Understanding Polygenic Risk Scores in Psychiatric Genetics

Precision medicine promises individualized care based on genomic architecture. Over recent years, clinicians and researchers turned to polygenic risk scores to evaluate inherited susceptibility to psychiatric illness. These complex mathematical scores quantify cumulative genetic liability by aggregating small effect sizes across millions of single nucleotide polymorphisms. In psychiatry, investigators hoped these computational metrics would soon improve early risk detection, refine diagnostic categories, and guide targeted pharmacological therapy for major mental illnesses.

However, psychiatric conditions feature complex polygenicity, high genetic overlap, and profound environmental influences. Severe social adversity, early trauma, and lifestyle factors play decisive roles in symptom expression. Consequently, clinical enthusiasm has outpaced empirical validation regarding how well these genomic metrics perform in daily practice. Before clinicians adopt genomic profiling into psychiatric workflows, they must evaluate two foundational measurement properties: algorithmic stability and diagnostic specificity. Stability ensures that an individual receives consistent risk classifications across differing analytical models. Specificity ensures that genetic profiles accurately differentiate affected individuals from healthy controls. Without rigorous performance in both areas, genomic stratification remains clinically unviable and potentially misleading for patients.

Evaluating Genomic Prediction Across the UK Biobank Cohort

To rigorously test these parameters, investigators analyzed extensive data from the UK Biobank. The massive community cohort initially enrolled 502,534 volunteers aged 37 to 73 years across the United Kingdom. Researchers selected 408,853 White British participants to minimize confounding from ancestral heterogeneity. Next, they calculated polygenic scores using a continuous shrinkage method. This advanced approach adjusts single nucleotide polymorphism effect sizes based on linkage disequilibrium patterns. The researchers generated multiple scores for major depressive disorder, alcohol use disorder, schizophrenia, and type 2 diabetes.

Additionally, the investigators used summary statistics from distinct, well-powered genome-wide association studies to formulate separate scores for each condition. Specifically, they constructed three alternative scoring algorithms for depression, alcohol misuse, and type 2 diabetes, along with two algorithms for schizophrenia. They quantified score stability through correlation coefficients and measured how consistently individuals landed in the top 5% risk tier. Furthermore, the researchers calculated clinical sensitivity and specificity against documented diagnoses in electronic health records. This systematic approach tested whether independent algorithms assign patients to the same clinical categories. Consequently, the study provides valuable real-world benchmarks for clinical genomics.

Low Stability and Algorithmic Discordance in Patient Stratification

The study demonstrated remarkable instability among polygenic scores that targeted identical disorders. Statistical correlations between distinct scoring models varied widely across psychiatric conditions. For schizophrenia, the correlation coefficient between the two scores reached 0.78. However, correlations fluctuated between 0.16 and 0.78 for major depressive disorder. Similarly, scores for alcohol use disorder yielded correlations between 0.13 and 0.90. These numbers reveal that alternative computational pipelines prioritize divergent genetic markers. As a result, different algorithms do not capture the same underlying biological signal.

This technical divergence creates severe discordance when classifying individual patients. In clinical practice, physicians often stratify high-risk patients using the top 5th percentile. Strikingly, only 47.7% of individuals flagged as high risk for schizophrenia by one score retained that status under the other. Thus, more than half of high-risk patients lost their designation under an alternative model. Concordance proved even worse for other psychiatric conditions. Consistent high-risk identification ranged from 9.5% to 47.0% for depression and from 8.3% to 63.5% for alcohol misuse. Therefore, a patient classified as vulnerable under one test would receive a normal result under another. Such algorithmic instability undermines diagnostic reliability and prevents safe clinical adoption.

Poor Specificity and Somatic Benchmarking Against Diabetes

Beyond score stability, clinical diagnostic tests require adequate sensitivity and high specificity. In this study, sensitivity varied substantially across conditions. Schizophrenia models achieved good sensitivity, ranging from 90.2% to 93.3%. Major depressive disorder demonstrated moderate sensitivity of 66.1% to 74.4%, whereas alcohol use disorder achieved 72.3% to 74.2%. Type 2 diabetes scores demonstrated moderate to good sensitivity, spanning 77.3% to 96.3%. Although high sensitivity detects many true cases, low specificity fundamentally limits clinical applicability.

