the SZ PRS has been more consistent than that of the BP PRS, even when considering the size of the discovery and validation samples, indicating that statistical power alone does not explain the discrepancy. We hypothesize that inconsistent associations between the BP PRS and clinical outcomes reflects selection bias. Selection bias occurs when the process of sample selection changes the distribution of a parameter of interest. In the context of genetic prediction, this may occur when samples are selected on the basis of phenotypes that reflect genetic risk, such as in first-admission or case-only studies. Selection therefore changes the distribution of genetic risk, and subsequently, associations between PRSs and outcomes (see Fig. 1). Selection bias has been shown to affect associations between PRSs and phenotype in both simulations ( 33 ) and in electronic medical record data ( 34 ). However, prior analyses have not explored how these effects might impact the potential use of PRS for prognostication in first-admission psychosis. We focus on the BP and schizophrenia (SZ) PRS because bipolar disorder and schizophrenia are closely correlated genetically (rg = 0.68 36), and both the BP and SZ PRS increase the odds of being selected into a first-admission sample. For these reasons we hypothesize that 1) selection on clinical status attenuates the correlation between the schizophrenia (SZ) and BP PRS (Fig. 1A and 1B), and; 2) because bipolar disorder has a more favorable course than schizophrenia, selection on case status markedly changes, and can even reverse the effect of BP PRS on outcomes (Fig. 1C). We investigate how selection impacts the correlation between the SZ and BP PRS in three independent samples. We also demonstrate that the association between the BP and SZ PRS and clinical outcomes is affected by sample selection in a longitudinal cohort of first-admission psychosis and demographically matched, never-psychotic controls. Methods UK Biobank Data Subjects Data are drawn from UK Biobank, a population-based cohort of approximately 500,000 individuals recruited from the UK ( 35 ). Participants were between 40–69 years old during the recruitment phase, which spanned 2006–2010. All participants provided written informed consent. The protocol was approved by the North West Multi-Centre Ethics Committee. These analyses were conducted under UK Biobank project 55741. Phenotypes Since participants were recruited through their registration with the National Health Service, all UK Biobank participants had linked medical record data. Psychotic disorder diagnoses were abstracted from inpatient hospitalizations (fields 41202 and 41203 for primary diagnoses and 41204 and 41205 for secondary diagnoses), death certificates (fields 40001 and 40002), and self-reported diagnoses (fields 20002 and 20544). Participants were defined as hospitalized for a psychotic disorder if a psychotic disorder diagnostic code (including schizophrenia spectrum disorders, bipolar disorder with psychosis, major depression with psychosis, delusional disorder, or substance induced psychotic disorder) appeared on a primary inpatient hospital record. A broader index of severity was operationalized as the total number of times a psychotic disorder diagnosis appeared across any of the six diagnostic variables listed. Genotypes Genotyping and quality control procedures for UK Biobank are described in ( 36 ). In summary, participants were genotyped on either the Applied Biosystems UK BiLEVE Axiom Array or the Applied Biosystems UK Biobank Axiom Array by Affymetrix. DNA was extracted from whole blood collected on participants’ visit to the UK Biobank assessment center. Samples were genotyped in 106 batches, and checks were performed to exclude variants (0.97%) that differed between arrays or batches. Samples with a high degree of missingness, heterozygosity, or sex aneuploidy were excluded (0.3%). Principal components analysis was used to identify a subset of individuals with relatively similar ancestry, who self-reported as white British (fields 22020 and 22006, respectively). This subsample of 337,426 unrelated individuals were used in the reported analyses. Genotype data provided by UK Biobank were filtered to exclude rare variants (MAF 2% missing), and variants out of Hardy-Weinberg equilibrium (HWE < 5 x 10 − 6 ). PRS were calculated from the largest schizophrenia and bipolar GWAS available ( 37 , 38 ), from which UK Biobank data was excluded. The two sets of summary statistics were first reduced to a set of 7,562,571 variants that appeared in both GWAS. Genotype data were clumped, removing variants correlated > 0.1 within a 250 kilobase window, based on MAF rather than p-value to avoid prioritizing one PRS over the other. PRS were calculated based on the intersection of 222,416 variants appearing in both GWAS summary stats and cleaned genotype calls. PRS were regressed on the first 10 principal components of population stratification, and the residuals normalized for subsequent analyses. PsyCourse Data Subjects Replication data are drawn from the PsyCourse Study, a longitudinal study of individuals with severe mental disorders from the psychotic-to-affective continuum and community controls conducted from a network of sites in Germany and Austria ( 39 ). The study protocol was approved by the respective ethics committee for each study center, and written informed consent was obtained from each study participant. The sample used in this study consists of 1308 individuals with a schizophrenia spectrum diagnosis (schizophrenia, schizoaffective disorder, or brief psychotic disorder), bipolar disorder I and II, or recurrent major depression, and 466 control individuals. For details see Heilbronner and colleagues ( 40 ). Phenotypes Participants were assessed four times at 6-month intervals, covering the 18 months following enrollment. Participants’ current level of psychiatric treatment was assessed at each follow-up (variables v1_cur_psy_trm, v2_cur_psy_trm, v3_cur_psy_trm, and v4_cur_psy_trm). This variable was coded as 1 if a participant received inpatient care, and 0 otherwise, and summed across follow-ups, creating an ordinal variable reflecting the cumulative number of assessments at which the participant was receiving inpatient treatment. Among cases, 44.8% had experienced one or more psychiatric hospitalizations during the study interval, and 4.4% had experienced two or more psychiatric hospitalizations. Among cases, 67.2% had experienced psychosis as assessed by the SCID-IV ( 41 ). Genotypes Genetic data was derived from venous blood. Samples were genotyped on the Infinium Global Screening Array (versions 1 and 3). After standard quality control procedures, genotypes were imputed against the 1000 Genomes Phase 3 reference panel ( 42 ) using SHAPEIT2 and IMPUTE2 ( 44 , 45 ). Variants with imputation quality < 0.8 were not included in downstream analyses. Principal components of genetic covariance were computed using Plink 1.9 ( 43 ). Individuals more than three standard deviations from the mean on any of the first three principal components of ancestry were excluded from the analysis, yielding a final sample size of 1,594 (1190 cases and 404 controls). SZ and BP PRS were estimated using PRSice ( 44 ), based on the same schizophrenia and bipolar GWAS as the Suffolk County sample ( 37 , 38 ). PRS were regressed on the first 10 principal components of population stratification, and the residuals normalized for subsequent analyses. Suffolk County Data Subjects Data are drawn from the Suffolk Health County Mental Health Project, a first-admission psychosis cohort recruited between 1989 and 1995 from all 12 inpatient psychiatric units located in Suffolk County, New York ( 45 ). The baseline wave includes 628 participants, representing a 72% response rate among eligible individuals. The inclusion criteria were: first admission for psychosis in the past 6 months, English-language comprehension, residence in Suffolk County, IQ > 70, and age 15 to 60 years. The Stony Brook University Committee on Research Involving Human Subjects and the participating hospital’s review boards authorized the study every year. Written consent was obtained from all the participants or parents of a participant in case of a minor aged from 15–17 years. At the 20-year follow-up, a sample of 261 demographically matched, never-psychotic adults was recruited from the same zip codes as cases. The present analyses are based on DNA collected at the 20-year follow-up. Supplemental Table 1 reports a comparison between those cases who were genotyped and those who were not (all never-psychotic controls were genotyped). These groups did not differ except in terms of age, as older participants were more likely to have died prior to the 20-year follow-up, and therefore were not genotyped. Outcomes were assessed at the 25-year follow-up. Ratings were made based on all data collected during the follow-up assessment, including the structured clinical interview, neuropsychological assessment, collateral interview, and review of medical records. Phenotypes Remission. Symptomatic remission was operationalized according Andreasen’s definition ( 46 ). Symptom severity was measured using the Scale for the Assessment of Positive Symptoms (SAPS; 47) and the Scale for the Assessment of Negative Symptoms (SANS; 48). To be considered remitted, the following symptoms had to be rated as mild of better (2 on a scale where 0 corresponds to no symptoms and 5 corresponds to severe symptoms): hallucinations, delusions, bizarre behavior, formal thought disorder, affective flattening, alogia, avolition-apathy, and anhedonia-asociality. Whereas Andreasen’s definition stipulates symptoms be rated over the past 6 months, symptoms in this sample were rated over the past month. Recovery. Recovery was operationalized according to Liberman’s definition ( 49 ). Symptoms were assessed using eight ratings from the Brief Psychiatric Rating Scale (BPRS; 50). For a participant to be defined as “recovered”, they had to have scores of “moderate” or better on BPRS ratings of conceptual disorganization, mannerisms and posturing, grandiosity, suspiciousness, hallucinatory behavior, unusual thought content, blunted affect, and emotional withdrawal. Psychosocial functioning was assessed based on two ratings from the Quality of Life in Schizophrenia scale (QLS; 51). Participants needed a rating of three or higher on ratings of social activity and accomplishment to be considered psychosocially recovered. Being employed part-time or being a part-time student was also considered evidence of adequate role function. Symptoms and functioning were assessed over the past month, rather than the prior two years as Liberman’s definition stipulates. Global Assessment of Functioning . Global Assessment of Functioning (GAF) was measured by consensus rating of psychiatrists using all available information. GAF ratings were made for the best month of the year preceding the interview. Symptom Severity. Symptom severity at the 25-year follow up was assessed using the Scale for the Assessment of Negative Symptoms (SANS; 48) and Scale for the Assessment of Positive Symptoms (SAPS; 47). Items from the SAPS and SANS were scored into 4 factor-analytically derived subscales, described in Kotov ( 52 ). Reliability of these subscales was high (α for reality distortion=[0.82–0.85], disorganization=[0.70–0.77], apathy/asociality=[0.78–0.82], inexpressivity=[0.84–0.88]). Role Functioning. Role functioning was assessed using item 4, level of accomplishment, from the QLS. Activities of Daily Living. Ability to manage day-to-day activities was measured using the University of California San Diego Performance-Based Skill Assessment (UPSA), an behavioral test of community living ( 53 ). Participants role-play tasks such as making daily and urgent calls, performing shopping tasks, money manipulation, and reading maps and schedules. Total scores range from 1–100 points. The distribution of scores was left-skewed, so scores were inversed to meet the assumptions of the negative binomial regression model (described in Statistical Analyses, below). Social Function. Social functioning was quantified as a composite of three items from the QLS ( 54 ). Ratings used to derive the social functioning composite included social activity, social sexual relationships, and relationships with friends. The composite score ranged from 0 (worst functioning) to 17 (best functioning). Reliability of the composite was 0.73. Residential & Economic Independence. Residential independence was defined as not being reliant on agencies or other people to arrange for a residence. Economic independence was similarly defined as not being reliant on another person or agency to maintain one’s finances. Genotypes DNA was extracted from peripheral lymphocytes and genotyped using the Illumina PsychArray-8 platform containing 571,054 markers. Standard quality control procedures were performed to exclude SNPs with minor allele frequency (MAF) 5%, Hardy-Weinberg equilibrium p .20, in which case the relative with the lower call rate was dropped). SNP imputation was conducted with Beagle version 5.4 ( 55 ), against the full 1000 Genomes phase 3 reference panel ( 42 ). The imputed SNPs underwent another round of quality control in which SNPs with missing data > 10%, Hardy-Weinberg equilibrium p < 10 − 6, and imputation information score < 0.8 were excluded, yielding 6.58M high quality biallelic SNPs. The schizophrenia polygenic risk score (PRS) was calculated based on summary statistics from Trubetskoy and colleagues ( 38 ). The BP PRS was calculated based on summary statistics from Mullins and colleagues ( 37 ). A schizophrenia versus bipolar PRS was calculated based results on from Ruderfer and colleagues ( 56 ). This score was reversed so that higher scores indicate greater risk for bipolar disorder and lower scores indicate greater risk for schizophrenia. PRS were estimated using PRS-ice ( 57 ), with variants correlated > 0.1 within a 500 kilobase window clumped according to p-value. The resulting BP and SZ PRS were based on 167,552 and 167,764 variants, respectively. The first ten principal components of ancestry were estimated using Plink 1.9 ( 43 ) and ancestry was assessed using ADMIXTURE ( 58 ). Analyses were restricted to participants of European ancestry, defined as being within the first 3 standard deviations of the first ten principal components (PCs), and participants' EUR ancestry score < 0.8. PRS were regressed on the first 10 principal components of population stratification, and