Evaluating phecodes, clinical classification software, and ICD-9-CM codes for phenome-wide association studies in the electronic health record

Proteomic insights into metabolic dysfunction-associated steatotic disease: Identifying therapeutic targets and assessing on-target side effects

AIMS

The prevalence of metabolic dysfunction-associated steatotic liver disease (MASLD) is rising sharply, yet treatment options remain inadequate. To uncover new therapeutic targets for MASLD, we conducted a comprehensive proteome-wide Mendelian randomization (MR) and phenome-wide association study (PheWAS).

MATERIALS AND METHODS

Discovery MR utilized protein quantitative trait loci (pQTL) data on 4907 plasma protein levels from 35,559 individuals, alongside genome-wide association study (GWAS) on MASLD from the Million Veteran Program (68,725 cases / 95,482 controls). Validation comprised five pairwise combinations of these discovery datasets with three additional datasets: pQTL data for 2923 proteins from the UK Biobank, and liver biopsy-confirmed MASLD GWAS (1483 cases/17,781 controls) and MRI-liver fat GWAS (31,377 subjects) (excluding discovery pair). Candidate proteins underwent druggability assessment and on-target side effect evaluation via PheWAS.

KEY FINDINGS

We identified 26 proteins associated with MASLD after Bonferroni correction (P < 1.16 × 10-5), with 19 of them showing no significant reverse association. Interleukin-6 (IL-6), alpha-1-antitrypsin (α1-antitrypsin), 5-hydroxytryptamine receptor 7 (5-HT7R), ephrin-B1 (EFNB1), and protein MENT (CA056) were replicated. Notably, IL-6 (OR = 2.02; 95 % CI 1.54-2.64), 5-HT7R (OR = 2.73; 95 % CI 1.96-3.80), and EFNB1 (OR = 1.82; 95 % CI 1.59-2.08) were positively associated with MASLD risk, whereas α1-antitrypsin (OR = 0.84; 95 % CI 0.78-0.90) and CA056 (OR = 0.90; 95 % CI 0.86-0.94) appeared protective. Among these, IL-6, 5-HT7R, and α1-antitrypsin were druggable. PheWAS identified potential cardiovascular side effects for 5-HT7R and α1-antitrypsin.

SIGNIFICANCE

The integrative study identified several plasma proteins associated with MASLD. IL-6, α1-antitrypsin, 5-HT7R, EFNB1 and CA056 deserve further investigation as potential drug targets for MASLD.

Beyond Phecodes: leveraging PheMAP to identify patients lacking diagnosis codes in electronic health records

OBJECTIVE

Diagnosis codes documented in electronic health records (EHR) are often relied upon to clinically phenotype patients for biomedical research. However, these diagnoses can be incomplete and inaccurate, leading to false negatives when searching for patients with phenotypes of interest. This study aims to determine whether PheMAP, a comprehensive knowledgebase integrating multiple clinical terminologies beyond diagnosis to capture phenotypes, can effectively identify patients lacking relevant EHR diagnosis codes.

MATERIALS AND METHODS

We investigated a collection of 3.5 million patient records from Vanderbilt University Medical Center's EHR and focused on 4 well-studied phenotypes: (1) type 2 diabetes mellitus (T2DM), (2) dementia, (3) prostate cancer, and (4) sensorineural hearing loss. We applied PheMAP to match structured concepts in patient records and calculated a phenotype risk score (PheScore) to indicate patient-phenotype similarity. Patients meeting predefined PheScore criteria but lacking diagnosis codes were identified. Clinically knowledgeable experts adjudicated randomly selected patients per phenotype as Positive, Possibly Positive, or Negative.

RESULTS

Our approach indicated that 5.3% of patients lacked a diagnosis for T2DM, 4.5% for dementia, 2.2% for prostate cancer, and 0.2% for sensorineural hearing loss. The expert review indicated 100% precision (for Possibly Positive or Positive cases) for dementia and sensorineural hearing loss, and 90.0% and 85.0% precision for T2DM and prostate cancer, respectively. Excluding Possibly Positive cases, the precision for T2DM and prostate cancer was 88.9% and 81.3%, respectively.

CONCLUSIONS

Leveraging clinical terminologies incorporated by PheMAP can effectively identify patients with phenotypes who lack EHR diagnosis codes, thereby enhancing phenotyping quality and related research reliability.

Association of Physical Activity from Wearable Devices and Chronic Disease Risk: Insights from the All of Us Research Program

Physical activity is a modifiable factor influencing chronic disease risk. Previous studies often relied on self-reported activity measures or short-term assessments, limiting their accuracy. Leveraging Fitbit-derived data from the All of Us Research Program, we investigated associations between long-term physical activity patterns and chronic disease incidence in a diverse cohort. The study included 22,019 participants with at least six months of Fitbit monitoring and linked electronic health records. Key activity metrics included daily step count, activity calories, elevation gain, and activity duration at different intensities. Higher physical activity levels were associated with a lower risk of multiple chronic diseases. A 2,000-step increase in daily step count was linked to a reduced risk of obesity (hazard ratio [HR] = 0.85, 95% confidence interval [CI]: 0.80-0.90), type 2 diabetes (HR = 0.78, CI: 0.72-0.84), and major depressive disorder (HR = 0.83, CI: 0.77-0.90). Elevation gain was inversely associated with obesity (HR = 0.86, CI: 0.78-0.95) and type 2 diabetes (HR = 0.65, CI: 0.53-0.80). Increased time spent in very active intensity correlated with a lower risk of multiple conditions, including obstructive sleep apnea and morbid obesity. Conversely, prolonged sedentary time was associated with an increased risk of cardiometabolic diseases, including obesity (HR = 1.08, CI: 1.06-1.10) and essential hypertension (HR = 1.05, CI: 1.04-1.07). A sensitivity analysis using BMI-defined obesity instead of EHR-based diagnoses confirmed the robustness of these associations. These findings underscore the protective role of increased physical activity and reduced sedentary time in mitigating chronic disease risk. They support the development of personalized physical activity recommendations and targeted public health interventions aimed at improving long-term health outcomes. Future research integrating machine learning approaches could further refine activity-based disease prevention strategies.