Specifically, all evaluated polygenic scores exhibited poor specificity, ranging between 50.7% and 56.4%. These rates barely exceed pure chance. Consequently, approximately half of all healthy individuals would test positive on a genetic screening panel. Comparing psychiatric traits with type 2 diabetes highlights why this failure is uniquely problematic in psychiatry. For diabetes, different scores also showed modest correlations (0.29 to 0.77) and low top-tier concordance (14.1% to 45.2%). However, clinicians diagnose diabetes using objective serum biomarkers like glycated hemoglobin. Conversely, psychiatry lacks objective confirmatory biomarkers. Generating massive false-positive results in healthy patients induces psychological distress, diagnostic confusion, and social stigmatization. Therefore, current polygenic scores cannot serve as reliable screening instruments.

Clinical Implications and Recommendations for Practitioners

These findings carry immediate implications for clinical practice and patient counseling. Direct-to-consumer testing companies increasingly market disease susceptibility reports directly to the public. However, practicing physicians must explain that polygenic profiling remains an investigative research tool rather than an actionable diagnostic assay. Because different commercial algorithms produce discordant rankings and high false-positive rates, clinicians should counsel patients against altering psychotropic treatments based on consumer genomic kits. Premature clinical adoption risks misallocating healthcare resources and causing undue anxiety.

Furthermore, generalizability presents a critical challenge for global health systems, particularly across diverse populations in India and South Asia. The UK Biobank findings reflect a homogenous White British cohort. When researchers apply European-derived scoring algorithms to non-European populations, predictive accuracy declines even further due to allele frequency differences. Consequently, practitioners must continue to base psychiatric management on established clinical methods. Comprehensive psychiatric histories, family pedigrees, mental state examinations, and longitudinal follow-ups remain the undisputed gold standard of care. Until computational geneticists establish standardized algorithmic consensus, validate diverse cohorts, and resolve low specificity, polygenic scores cannot safely guide clinical decision-making in psychiatric practice.

Frequently Asked Questions

Can polygenic risk scores reliably confirm a psychiatric diagnosis?

No, polygenic risk scores cannot confirm psychiatric conditions. Their diagnostic specificity remains extremely low, between 50.7% and 56.4%, generating an unacceptable volume of false-positive results. Clinicians must establish psychiatric diagnoses through thorough clinical evaluations, longitudinal observations, and standardized diagnostic criteria rather than relying on current polygenic scoring tools.

Why do different algorithms assign conflicting risk tiers to the same patient?

Different algorithms utilize varying discovery cohorts, variant weighting methods, and continuous shrinkage parameters. Consequently, statistical correlations between alternative scores remain inconsistent. In large biobank studies, less than half of individuals categorized as high risk by one formula maintained that status under another, demonstrating substantial technical discordance across models.

Are direct-to-consumer psychiatric polygenic tests clinically actionable?

Direct-to-consumer genetic tests for psychiatric risk lack clinical utility and regulatory validation. Because specificity is low and stability across algorithms is erratic, commercial results can easily mislead patients. Clinicians should advise patients against modifying treatment or making life decisions based on these premature direct-to-consumer genomic profiles.

Disclaimer: This content is for informational and educational purposes only... Refer to the latest local and national guidelines for clinical practice.

References

1. Mollon J et al. Low Stability and Specificity of Polygenic Risk Scores for Major Psychiatric Disorders Limit Their Clinical Utility. Biol Psychiatry. 2025 Sep 15. doi: 10.1016/j.biopsych.2025.03.006. PMID: 40113122.


2. Allegrini AG, Cheesman R, Rimfeld K, et al. Specificity of Polygenic Scores for Psychiatric Disorders Beyond Transdiagnostic Genetic Risk. JAMA Netw Open. 2026;9(1):e2548518.


3. Murray GK, Lin T, Austin J, et al. Polygenic risk scores in psychiatry: Will they be useful for clinicians? F1000Res. 2019;8:1211.