the residuals normalized for subsequent analyses. Statistical Analysis In all samples, demographic differences were analyzed using the Student t-test for continuous variables and the chi-square test for categorical variables. The association between the SZ and BP PRS and factors on which samples were stratified was evaluated using t-tests between diagnostic groups, with the effect size expressed as Cohen’s d, and Pearson correlations ( r ) between PRS and ordinal phenotypes. We evaluated the effect of sample stratification on the correlation between the SZ and BP PRS by estimating Pearson correlations between the two PRS. In UK Biobank, this correlation was estimated in the full sample (N = 322,774), individuals with a primary diagnosis of a psychotic disorder from an inpatient encounter (N = 934), and individuals with at least 1 (N = 1,695), 2 (N = 643), 3 (N = 264), or 4 (N = 100) psychotic disorder diagnoses recorded in the electronic medical record, death certificate, or by self-report. In PsyCourse, this correlation was estimated in the combined case-control sample (N = 1,594), among those who experienced psychotic symptoms and had at least one (N = 427) and at least two (N = 55) psychiatric hospitalizations. In data from the Suffolk County Mental Health Project, associations between 25-year outcomes and genetic risk were analyzed using Poisson and negative binomial regression for count outcomes. When exponentiated, regression coefficients from these models are equivalent to relative risks (RR), and are reported as such. Logistic regression was used for binary outcomes. Exponentiated regression coefficients from logistic regressions are equivalent to odds ratios (OR), and are reported as such. All ratios are reported with 95% confidence intervals. All regressions were univariate regressions. Results Effect of selection bias on correlation of SZ and BP PRS The effect of selection on the correlation between the SZ and BP PRS was evaluated in UK Biobank and replicated in data from the PsyCourse study. The UK Biobank analysis sample was 53.8% female (N female = 173,807) and age 56.9 (SD = 8.0) at recruitment, on average. Both the BP and SZ PRS were associated with a primary inpatient diagnosis of a psychotic disorder (Cohen’s d = 0.37 and 0.46 respectively, both p < 0.01), as well as the total number of psychotic disorder diagnoses recorded in the medical record, on death certificates, and by self-report ( r = 0.02 and 0.03 respectively, both p < 0.01). The correlation between the BP and SZ PRS in the full sample (N = 322,774) was 0.33. Among those with a primary inpatient psychotic disorder diagnosis the correlation was 0.25 (N = 934; 95% CI 0.19–0.31). Among those with any psychotic disorder diagnosis, the correlation was 0.29 (N = 1,695; [0.25–0.34]). Among those with two or more recorded diagnoses, the correlation was 0.24 (N = 643; [0.16–0.31]), among those with three or more recorded diagnoses was 0.25 (N = 264; [0.14–0.36]), and was lowest among those with four or more diagnoses (N = 100; r = 0.13; [-0.07-0.32]). Demographics of the PsyCourse Study are reported in Table 2. Both the SZ and BP PRS were correlated with the number of psychiatric hospitalizations ( r = 0.08 and 0.03, respectively). Within the combined sample of cases and controls, the SZ and BP PRS were closely correlated (N = 1,594, r = 0.50 [0.46–0.53]). The strength of this association among those who had a lifetime history of psychosis and had at least one psychiatric hospitalization was 0.48 (N = 427, [0.38–0.54]). This correlation was lowest among those with a history of psychosis and two or more psychiatric hospitalizations over the 18 months of the study (N = 55, r = 0.33 [0.07–0.60]). Demographic and clinical characteristics of the Suffolk County Mental Health Project participants are reported in Table 3. Cases were two years younger than controls, on average. There were no statistically significant differences in gender, race, or socioeconomic status. Cases had more severe symptoms and worse functioning on all outcomes included in the analysis. In the case-control cohort, the correlation between the SZ and BP PRS was 0.40, which decreased r = 0.20 in cases only. Similarly, variance of the schizophrenia PRS decreased from 1 to 0.74. Variance of the BP PRS did not change (1.00 to 1.01). Table 4 reports the association between 25-year outcomes and the schizophrenia (SZ) and bipolar (BP) PRS in the combined sample of cases and controls (right side). Effect sizes are exponentiated beta coefficients, for which a value greater than one indicates a positive association and a value less than one indicates a negative association. The SZ PRS was associated with lower odds of recovery (OR = 0.61, 95% CI=[0.47–0.78]) and remission (OR = 0.52 [0.40–0.68]), and reduced odds of attaining residential (OR = 0.58 [0.43–0.79]) and economic (OR = 0.59 [0.46–0.76]) independence at the 25-year follow-up. The SZ PRS predicted poorer functioning and more severe symptoms across all domains assessed. The BP PRS had a narrower range of associations but was associated with lower odds of remission (OR = 0.77 [0.61–0.98]) and economic independence (OR = 0.79 [0.63–0.99]), poorer functioning (GAF; RR = 0.95 [0.91–0.99]), and more severe psychotic symptoms (RR = 1.54 [1.09–2.23]). Associations between genetic risk and 25-year outcomes estimated among cases only are reported on the left side of Table 4. In this sample, the SZ PRS still predicts lower odds of remission (OR = 0.61 [0.41–0.89]), but other associations were no longer statistically significant. Associations between BP PRS and 25-year outcomes changed directions for all outcomes except inexpressivity. Among cases, the BP PRS was associated with increased odds of economic independence (OR = 1.46 [1.03–2.06]) and better social performance (RR = 1.11 [1.04–1.19]). The impact of sample stratification on genetic associations is depicted in Figs. 1A and 1B. Supplemental Table 2 reports the results of a sensitivity analysis estimating associations between the case-case BP versus SZ PRS on 25-year outcomes. The observed pattern of results is similar to that of the BP PRS, with greater BP versus SZ PRS associated with better 25-year outcomes among cases. Discussion We assessed the degree to which selection bias impacts the association between the BP and SZ PRS when selection is based on phenotypes that are themselves impacted by genetic risk. In two samples, selecting samples on the basis of a psychotic disorder diagnosis or correlates thereof significantly attenuated the association between the BP and SZ PRS. We furthermore evaluated the impact of selection bias on the associations between the BP and SZ PRS and an array of outcomes in a first-admission psychosis cohort followed for 25 years and demographically matched, never-psychotic controls, in order to understand how this phenomenon impacts the potential clinical application of PRS in first-admission psychosis. In the combined sample of both cases and never-psychotic controls assessed 25 years after first admission, both the SZ and BP PRS were associated with more severe symptoms and worse outcomes. However, when analyses were limited to cases, these associations were dramatically attenuated. In some cases, the direction of the BP PRS effects reversed, with greater genetic risk predicting better outcomes. Below, we describe how these results are consistent with selection bias, and the implications for clinical translation of PRS in psychiatry. Selection bias occurs when the process of sample ascertainment changes the distribution of the parameter of interest. In two independent samples, both the SZ and BP PRS were associated with psychotic disorder diagnoses and hospitalization for psychosis. Selection based on these phenotypes therefore truncates the joint distribution of the SZ and BP PRS, attenuating the correlation between these two predictors, (see Fig. 1 for a schematic representation). The correlation between the SZ and BP PRS was shown to decrease as a function of the number of inpatient hospitalizations (from r = 0.50 in the case-control sample to 0.33 among those with multiple hospitalizations), and as a function of the number of documented psychotic disorder diagnoses (from r = 0.33 to 0.13 among those with four or more diagnoses). One consequence of this selection effect is that, whereas the BP PRS indicates greater genetic risk for psychosis broadly in the case-control sample, in the case-only sample it functions similarly to the case-case BP versus SZ PRS (see Supplemental Table 2), indicating greater genetic risk specifically for bipolar disorder, conditional on high genetic risk for psychosis generally. Since bipolar disorder has the most favorable outcomes among psychotic disorders ( 59 – 63 ), the BP PRS predicts better outcomes in the context of a cohort with a high genetic loading for psychosis broadly. These findings may explain the inconsistent associations between the BP PRS and outcomes in clinical samples (see Table 1 for a summary of this literature). Associations that are significant in the general population may be non-significant ( 33 ), or even reversed, among cases. The same dynamics likely affect analyses of genetic structure. Genetic correlations are likely to be attenuated among case-only samples, relative to case-control or population-based samples ( 64 , 65 ). Researchers should be especially cautious in drawing etiological inferences based on studies of cases only, as selection bias may obscure true causal pathways, if they exist, or reverse the direction of the effect. Research investigating etiological mechanisms of genetic risk are best performed in samples that capture the full spectrum of genetic risk. To mitigate the impact of selection bias, researchers should carefully consider and account for causal relationships among the variables under investigation. Sensitivity analyses, causal modeling, and cautious interpretation of results can help minimize the impact of selection bias on etiological research when population samples are unavailable or infeasible. That selection bias impacts genetic prediction should not be taken to mean PRS cannot be useful in case-only samples. While etiological research is confounded by selection bias, predictive models are not, even though the direction of effects may be counterintuitive. Translational research on potential clinical applications of PRS need only ensure that the same forms of selection occur in both research cohorts and clinical settings. These results indicate, for example, that a relatively high BP PRS may be a marker of good prognosis among those admitted for first-episode psychosis. Similarly, even though inferences about genetic mechanisms of treatment resistance are likely confounded by sampling bias, SZ and BP PRS may be accurate predictors of response to treatment in clinical cohorts ( 32 , 66 ). Limitations This research is limited in three ways. First, the Suffolk County sample is relatively small, which may have prevented the detection of small effects. While longitudinal analyses could not be replicated in an independent sample due to lack of comparable data, the general effect of sample stratification on genetic risk were replicated. Second, prior research has demonstrated that UK Biobank participants are healthier than the UK population, and that this can bias genetic correlations ( 67 ). Since the effects observed in UK Biobank were replicated in PsyCourse, it does not appear that this has impact the conclusions that should be drawn from the analyses. Lastly, given the predominantly European ancestry composition of all three samples, the analytical scope of this study was limited to individuals of European ancestry. GWAS in other ancestries samples are available, facilitating the extension of these results to other ancestries. Conclusion Selection bias substantially impacts the association between schizophrenia and bipolar disorder PRS, as well as the association between PRS and clinical outcomes. Whereas greater genetic risk is associated with worse outcomes in a case-control sample, genetic risk for bipolar disorder predicts better outcomes among cases only. Selection bias may explain the heterogeneity of effects linking the bipolar PRS to clinical outcomes. It also complicates the use of PRS for both etiological and translational research. Declarations Availability of Data and Materials Statement Data from UK Biobank are available to qualified researchers. The application process is outlined at https://community.ukbiobank.ac.uk/hc/en-gb/categories/14494598931229-Enable-your-research. Data from PsyCourse are available to qualified researchers. The application process is outlined at http://www.psycourse.de/index-en.html. Data from the Suffolk County Mental Health Project are available from the NIMH Data Archive, collection number 2477. Analytic syntax is available from the corresponding author upon request. Conflict of Interest The authors would like to state there are no financial conflicts of interest associated with the work described. Acknowledgments This research was supported by National Institutes of Health (MH44801, MH094398, MH110434), and a NARSAD Young Investigator Grant to R.K. This study received funding from the National Institutes of Health under grant number R21MH123908, awarded to K. J. The authors gratefully acknowledge the support of the participants and mental health community of Suffolk County for contributing their time and energy to this project. They are also indebted to the study coordinators for their dedicated efforts, the interviewers for their careful assessments, and the psychiatrists who derived the consensus diagnoses. This research has been conducted using data from UK Biobank, a major biomedical database, under project ID 55741. U.H. was supported by European Union’s Horizon 2020 Research and Innovation Program (PSY-PGx, grant agreement No 945151) and the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation, project number 514201724). Thomas G. Schulze was supported by the Deutsche Forschungsgemeinschaft (KFO241/PsyCourse, SCHU 1603/4-1, 5-1, 7-1), the German Ministry of Education and Research (IntegraMent: 01ZX1614K; BipoLife: 01EE1404H; the German Center for Mental Health [DZPG]: 01EE2303A/01EE2303F), and the European Union (ERA-NET NEURON - MulioBio: 01EW2009; GEPI-BIOPSY: 01EW2005). Supplementary information is available at MP’s website References Fullerton JM, Nurnberger JI. Polygenic risk scores in psychiatry: Will they be useful for clinicians? F1000Research. 