Associations of a multidimensional polygenic sleep health score and a sleep lifestyle index with disease outcomes and their interaction in a clinical biobank

OBJECTIVES

Sleep is a complex behavior regulated by genetic and environmental factors impacting disease outcomes. However, the effect of multidimensional sleep encompassing several sleep dimensions on common diseases, specifically mental health disorders, has yet to be fully elucidated. Using the Mass General Brigham Biobank, we examined the association of multidimensional sleep with disease outcomes and investigated whether sleep behaviors modulate genetic predisposition to unfavorable sleep on mental health diseases.

METHODS

We generated a Polygenic Sleep Health Score using previously identified single nucleotide polymorphisms and constructed a Sleep Lifestyle Index based on self-reported questions and electronic health records; tested their association; performed phenome-wide association analyses between these indexes and clinical phenotypes; and analyzed their interaction on prevalent mental health diseases. A total of 15,884 participants were included in the analysis (mean age 54.4; 58.6% female).

RESULTS

The Polygenic Sleep Health Score was associated with the Sleep Lifestyle Index (β=0.050, 95% CI=0.032, 0.068) and with 114 disease outcomes spanning 12 disease groups, including obesity, sleep, and substance use disease outcomes (p<3.3×10-5). The Sleep Lifestyle Index was associated with 458 disease outcomes spanning 17 groups, including sleep, mood, and anxiety disease outcomes (p<5.1×10-5). A total of 108 disease outcomes were associated with both indexes, spanning 12 disease groups. No interactions were found between the indexes on mental health diseases.

CONCLUSIONS

Favorable sleep behaviors and genetic predisposition to healthy sleep may independently protect against disease, underscoring the impact of multidimensional sleep on population health and the need for prevention strategies focused on healthy sleep habits.

Familial Renal Glucosuria and Potential Pharmacogenetic Impact on Sodium-Glucose Cotransporter-2 Inhibitors

Key Points

A significant knowledge gap exists in SLC5A2 's role in familial renal glycosuria and sodium-glucose cotransporter-2 inhibitors' efficacy. Two percent of individuals in the All-of-Us cohort harbor rare genetic variants in SLC5A2 , potentially increasing the risk of familial renal glycosuria. Our trial suggests differential responses to sodium-glucose cotransporter-2 inhibitors in individuals with rare SLC5A2 alleles compared with wild types.

Background

Renal glucosuria is a rare inheritable trait caused by loss-of-function variants in the gene that encodes sodium-glucose cotransporter-2 (SGLT2) (i.e ., SLC5A2 ). The genetics of renal glucosuria is poorly understood, and even less is known on how loss-of-function variants in SLC5A2 may affect response to SGLT2 inhibitors, a new class of medication gaining popularity to treat diabetes by artificially inducing glucosuria.

Methods

We used two biobanks that link genomic with electronic health record data to study the genetics of renal glucosuria. This included 245,394 participants enrolled in the All of Us Research Program and 11,011 enrolled in Marshfield Clinic's Personalized Research Project (PMRP). Association studies in All of Us and PMRP identified ten variants that reached an experiment-wise Bonferroni threshold in either cohort, of which nine were novel. PMRP was further used as a recruitment source for a prospective SGLT2 pharmacogenetic trial. During a glucose tolerance test, the trial measured urine glucose concentrations in 15 SLC5A2 variant–positive individuals and 15 matched wild types with and without an SGLT2 inhibitor.

Results

This trial demonstrated that carriers of SLC5A2 risk variants may be more sensitive to SGLT2 inhibitors compared with wild types (P = 0.075). On the basis of population data, 2% of an ethnically diverse population carried rare variants in SLC5A2 and are at risk of renal glucosuria.

Conclusions

As a result, 2% of individuals being treated with SGLT2 inhibitors may respond differently to this new class of medication compared with the general population, suggesting that a larger investigation into SLC5A2 variants and SGLT2 inhibitors is needed.

Genome-wide association meta-analysis identifies 126 novel loci for diverticular disease and implicates connective tissue and colonic motility

Diverticular disease is a common and morbid complex phenotype influenced by both innate and environmental risk factors. We performed the largest genome-wide association study meta-analysis for diverticular disease, identifying 126 novel loci. Employing multiple downstream analytic strategies, including tissue and pathway enrichment, statistical fine-mapping, allele-specific expression, protein quantitative trait loci and drug-target investigations, and linkage disequilibrium score regression, we prioritized causal genes and produced several lines of evidence linking diverticular disease to connective tissue biology and colonic motility. We substantiated these findings by integrating single-cell RNA sequencing data, showing that prioritized diverticular disease-associated genes are enriched for expression in colonic smooth muscle, fibroblasts, and interstitial cells of Cajal. In quantitative analysis of surgical specimens, we found a substantial reduction in the density of elastin present in the sigmoid colon in severe diverticulitis.

Transformer-based artificial intelligence on single-cell clinical data for homeostatic mechanism inference and rational biomarker discovery

Artificial intelligence (AI) applied to single-cell data has the potential to transform our understanding of biological systems by revealing patterns and mechanisms that simpler traditional methods miss. Here, we develop a general-purpose, interpretable AI pipeline consisting of two deep learning models: the Multi- Input Set Transformer++ (MIST) model for prediction and the single-cell FastShap model for interpretability. We apply this pipeline to a large set of routine clinical data containing single-cell measurements of circulating red blood cells (RBC), white blood cells (WBC), and platelets (PLT) to study population fluxes and homeostatic hematological mechanisms. We find that MIST can use these single-cell measurements to explain 70-82% of the variation in blood cell population sizes among patients (RBC count, PLT count, WBC count), compared to 5-20% explained with current approaches. MIST's accuracy implies that substantial information on cellular production and clearance is present in the single-cell measurements. MIST identified substantial crosstalk among RBC, WBC, and PLT populations, suggesting co-regulatory relationships that we validated and investigated using interpretability maps generated by single-cell FastShap. The maps identify granular single-cell subgroups most important for each population's size, enabling generation of evidence-based hypotheses for co-regulatory mechanisms. The interpretability maps also enable rational discovery of a single-WBC biomarker, "Down Shift", that complements an existing marker of inflammation and strengthens diagnostic associations with diseases including sepsis, heart disease, and diabetes. This study illustrates how single-cell data can be leveraged for mechanistic inference with potential clinical relevance and how this AI pipeline can be applied to power scientific discovery.

