Considering social risk alongside genetic risk for bipolar disorder in the All of Us Research Program

In a large-scale study of the All of Us Research Program, researchers found that social risk factors, such as perceived stress and adverse childhood experiences, exhibit stronger associations with bipolar disorder than genetic risk scores alone, demonstrating that integrating social and environmental data with genetics is essential for more accurate and equitable psychiatric risk prediction.

The Gist

The Big Picture: It's Not Just About the "Genetic Lottery"

Imagine your risk for developing Bipolar Disorder (BD) is like trying to predict the weather. For a long time, scientists thought the only thing that mattered was the genetic blueprint you were born with—like looking at the barometer to see if a storm is coming.

This study, using data from over 115,000 people in the "All of Us" research program, asks a simple but revolutionary question: "What if the weather isn't just about the barometer? What if it's also about the wind, the humidity, and the temperature?"

The researchers found that while your genes do matter, your social environment (stress, neighborhood, discrimination, childhood trauma) is often a much stronger predictor of whether someone will develop Bipolar Disorder than their genes alone.

---

The Analogy: The "Seed" and the "Soil"

Think of Bipolar Disorder risk like a seed.

• The Genes (PRS): This is the seed itself. Some seeds are "high risk" (they are more likely to grow into a stormy plant), and some are "low risk."
• The Social Environment: This is the soil, water, and sunlight.

The Old Way: Scientists used to focus almost entirely on the seed. They tried to predict the storm just by looking at the seed's DNA.
The New Finding: This study shows that even a "low-risk" seed planted in toxic soil (high stress, discrimination, poverty, bad neighborhoods) is likely to grow into a stormy plant. Conversely, a "high-risk" seed planted in rich, healthy soil (low stress, strong community, good support) might stay small and manageable.

In fact, the study found that a person with the lowest genetic risk but a high social burden (stress, trauma, etc.) was just as likely to have Bipolar Disorder as someone with the highest genetic risk but a low social burden.

The Key Players: What Did They Measure?

The researchers looked at two main things:

1. The Genetic Score (The Seed): They calculated a "Polygenic Risk Score" (PRS). Think of this as a genetic report card.

   • The Result: It works, but it's not perfect. It explains only a small slice of the puzzle (about 1-2% in real-world data, compared to the 9% seen in perfect lab settings). It's like a weather forecast that gets the day right only 20% of the time.

2. The Social Factors (The Soil): They measured six specific "social risks":

   • Perceived Stress: How overwhelmed you feel.
   • Discrimination in Medical Settings: Feeling judged or mistreated by doctors.
   • Neighborhood Social Cohesion: Do you trust your neighbors? Do they look out for each other?
   • Neighborhood Disorder: Is your neighborhood safe, or is it chaotic and unsafe?
   • Medication Non-adherence due to Cost: Can you afford your pills?
   • Adverse Childhood Experiences (ACEs): Trauma or abuse you faced as a kid.

The Shocking Result: Every single one of these social factors was a stronger predictor of Bipolar Disorder than the genetic score.

• Example: If you have high perceived stress, your risk doubles. If you have 4+ childhood traumas, your risk nearly triples. These numbers were much higher than the risk increase seen from having "bad genes."

The "Interaction" Question: Do They Mix?

The researchers wondered: Does a bad seed grow even worse in bad soil? (This is called a "Gene-Environment Interaction").

They found that the factors mostly work additively, not multiplicatively.

• Analogy: Imagine you are driving a car.
  • Genes are the engine.
  • Social Risk is the road conditions.
  • If you have a weak engine (low genetic risk) but you are driving on a road covered in ice and potholes (high social risk), you will crash.
  • If you have a powerful engine (high genetic risk) but you are driving on a smooth, sunny highway (low social risk), you might be fine.
  • The study suggests that the "road conditions" (social factors) are so powerful that they can overwhelm the "engine" (genes).

Why This Matters for Everyone

1. It's Not Just "Bad Luck" in the Genes: If you have Bipolar Disorder, it's not just because you "inherited bad genes." It's often because of the world you live in.
2. Equity is Key: The study showed that genetic tests work best for people of European ancestry and are less accurate for others. But social factors (like stress and discrimination) affect everyone. By focusing on social factors, we can help everyone, not just those with the "right" genetic background.
3. Actionable Change: You can't change your DNA. But you can change your environment.
   • We can fix the "soil."
   • We can reduce stress.
   • We can make healthcare less discriminatory.
   • We can ensure people can afford their meds.
   • We can support children who face trauma.