2019;8:F1000 Faculty Rev-1293. Lewis CM, Vassos E. Polygenic risk scores: from research tools to clinical instruments. Genome Med. 2020;12:44. Wray NR, Lin T, Austin J, McGrath JJ, Hickie IB, Murray GK, et al. From Basic Science to Clinical Application of Polygenic Risk Scores: A Primer. JAMA Psychiatry. 2021;78(1):101–9. Khera AV, Chaffin M, Aragam KG, Haas ME, Roselli C, Choi SH, et al. Genome-wide polygenic scores for common diseases identify individuals with risk equivalent to monogenic mutations. Nat Genet. 2018;50(9):1219–24. Lambert SA, Abraham G, Inouye M. Towards clinical utility of polygenic risk scores. Hum Mol Genet. 2019;28(R2):R133–42. Mars N, Koskela JT, Ripatti P, Kiiskinen TTJ, Havulinna AS, Lindbohm JV, et al. Polygenic and clinical risk scores and their impact on age at onset and prediction of cardiometabolic diseases and common cancers. Nat Med. 2020;26(4):549–57. Mavaddat N, Michailidou K, Dennis J, Lush M, Fachal L, Lee A, et al. Polygenic Risk Scores for Prediction of Breast Cancer and Breast Cancer Subtypes. Am J Hum Genet. 2019;104(1):21–34. Riveros-Mckay F, Weale ME, Moore R, Selzam S, Krapohl E, Sivley RM, et al. Integrated Polygenic Tool Substantially Enhances Coronary Artery Disease Prediction. Circ Genomic Precis Med. 2021;14(2):e003304. Weale ME, Riveros-Mckay F, Selzam S, Seth P, Moore R, Tarran WA, et al. Validation of an Integrated Risk Tool, Including Polygenic Risk Score, for Atherosclerotic Cardiovascular Disease in Multiple Ethnicities and Ancestries. Am J Cardiol. 2021;148:157–64. Anderson JS, Shade J, DiBlasi E, Shabalin AA, Docherty AR. Polygenic Risk Scoring and Prediction of Mental Health Outcomes. Curr Opin Psychol. 2019;27:77–81. Smigielski L, Papiol S, Theodoridou A, Heekeren K, Gerstenberg M, Wotruba D, et al. Polygenic risk scores across the extended psychosis spectrum. Transl Psychiatry. 2021;11(1):1–11. Gui Y, Zhou X, Wang Z, Zhang Y, Wang Z, Zhou G, et al. Sex-specific genetic association between psychiatric disorders and cognition, behavior and brain imaging in children and adults. Transl Psychiatry. 2022;12(1):1–8. Loughnan RJ, Palmer CE, Makowski C, Thompson WK, Barch DM, Jernigan TL, et al. Unique prediction of developmental psychopathology from genetic and familial risk. J Child Psychol Psychiatry. 2022;63(12):1631–43. Mistry S, Escott-Price V, Florio AD, Smith DJ, Zammit S. Genetic risk for bipolar disorder and psychopathology from childhood to early adulthood. J Affect Disord. 2019;246:633–9. Liebers DT, Pirooznia M, Seiffudin F, Musliner KL, Zandi PP, Goes FS. Polygenic Risk of Schizophrenia and Cognition in a Population-Based Survey of Older Adults. Schizophr Bull. 2016;42(4):984–91. Hosang GM, Martin J, Karlsson R, Lundström S, Larsson H, Ronald A, et al. Association of Etiological Factors for Hypomanic Symptoms, Bipolar Disorder, and Other Severe Mental Illnesses. JAMA Psychiatry. 2022;79(2):143–50. Liebers DT, Pirooznia M, Ganna A, Study (BiGS) BG, Goes FS. Discriminating bipolar depression from major depressive disorder with polygenic risk scores. Psychol Med. 2021;51(9):1451–8. Kalman JL, Papiol S, Grigoroiu-Serbanescu M, Adorjan K, Anderson-Schmidt H, Brosch K, et al. Genetic risk for psychiatric illness is associated with the number of hospitalizations of bipolar disorder patients. J Affect Disord. 2022;296:532–40. Ahangari M, Bustamante D, Kirkpatrick R, Nguyen TH, Verrelli BC, Fanous A, et al. Relationship between polygenic risk scores and symptom dimensions of schizophrenia and schizotypy in multiplex families with schizophrenia. Br J Psychiatry J Ment Sci. 2022;1–8. David FS, Stein F, Andlauer TFM, Streit F, Witt SH, Herms S, et al. Genetic contributions to transdiagnostic symptom dimensions in patients with major depressive disorder, bipolar disorder, and schizophrenia spectrum disorders. Schizophr Res. 2023;252:161–71. Richards AL, Pardiñas AF, Frizzati A, Tansey KE, Lynham AJ, Holmans P, et al. The Relationship Between Polygenic Risk Scores and Cognition in Schizophrenia. Schizophr Bull. 2020;46(2):336–44. Kalman JL, Papiol S, Forstner AJ, Heilbronner U, Degenhardt F, Strohmaier J, et al. Investigating polygenic burden in age at disease onset in bipolar disorder: Findings from an international multicentric study. Bipolar Disord. 2019;21(1):68–75. Schubert KO, Thalamuthu A, Amare AT, Frank J, Streit F, Adl M, et al. Combining schizophrenia and depression polygenic risk scores improves the genetic prediction of lithium response in bipolar disorder patients. Transl Psychiatry. 2021;11(1):1–8. Lopes FL, Zhu K, Purves KL, Song C, Ahn K, Hou L, et al. Polygenic risk for anxiety influences anxiety comorbidity and suicidal behavior in bipolar disorder. Transl Psychiatry. 2020;10(1):1–7. Dwyer DB, Kalman JL, Budde M, Kambeitz J, Ruef A, Antonucci LA, et al. An Investigation of Psychosis Subgroups With Prognostic Validation and Exploration of Genetic Underpinnings: The PsyCourse Study. JAMA Psychiatry. 2020;77(5):523–33. Coombes BJ, Markota M, Mann JJ, Colby C, Stahl E, Talati A, et al. Dissecting clinical heterogeneity of bipolar disorder using multiple polygenic risk scores. Transl Psychiatry. 2020;10(1):1–8. Song J, Jonsson L, Lu Y, Bergen SE, Karlsson R, Smedler E, et al. Key subphenotypes of bipolar disorder are differentially associated with polygenic liabilities for bipolar disorder, schizophrenia, and major depressive disorder. Mol Psychiatry. 2024;1–10. Rodriguez V, Alameda L, Quattrone D, Tripoli G, Gayer-Anderson C, Spinazzola E, et al. Use of multiple polygenic risk scores for distinguishing schizophrenia-spectrum disorder and affective psychosis categories in a first-episode sample; the EU-GEI study. Psychol Med. 2022;1–10. Ferraro L, Quattrone D, La Barbera D, La Cascia C, Morgan C, Kirkbride JB, et al. First-Episode Psychosis Patients Who Deteriorated in the Premorbid Period Do Not Have Higher Polygenic Risk Scores Than Others: A Cluster Analysis of EU-GEI Data. Schizophr Bull [Internet]. 2022 Aug 10 [cited 2022 Oct 13]; Available from: https://doi.org/10.1093/schbul/sbac100 Frank J, Lang M, Witt SH, Strohmaier J, Rujescu D, Cichon S, et al. Identification of increased genetic risk scores for schizophrenia in treatment-resistant patients. Mol Psychiatry. 2015;20(2):150–1. Jonas KG, Lencz T, Li K, Malhotra AK, Perlman G, Fochtmann LJ, et al. Schizophrenia polygenic risk score and 20-year course of illness in psychotic disorders. Transl Psychiatry. 2019;9(1):300. Zhang JP, Robinson D, Yu J, Gallego J, Wolfgang Fleischhacker W, Kahn RS, et al. Schizophrenia Polygenic Risk Score as a Predictor of Antipsychotic Efficacy in First Episode Psychosis. Am J Psychiatry. 2019;176(1):21–8. Munafò MR, Tilling K, Taylor AE, Evans DM, Davey Smith G. Collider scope: when selection bias can substantially influence observed associations. Int J Epidemiol. 2018;47(1):226–35. Lee YH, Thaweethai T, Sheu YH, Feng YCA, Karlson EW, Ge T, et al. Impact of selection bias on polygenic risk score estimates in healthcare settings. Psychol Med. 2023;1–11. Sudlow C, Gallacher J, Allen N, Beral V, Burton P, Danesh J, et al. UK biobank: an open access resource for identifying the causes of a wide range of complex diseases of middle and old age. PLoS Med. 2015;12(3):e1001779. Bycroft C, Freeman C, Petkova D, Band G, Elliott LT, Sharp K, et al. The UK Biobank resource with deep phenotyping and genomic data. Nature. 2018;562(7726):203–9. Mullins N, Forstner AJ, O’Connell KS, Coombes B, Coleman JRI, Qiao Z, et al. Genome-wide association study of more than 40,000 bipolar disorder cases provides new insights into the underlying biology. Nat Genet. 2021;53(6):817–29. Trubetskoy V, Pardiñas AF, Qi T, Panagiotaropoulou G, Awasthi S, Bigdeli TB, et al. Mapping genomic loci implicates genes and synaptic biology in schizophrenia. Nature. 2022;604(7906):502–8. Budde M, Anderson-Schmidt H, Gade K, Reich‐Erkelenz D, Adorjan K, Kalman JL, et al. A longitudinal approach to biological psychiatric research: The PsyCourse study. Am J Med Genet. 2019;180(2):89–102. Heilbronner U, Adorjan K, Anderson-Schmidt H, Budde M, Comes AL, Gade K, et al. The PsyCourse Codebook, Version 6.0 [Internet]. 2023 [cited 2024 Jun 4]. Available from: https://data.ub.uni-muenchen.de/390/ First MB, Gibbon M. The Structured Clinical Interview for DSM-IV Axis I Disorders (SCID-I) and the Structured Clinical Interview for DSM-IV Axis II Disorders (SCID-II). In: Comprehensive handbook of psychological assessment, Vol 2: Personality assessment. Hoboken, NJ, US: John Wiley & Sons, Inc.; 2004. p. 134–43. 1000 Genomes Project Consortium, Abecasis GR, Auton A, Brooks LD, DePristo MA, Durbin RM, et al. An integrated map of genetic variation from 1,092 human genomes. Nature. 2012;491(7422):56–65. Purcell S, Neale B, Todd-Brown K, Thomas L, Ferreira MAR, Bender D, et al. PLINK: A Tool Set for Whole-Genome Association and Population-Based Linkage Analyses. Am J Hum Genet. 2007;81(3):559–75. Euesden J, Lewis CM, O’Reilly PF. PRSice: Polygenic Risk Score software. Bioinforma Oxf Engl. 2015;31(9):1466–8. Bromet EJ, Schwartz JE, Fennig S, Geller L, Jandorf L, Kovasznay B, et al. The epidemiology of psychosis: the Suffolk County Mental Health Project. Schizophr Bull. 1992;18(2):243–55. Andreasen NC, Carpenter WT, Kane JM, Lasser RA, Marder SR, Weinberger DR. Remission in schizophrenia: proposed criteria and rationale for consensus. Am J Psychiatry. 2005;162(3):441–9. Andreasen NC. Scale for the Assessment of Positive Symptoms: SAPS. Department of Psychiatry, College of Medicine, the University of Iowa; 1984. 40 p. Andreasen NC. The Scale for the Assessment of Negative Symptoms (SANS): Conceptual and Theoretical Foundations. Br J Psychiatry. 1989;155(S7):49–52. Liberman RP, Kopelowicz A. Recovery from schizophrenia: a concept in search of research. Psychiatr Serv Wash DC. 2005;56(6):735–42. Overall JE, Gorham DR. The Brief Psychiatric Rating Scale. Psychol Rep. 1962;10:799–812. Heinrichs DW, Hanlon TE, Carpenter WT Jr. The Quality of Life Scale: An Instrument for Rating the Schizophrenic Deficit Syndrome. Schizophr Bull. 1984;10(3):388–98. Kotov R, Foti D, Li K, Bromet EJ, Hajcak G, Ruggero CJ. Validating dimensions of psychosis symptomatology: Neural correlates and 20-year outcomes. J Abnorm Psychol. 2016;125(8):1103–19. Harvey PD, Velligan DI, Bellack AS. Performance-Based Measures of Functional Skills: Usefulness in Clinical Treatment Studies. Schizophr Bull. 2007;33(5):1138–48. Velthorst E, Fett AKJ, Reichenberg A, Perlman G, van Os J, Bromet EJ, et al. The 20-Year Longitudinal Trajectories of Social Functioning in Individuals With Psychotic Disorders. Am J Psychiatry. 2017;174(11):1075–85. Browning BL, Zhou Y, Browning SR. A One-Penny Imputed Genome from Next-Generation Reference Panels. Am J Hum Genet. 2018;103(3):338–48. Ruderfer DM, Ripke S, McQuillin A, Boocock J, Stahl EA, Pavlides JMW, et al. Genomic Dissection of Bipolar Disorder and Schizophrenia, Including 28 Subphenotypes. Cell. 2018;173(7):1705–1715.e16. Choi SW, O’Reilly PF. PRSice-2: Polygenic Risk Score software for biobank-scale data. GigaScience. 2019;8(7):giz082. Alexander DH, Novembre J, Lange K. Fast model-based estimation of ancestry in unrelated individuals. Genome Res. 2009;19(9):1655–64. Kotov R, Fochtmann L, Li K, Tanenberg-Karant M, Constantino EA, Rubinstein J, et al. Declining Clinical Course of Psychotic Disorders Over the Two Decades Following First Hospitalization: Evidence From the Suffolk County Mental Health Project. Am J Psychiatry. 2017;174(11):1064–74. Jonas K, Lian W, Callahan J, Ruggero CJ, Clouston S, Reichenberg A, et al. The Course of General Cognitive Ability in Individuals With Psychotic Disorders. JAMA Psychiatry. 2022;79(7):659–66. Tramazzo S, Lian W, Ajnakina O, Carlson G, Bromet E, Kotov R, et al. The Long-Term Course of Remission and Recovery in Psychotic Disorder [Internet]. 2022 [cited 2023 Feb 7]. Available from: https://doi.org/10.31234/osf.io/yh74r Bowie CR, Depp C, McGrath JA, Wolyniec P, Mausbach BT, Thornquist MH, et al. Prediction of Real World Functional Disability in Chronic Mental Disorders: A Comparison of Schizophrenia and Bipolar Disorder. Am J Psychiatry. 2010;167(9):1116–24. Harrow M, Grossman LS, Herbener ES, Davies EW. Ten-year outcome: patients with schizoaffective disorders, schizophrenia, affective disorders and mood-incongruent psychotic symptoms. Br J Psychiatry J Ment Sci. 2000;177:421–6. Mundy J, Hübel C, Adey BN, Davies HL, Davies MR, Coleman JRI, et al. Genetic examination of the Mood Disorder Questionnaire and its relationship with bipolar disorder. Am J Med Genet. 2023;192(7–8):147–60. Grotzinger AD, Mallard TT, Akingbuwa WA, Ip HF, Adams MJ, Lewis CM, et al. Genetic architecture of 11 major psychiatric disorders at biobehavioral, functional genomic and molecular genetic levels of analysis. Nat Genet. 2022;54(5):548–59. Cearns M, Amare AT, Schubert KO, Thalamuthu A, Frank J, Streit F, et al. Using polygenic scores and clinical data for bipolar disorder patient stratification and lithium response prediction: machine learning approach. Br J Psychiatry J Ment Sci. 2022;1–10. Schoeler T, Speed D, Porcu E, Pirastu N, Pingault JB, Kutalik Z. Participation bias in the UK Biobank distorts genetic associations and downstream analyses. Nat Hum Behav. 2023;7(7):1216–27. Lim K, Lam M, Huang H, Liu J, Lee J. Genetic liability in individuals at ultra-high risk of psychosis: A comparison study of 9 psychiatric traits. PLOS ONE. 2020;15(12):e0243104. Warren TL, Tubbs JD, Lesh TA, Corona MB, Pakzad SS, Albuquerque MD, et al. Association of neurotransmitter pathway polygenic risk with specific symptom profiles in psychosis. Mol Psychiatry [Internet]. 2024; Available from: https://doi.org/10.1038/s41380-024-02457-0 Sanchez-Ruiz JA, Coombes BJ, Pazdernik VM, Melhuish Beaupre LM, Jenkins GD, Pendegraft RS, et al. Clinical and genetic contributions to medical comorbidity in bipolar disorder: a study using electronic health records-linked biobank data. Mol Psychiatry. 2024; Mistry S, Escott-Price V, Florio AD, Smith DJ, Zammit S. Genetic risk for bipolar disorder and psychopathology from childhood to early adulthood. J Affect Disord. 2019;246:633–9. Bigdeli TB, Barr PB, Rajeevan N, Graham DP, Li Y, Meyers JL, et al. Correlates of suicidal behaviors and genetic risk among United States veterans with schizophrenia or bipolar I disorder. Mol Psychiatry. 2024; Kämpe A, Suvisaari J, Lähteenvuo M, Singh T, Ahola-Olli A, Urpa L, et al. Genetic contribution to disease-course severity and progression in the SUPER-Finland study, a cohort of 10,403 individuals with psychotic disorders. Mol Psychiatry. 