Longitudinal Analysis of Electronic Health Records Reveals Medical Conditions Associated with Subsequent Alzheimer's Disease Development

Background

Several health conditions are known to increase the risk of Alzheimer's disease (AD). We aim to systematically identify medical conditions that are associated with subsequent development of AD by leveraging the growing resources of electronic health records (EHRs).

Methods

This retrospective cohort study used de-identified EHRs from two independent databases (MarketScan and VUMC) with 153 million individuals to identify AD cases and age- and gender-matched controls. By tracking their EHRs over a 10-year window before AD diagnosis and comparing the EHRs between AD cases and controls, we identified medical conditions that occur more likely in those who later develop AD. We further assessed the genetic underpinnings of these conditions in relation to AD genetics using data from two large-scale biobanks (BioVU and UK Biobank, total N=450,000).

Results

We identified 43,508 AD cases and 419,455 matched controls in MarketScan, and 1,320 AD cases and 12,720 matched controls in VUMC. We detected 406 and 102 medical phenotypes that are significantly enriched among the future AD cases in MarketScan and VUMC databases, respectively. In both EHR databases, mental disorders and neurological disorders emerged as the top two most enriched clinical categories. More than 70 medical phenotypes are replicated in both EHR databases, which are dominated by mental disorders (e.g., depression), neurological disorders (e.g., sleep orders), circulatory system disorders (e.g. cerebral atherosclerosis) and endocrine/metabolic disorders (e.g., type 2 diabetes). We identified 19 phenotypes that are either associated with individual risk variants of AD or a polygenic risk score of AD.

Conclusions

In this study, analysis of longitudinal EHRs from independent large-scale databases enables robust identification of health conditions associated with subsequent development of AD, highlighting potential opportunities of therapeutics and interventions to reduce AD risk.

Genetic association studies using disease liabilities from deep neural networks

The case-control study is a widely used method for investigating the genetic underpinnings of binary traits. However, long-term, prospective cohort studies often grapple with absent or evolving health-related outcomes. Here, we propose two methods, liability and meta, for conducting genome-wide association studies (GWASs) that leverage disease liabilities calculated from deep patient phenotyping. Analyzing 38 common traits in ∼300,000 UK Biobank participants, we identified an increased number of loci in comparison to the number identified by the conventional case-control approach, and there were high replication rates in larger external GWASs. Further analyses confirmed the disease specificity of the genetic architecture; the meta method demonstrated higher robustness when phenotypes were imputed with low accuracy. Additionally, polygenic risk scores based on disease liabilities more effectively predicted newly diagnosed cases in the 2022 dataset, which were controls in the earlier 2019 dataset. Our findings demonstrate that integrating high-dimensional phenotypic data into deep neural networks enhances genetic association studies while capturing disease-relevant genetic architecture.

Evaluation of polygenic scores for hypertrophic cardiomyopathy in the general population and across clinical settings

Hypertrophic cardiomyopathy (HCM) is an important cause of morbidity and mortality, with pathogenic variants found in about a third of cases. Large-scale genome-wide association studies (GWAS) demonstrate that common genetic variation contributes to HCM risk. Here we derive polygenic scores (PGS) from HCM GWAS and genetically correlated traits and test their performance in the UK Biobank, 100,000 Genomes Project, and clinical cohorts. We show that higher PGS significantly increases the risk of HCM in the general population, particularly among pathogenic variant carriers, where HCM penetrance differs 10-fold between those in the highest and lowest PGS quintiles. Among relatives of HCM probands, PGS stratifies risks of developing HCM and adverse outcomes. Finally, among HCM cases, PGS strongly predicts the risk of adverse outcomes and death. These findings support the broad utility of PGS across clinical settings, enabling tailored screening and surveillance and stratification of risk of adverse outcomes.

Unsupervised Learning-Derived Complex Metabolic Signatures Refine Cardiometabolic Risk

BACKGROUND

Cardiometabolic diseases have become a leading cause of morbidity and mortality globally. Nuclear magnetic resonance metabolomics represents a precise tool for assessing metabolic individuality.

OBJECTIVES

This study aimed to use unsupervised learning to decode plasma metabolomic profiles, providing new insights into the etiology of cardiometabolic diseases.

METHODS

We applied unsupervised learning to generate robust metabolic signatures from the plasma profiles of 118,001 UK Biobank participants. Phenome-wide and genome-wide association studies were conducted to reveal their phenomic and genetic architectures. Integrated prospective cohort analyses and Mendelian randomization clarified their roles in cardiometabolic risks.

RESULTS

Eleven metabolic clusters were identified, linked to 101 loci and 445 phenotypes, mostly cardiometabolic diseases. These novel signatures partially outperformed traditional lipids in cardiometabolic risk prediction. Triglyceride-rich lipoproteins demonstrated superior predictive power for ischemic heart disease, type 2 diabetes, and hypertension, compared with apolipoprotein B and lipoprotein(a). Non-high-density lipoprotein cholesterol was found to increase the risk of hyperlipidemia and ischemic heart disease while offering a protective effect against type 2 diabetes. Furthermore, different high-density lipoprotein clusters showed heterogeneous associations with cardiometabolic diseases, with high-density lipoprotein subpopulations enriched in free cholesterol or triglyceride increasing risk, and those enriched in cholesterol esters providing protection.

CONCLUSIONS

These metabolic signatures extract comprehensive information from the metabolomic profile while maintaining clarity and interpretability, facilitating clinical translation. The findings emphasize the crucial roles of lipid subpopulations in cardiometabolic risks, encouraging clinicians to take a more nuanced approach to managing blood lipids and balancing different disease risks.