The Bottom Line

This paper tells us that to understand and predict mental health, we need to stop looking at the seed in isolation. We have to look at the garden.

If we want to prevent or manage Bipolar Disorder, we need to build better gardens (communities, healthcare systems, and support networks) for everyone, regardless of what seeds they were born with. The "social soil" is just as important as the genetic seed.

Technical Summary

1. Problem Statement

Polygenic risk scores (PRS) have emerged as a tool for stratifying risk in complex diseases, but their utility in psychiatric disorders like Bipolar Disorder (BD) is limited. Current PRS for BD explain only a small fraction of phenotypic variance (~1–9% in curated cohorts) and suffer from significantly reduced predictive performance in non-European populations due to Eurocentric bias in Genome-Wide Association Studies (GWAS). Furthermore, psychiatric disorders are heavily influenced by social determinants of health (SDOH), yet most risk prediction models rely solely on genetic data, omitting critical environmental and social contexts. This omission can bias risk estimates and fails to capture the holistic nature of disease liability. The study aims to evaluate whether integrating social risk factors with genetic liability improves risk prediction and to understand how these factors interact in a diverse, real-world clinical cohort.

2. Methodology

Cohort and Data Source:

• Dataset: The study utilized data from the All of Us Research Program (AoU), specifically Data Release Version 8 (enrolled by Oct 1, 2023).
• Sample Size: From 414,830 participants with whole-genome sequencing (WGS) data, 115,275 individuals (7,000 BD cases, 108,275 controls) were retained after quality control (QC), Electronic Health Record (EHR) filtering, and phenotyping.
• Ancestry: The cohort included European (58.5%), African (22.5%), Latino/Admixed American (16.7%), and East Asian (2.3%) participants. Analyses were stratified by these groups.

Phenotyping (Case Definition):

• Cases: Defined via EHR data using a stringent, clinician-informed schema requiring:
  • At least 2 BD diagnostic codes (ICD-9/10 or SNOMED) >30 days apart; OR
  • 1 BD code + a prescription for a mood stabilizer (Lithium, Valproate, Lamotrigine, Carbamazepine) within one year; OR
  • 1 BD code + high volume of mood stabilizer records.
  • Exclusion of participants with conflicting diagnoses (e.g., schizophrenia) unless BD codes were more recent.
• Controls: Participants with no BD codes and no exclusionary psychiatric diagnoses.

Genetic Risk Assessment:

• PRS Calculation: A Polygenic Risk Score for BD ($PRS_{BD}$) was calculated using summary statistics from the largest multi-ancestry PGC meta-analysis (excluding self-reported cases to improve clinical validity).
• Method: Bayesian regression framework (PRS-CS-auto) was used with a 1000 Genomes European reference panel.
• Adjustment: Raw scores were adjusted for the top 16 genetic principal components (PCs) to correct for population stratification.

Social Risk Measures:
Six social risk factors were derived from AoU survey modules, covering domains known to impact BD:

1. Perceived Stress (PSS-10): Continuous score.
2. Discrimination in Medical Settings (DMS): Continuous score.
3. Neighborhood Social Cohesion (SCNS): Continuous score.
4. Perceived Neighborhood Disorder (PNDS): Continuous score.
5. Cost-Related Medication Non-adherence (CRMN): Binary (any non-adherence vs. none).
6. Adverse Childhood Experiences (ACE): Binary (≥4 ACEs vs. <4).

• Preprocessing: Missing data (<70% completion) was excluded; remaining missing items were imputed using k-nearest neighbors. Scores were standardized (Z-scores) or binarized as appropriate.

Statistical Analysis:

• Models: Logistic regression models were fitted to predict BD case status.
• Covariates: Age, age², sex/gender, EHR record length/depth, and 16 genetic PCs.
• Metrics:
  • Variance Explained: Liability-scale $R^2$ estimated via bootstrapping.
  • Discrimination: Area Under the Curve (AUC).
  • Interactions: Likelihood Ratio Tests (LRT) were used to test for multiplicative interactions between $PRS_{BD}$ and each social risk factor.
  • Correction: Benjamini-Hochberg procedure for multiple testing.