2024; Wilson JE, Sealock J, Straub P, Raman R, Kipp AM, Dittus RS, et al. Exploring genetic risk for catatonia in a genome wide association study and polygenic risk score analysis. Schizophr Res. 2024;263:178–90. Askeland RB, Hannigan LJ, O’Connell KS, Corfield EC, Frei O, Thapar A, et al. Developmental manifestations of polygenic risk for bipolar disorder from infancy to middle childhood. Transl Psychiatry. 2023;13(1):222. Mistry S, Escott-Price V, Florio AD, Smith DJ, Zammit S. Investigating associations between genetic risk for bipolar disorder and cognitive functioning in childhood. J Affect Disord. 2019;259:112–20. Jiang X, Zai CC, Kennedy KG, Zou Y, Nikolova YS, Felsky D, et al. Association of polygenic risk for bipolar disorder with grey matter structure and white matter integrity in youth. Transl Psychiatry. 2023;13:322. Thiel K, Lemke H, Winter A, Flinkenflügel K, Waltemate L, Bonnekoh L, et al. White and gray matter alterations in bipolar I and bipolar II disorder subtypes compared with healthy controls - exploring associations with disease course and polygenic risk. Neuropsychopharmacol Off Publ Am Coll Neuropsychopharmacol. 2024;49(5):814–23. Tables Table 1 to 4 are available in the Supplementary Files section. Additional Declarations The authors have declared there is NO conflict of interest to disclose Supplementary Files NewTable1.xlsx MolecularPsyTable2.xlsx MolecularPsyTable3.xlsx MolecularPsyTable4.xlsx MolecularPsySupplementaryInformationTable1.xlsx MolecularPsySupplementaryInformationTable2.xlsx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-4536236","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":315997462,"identity":"4d521310-e335-43d1-9f6c-4ece0e29e869","order_by":0,"name":"Katherine Jonas","email":"data:image/png;base64,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","orcid":"https://orcid.org/0000-0002-1910-223X","institution":"Stony Brook University","correspondingAuthor":true,"prefix":"","firstName":"Katherine","middleName":"","lastName":"Jonas","suffix":""},{"id":315997463,"identity":"04397733-020c-4818-bec1-cd66ebf8dc83","order_by":1,"name":"Amna Asim","email":"","orcid":"","institution":"Stony Brook University","correspondingAuthor":false,"prefix":"","firstName":"Amna","middleName":"","lastName":"Asim","suffix":""},{"id":315997464,"identity":"83ada356-0167-4318-86d6-9efa8c4608ca","order_by":2,"name":"Yuan Yang","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Yuan","middleName":"","lastName":"Yang","suffix":""},{"id":315997465,"identity":"b23ecd41-e0ef-4f0f-b460-9314bdd3fc3b","order_by":3,"name":"Urs Heilbronner","email":"","orcid":"https://orcid.org/0000-0001-7135-762X","institution":"Institute of Psychiatric Phenomics and Genomics, University Hospital, LMU Munich","correspondingAuthor":false,"prefix":"","firstName":"Urs","middleName":"","lastName":"Heilbronner","suffix":""},{"id":315997466,"identity":"dd7f8646-1321-4564-8703-fdbccc359447","order_by":4,"name":"Thomas Schulze","email":"","orcid":"","institution":"SUNY Upstate Medical University","correspondingAuthor":false,"prefix":"","firstName":"Thomas","middleName":"","lastName":"Schulze","suffix":""},{"id":315997467,"identity":"1daf7f64-51b6-4639-9ac8-655ccdce6245","order_by":5,"name":"Todd Lencz","email":"","orcid":"https://orcid.org/0000-0001-8586-338X","institution":"Zucker School of Medicine at Hofstra/Northwell","correspondingAuthor":false,"prefix":"","firstName":"Todd","middleName":"","lastName":"Lencz","suffix":""},{"id":315997468,"identity":"1cd6c579-44b5-4569-8ce9-bcb6133eea1a","order_by":6,"name":"Evangelos Vassos","email":"","orcid":"https://orcid.org/0000-0001-6363-0438","institution":"King's College London, Institute of Psychiatry Psychology \u0026 Neuroscience","correspondingAuthor":false,"prefix":"","firstName":"Evangelos","middleName":"","lastName":"Vassos","suffix":""},{"id":315997469,"identity":"2b167c3b-c55a-4f51-9e5c-2d324989463a","order_by":7,"name":"Sean Clouston","email":"","orcid":"https://orcid.org/0000-0002-6124-0329","institution":"Stony Brook University","correspondingAuthor":false,"prefix":"","firstName":"Sean","middleName":"","lastName":"Clouston","suffix":""},{"id":315997470,"identity":"90c69ddf-45df-41b6-84ed-28f245c3d48d","order_by":8,"name":"Roman Kotov","email":"","orcid":"","institution":"Stony Brook University","correspondingAuthor":false,"prefix":"","firstName":"Roman","middleName":"","lastName":"Kotov","suffix":""}],"badges":[],"createdAt":"2024-06-05 21:20:28","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4536236/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4536236/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":60428794,"identity":"4d8bdca1-bdd2-4510-b19e-cab7a2a245e1","added_by":"auto","created_at":"2024-07-16 15:59:06","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":539436,"visible":true,"origin":"","legend":"\u003cp\u003eSee image above for figure legend\u003c/p\u003e","description":"","filename":"Figure2formanuscript.jpg","url":"https://assets-eu.researchsquare.com/files/rs-4536236/v1/c530f1f5cb08f573327f2ee5.jpg"},{"id":64608815,"identity":"e78a0659-7b6d-4dd7-b3ec-ea2bd7bd20a8","added_by":"auto","created_at":"2024-09-16 13:34:01","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1029600,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4536236/v1/399b28fe-60da-4b00-9373-a1a403f29cae.pdf"},{"id":60429712,"identity":"4c979091-5018-4f19-834d-17c22d92d561","added_by":"auto","created_at":"2024-07-16 16:07:06","extension":"xlsx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":14455,"visible":true,"origin":"","legend":"","description":"","filename":"NewTable1.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-4536236/v1/a764506c55e4664400d5615e.xlsx"},{"id":60428798,"identity":"8cd959e0-42df-4508-bc8c-68b44a0b4756","added_by":"auto","created_at":"2024-07-16 15:59:06","extension":"xlsx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":10705,"visible":true,"origin":"","legend":"","description":"","filename":"MolecularPsyTable2.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-4536236/v1/309c8770b58686345f40665c.xlsx"},{"id":60428801,"identity":"f9d03719-3e57-4c19-b7d6-5f9f480bdef9","added_by":"auto","created_at":"2024-07-16 15:59:07","extension":"xlsx","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":11640,"visible":true,"origin":"","legend":"","description":"","filename":"MolecularPsyTable3.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-4536236/v1/0095aeffb6469cc288e4baf0.xlsx"},{"id":60429711,"identity":"20e4a00f-7283-47ad-ad58-f48b1bb35e21","added_by":"auto","created_at":"2024-07-16 16:07:06","extension":"xlsx","order_by":4,"title":"","display":"","copyAsset":false,"role":"supplement","size":11362,"visible":true,"origin":"","legend":"","description":"","filename":"MolecularPsyTable4.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-4536236/v1/5147cd1f071e1a16bbcf876d.xlsx"},{"id":60428799,"identity":"7efa8ff5-641d-4d70-bbf9-02a8d7e3f765","added_by":"auto","created_at":"2024-07-16 15:59:06","extension":"xlsx","order_by":5,"title":"","display":"","copyAsset":false,"role":"supplement","size":12216,"visible":true,"origin":"","legend":"","description":"","filename":"MolecularPsySupplementaryInformationTable1.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-4536236/v1/63b7e5236dd39f9d2498fa55.xlsx"},{"id":60428802,"identity":"e1081c7a-dd0a-4a8c-9809-0c2a0d9f0c3f","added_by":"auto","created_at":"2024-07-16 15:59:07","extension":"xlsx","order_by":6,"title":"","display":"","copyAsset":false,"role":"supplement","size":10706,"visible":true,"origin":"","legend":"","description":"","filename":"MolecularPsySupplementaryInformationTable2.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-4536236/v1/043168cee911756a51062ee9.xlsx"}],"financialInterests":"The authors have declared there is \u003cb\u003eNO\u003c/b\u003e conflict of interest to disclose","formattedTitle":"The Impact of Selection Bias on Genetic Prediction Using the Bipolar Polygenic Risk Score in First-Admission Psychosis","fulltext":[{"header":"Introduction","content":"\u003cp\u003ePolygenic risk scores (PRS) have promise as biomarkers for psychiatric disorders. PRS are the weighted sum of single nucleotide polymorphisms (SNPs) carried by an individual, quantifying an individual\u0026rsquo;s predisposition to a given phenotype explained by common genetic variants (\u003cspan additionalcitationids=\"CR2\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e). PRS have proved to be a clinically useful biomarker of inflammatory bowel disease, coronary artery disease, breast cancer, and type II diabetes (\u003cspan additionalcitationids=\"CR5\" citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e). Women of European descent with PRS in the top percentile have 3-fold increased risk of breast cancer compared to those 40-60th percentile (\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e). PRS for coronary artery disease provide incremental predictive utility relative to well-validated clinical prediction models (\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e). With the advancement of new low-cost and time-efficient genotyping techniques, practical barriers to the deployment of PRS are eroding, and there is hope that PRS may eventually improve the prediction diagnosis and treatment of psychiatric illness (\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eIn population and case-control studies, the bipolar disorder polygenic risk score (BP PRS) has been consistently associated with diagnostic status and its correlates (see Table\u0026nbsp;1 for a review of this literature). The BP PRS discriminates between controls and both individuals at risk for psychosis and with psychotic disorders (\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e). Multiple studies identify an association between the BP PRS and intelligence in population samples (\u003cspan additionalcitationids=\"CR13\" citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e), with a similar trend-level effect observed in older adults (\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e). Associations with clinical characteristics of bipolar disorder are less consistent (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e), likely due to low base-rates of these phenotypes in the population. Overall, of 25 population-based or case control studies in Table\u0026nbsp;1, 16 (64%) reported significant results in the anticipated direction.\u003c/p\u003e \u003cp\u003eIn contrast to population and case-control samples, in clinical cohorts, associations between the BP PRS and correlates of the disorder have been inconsistent. In clinical cohorts, the BP PRS differentiates diagnostic categories (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e), and number of hospitalizations (\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e). The BP PRS has also been associated with positive formal thought disorder and mania (\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e). However, a majority of studies in cases have observed no effect of the BP PRS on clinical outcomes (6 of 22 studies listed in Table\u0026nbsp;1 report significant effects, 27%). For example, studies of clinical samples have detected no association between the BP PRS and general intelligence (\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e), age of onset (\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e), response to lithium (\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e), suicide attempts (\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e), or clinical subgroups derived from symptom and cognitive profiles (\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e). A number of studies report effects in the direction opposite of what clinical research would hypothesize. In one analysis, the BP PRS was \u003cem\u003einversely\u003c/em\u003e associated with rapid cycling (\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e), such that those with more genetic risk were less likely to experience rapid cycling, and in another sample the BP PRS was associated with less severe depression symptoms (\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e), despite the common comorbidity of mania and depression. In another large study of cases, a higher BP PRS was associated with higher odds of remission between episodes, and better functioning (\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eIn general, results linking the BP PRS to phenotypes associated with bipolar disorder are more robust in population and case-control samples than among cases. This trend is highlighted in two case-control studies in which analyses were conducted in both the full sample and in case-only analyses. The BP PRS predicted both affective and non-affective diagnoses in the full sample of both cases and controls, but among cases was associated with \u003cem\u003elower\u003c/em\u003e odds of bipolar disorder relative to major depression with psychotic features (\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e). Similarly, the BP PRS was sensitive to differences in cognitive trajectories in a case-control sample, but not among cases (\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e). Both studies had a large number of cases (N\u0026thinsp;\u0026gt;\u0026thinsp;800), making it unlikely the lack of significant effects is an issue of statistical power. Notably, in many of these and other case-only samples, the schizophrenia (SZ) PRS remained significantly associated with poor clinical (\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan additionalcitationids=\"CR31\" citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e) outcomes. A lack of statistical power also fails to account for reverse effects observed in a number of studies. In general, performance of the SZ PRS has been more consistent than that of the BP PRS, even when considering the size of the discovery and validation samples, indicating that statistical power alone does not explain the discrepancy.\u003c/p\u003e \u003cp\u003eWe hypothesize that inconsistent associations between the BP PRS and clinical outcomes reflects selection bias. Selection bias occurs when the process of sample selection changes the distribution of a parameter of interest. In the context of genetic prediction, this may occur when samples are selected on the basis of phenotypes that reflect genetic risk, such as in first-admission or case-only studies. Selection therefore changes the distribution of genetic risk, and subsequently, associations between PRSs and outcomes (see Fig.