Improving topic modeling performance on social media through semantic relationships within biomedical terminology

Topic modeling utilizes unsupervised machine learning to detect underlying themes within texts and has been deployed routinely to analyze social media for insights into healthcare issues. However, the inherent messiness of social media hinders the full realization of this technique's potential. As such, we hypothesized that restricting medical concepts in social media texts to specific related semantic types and applying topic modeling to these concepts could be a feasible approach to overcome the challenge of traditional topic modeling for social media texts. Therefore, we developed a semantic-type-based topic modeling pipeline to discover self-reported health-related topics. This pipeline integrated semantic type information and Systematized Medical Nomenclature for Medicine (SNOMED) precoordinated expressions into a traditional topic modeling approach to enhance effectiveness in clustering meaningful, distinct topics. Using social media texts regarding statins for illustration, we evaluated the efficacy of this new approach and validated a newly identified topic using real-world clinical data. Based on expert evaluations, this approach resulted in more novel, distinguishable, and meaningful health-related topics compared to traditional topic modeling. In addition, our electronic health record validation for a newly identified topic in two real-world clinical databases indicated that statin users had a higher prevalence of depression or anxiety compared to matched non-users. Our results indicate that this new topic modeling pipeline can improve the extraction of themes from noisy online discussions, thereby contributing to deeper insights for healthcare research.

A Multimodal Sleep Foundation Model Developed with 500K Hours of Sleep Recordings for Disease Predictions

Sleep is a fundamental biological process with profound implications for physical and mental health, yet our understanding of its complex patterns and their relationships to a broad spectrum of diseases remains limited. While polysomnography (PSG), the gold standard for sleep analysis, captures rich multimodal physiological data, analyzing these measurements has been challenging due to limited flexibility across recording environments, poor generalizability across cohorts, and difficulty in leveraging information from multiple signals simultaneously. To address this gap, we curated over 585,000 hours of high-quality sleep recordings from approximately 65,000 participants across multiple cohorts and developed SleepFM, a multimodal sleep foundation model trained with a novel contrastive learning approach, designed to accommodate any PSG montage. SleepFM produces informative sleep embeddings that enable predictions of future diseases. We systematically demonstrate that SleepFM embeddings can predict 130 future diseases, as modeled by Phecodes, with C-Index and AUROC of at least 0.75 on held-out participants (Bonferroni-corrected p < 0.01). This includes accurate predictions for death (C-Index: 0.84 [95% CI: 0.81-0.87]), heart failure (C-Index: 0.80 [95% CI: 0.77-0.83]), chronic kidney disease (C-Index: 0.79 [95% CI: 0.77-0.81]), dementia (C-Index: 0.85 [95% CI: 0.82-0.87]), stroke (C-Index: 0.78 [95% CI: 0.76-0.81]), atrial fibrillation (C-Index: 0.78 [95% CI: 0.75-0.81]), and myocardial infarction (C-Index: 0.81 [95% CI: 0.78-0.84]). The model's generalizability was further validated through strong performance on the Sleep Heart Health Study (SHHS), a dataset unseen during pre-training. Additionally, SleepFM demonstrates strong performance on traditional sleep analysis tasks, achieving competitive results in both sleep staging (mean F1 scores: 0.70-0.78) and sleep apnea diagnosis (AUROC: 0.90-0.94). Beyond these standard applications, our analysis reveals that specific sleep stages and physiological signals carry distinct predictive power for different diseases. This work demonstrates how foundation models can leverage sleep polysomnography data to uncover the extensive relationship between sleep physiology and future disease risk.

Exploring beyond diagnoses in electronic health records to improve discovery: a review of the phenome-wide association study

Objective

The phenome-wide association study (PheWAS) systematically examines the phenotypic spectrum extracted from electronic health records (EHRs) to uncover correlations between phenotypes and exposures. This review explores methodologies, highlights challenges, and outlines future directions for EHR-driven PheWAS.

Materials and Methods

We searched the PubMed database for articles spanning from 2010 to 2023, and we collected data regarding exposures, phenotypes, cohorts, terminologies, replication, and ancestry.

Results

Our search yielded 690 articles. Following exclusion criteria, we identified 291 articles published between January 1, 2010, and December 31, 2023. A total number of 162 (55.6%) articles defined phenomes using phecodes, indicating that research is reliant on the organization of billing codes. Moreover, 72.8% of articles utilized exposures consisting of genetic data, and the majority (69.4%) of PheWAS lacked replication analyses.

Discussion

Existing literature underscores the need for deeper phenotyping, variability in PheWAS exposure variables, and absence of replication in PheWAS. Current applications of PheWAS mainly focus on cardiovascular, metabolic, and endocrine phenotypes; thus, applications of PheWAS in uncommon diseases, which may lack structured data, remain largely understudied.

Conclusions

With modern EHRs, future PheWAS should extend beyond diagnosis codes and consider additional data like clinical notes or medications to create comprehensive phenotype profiles that consider severity, temporality, risk, and ancestry. Furthermore, data interoperability initiatives may help mitigate the paucity of PheWAS replication analyses. With the growing availability of data in EHR, PheWAS will remain a powerful tool in precision medicine.

The impact of imprecise case definitions in electronic health record research: a melanoma case-study from the Million Veteran Program

Cases for a disease can be defined broadly using diagnostic codes, or narrowly using gold-standard confirmation that often is not available in large administrative datasets. These different definitions can have significant impacts on the results and conclusions of studies. We conducted this study to assess how using melanoma phecodes versus histologic confirmation for invasive or in situ melanoma impacts the results of a genome-wide association study (GWAS) using the Million Veteran Program. Melanoma status was determined three ways: (1) by the presence of two or more phecodes, (2) histologically-confirmed invasive melanoma, and (3) histologically-confirmed melanoma in situ. We conducted a GWAS for variants with minor allele frequencies of 1% or greater. There were 45,665 cases in the phecode cohort, 5364 cases in the confirmed invasive melanoma cohort, and 4792 cases in the confirmed melanoma in situ cohort. There were 20,457 variants significant at the genome-wide level in the phecode cohort, 2582 in the invasive melanoma cohort, and 1989 in the melanoma in situ cohort. Most of the variants identified in the phecode cohort did not replicate in the histologically-confirmed cohorts. The different case definitions led to large differences in sample size and variants associated at the genome-wide level. Unvalidated and imprecise case definitions can lead to less accurate results. Investigators should use validated phenotypes when gold-standard definitions are not available.

