19. GENETICS OF EHR-DERIVED ANTIDEPRESSANT NONRESPONSE ACROSS COUNTRIES AND ANCESTRY GROUPS

Background Genetic predictors of antidepressant (AD) response can advance mechanistic understanding and personalised medicine, but clinical pharmacogenomic trials are underpowered to detect individual genetic variants. Electronic health records (EHRs) offer scalable phenotypes of treatment outcomes, though heterogeneity across biobanks and differences in clinical guidelines and practice pose challenges. Here, we derived proxy measures of nonresponse to the first SSRI (selective serotonin reuptake inhibitor) or SNRI (serotonin and norepinephrine reuptake inhibitor) trial and conducted GWAS meta-analyses stratified by country and ancestry group to parse the shared and divergent genetic architecture of AD nonresponse. Methods We reviewed the US and UK clinical guidelines to inform measures of AD nonresponse in EHR data from All of Us (AoU), Vanderbilt University Biorepository (BioVU), Mass General Brigham Biobank (MGBB), UK Biobank (UKB), and Generation Scotland (GenScot). The US guidelines consider second-generation ADs as broadly equivalent in efficacy, whereas the UK guidelines explicitly recommend SSRIs as first-line ADs. We therefore identified treatment outcomes following the first SSRI/SNRI trial in the US and the first SSRI trial in the UK. Given similar treatment-resistant pathways across countries, we defined nonresponse uniformly as AD switching and/or augmentation with another AD or antipsychoticWe examined cross-ancestry differences in nonresponse indicators in AoU. Preliminary GWAS in European-ancestry (EUR-like) participants from AoU, BioVU, and UKB were followed by a US meta-analysis and a cross-national meta-analysis including UKB. Results In AoU (N ≈ 47k; 16% nonresponse), African-ancestry (AFR-like) participants had higher odds of nonresponse than EUR-like (OR = 1.12; 95% CI 1.02–1.23; p = 0.013; excluding individuals ever diagnosed with schizophrenia or bipolar disorder), driven by higher rates of antipsychotic augmentation despite lower AD switchingPreliminary EUR-like GWAS showed similar SNP-h2 estimates (3-4%) in AoU (Neff ≈ 17k), BioVU (Neff ≈ 16k), and UKB (Neff ≈ 17k). The AoU+BioVU meta-analysis (h2 ≈ 2%) yielded no genome-wide significant associations, but 14 independent loci reached suggestive significance (p < 5e-6), including loci previously linked to cognitive ability, lamotrigine treatment response, and neuroticism. GWAS in UKB identified 8 suggestive loci, and the cross-national meta-analysis (h2 < 0.1%) identified 9, only one of which overlapped with the US-only results, underscoring cross-cohort heterogeneityThis work is being extended to Mount Sinai Million (US) and Our Future Health (UK), with ongoing ancestry-stratified GWAS in AoU (AFR-like, Admixed American [AMR]-like), BioVU (AFR-like), and MGBB and GenScot (EUR-like). Discussion Early GWAS results highlight genomic loci linked to both treatment response and mental health traits. Qualitative differences in how AD nonresponse manifests across ancestry groups in real-world settings may partly reflect differences in clinical decision-making, including possible implicit biases. The striking drop in heritability upon adding UKB is likely due to differences in both healthcare settings and cohort characteristics. Ongoing work will provide more robust evidence for the shared and diverging genetic architecture of AD nonresponse across ancestries and healthcare settings, informing efforts towards more equitable and personalised antidepressant prescribing.