\u0026nbsp;1). Selection bias has been shown to affect associations between PRSs and phenotype in both simulations (\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e) and in electronic medical record data (\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e). However, prior analyses have not explored how these effects might impact the potential use of PRS for prognostication in first-admission psychosis. We focus on the BP and schizophrenia (SZ) PRS because bipolar disorder and schizophrenia are closely correlated genetically (rg\u0026thinsp;=\u0026thinsp;0.68 36), and both the BP and SZ PRS increase the odds of being selected into a first-admission sample. For these reasons we hypothesize that 1) selection on clinical status attenuates the correlation between the schizophrenia (SZ) and BP PRS (Fig.\u0026nbsp;1A and 1B), and; 2) because bipolar disorder has a more favorable course than schizophrenia, selection on case status markedly changes, and can even reverse the effect of BP PRS on outcomes (Fig.\u0026nbsp;1C). We investigate how selection impacts the correlation between the SZ and BP PRS in three independent samples. We also demonstrate that the association between the BP and SZ PRS and clinical outcomes is affected by sample selection in a longitudinal cohort of first-admission psychosis and demographically matched, never-psychotic controls.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eUK Biobank Data\u003c/h2\u003e \u003cdiv id=\"Sec4\" class=\"Section3\"\u003e \u003ch2\u003eSubjects\u003c/h2\u003e \u003cp\u003eData are drawn from UK Biobank, a population-based cohort of approximately 500,000 individuals recruited from the UK (\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e). Participants were between 40\u0026ndash;69 years old during the recruitment phase, which spanned 2006\u0026ndash;2010. All participants provided written informed consent. The protocol was approved by the North West Multi-Centre Ethics Committee. These analyses were conducted under UK Biobank project 55741.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003ePhenotypes\u003c/h2\u003e \u003cp\u003eSince participants were recruited through their registration with the National Health Service, all UK Biobank participants had linked medical record data. Psychotic disorder diagnoses were abstracted from inpatient hospitalizations (fields 41202 and 41203 for primary diagnoses and 41204 and 41205 for secondary diagnoses), death certificates (fields 40001 and 40002), and self-reported diagnoses (fields 20002 and 20544). Participants were defined as hospitalized for a psychotic disorder if a psychotic disorder diagnostic code (including schizophrenia spectrum disorders, bipolar disorder with psychosis, major depression with psychosis, delusional disorder, or substance induced psychotic disorder) appeared on a primary inpatient hospital record. A broader index of severity was operationalized as the total number of times a psychotic disorder diagnosis appeared across any of the six diagnostic variables listed.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003eGenotypes\u003c/h2\u003e \u003cp\u003eGenotyping and quality control procedures for UK Biobank are described in (\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e). In summary, participants were genotyped on either the Applied Biosystems UK BiLEVE Axiom Array or the Applied Biosystems UK Biobank Axiom Array by Affymetrix. DNA was extracted from whole blood collected on participants\u0026rsquo; visit to the UK Biobank assessment center. Samples were genotyped in 106 batches, and checks were performed to exclude variants (0.97%) that differed between arrays or batches. Samples with a high degree of missingness, heterozygosity, or sex aneuploidy were excluded (0.3%). Principal components analysis was used to identify a subset of individuals with relatively similar ancestry, who self-reported as white British (fields 22020 and 22006, respectively). This subsample of 337,426 unrelated individuals were used in the reported analyses.\u003c/p\u003e \u003cp\u003eGenotype data provided by UK Biobank were filtered to exclude rare variants (MAF\u0026thinsp;\u0026lt;\u0026thinsp;1%), poorly genotyped variants (\u0026gt;\u0026thinsp;2% missing), and variants out of Hardy-Weinberg equilibrium (HWE\u0026thinsp;\u0026lt;\u0026thinsp;5 x 10\u003csup\u003e\u0026minus;\u0026thinsp;6\u003c/sup\u003e). PRS were calculated from the largest schizophrenia and bipolar GWAS available (\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e), from which UK Biobank data was excluded. The two sets of summary statistics were first reduced to a set of 7,562,571 variants that appeared in both GWAS. Genotype data were clumped, removing variants correlated\u0026thinsp;\u0026gt;\u0026thinsp;0.1 within a 250 kilobase window, based on MAF rather than p-value to avoid prioritizing one PRS over the other. PRS were calculated based on the intersection of 222,416 variants appearing in both GWAS summary stats and cleaned genotype calls. PRS were regressed on the first 10 principal components of population stratification, and the residuals normalized for subsequent analyses.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003ePsyCourse Data\u003c/h2\u003e \u003cdiv id=\"Sec8\" class=\"Section3\"\u003e \u003ch2\u003eSubjects\u003c/h2\u003e \u003cp\u003eReplication data are drawn from the PsyCourse Study, a longitudinal study of individuals with severe mental disorders from the psychotic-to-affective continuum and community controls conducted from a network of sites in Germany and Austria (\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e). The study protocol was approved by the respective ethics committee for each study center, and written informed consent was obtained from each study participant. The sample used in this study consists of 1308 individuals with a schizophrenia spectrum diagnosis (schizophrenia, schizoaffective disorder, or brief psychotic disorder), bipolar disorder I and II, or recurrent major depression, and 466 control individuals. For details see Heilbronner and colleagues (\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003ePhenotypes\u003c/h2\u003e \u003cp\u003eParticipants were assessed four times at 6-month intervals, covering the 18 months following enrollment. Participants\u0026rsquo; current level of psychiatric treatment was assessed at each follow-up (variables v1_cur_psy_trm, v2_cur_psy_trm, v3_cur_psy_trm, and v4_cur_psy_trm). This variable was coded as 1 if a participant received inpatient care, and 0 otherwise, and summed across follow-ups, creating an ordinal variable reflecting the cumulative number of assessments at which the participant was receiving inpatient treatment. Among cases, 44.8% had experienced one or more psychiatric hospitalizations during the study interval, and 4.4% had experienced two or more psychiatric hospitalizations. Among cases, 67.2% had experienced psychosis as assessed by the SCID-IV (\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e).\u003c/p\u003e \u003cdiv id=\"Sec10\" class=\"Section3\"\u003e \u003ch2\u003eGenotypes\u003c/h2\u003e \u003cp\u003eGenetic data was derived from venous blood. Samples were genotyped on the Infinium Global Screening Array (versions 1 and 3). After standard quality control procedures, genotypes were imputed against the 1000 Genomes Phase 3 reference panel (\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e) using SHAPEIT2 and IMPUTE2 (\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e, \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e). Variants with imputation quality\u0026thinsp;\u0026lt;\u0026thinsp;0.8 were not included in downstream analyses. Principal components of genetic covariance were computed using Plink 1.9 (\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e). Individuals more than three standard deviations from the mean on any of the first three principal components of ancestry were excluded from the analysis, yielding a final sample size of 1,594 (1190 cases and 404 controls). SZ and BP PRS were estimated using PRSice (\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e), based on the same schizophrenia and bipolar GWAS as the Suffolk County sample (\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e). PRS were regressed on the first 10 principal components of population stratification, and the residuals normalized for subsequent analyses.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eSuffolk County Data\u003c/h2\u003e \u003cdiv id=\"Sec12\" class=\"Section3\"\u003e \u003ch2\u003eSubjects\u003c/h2\u003e \u003cp\u003eData are drawn from the Suffolk Health County Mental Health Project, a first-admission psychosis cohort recruited between 1989 and 1995 from all 12 inpatient psychiatric units located in Suffolk County, New York (\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e). The baseline wave includes 628 participants, representing a 72% response rate among eligible individuals. The inclusion criteria were: first admission for psychosis in the past 6 months, English-language comprehension, residence in Suffolk County, IQ\u0026thinsp;\u0026gt;\u0026thinsp;70, and age 15 to 60 years. The Stony Brook University Committee on Research Involving Human Subjects and the participating hospital\u0026rsquo;s review boards authorized the study every year. Written consent was obtained from all the participants or parents of a participant in case of a minor aged from 15\u0026ndash;17 years. At the 20-year follow-up, a sample of 261 demographically matched, never-psychotic adults was recruited from the same zip codes as cases. The present analyses are based on DNA collected at the 20-year follow-up. Supplemental Table\u0026nbsp;1 reports a comparison between those cases who were genotyped and those who were not (all never-psychotic controls were genotyped). These groups did not differ except in terms of age, as older participants were more likely to have died prior to the 20-year follow-up, and therefore were not genotyped. Outcomes were assessed at the 25-year follow-up. Ratings were made based on all data collected during the follow-up assessment, including the structured clinical interview, neuropsychological assessment, collateral interview, and review of medical records.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003ePhenotypes\u003c/h2\u003e \u003cp\u003e \u003cb\u003eRemission.\u003c/b\u003e Symptomatic remission was operationalized according Andreasen\u0026rsquo;s definition (\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e). Symptom severity was measured using the Scale for the Assessment of Positive Symptoms (SAPS; 47) and the Scale for the Assessment of Negative Symptoms (SANS; 48). To be considered remitted, the following symptoms had to be rated as mild of better (2 on a scale where 0 corresponds to no symptoms and 5 corresponds to severe symptoms): hallucinations, delusions, bizarre behavior, formal thought disorder, affective flattening, alogia, avolition-apathy, and anhedonia-asociality. Whereas Andreasen\u0026rsquo;s definition stipulates symptoms be rated over the past 6 months, symptoms in this sample were rated over the past month.\u003c/p\u003e \u003cp\u003e \u003cb\u003eRecovery.\u003c/b\u003e Recovery was operationalized according to Liberman\u0026rsquo;s definition (\u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e). Symptoms were assessed using eight ratings from the Brief Psychiatric Rating Scale (BPRS; 50). For a participant to be defined as \u0026ldquo;recovered\u0026rdquo;, they had to have scores of \u0026ldquo;moderate\u0026rdquo; or better on BPRS ratings of conceptual disorganization, mannerisms and posturing, grandiosity, suspiciousness, hallucinatory behavior, unusual thought content, blunted affect, and emotional withdrawal. Psychosocial functioning was assessed based on two ratings from the Quality of Life in Schizophrenia scale (QLS; 51). Participants needed a rating of three or higher on ratings of social activity and accomplishment to be considered psychosocially recovered. Being employed part-time or being a part-time student was also considered evidence of adequate role function. Symptoms and functioning were assessed over the past month, rather than the prior two years as Liberman\u0026rsquo;s definition stipulates.\u003c/p\u003e \u003cp\u003e \u003cb\u003eGlobal Assessment of Functioning\u003c/b\u003e. Global Assessment of Functioning (GAF) was measured by consensus rating of psychiatrists using all available information. GAF ratings were made for the best month of the year preceding the interview.\u003c/p\u003e \u003cp\u003e \u003cb\u003eSymptom Severity.\u003c/b\u003e Symptom severity at the 25-year follow up was assessed using the Scale for the Assessment of Negative Symptoms (SANS; 48) and Scale for the Assessment of Positive Symptoms (SAPS; 47). Items from the SAPS and SANS were scored into 4 factor-analytically derived subscales, described in Kotov (\u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e52\u003c/span\u003e). Reliability of these subscales was high (α for reality distortion=[0.82\u0026ndash;0.85], disorganization=[0.70\u0026ndash;0.77], apathy/asociality=[0.78\u0026ndash;0.82], inexpressivity=[0.84\u0026ndash;0.88]).\u003c/p\u003e \u003cp\u003e \u003cb\u003eRole Functioning.\u003c/b\u003e Role functioning was assessed using item 4, level of accomplishment, from the QLS.\u003c/p\u003e \u003cp\u003e \u003cb\u003eActivities of Daily Living.\u003c/b\u003e Ability to manage day-to-day activities was measured using the University of California San Diego Performance-Based Skill Assessment (UPSA), an behavioral test of community living (\u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e53\u003c/span\u003e). Participants role-play tasks such as making daily and urgent calls, performing shopping tasks, money manipulation, and reading maps and schedules. Total scores range from 1\u0026ndash;100 points. The distribution of scores was left-skewed, so scores were inversed to meet the assumptions of the negative binomial regression model (described in Statistical Analyses, below).\u003c/p\u003e \u003cp\u003e \u003cb\u003eSocial Function.\u003c/b\u003e Social functioning was quantified as a composite of three items from the QLS (\u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e54\u003c/span\u003e). Ratings used to derive the social functioning composite included social activity, social sexual relationships, and relationships with friends. The composite score ranged from 0 (worst functioning) to 17 (best functioning). Reliability of the composite was 0.73.\u003c/p\u003e \u003cp\u003e \u003cb\u003eResidential \u0026amp; Economic Independence.