Medical history predicts phenome-wide disease onset and enables the rapid response to emerging health threats

The COVID-19 pandemic exposed a global deficiency of systematic, data-driven guidance to identify high-risk individuals. Here, we illustrate the utility of routinely recorded medical history to predict the risk for 1741 diseases across clinical specialties and support the rapid response to emerging health threats such as COVID-19. We developed a neural network to learn from health records of 502,489 UK Biobank participants. Importantly, we observed discriminative improvements over basic demographic predictors for 1546 (88.8%) endpoints. After transferring the unmodified risk models to the All of US cohort, we replicated these improvements for 1115 (78.9%) of 1414 investigated endpoints, demonstrating generalizability across healthcare systems and historically underrepresented groups. Ultimately, we showed how this approach could have been used to identify individuals vulnerable to severe COVID-19. Our study demonstrates the potential of medical history to support guidance for emerging pandemics by systematically estimating risk for thousands of diseases at once at minimal cost.

Comparison of methods for building polygenic scores for diverse populations

Polygenic scores (PGSs) are a promising tool for estimating individual-level genetic risk of disease based on the results of genome-wide association studies (GWASs). However, their promise has yet to be fully realized because most currently available PGSs were built with genetic data from predominantly European-ancestry populations, and PGS performance declines when scores are applied to target populations different from the populations from which they were derived. Thus, there is a great need to improve PGS performance in currently under-studied populations. In this work we leverage data from two large and diverse cohorts the Million Veterans Program (MVP) and All of Us (AoU), providing us the unique opportunity to compare methods for building PGSs for multi-ancestry populations across multiple traits. We build PGSs for five continuous traits and five binary traits using both multi-ancestry and single-ancestry approaches with popular Bayesian PGS methods and both MVP META GWAS results and population-specific GWAS results from the respective African, European, and Hispanic MVP populations. We evaluate these scores in three AoU populations genetically similar to the respective African, Admixed American, and European 1000 Genomes Project superpopulations. Using correlation-based tests, we make formal comparisons of the PGS performance across the multiple AoU populations. We conclude that approaches that combine GWAS data from multiple populations produce PGSs that perform better than approaches that utilize smaller single-population GWAS results matched to the target population, and specifically that multi-ancestry scores built with PRS-CSx outperform the other approaches in the three AoU populations.

Phenotypic findings associated with variation in elastin

Variation in the elastin gene (ELN) may contribute to connective tissue disease beyond the known disease associations of supravalvar aortic stenosis and cutis laxa. Exome data from MyCode Community Health Initiative participants were analyzed for ELN rare variants (mean allele frequency <1%, not currently annotated as benign). Participants with variants of interest underwent phenotyping by dual chart review using a standardized abstraction tool. Additionally, all rare variants that met inclusion criteria were collapsed into an ELN gene burden score to perform a phenome-wide association study (PheWAS). Two hundred and ninety-six eligible participants with relevant ELN variants were identified from 184,293 MyCode participants. One hundred and three of 254 living participants (41%) met phenotypic criteria, most commonly aortic hypoplasia, arterial dilation, aneurysm, and dissection, and connective tissue abnormalities. ELN variation was significantly (p < 2.8 × 10-5) associated with "arterial dissection" in the PheWAS and two connective tissue Phecodes approached significance. Variation in ELN is associated with connective tissue pathology beyond classic phenotypes.

The impact of common and rare genetic variants on bradyarrhythmia development

To broaden our understanding of bradyarrhythmias and conduction disease, we performed common variant genome-wide association analyses in up to 1.3 million individuals and rare variant burden testing in 460,000 individuals for sinus node dysfunction (SND), distal conduction disease (DCD) and pacemaker (PM) implantation. We identified 13, 31 and 21 common variant loci for SND, DCD and PM, respectively. Four well-known loci (SCN5A/SCN10A, CCDC141, TBX20 and CAMK2D) were shared for SND and DCD, while others were more specific for SND or DCD. SND and DCD showed a moderate genetic correlation (rg = 0.63). Cardiomyocyte-expressed genes were enriched for contributions to DCD heritability. Rare-variant analyses implicated LMNA for all bradyarrhythmia phenotypes, SMAD6 and SCN5A for DCD and TTN, MYBPC3 and SCN5A for PM. These results show that variation in multiple genetic pathways (for example, ion channel function, cardiac developmental programs, sarcomeric structure and cellular homeostasis) appear critical to the development of bradyarrhythmias.

Clinical and genetic associations for night eating syndrome in a patient biobank

OBJECTIVE

Night eating syndrome (NES) is an eating disorder characterized by evening hyperphagia. Despite having a prevalence comparable to some other eating disorders, NES remains sparsely investigated and poorly characterized. The present study examined the phenotypic and genetic associations for NES in the clinical Mass General Brigham Biobank.

METHOD

Cases of NES were identified through relevant billing codes for eating disorders (F50.89/F50.9) and subsequent chart review; patients likely without NES were set as controls. Other diagnoses were determined from billing codes and collapsed into one of 1,857 distinct phenotypes based on clinical similarity. NES associations with diagnoses were systematically conducted in phenome-wide association scans using logistic regression models with adjustments for age, sex, race, and ethnicity. Polygenic scores for six related traits, namely for anorexia nervosa, depression, insomnia, sleep apnea, obesity, and type 2 diabetes were tested for associations with NES among participants of European ancestry using adjusted logistic regression models.