3. Key Results

Polygenic Risk Score Performance:

• Association: $PRS_{BD}$ was significantly associated with BD case status across all ancestry groups ($P < 5.91 \times 10^{-52}$ for All Ancestries).
• Predictive Power: The variance explained ($R^2$) was substantially lower than in curated research cohorts:
  • European: 1.86%
  • Latino/Admixed American: 1.65%
  • African: 0.60%
• Decile Analysis: Individuals in the top PRS decile had roughly double the odds of BD compared to the median decile (OR ~1.45), but the absolute risk difference was modest (4.4% to 8.6% prevalence in the All Ancestries sample).

Social Risk Factor Associations:

• Effect Sizes: All six social risk factors showed significant associations with BD. Crucially, social risk factors exhibited larger effect sizes than the PRS.
  • Perceived Stress (PSS): OR = 2.05 per SD increase (All Ancestries).
  • Adverse Childhood Experiences (ACE): OR = 2.68 for ≥4 ACEs (All Ancestries); OR = 2.89 in European ancestry.
  • Other factors (DMS, PNDS, SCNS, CRMN) showed ORs ranging from 1.47 to 1.55 per SD.
• Ancestry Differences: Effect sizes for social risks were generally lower in the African ancestry group compared to European and Latino groups, potentially due to diagnostic disparities (misdiagnosis of BD as schizophrenia) or survey response biases.

Combined Genetic and Social Risk:

• Additivity: The study found primarily additive effects rather than strong multiplicative interactions.
• Interaction Tests: Interaction terms between PRS and social risk were generally small and did not survive multiple testing correction (FDR > 0.05). Nominally significant negative interactions were observed for SCNS and CRMN, suggesting a slight attenuation of genetic risk effects in high-social-risk groups.
• Prevalence Stratification: Individuals in the lowest genetic risk decile but with high social burden (e.g., high stress, low cohesion) exhibited BD prevalence rates comparable to or exceeding those in the highest genetic risk decile with low social burden. This highlights that high social risk can override low genetic risk.

4. Key Contributions

1. Quantification of Social vs. Genetic Risk: Demonstrated that in a real-world clinical cohort, specific social determinants (stress, ACEs, discrimination) have larger individual effect sizes on BD liability than current polygenic risk scores.
2. Real-World PRS Performance: Provided a realistic assessment of PRS performance in a diverse, EHR-derived cohort, showing significant attenuation ($R^2 \approx 1.86\%$ in Europeans) compared to curated research cohorts ($R^2 \approx 9\%$). This "phenotyping gap" is attributed to EHR noise, misdiagnosis, and the inclusion of heterogeneous cases.
3. Equity and Portability: Highlighted the "portability gap" where PRS performance drops significantly in African ancestry groups, while also noting that social risk effect sizes vary by ancestry, likely due to structural healthcare disparities and diagnostic bias.
4. Integrated Risk Framework: Provided empirical evidence that risk prediction models ignoring social context are incomplete. The data supports a shift toward integrated genetic-social frameworks for more equitable psychiatric epidemiology.

5. Significance and Implications

• Clinical Translation: The findings suggest that relying solely on PRS for clinical risk stratification in psychiatry is insufficient and potentially inequitable. Incorporating modifiable social risk factors (e.g., stress management, addressing medication cost barriers) could significantly improve risk prediction accuracy.
• Public Health: The observation that high social burden can elevate risk to levels comparable to high genetic risk underscores the importance of addressing social determinants of health as a primary intervention strategy for mental health.
• Methodological Insight: The study emphasizes the need for deep, specific phenotyping in genomic research. The discrepancy between curated and EHR-based PRS performance indicates that "real-world" genetic signals are noisier, necessitating larger, more diverse cohorts and ancestry-specific pipelines to achieve clinical utility.
• Future Directions: The authors call for future studies to expand to larger, more diverse cohorts to robustly test gene-environment interactions and to develop risk models that jointly account for genetic liability and social context to enable earlier, more targeted interventions.