\u003c/b\u003e Residential independence was defined as not being reliant on agencies or other people to arrange for a residence. Economic independence was similarly defined as not being reliant on another person or agency to maintain one\u0026rsquo;s finances.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eGenotypes\u003c/h2\u003e \u003cp\u003eDNA was extracted from peripheral lymphocytes and genotyped using the Illumina PsychArray-8 platform containing 571,054 markers. Standard quality control procedures were performed to exclude SNPs with minor allele frequency (MAF)\u0026thinsp;\u0026lt;\u0026thinsp;1%, genotyping failure\u0026thinsp;\u0026gt;\u0026thinsp;5%, Hardy-Weinberg equilibrium p\u0026thinsp;\u0026lt;\u0026thinsp;1E-6, mismatch between recorded and genotyped sex, as well as related individuals (π̂\u0026gt;.20, in which case the relative with the lower call rate was dropped). SNP imputation was conducted with Beagle version 5.4 (\u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e55\u003c/span\u003e), against the full 1000 Genomes phase 3 reference panel (\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e). The imputed SNPs underwent another round of quality control in which SNPs with missing data\u0026thinsp;\u0026gt;\u0026thinsp;10%, Hardy-Weinberg equilibrium p\u0026thinsp;\u0026lt;\u0026thinsp;10\u0026thinsp;\u0026minus;\u0026thinsp;6, and imputation information score\u0026thinsp;\u0026lt;\u0026thinsp;0.8 were excluded, yielding 6.58M high quality biallelic SNPs.\u003c/p\u003e \u003cp\u003eThe schizophrenia polygenic risk score (PRS) was calculated based on summary statistics from Trubetskoy and colleagues (\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e). The BP PRS was calculated based on summary statistics from Mullins and colleagues (\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e). A schizophrenia versus bipolar PRS was calculated based results on from Ruderfer and colleagues (\u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e56\u003c/span\u003e). This score was reversed so that higher scores indicate greater risk for bipolar disorder and lower scores indicate greater risk for schizophrenia. PRS were estimated using PRS-ice (\u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e57\u003c/span\u003e), with variants correlated\u0026thinsp;\u0026gt;\u0026thinsp;0.1 within a 500 kilobase window clumped according to p-value. The resulting BP and SZ PRS were based on 167,552 and 167,764 variants, respectively. The first ten principal components of ancestry were estimated using Plink 1.9 (\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e) and ancestry was assessed using ADMIXTURE (\u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e58\u003c/span\u003e). Analyses were restricted to participants of European ancestry, defined as being within the first 3 standard deviations of the first ten principal components (PCs), and participants' EUR ancestry score\u0026thinsp;\u0026lt;\u0026thinsp;0.8. PRS were regressed on the first 10 principal components of population stratification, and the residuals normalized for subsequent analyses.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eStatistical Analysis\u003c/h2\u003e \u003cp\u003eIn all samples, demographic differences were analyzed using the Student t-test for continuous variables and the chi-square test for categorical variables.\u003c/p\u003e \u003cp\u003eThe association between the SZ and BP PRS and factors on which samples were stratified was evaluated using t-tests between diagnostic groups, with the effect size expressed as Cohen\u0026rsquo;s d, and Pearson correlations (\u003cem\u003er\u003c/em\u003e) between PRS and ordinal phenotypes. We evaluated the effect of sample stratification on the correlation between the SZ and BP PRS by estimating Pearson correlations between the two PRS. In UK Biobank, this correlation was estimated in the full sample (N\u0026thinsp;=\u0026thinsp;322,774), individuals with a primary diagnosis of a psychotic disorder from an inpatient encounter (N\u0026thinsp;=\u0026thinsp;934), and individuals with at least 1 (N\u0026thinsp;=\u0026thinsp;1,695), 2 (N\u0026thinsp;=\u0026thinsp;643), 3 (N\u0026thinsp;=\u0026thinsp;264), or 4 (N\u0026thinsp;=\u0026thinsp;100) psychotic disorder diagnoses recorded in the electronic medical record, death certificate, or by self-report. In PsyCourse, this correlation was estimated in the combined case-control sample (N\u0026thinsp;=\u0026thinsp;1,594), among those who experienced psychotic symptoms and had at least one (N\u0026thinsp;=\u0026thinsp;427) and at least two (N\u0026thinsp;=\u0026thinsp;55) psychiatric hospitalizations.\u003c/p\u003e \u003cp\u003eIn data from the Suffolk County Mental Health Project, associations between 25-year outcomes and genetic risk were analyzed using Poisson and negative binomial regression for count outcomes. When exponentiated, regression coefficients from these models are equivalent to relative risks (RR), and are reported as such. Logistic regression was used for binary outcomes. Exponentiated regression coefficients from logistic regressions are equivalent to odds ratios (OR), and are reported as such. All ratios are reported with 95% confidence intervals. All regressions were univariate regressions.\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003eEffect of selection bias on correlation of SZ and BP PRS\u003c/h2\u003e \u003cp\u003eThe effect of selection on the correlation between the SZ and BP PRS was evaluated in UK Biobank and replicated in data from the PsyCourse study. The UK Biobank analysis sample was 53.8% female (N female\u0026thinsp;=\u0026thinsp;173,807) and age 56.9 (SD\u0026thinsp;=\u0026thinsp;8.0) at recruitment, on average. Both the BP and SZ PRS were associated with a primary inpatient diagnosis of a psychotic disorder (Cohen\u0026rsquo;s d\u0026thinsp;=\u0026thinsp;0.37 and 0.46 respectively, both p\u0026thinsp;\u0026lt;\u0026thinsp;0.01), as well as the total number of psychotic disorder diagnoses recorded in the medical record, on death certificates, and by self-report (\u003cem\u003er\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.02 and 0.03 respectively, both p\u0026thinsp;\u0026lt;\u0026thinsp;0.01). The correlation between the BP and SZ PRS in the full sample (N\u0026thinsp;=\u0026thinsp;322,774) was 0.33. Among those with a primary inpatient psychotic disorder diagnosis the correlation was 0.25 (N\u0026thinsp;=\u0026thinsp;934; 95% CI 0.19\u0026ndash;0.31). Among those with any psychotic disorder diagnosis, the correlation was 0.29 (N\u0026thinsp;=\u0026thinsp;1,695; [0.25\u0026ndash;0.34]). Among those with two or more recorded diagnoses, the correlation was 0.24 (N\u0026thinsp;=\u0026thinsp;643; [0.16\u0026ndash;0.31]), among those with three or more recorded diagnoses was 0.25 (N\u0026thinsp;=\u0026thinsp;264; [0.14\u0026ndash;0.36]), and was lowest among those with four or more diagnoses (N\u0026thinsp;=\u0026thinsp;100; \u003cem\u003er\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.13; [-0.07-0.32]).\u003c/p\u003e \u003cp\u003eDemographics of the PsyCourse Study are reported in Table\u0026nbsp;2. Both the SZ and BP PRS were correlated with the number of psychiatric hospitalizations (\u003cem\u003er\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.08 and 0.03, respectively). Within the combined sample of cases and controls, the SZ and BP PRS were closely correlated (N\u0026thinsp;=\u0026thinsp;1,594, r\u0026thinsp;=\u0026thinsp;0.50 [0.46\u0026ndash;0.53]). The strength of this association among those who had a lifetime history of psychosis and had at least one psychiatric hospitalization was 0.48 (N\u0026thinsp;=\u0026thinsp;427, [0.38\u0026ndash;0.54]). This correlation was lowest among those with a history of psychosis and two or more psychiatric hospitalizations over the 18 months of the study (N\u0026thinsp;=\u0026thinsp;55, r\u0026thinsp;=\u0026thinsp;0.33 [0.07\u0026ndash;0.60]).\u003c/p\u003e \u003cp\u003eDemographic and clinical characteristics of the Suffolk County Mental Health Project participants are reported in Table\u0026nbsp;3. Cases were two years younger than controls, on average. There were no statistically significant differences in gender, race, or socioeconomic status. Cases had more severe symptoms and worse functioning on all outcomes included in the analysis. In the case-control cohort, the correlation between the SZ and BP PRS was 0.40, which decreased r\u0026thinsp;=\u0026thinsp;0.20 in cases only. Similarly, variance of the schizophrenia PRS decreased from 1 to 0.74. Variance of the BP PRS did not change (1.00 to 1.01).\u003c/p\u003e \u003cp\u003eTable\u0026nbsp;4 reports the association between 25-year outcomes and the schizophrenia (SZ) and bipolar (BP) PRS in the combined sample of cases and controls (right side). Effect sizes are exponentiated beta coefficients, for which a value greater than one indicates a positive association and a value less than one indicates a negative association. The SZ PRS was associated with lower odds of recovery (OR\u0026thinsp;=\u0026thinsp;0.61, 95% CI=[0.47\u0026ndash;0.78]) and remission (OR\u0026thinsp;=\u0026thinsp;0.52 [0.40\u0026ndash;0.68]), and reduced odds of attaining residential (OR\u0026thinsp;=\u0026thinsp;0.58 [0.43\u0026ndash;0.79]) and economic (OR\u0026thinsp;=\u0026thinsp;0.59 [0.46\u0026ndash;0.76]) independence at the 25-year follow-up. The SZ PRS predicted poorer functioning and more severe symptoms across all domains assessed. The BP PRS had a narrower range of associations but was associated with lower odds of remission (OR\u0026thinsp;=\u0026thinsp;0.77 [0.61\u0026ndash;0.98]) and economic independence (OR\u0026thinsp;=\u0026thinsp;0.79 [0.63\u0026ndash;0.99]), poorer functioning (GAF; RR\u0026thinsp;=\u0026thinsp;0.95 [0.91\u0026ndash;0.99]), and more severe psychotic symptoms (RR\u0026thinsp;=\u0026thinsp;1.54 [1.09\u0026ndash;2.23]).\u003c/p\u003e \u003cp\u003eAssociations between genetic risk and 25-year outcomes estimated among cases only are reported on the left side of Table\u0026nbsp;4. In this sample, the SZ PRS still predicts lower odds of remission (OR\u0026thinsp;=\u0026thinsp;0.61 [0.41\u0026ndash;0.89]), but other associations were no longer statistically significant. Associations between BP PRS and 25-year outcomes changed directions for all outcomes except inexpressivity. Among cases, the BP PRS was associated with increased odds of economic independence (OR\u0026thinsp;=\u0026thinsp;1.46 [1.03\u0026ndash;2.06]) and better social performance (RR\u0026thinsp;=\u0026thinsp;1.11 [1.04\u0026ndash;1.19]). The impact of sample stratification on genetic associations is depicted in Figs.\u0026nbsp;1A and 1B.\u003c/p\u003e \u003cp\u003eSupplemental Table\u0026nbsp;2 reports the results of a sensitivity analysis estimating associations between the case-case BP versus SZ PRS on 25-year outcomes. The observed pattern of results is similar to that of the BP PRS, with greater BP versus SZ PRS associated with better 25-year outcomes among cases.\u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eWe assessed the degree to which selection bias impacts the association between the BP and SZ PRS when selection is based on phenotypes that are themselves impacted by genetic risk. In two samples, selecting samples on the basis of a psychotic disorder diagnosis or correlates thereof significantly attenuated the association between the BP and SZ PRS. We furthermore evaluated the impact of selection bias on the associations between the BP and SZ PRS and an array of outcomes in a first-admission psychosis cohort followed for 25 years and demographically matched, never-psychotic controls, in order to understand how this phenomenon impacts the potential clinical application of PRS in first-admission psychosis. In the combined sample of both cases and never-psychotic controls assessed 25 years after first admission, both the SZ and BP PRS were associated with more severe symptoms and worse outcomes. However, when analyses were limited to cases, these associations were dramatically attenuated. In some cases, the direction of the BP PRS effects reversed, with greater genetic risk predicting better outcomes. Below, we describe how these results are consistent with selection bias, and the implications for clinical translation of PRS in psychiatry.\u003c/p\u003e \u003cp\u003eSelection bias occurs when the process of sample ascertainment changes the distribution of the parameter of interest. In two independent samples, both the SZ and BP PRS were associated with psychotic disorder diagnoses and hospitalization for psychosis. Selection based on these phenotypes therefore truncates the joint distribution of the SZ and BP PRS, attenuating the correlation between these two predictors, (see Fig.\u0026nbsp;1 for a schematic representation). The correlation between the SZ and BP PRS was shown to decrease as a function of the number of inpatient hospitalizations (from \u003cem\u003er\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.50 in the case-control sample to 0.33 among those with multiple hospitalizations), and as a function of the number of documented psychotic disorder diagnoses (from \u003cem\u003er\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.33 to 0.13 among those with four or more diagnoses). One consequence of this selection effect is that, whereas the BP PRS indicates greater genetic risk for psychosis broadly in the case-control sample, in the case-only sample it functions similarly to the case-case BP versus SZ PRS (see Supplemental Table\u0026nbsp;2), indicating greater genetic risk specifically for bipolar disorder, conditional on high genetic risk for psychosis generally. Since bipolar disorder has the most favorable outcomes among psychotic disorders (\u003cspan additionalcitationids=\"CR60 CR61 CR62\" citationid=\"CR59\" class=\"CitationRef\"\u003e59\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR63\" class=\"CitationRef\"\u003e63\u003c/span\u003e), the BP PRS predicts \u003cem\u003ebetter\u003c/em\u003e outcomes in the context of a cohort with a high genetic loading for psychosis broadly.