RESULTS

Phenome-wide scans comparing patients with NES against controls (cases n = 88; controls n = 64,539) identified associations with 159 clinical diagnoses spanning 13 broad disease groups including endocrine/metabolic and digestive diseases. Notable associations were evident for bariatric surgery, vitamin D deficiency, sleep disorders (sleep apnea, insomnia, and restless legs syndrome), and attention deficit hyperactivity disorder. The polygenic scores for insomnia and obesity were associated with higher odds of NES (insomnia: odds ratio [OR], 1.24; 95% CI, 1.07, 1.43; obesity: 1.98; 95% CI, 1.71, 2.28).

DISCUSSION

Complementary phenome-wide and genetic exploratory analyses provided information on unique and shared features of NES, offering insights that may facilitate its precise definition, diagnosis, and the development of targeted therapeutic interventions.

Pain can't be carved at the joints: defining function-based pain profiles and their relevance to chronic disease management in healthcare delivery design

BACKGROUND

Pain is a complex problem that is triaged, diagnosed, treated, and billed based on which body part is painful, almost without exception. While the "body part framework" guides the organization and treatment of individual patients' pain conditions, it remains unclear how to best conceptualize, study, and treat pain conditions at the population level. Here, we investigate (1) how the body part framework agrees with population-level, biologically derived pain profiles; (2) how do data-derived pain profiles interface with other symptom domains from a whole-body perspective; and (3) whether biologically derived pain profiles capture clinically salient differences in medical history.

METHODS

To understand how pain conditions might be best organized, we applied a carefully designed a multi-variate pattern-learning approach to a subset of the UK Biobank (n = 34,337), the largest publicly available set of real-world pain experience data to define common population-level profiles. We performed a series of post hoc analyses to validate that each pain profile reflects real-world, clinically relevant differences in patient function by probing associations of each profile across 137 medication categories, 1425 clinician-assigned ICD codes, and 757 expert-curated phenotypes.

RESULTS

We report four unique, biologically based pain profiles that cut across medical specialties: pain interference, depression, medical pain, and anxiety, each representing different facets of functional impairment. Importantly, these profiles do not specifically align with variables believed to be important to the standard pain evaluation, namely painful body part, pain intensity, sex, or BMI. Correlations with individual-level clinical histories reveal that our pain profiles are largely associated with clinical variables and treatments of modifiable, chronic diseases, rather than with specific body parts. Across profiles, notable differences include opioids being associated only with the pain interference profile, while antidepressants linked to the three complimentary profiles. We further provide evidence that our pain profiles offer valuable, additional insights into patients' wellbeing that are not captured by the body-part framework and make recommendations for how our pain profiles might sculpt the future design of healthcare delivery systems.

CONCLUSION

Overall, we provide evidence for a shift in pain medicine delivery systems from the conventional, body-part-based approach to one anchored in the pain experience and holistic profiles of patient function. This transition facilitates a more comprehensive management of chronic diseases, wherein pain treatment is integrated into broader health strategies. By focusing on holistic patient profiles, our approach not only addresses pain symptoms but also supports the management of underlying chronic conditions, thereby enhancing patient outcomes and improving quality of life. This model advocates for a seamless integration of pain management within the continuum of care for chronic diseases, emphasizing the importance of understanding and treating the interdependencies between chronic conditions and pain.

Social and Polygenic Risk Factors for Time to Comorbid Diagnoses in Individuals with Substance Use Disorders: A Phenome-Wide Survival Analysis

Importance

Persons with substance use disorders (SUD) often suffer from additional comorbidities, including psychiatric conditions and physical health problems. Researchers have explored this overlap in electronic health records (EHR) using phenome wide association studies (PheWAS) to characterize how different indicators are related to all conditions in an individual's EHR. However, analyses have been largely cross-sectional in nature.

Objective

To characterize whether various social and genetic risk factors are associated with time to comorbid diagnoses in electronic health records (EHR) after the first diagnosis of SUD.

Design

Leveraging those with EHR and whole-genome sequencing data in All of Us (N = 287,012), we explored whether social determinants of health are associated with lifetime risk of SUD. Next, within those with a diagnosed SUD (N = 17,460), we examined whether polygenic scores (PGS) were associated with time to comorbid diagnoses performing a phenome-wide survival analysis.

Setting

Participating health care organizations across the United States.

Participants

Participants in the All of Us Research Program with available EHR and genomic data.

Exposures

Social determinants of health and polygenic scores (PGS) for psychiatric and substance use disorders.

Main Outcomes and Measures

Phecodes for diagnoses derived from International Statistical Classification of Diseases, Ninth and Tenth Revisions, Clinical Modification, codes from EHR.

Results

Multiple social and demographic risk factors were associated with lifetime SUD diagnosis. Most strikingly, those reporting an annual income <$10K had 4.5 times the odds of having an SUD diagnosis compared to those reporting $100-$150K annually (OR = 4.48, 95% CI = 4.01, 5.01). PGSs for alcohol use disorders, schizophrenia, and post-traumatic stress disorder were associated with time to their respective diagnoses (HRAUD = 1.10, 95% CI = 1.06, 1.14; HRSCZ = 1.13, 95% CI = 1.06, 1.20; HRPTSD = 1.15, 95% CI = 1.08, 1.22). A PGS for ever-smoking was associated with time to subsequent smoking related comorbidities and additional SUD diagnoses HRSMOK = 1.6 to 1.16).

Conclusions and Relevance

Social determinants, especially those related to income have profound associations with lifetime SUD risk. Additionally, PGS may include information related to outcomes above and beyond lifetime risk, including timing and severity.

Unified Clinical Vocabulary Embeddings for Advancing Precision Medicine

Integrating clinical knowledge into AI remains challenging despite numerous medical guidelines and vocabularies. Medical codes, central to healthcare systems, often reflect operational patterns shaped by geographic factors, national policies, insurance frameworks, and physician practices rather than the precise representation of clinical knowledge. This disconnect hampers AI in representing clinical relationships, raising concerns about bias, transparency, and generalizability. Here, we developed a resource of 67,124 clinical vocabulary embeddings derived from a clinical knowledge graph tailored to electronic health record vocabularies, spanning over 1.3 million edges. Using graph transformer neural networks, we generated clinical vocabulary embeddings that provide a new representation of clinical knowledge by unifying seven medical vocabularies. These embeddings were validated through a phenotype risk score analysis involving 4.57 million patients from Clalit Healthcare Services, effectively stratifying individuals based on survival outcomes. Inter-institutional panels of clinicians evaluated the embeddings for alignment with clinical knowledge across 90 diseases and 3,000 clinical codes, confirming their robustness and transferability. This resource addresses gaps in integrating clinical vocabularies into AI models and training datasets, paving the way for knowledge-grounded population and patient-level models.