\u003c/p\u003e \u003cp\u003eThese findings may explain the inconsistent associations between the BP PRS and outcomes in clinical samples (see Table\u0026nbsp;1 for a summary of this literature). Associations that are significant in the general population may be non-significant (\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e), or even reversed, among cases. The same dynamics likely affect analyses of genetic structure. Genetic correlations are likely to be attenuated among case-only samples, relative to case-control or population-based samples (\u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e64\u003c/span\u003e, \u003cspan citationid=\"CR65\" class=\"CitationRef\"\u003e65\u003c/span\u003e). Researchers should be especially cautious in drawing etiological inferences based on studies of cases only, as selection bias may obscure true causal pathways, if they exist, or reverse the direction of the effect. Research investigating etiological mechanisms of genetic risk are best performed in samples that capture the full spectrum of genetic risk. To mitigate the impact of selection bias, researchers should carefully consider and account for causal relationships among the variables under investigation. Sensitivity analyses, causal modeling, and cautious interpretation of results can help minimize the impact of selection bias on etiological research when population samples are unavailable or infeasible.\u003c/p\u003e \u003cp\u003eThat selection bias impacts genetic prediction should not be taken to mean PRS cannot be useful in case-only samples. While etiological research is confounded by selection bias, predictive models are not, even though the direction of effects may be counterintuitive. Translational research on potential clinical applications of PRS need only ensure that the same forms of selection occur in both research cohorts and clinical settings. These results indicate, for example, that a relatively high BP PRS may be a marker of good prognosis among those admitted for first-episode psychosis. Similarly, even though inferences about genetic mechanisms of treatment resistance are likely confounded by sampling bias, SZ and BP PRS may be accurate predictors of response to treatment in clinical cohorts (\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e, \u003cspan citationid=\"CR66\" class=\"CitationRef\"\u003e66\u003c/span\u003e).\u003c/p\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003eLimitations\u003c/h2\u003e \u003cp\u003eThis research is limited in three ways. First, the Suffolk County sample is relatively small, which may have prevented the detection of small effects. While longitudinal analyses could not be replicated in an independent sample due to lack of comparable data, the general effect of sample stratification on genetic risk were replicated. Second, prior research has demonstrated that UK Biobank participants are healthier than the UK population, and that this can bias genetic correlations (\u003cspan citationid=\"CR67\" class=\"CitationRef\"\u003e67\u003c/span\u003e). Since the effects observed in UK Biobank were replicated in PsyCourse, it does not appear that this has impact the conclusions that should be drawn from the analyses. Lastly, given the predominantly European ancestry composition of all three samples, the analytical scope of this study was limited to individuals of European ancestry. GWAS in other ancestries samples are available, facilitating the extension of these results to other ancestries.\u003c/p\u003e \u003c/div\u003e"},{"header":"Conclusion","content":"\u003cp\u003eSelection bias substantially impacts the association between schizophrenia and bipolar disorder PRS, as well as the association between PRS and clinical outcomes. Whereas greater genetic risk is associated with worse outcomes in a case-control sample, genetic risk for bipolar disorder predicts better outcomes among cases only. Selection bias may explain the heterogeneity of effects linking the bipolar PRS to clinical outcomes. It also complicates the use of PRS for both etiological and translational research.\u003c/p\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAvailability of Data and Materials Statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eData from UK Biobank are available to qualified researchers. The application process is outlined at https://community.ukbiobank.ac.uk/hc/en-gb/categories/14494598931229-Enable-your-research. Data from PsyCourse are available to qualified researchers. The application process is outlined at http://www.psycourse.de/index-en.html. Data from the Suffolk County Mental Health Project are available from the NIMH Data Archive, collection number 2477. Analytic syntax is available from the corresponding author upon request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflict of Interest\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors would like to state there are no financial conflicts of interest associated with the work described.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis research was supported by National Institutes of Health (MH44801, MH094398, MH110434), and a NARSAD Young Investigator Grant to R.K. This study received funding from the National Institutes of Health under grant number R21MH123908, awarded to K. J. The authors gratefully acknowledge the support of the participants and mental health community of Suffolk County for contributing their time and energy to this project. They are also indebted to the study coordinators for their dedicated efforts, the interviewers for their careful assessments, and the psychiatrists who derived the consensus diagnoses. This research has been conducted using data from UK Biobank, a major biomedical database, under project ID 55741. U.H. was supported by European Union\u0026rsquo;s Horizon 2020 Research and Innovation Program (PSY-PGx, grant agreement No 945151) and the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation, project number 514201724). Thomas G. Schulze was supported by the Deutsche Forschungsgemeinschaft (KFO241/PsyCourse, SCHU 1603/4-1, 5-1, 7-1), the German Ministry of Education and Research (IntegraMent: 01ZX1614K; BipoLife: 01EE1404H; the German Center for Mental Health [DZPG]: 01EE2303A/01EE2303F), and the European Union (ERA-NET NEURON - MulioBio: 01EW2009; GEPI-BIOPSY: 01EW2005).\u003c/p\u003e\n\u003cp\u003eSupplementary information is available at MP\u0026rsquo;s website\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eFullerton JM, Nurnberger JI. Polygenic risk scores in psychiatry: Will they be useful for clinicians? F1000Research. 2019;8:F1000 Faculty Rev-1293.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLewis CM, Vassos E. Polygenic risk scores: from research tools to clinical instruments. Genome Med. 2020;12:44.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWray NR, Lin T, Austin J, McGrath JJ, Hickie IB, Murray GK, et al. From Basic Science to Clinical Application of Polygenic Risk Scores: A Primer. JAMA Psychiatry. 2021;78(1):101\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKhera AV, Chaffin M, Aragam KG, Haas ME, Roselli C, Choi SH, et al. Genome-wide polygenic scores for common diseases identify individuals with risk equivalent to monogenic mutations. Nat Genet. 2018;50(9):1219\u0026ndash;24.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLambert SA, Abraham G, Inouye M. Towards clinical utility of polygenic risk scores. Hum Mol Genet. 2019;28(R2):R133\u0026ndash;42.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMars N, Koskela JT, Ripatti P, Kiiskinen TTJ, Havulinna AS, Lindbohm JV, et al. Polygenic and clinical risk scores and their impact on age at onset and prediction of cardiometabolic diseases and common cancers. Nat Med. 2020;26(4):549\u0026ndash;57.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMavaddat N, Michailidou K, Dennis J, Lush M, Fachal L, Lee A, et al. Polygenic Risk Scores for Prediction of Breast Cancer and Breast Cancer Subtypes. Am J Hum Genet. 2019;104(1):21\u0026ndash;34.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRiveros-Mckay F, Weale ME, Moore R, Selzam S, Krapohl E, Sivley RM, et al. Integrated Polygenic Tool Substantially Enhances Coronary Artery Disease Prediction. Circ Genomic Precis Med. 2021;14(2):e003304.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWeale ME, Riveros-Mckay F, Selzam S, Seth P, Moore R, Tarran WA, et al. Validation of an Integrated Risk Tool, Including Polygenic Risk Score, for Atherosclerotic Cardiovascular Disease in Multiple Ethnicities and Ancestries. Am J Cardiol. 2021;148:157\u0026ndash;64.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAnderson JS, Shade J, DiBlasi E, Shabalin AA, Docherty AR. Polygenic Risk Scoring and Prediction of Mental Health Outcomes. Curr Opin Psychol. 2019;27:77\u0026ndash;81.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSmigielski L, Papiol S, Theodoridou A, Heekeren K, Gerstenberg M, Wotruba D, et al. Polygenic risk scores across the extended psychosis spectrum. Transl Psychiatry. 2021;11(1):1\u0026ndash;11.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGui Y, Zhou X, Wang Z, Zhang Y, Wang Z, Zhou G, et al. Sex-specific genetic association between psychiatric disorders and cognition, behavior and brain imaging in children and adults. Transl Psychiatry. 2022;12(1):1\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLoughnan RJ, Palmer CE, Makowski C, Thompson WK, Barch DM, Jernigan TL, et al. Unique prediction of developmental psychopathology from genetic and familial risk. J Child Psychol Psychiatry. 2022;63(12):1631\u0026ndash;43.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMistry S, Escott-Price V, Florio AD, Smith DJ, Zammit S. Genetic risk for bipolar disorder and psychopathology from childhood to early adulthood. J Affect Disord. 2019;246:633\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiebers DT, Pirooznia M, Seiffudin F, Musliner KL, Zandi PP, Goes FS. Polygenic Risk of Schizophrenia and Cognition in a Population-Based Survey of Older Adults. Schizophr Bull. 2016;42(4):984\u0026ndash;91.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHosang GM, Martin J, Karlsson R, Lundstr\u0026ouml;m S, Larsson H, Ronald A, et al. Association of Etiological Factors for Hypomanic Symptoms, Bipolar Disorder, and Other Severe Mental Illnesses. JAMA Psychiatry. 2022;79(2):143\u0026ndash;50.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiebers DT, Pirooznia M, Ganna A, Study (BiGS) BG, Goes FS. Discriminating bipolar depression from major depressive disorder with polygenic risk scores. Psychol Med. 2021;51(9):1451\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKalman JL, Papiol S, Grigoroiu-Serbanescu M, Adorjan K, Anderson-Schmidt H, Brosch K, et al. Genetic risk for psychiatric illness is associated with the number of hospitalizations of bipolar disorder patients. J Affect Disord. 2022;296:532\u0026ndash;40.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAhangari M, Bustamante D, Kirkpatrick R, Nguyen TH, Verrelli BC, Fanous A, et al. Relationship between polygenic risk scores and symptom dimensions of schizophrenia and schizotypy in multiplex families with schizophrenia. Br J Psychiatry J Ment Sci. 2022;1\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDavid FS, Stein F, Andlauer TFM, Streit F, Witt SH, Herms S, et al. Genetic contributions to transdiagnostic symptom dimensions in patients with major depressive disorder, bipolar disorder, and schizophrenia spectrum disorders. Schizophr Res. 2023;252:161\u0026ndash;71.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRichards AL, Pardi\u0026ntilde;as AF, Frizzati A, Tansey KE, Lynham AJ, Holmans P, et al. The Relationship Between Polygenic Risk Scores and Cognition in Schizophrenia. Schizophr Bull. 2020;46(2):336\u0026ndash;44.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKalman JL, Papiol S, Forstner AJ, Heilbronner U, Degenhardt F, Strohmaier J, et al. Investigating polygenic burden in age at disease onset in bipolar disorder: Findings from an international multicentric study. Bipolar Disord. 2019;21(1):68\u0026ndash;75.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchubert KO, Thalamuthu A, Amare AT, Frank J, Streit F, Adl M, et al. Combining schizophrenia and depression polygenic risk scores improves the genetic prediction of lithium response in bipolar disorder patients. Transl Psychiatry. 2021;11(1):1\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLopes FL, Zhu K, Purves KL, Song C, Ahn K, Hou L, et al. Polygenic risk for anxiety influences anxiety comorbidity and suicidal behavior in bipolar disorder. Transl Psychiatry. 2020;10(1):1\u0026ndash;7.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDwyer DB, Kalman JL, Budde M, Kambeitz J, Ruef A, Antonucci LA, et al. An Investigation of Psychosis Subgroups With Prognostic Validation and Exploration of Genetic Underpinnings: The PsyCourse Study. JAMA Psychiatry. 2020;77(5):523\u0026ndash;33.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCoombes BJ, Markota M, Mann JJ, Colby C, Stahl E, Talati A, et al. Dissecting clinical heterogeneity of bipolar disorder using multiple polygenic risk scores. Transl Psychiatry. 2020;10(1):1\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSong J, Jonsson L, Lu Y, Bergen SE, Karlsson R, Smedler E, et al. Key subphenotypes of bipolar disorder are differentially associated with polygenic liabilities for bipolar disorder, schizophrenia, and major depressive disorder. Mol Psychiatry. 2024;1\u0026ndash;10.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRodriguez V, Alameda L, Quattrone D, Tripoli G, Gayer-Anderson C, Spinazzola E, et al. Use of multiple polygenic risk scores for distinguishing schizophrenia-spectrum disorder and affective psychosis categories in a first-episode sample; the EU-GEI study. Psychol Med. 2022;1\u0026ndash;10.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFerraro L, Quattrone D, La Barbera D, La Cascia C, Morgan C, Kirkbride JB, et al. First-Episode Psychosis Patients Who Deteriorated in the Premorbid Period Do Not Have Higher Polygenic Risk Scores Than Others: A Cluster Analysis of EU-GEI Data. Schizophr Bull [Internet]. 2022 Aug 10 [cited 2022 Oct 13]; Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/schbul/sbac100\u003c/span\u003e\u003cspan address=\"10.1093/schbul/sbac100\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFrank J, Lang M, Witt SH, Strohmaier J, Rujescu D, Cichon S, et al. Identification of increased genetic risk scores for schizophrenia in treatment-resistant patients. Mol Psychiatry. 2015;20(2):150\u0026ndash;1.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJonas KG, Lencz T, Li K, Malhotra AK, Perlman G, Fochtmann LJ, et al. Schizophrenia polygenic risk score and 20-year course of illness in psychotic disorders. Transl Psychiatry. 2019;9(1):300.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang JP, Robinson D, Yu J, Gallego J, Wolfgang Fleischhacker W, Kahn RS, et al. Schizophrenia Polygenic Risk Score as a Predictor of Antipsychotic Efficacy in First Episode Psychosis. Am J Psychiatry. 