Mdpg: a novel multi-disease diagnosis prediction method based on patient knowledge graphs

Diagnosis prediction, a key factor in enhancing healthcare efficiency, remains a focal point in clinical decision support research. However, the time-series, sparse and multi-noise characteristics of electronic health record (EHR) data make it a great challenge. Existing methods commonly address these issues using RNNs and incorporating medical prior knowledge from medical knowledge bases, but they neglect the local spatial characteristics and spatial-temporal correlation of the data. Consequently, we propose MDPG, a diagnosis prediction model based on patient knowledge graphs. Initially, we represent the electronic visit records of patients as a patient-centered temporal knowledge graph, capturing the local spatial structure and temporal characteristics of the visit information. Subsequently, we design the spatial graph convolution block, temporal self-attention block, and spatial-temporal synchronous graph convolution block to capture the spatial, temporal, and spatial-temporal correlations embedded in them, respectively. Ultimately, we accomplish the prediction of patients' future states through multi-label classification. We conduct comprehensive experiments on two real-world datasets independently and evaluate the results using visit-level precision@k and code-level accuracy@k metrics. The experimental results demonstrate that MDPG outperforms all baseline models, yielding the best performance.

Genome-wide association analysis provides insights into the molecular etiology of dilated cardiomyopathy

Dilated cardiomyopathy (DCM) is a leading cause of heart failure and cardiac transplantation. We report a genome-wide association study and multi-trait analysis of DCM (14,256 cases) and three left ventricular traits (36,203 UK Biobank participants). We identified 80 genomic risk loci and prioritized 62 putative effector genes, including several with rare variant DCM associations (MAP3K7, NEDD4L and SSPN). Using single-nucleus transcriptomics, we identify cellular states, biological pathways, and intracellular communications that drive pathogenesis. We demonstrate that polygenic scores predict DCM in the general population and modify penetrance in carriers of rare DCM variants. Our findings may inform the design of genetic testing strategies that incorporate polygenic background. They also provide insights into the molecular etiology of DCM that may facilitate the development of targeted therapeutics.

Clinical associations with treatment resistance in depression: An electronic health record study

Treatment resistance is common in major depressive disorder (MDD), yet clinical risk factors are not well understood. Using a discovery-replication design, we conducted phenome-wide association studies (PheWASs) of MDD treatment resistance in two electronic health record (EHR)-linked biobanks. The PheWAS included participants with an MDD diagnosis in the EHR and at least one antidepressant (AD) prescription. Participant lifetime diagnoses were mapped to phecodes. PheWASs were conducted for three treatment resistance outcomes based on AD prescription data: number of unique ADs prescribed, ≥1 and ≥2 CE switches. Of the 180 phecodes significantly associated with these outcomes in the discovery cohort (n = 12,558), 71 replicated (n = 8,206). In addition to identifying known clinical factors for treatment resistance in MDD, the total unique AD prescriptions was associated with additional clinical variables including irritable bowel syndrome, gastroesophageal reflux disease, symptomatic menopause, and spondylosis. We calculated polygenic risk of specific-associated conditions and tested their association with AD outcomes revealing that genetic risk for many of these conditions is also associated with the total unique AD prescriptions. The number of unique ADs prescribed, which is easily assessed in EHRs, provides a more nuanced measure of treatment resistance, and may facilitate future research and clinical application in this area.

Defining Appropriate Comparator Populations for Placental Pathology for Pregnant People With HIV

Background. Widespread adoption of antiretroviral therapy has reduced perinatal transmission of HIV; however, people living with HIV (PWH) have higher rates of preterm birth and hypertensive disorders of pregnancy. The placenta is the critical fetal support organ in pregnancy, and multiple investigations have sought associations in PWH between HIV and placental pathology. However, results have been inconclusive. We posit that selection of control group populations influences the apparent anomalies in placentas from PWH and examined the differences seen between these placentas and those of four comparator populations. Methods. Placentas from PWH were compared with those from all patients without HIV, controls from a recent study of severe acute respiratory syndrome coronavirus 2 in pregnancy, patients with a history of melanoma-an indication for examination relatively orthogonal to other problems in pregnancy, and patients paired with PWH using propensity score matching. Results. People living with HIV differ in demographics and comorbidities from comparator groups other than propensity score-matched patients. Placentas from PWH had higher rates of acute placental inflammation, including maternal inflammatory response and fetal inflammatory response, than multiple comparator groups. Placentas from PHW had lower rates of chronic placental inflammation than three of four comparator groups, including the largest comparator group and the group matched to PWH using propensity scores. Conclusion. Differences in placental pathology in PWH depend on the comparator group. Commonly used comparator groups have significantly different demographic and comorbidity profiles, suggesting they are inappropriate comparators for PWH. Propensity score matching may be useful in identifying comparator populations.

Association of Physical Activity from Wearable Devices and Chronic Disease Risk: Insights from the All of Us Research Program

Background

Physical activity is widely recognized as a key modifiable factor for reducing the risk of chronic diseases. Wearable devices such as Fitbit offer a unique opportunity to objectively measure physical activity metrics, providing insights into the association between different types of physical activity and chronic disease risk.

Objective

This study aims to examine the association between physical activity metrics derived from Fitbit devices and the incidence of various chronic diseases among participants from the All of Us Research Program.

Methods

Physical activity metrics included daily steps, elevation gain, and activity durations at different intensities (e.g., very active, lightly active, and sedentary). Cox proportional hazards models and multiple regression models were used to assess the relationship between these metrics and the incidence of chronic diseases represented by Phecodes. Age, sex, and body mass index (BMI) were included as covariates.