2019;176(1):21\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMunaf\u0026ograve; MR, Tilling K, Taylor AE, Evans DM, Davey Smith G. Collider scope: when selection bias can substantially influence observed associations. Int J Epidemiol. 2018;47(1):226\u0026ndash;35.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee YH, Thaweethai T, Sheu YH, Feng YCA, Karlson EW, Ge T, et al. Impact of selection bias on polygenic risk score estimates in healthcare settings. Psychol Med. 2023;1\u0026ndash;11.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSudlow C, Gallacher J, Allen N, Beral V, Burton P, Danesh J, et al. UK biobank: an open access resource for identifying the causes of a wide range of complex diseases of middle and old age. PLoS Med. 2015;12(3):e1001779.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBycroft C, Freeman C, Petkova D, Band G, Elliott LT, Sharp K, et al. The UK Biobank resource with deep phenotyping and genomic data. Nature. 2018;562(7726):203\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMullins N, Forstner AJ, O\u0026rsquo;Connell KS, Coombes B, Coleman JRI, Qiao Z, et al. Genome-wide association study of more than 40,000 bipolar disorder cases provides new insights into the underlying biology. Nat Genet. 2021;53(6):817\u0026ndash;29.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTrubetskoy V, Pardi\u0026ntilde;as AF, Qi T, Panagiotaropoulou G, Awasthi S, Bigdeli TB, et al. Mapping genomic loci implicates genes and synaptic biology in schizophrenia. Nature. 2022;604(7906):502\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBudde M, Anderson-Schmidt H, Gade K, Reich‐Erkelenz D, Adorjan K, Kalman JL, et al. A longitudinal approach to biological psychiatric research: The PsyCourse study. Am J Med Genet. 2019;180(2):89\u0026ndash;102.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHeilbronner U, Adorjan K, Anderson-Schmidt H, Budde M, Comes AL, Gade K, et al. The PsyCourse Codebook, Version 6.0 [Internet]. 2023 [cited 2024 Jun 4]. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://data.ub.uni-muenchen.de/390/\u003c/span\u003e\u003cspan address=\"https://data.ub.uni-muenchen.de/390/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFirst MB, Gibbon M. The Structured Clinical Interview for DSM-IV Axis I Disorders (SCID-I) and the Structured Clinical Interview for DSM-IV Axis II Disorders (SCID-II). In: Comprehensive handbook of psychological assessment, Vol 2: Personality assessment. Hoboken, NJ, US: John Wiley \u0026amp; Sons, Inc.; 2004. p. 134\u0026ndash;43.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003e1000 Genomes Project Consortium, Abecasis GR, Auton A, Brooks LD, DePristo MA, Durbin RM, et al. An integrated map of genetic variation from 1,092 human genomes. Nature. 2012;491(7422):56\u0026ndash;65.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePurcell S, Neale B, Todd-Brown K, Thomas L, Ferreira MAR, Bender D, et al. PLINK: A Tool Set for Whole-Genome Association and Population-Based Linkage Analyses. Am J Hum Genet. 2007;81(3):559\u0026ndash;75.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEuesden J, Lewis CM, O\u0026rsquo;Reilly PF. PRSice: Polygenic Risk Score software. Bioinforma Oxf Engl. 2015;31(9):1466\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBromet EJ, Schwartz JE, Fennig S, Geller L, Jandorf L, Kovasznay B, et al. The epidemiology of psychosis: the Suffolk County Mental Health Project. Schizophr Bull. 1992;18(2):243\u0026ndash;55.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAndreasen NC, Carpenter WT, Kane JM, Lasser RA, Marder SR, Weinberger DR. Remission in schizophrenia: proposed criteria and rationale for consensus. Am J Psychiatry. 2005;162(3):441\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAndreasen NC. Scale for the Assessment of Positive Symptoms: SAPS. Department of Psychiatry, College of Medicine, the University of Iowa; 1984. 40 p.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAndreasen NC. The Scale for the Assessment of Negative Symptoms (SANS): Conceptual and Theoretical Foundations. Br J Psychiatry. 1989;155(S7):49\u0026ndash;52.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiberman RP, Kopelowicz A. Recovery from schizophrenia: a concept in search of research. Psychiatr Serv Wash DC. 2005;56(6):735\u0026ndash;42.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOverall JE, Gorham DR. The Brief Psychiatric Rating Scale. Psychol Rep. 1962;10:799\u0026ndash;812.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHeinrichs DW, Hanlon TE, Carpenter WT Jr. The Quality of Life Scale: An Instrument for Rating the Schizophrenic Deficit Syndrome. Schizophr Bull. 1984;10(3):388\u0026ndash;98.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKotov R, Foti D, Li K, Bromet EJ, Hajcak G, Ruggero CJ. Validating dimensions of psychosis symptomatology: Neural correlates and 20-year outcomes. J Abnorm Psychol. 2016;125(8):1103\u0026ndash;19.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHarvey PD, Velligan DI, Bellack AS. Performance-Based Measures of Functional Skills: Usefulness in Clinical Treatment Studies. Schizophr Bull. 2007;33(5):1138\u0026ndash;48.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVelthorst E, Fett AKJ, Reichenberg A, Perlman G, van Os J, Bromet EJ, et al. The 20-Year Longitudinal Trajectories of Social Functioning in Individuals With Psychotic Disorders. Am J Psychiatry. 2017;174(11):1075\u0026ndash;85.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBrowning BL, Zhou Y, Browning SR. A One-Penny Imputed Genome from Next-Generation Reference Panels. Am J Hum Genet. 2018;103(3):338\u0026ndash;48.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRuderfer DM, Ripke S, McQuillin A, Boocock J, Stahl EA, Pavlides JMW, et al. Genomic Dissection of Bipolar Disorder and Schizophrenia, Including 28 Subphenotypes. Cell. 2018;173(7):1705\u0026ndash;1715.e16.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChoi SW, O\u0026rsquo;Reilly PF. PRSice-2: Polygenic Risk Score software for biobank-scale data. GigaScience. 2019;8(7):giz082.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlexander DH, Novembre J, Lange K. Fast model-based estimation of ancestry in unrelated individuals. Genome Res. 2009;19(9):1655\u0026ndash;64.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKotov R, Fochtmann L, Li K, Tanenberg-Karant M, Constantino EA, Rubinstein J, et al. Declining Clinical Course of Psychotic Disorders Over the Two Decades Following First Hospitalization: Evidence From the Suffolk County Mental Health Project. Am J Psychiatry. 2017;174(11):1064\u0026ndash;74.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJonas K, Lian W, Callahan J, Ruggero CJ, Clouston S, Reichenberg A, et al. The Course of General Cognitive Ability in Individuals With Psychotic Disorders. JAMA Psychiatry. 2022;79(7):659\u0026ndash;66.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTramazzo S, Lian W, Ajnakina O, Carlson G, Bromet E, Kotov R, et al. The Long-Term Course of Remission and Recovery in Psychotic Disorder [Internet]. 2022 [cited 2023 Feb 7]. Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.31234/osf.io/yh74r\u003c/span\u003e\u003cspan address=\"10.31234/osf.io/yh74r\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBowie CR, Depp C, McGrath JA, Wolyniec P, Mausbach BT, Thornquist MH, et al. Prediction of Real World Functional Disability in Chronic Mental Disorders: A Comparison of Schizophrenia and Bipolar Disorder. Am J Psychiatry. 2010;167(9):1116\u0026ndash;24.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHarrow M, Grossman LS, Herbener ES, Davies EW. Ten-year outcome: patients with schizoaffective disorders, schizophrenia, affective disorders and mood-incongruent psychotic symptoms. Br J Psychiatry J Ment Sci. 2000;177:421\u0026ndash;6.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMundy J, H\u0026uuml;bel C, Adey BN, Davies HL, Davies MR, Coleman JRI, et al. Genetic examination of the Mood Disorder Questionnaire and its relationship with bipolar disorder. Am J Med Genet. 2023;192(7\u0026ndash;8):147\u0026ndash;60.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGrotzinger AD, Mallard TT, Akingbuwa WA, Ip HF, Adams MJ, Lewis CM, et al. Genetic architecture of 11 major psychiatric disorders at biobehavioral, functional genomic and molecular genetic levels of analysis. Nat Genet. 2022;54(5):548\u0026ndash;59.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCearns M, Amare AT, Schubert KO, Thalamuthu A, Frank J, Streit F, et al. Using polygenic scores and clinical data for bipolar disorder patient stratification and lithium response prediction: machine learning approach. Br J Psychiatry J Ment Sci. 2022;1\u0026ndash;10.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchoeler T, Speed D, Porcu E, Pirastu N, Pingault JB, Kutalik Z. Participation bias in the UK Biobank distorts genetic associations and downstream analyses. Nat Hum Behav. 2023;7(7):1216\u0026ndash;27.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLim K, Lam M, Huang H, Liu J, Lee J. Genetic liability in individuals at ultra-high risk of psychosis: A comparison study of 9 psychiatric traits. PLOS ONE. 2020;15(12):e0243104.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWarren TL, Tubbs JD, Lesh TA, Corona MB, Pakzad SS, Albuquerque MD, et al. Association of neurotransmitter pathway polygenic risk with specific symptom profiles in psychosis. Mol Psychiatry [Internet]. 2024; Available from: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41380-024-02457-0\u003c/span\u003e\u003cspan address=\"10.1038/s41380-024-02457-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSanchez-Ruiz JA, Coombes BJ, Pazdernik VM, Melhuish Beaupre LM, Jenkins GD, Pendegraft RS, et al. Clinical and genetic contributions to medical comorbidity in bipolar disorder: a study using electronic health records-linked biobank data. Mol Psychiatry. 2024;\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMistry S, Escott-Price V, Florio AD, Smith DJ, Zammit S. Genetic risk for bipolar disorder and psychopathology from childhood to early adulthood. J Affect Disord. 2019;246:633\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBigdeli TB, Barr PB, Rajeevan N, Graham DP, Li Y, Meyers JL, et al. Correlates of suicidal behaviors and genetic risk among United States veterans with schizophrenia or bipolar I disorder. Mol Psychiatry. 2024;\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eK\u0026auml;mpe A, Suvisaari J, L\u0026auml;hteenvuo M, Singh T, Ahola-Olli A, Urpa L, et al. Genetic contribution to disease-course severity and progression in the SUPER-Finland study, a cohort of 10,403 individuals with psychotic disorders. Mol Psychiatry. 2024;\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWilson JE, Sealock J, Straub P, Raman R, Kipp AM, Dittus RS, et al. Exploring genetic risk for catatonia in a genome wide association study and polygenic risk score analysis. Schizophr Res. 2024;263:178\u0026ndash;90.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAskeland RB, Hannigan LJ, O\u0026rsquo;Connell KS, Corfield EC, Frei O, Thapar A, et al. Developmental manifestations of polygenic risk for bipolar disorder from infancy to middle childhood. Transl Psychiatry. 2023;13(1):222.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMistry S, Escott-Price V, Florio AD, Smith DJ, Zammit S. Investigating associations between genetic risk for bipolar disorder and cognitive functioning in childhood. J Affect Disord. 2019;259:112\u0026ndash;20.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJiang X, Zai CC, Kennedy KG, Zou Y, Nikolova YS, Felsky D, et al. Association of polygenic risk for bipolar disorder with grey matter structure and white matter integrity in youth. Transl Psychiatry. 2023;13:322.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eThiel K, Lemke H, Winter A, Flinkenfl\u0026uuml;gel K, Waltemate L, Bonnekoh L, et al. White and gray matter alterations in bipolar I and bipolar II disorder subtypes compared with healthy controls - exploring associations with disease course and polygenic risk. Neuropsychopharmacol Off Publ Am Coll Neuropsychopharmacol. 2024;49(5):814\u0026ndash;23.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003eTable 1 to 4 are available in the Supplementary Files section.\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-4536236/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4536236/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003ePolygenic risk scores (PRS) have potential utility as biomarkers of psychiatric disorders. However, while the schizophrenia (SZ) PRS has been consistently associated with case-control status and a more severe course of illness, the associations between the bipolar (BP) PRS and markers of bipolar disorder vary considerably between studies, with studies of population and case-control samples identifying many effects that cannot be replicated in case-only analyses. These analyses demonstrate that the heterogeneity in studies of the BP PRS is driven by selection bias. Specifically, selecting samples on the basis of diagnostic status or other phenotypes associated with genetic risk attenuates the correlation between the BP and SZ PRS. In such high-severity samples, while the SZ PRS predicts poor outcomes, the BP PRS predicts \u003cem\u003ebetter\u003c/em\u003e outcomes. These findings highlight the importance of understanding the impact of selection bias in translational research evaluating PRS as biomarkers of psychiatric disorders, particularly when the intended application is populations enriched for high levels of genetic risk.\u003c/p\u003e","manuscriptTitle":"The Impact of Selection Bias on Genetic Prediction Using the Bipolar Polygenic Risk Score in First-Admission Psychosis","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-07-16 15:59:02","doi":"10.21203/rs.3.rs-4536236/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"b90f1860-bf51-4081-aa1f-70c7886568b2","owner":[],"postedDate":"July 16th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":33416497,"name":"Biological sciences/Genetics"},{"id":33416498,"name":"Biological sciences/Psychology"}],"tags":[],"updatedAt":"2024-09-16T13:25:54+00:00","versionOfRecord":[],"versionCreatedAt":"2024-07-16 15:59:02","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-4536236","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4536236","identity":"rs-4536236","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}