Results

A total of 15,538 participants provided Fitbit activity data, of which 9,320 also had electronic health records (EHR). Increased daily step count, elevation gain, and very active minutes were significantly associated with a reduced risk of several chronic conditions, including obesity, Type 2 diabetes, and major depressive disorder. Conversely, increased sedentary time was linked to higher risks for conditions such as obesity, Type 2 diabetes, and essential hypertension. Multiple regression analyses confirmed these associations.

Conclusion

Our findings highlight the beneficial effects of increased physical activity, particularly daily steps and elevation gain, on reducing the risk of chronic diseases. Conversely, sedentary behavior remains a significant risk factor for the development of several conditions. These insights may inform personalized activity recommendations aimed at reducing disease burden and improving population health outcomes.

Development of electronic health record based algorithms to identify individuals with diabetic retinopathy

OBJECTIVES

To develop, validate, and implement algorithms to identify diabetic retinopathy (DR) cases and controls from electronic health care records (EHRs).

MATERIALS AND METHODS

We developed and validated electronic health record (EHR)-based algorithms to identify DR cases and individuals with type I or II diabetes without DR (controls) in 3 independent EHR systems: Vanderbilt University Medical Center Synthetic Derivative (VUMC), the VA Northeast Ohio Healthcare System (VANEOHS), and Massachusetts General Brigham (MGB). Cases were required to meet 1 of the following 3 criteria: (1) 2 or more dates with any DR ICD-9/10 code documented in the EHR, (2) at least one affirmative health-factor or EPIC code for DR along with an ICD9/10 code for DR on a different day, or (3) at least one ICD-9/10 code for any DR occurring within 24 hours of an ophthalmology examination. Criteria for controls included affirmative evidence for diabetes as well as an ophthalmology examination.

RESULTS

The algorithms, developed and evaluated in VUMC through manual chart review, resulted in a positive predictive value (PPV) of 0.93 for cases and negative predictive value (NPV) of 0.91 for controls. Implementation of algorithms yielded similar metrics in VANEOHS (PPV = 0.94; NPV = 0.86) and lower in MGB (PPV = 0.84; NPV = 0.76). In comparison, the algorithm for DR implemented in Phenome-wide association study (PheWAS) in VUMC yielded similar PPV (0.92) but substantially reduced NPV (0.48). Implementation of the algorithms to the Million Veteran Program identified over 62 000 DR cases with genetic data including 14 549 African Americans and 6209 Hispanics with DR.

CONCLUSIONS/DISCUSSION

We demonstrate the robustness of the algorithms at 3 separate healthcare centers, with a minimum PPV of 0.84 and substantially improved NPV than existing automated methods. We strongly encourage independent validation and incorporation of features unique to each EHR to enhance algorithm performance for DR cases and controls.

A Genome-Wide Association Study Reveals a Novel Susceptibility Locus for Pancreas Divisum at 3q29

INTRODUCTION

Pancreas divisum (PD) is a common congenital anomaly of the pancreas, but its genetic basis remains unknown. The purpose of this genome-wide association study was to identify genetic loci associated with PD.

METHODS

Using the Mass General Brigham Biobank, patients diagnosed with PD were identified. Quality control and imputation were performed using standard approaches. Single nucleotide polymorphisms (SNPs) with minor allele frequency (MAF) ≥ 5% were tested for association with PD using mixed linear model-based association analysis. The significance threshold was set at 5 × 10-8.

RESULTS

A total of 13,940 subjects were included, of which 251 (1.8%) were diagnosed with PD. A genetic locus in chromosome 3q29 was found to be associated with PD (lead SNP rs3850646, MAFPD = 34.6% vs. MAFcontrols = 26.4%, beta = 0.0106, P = 1.47 × 10-8). The identified locus is located in the phosphatidylinositol glycan anchor biosynthesis class Xand p21 activated kinase 2genes. The heritability of PD was estimated at 27.5%. (Expression quantitative trait loci) and chromatin interaction analysis found 12 genes whose expression may be regulated by SNPs in this genomic locus.

CONCLUSIONS

The results of this study suggest that a genetic locus at 3q29 is associated with PD. This locus is in the phosphatidylinositol glycan anchor biosynthesis class X and p21 activated kinase 2 genes. Twelve candidate genes were identified whose expression may be regulated by this locus. These findings may help us understand both normal and aberrant pancreatic development and may aid in clinical evaluation and genetic counseling of patients with PD and associated diseases, such as acute pancreatitis.

Phenome-Wide Analysis of Coffee Intake on Health over 20 Years of Follow-Up Among Adults in Hong Kong Osteoporosis Study

BACKGROUND/OBJECTIVES

There has been limited evidence on the long-term impacts of coffee intake on health. We aimed to investigate the association between coffee intake and the incidence of diseases and mortality risk over 20 years among community-dwelling Chinese adults.

METHODS

Participants were from the Hong Kong Osteoporosis Study who attended baseline assessments during 1995-2010. Coffee intake was self-reported through a food frequency questionnaire and was previously validated. Disease diagnoses, which were mapped into 1795 distinct phecodes, and mortality data were obtained from linkage with territory-wide electronic health records. Cox models were used to estimate the association between coffee intake and the incidence of each disease outcome and mortality among individuals without a history of the respective medical condition at baseline. All models were adjusted for age, sex, body mass index, smoking, alcohol drinking, and education.

RESULTS

Among the 7420 included participants (mean age 53.2 years, 72.2% women), 54.0% were non-coffee drinkers, and only 2.7% consumed more than one cup of coffee per day. Over a median follow-up of 20.0 years, any coffee intake was associated with a reduced risk of dementia, atrial fibrillation, painful respirations, infections, atopic dermatitis, and dizziness at a false discovery rate (FDR) of <0.05. Furthermore, any coffee intake was associated with an 18% reduced risk of all-cause mortality (95% confidence interval = 0.73-0.93).

CONCLUSION

In a population with relatively low coffee consumption, any coffee intake is linked to a lower risk of several neurological, circulatory, and respiratory diseases and symptoms, as well as mortality.

Genetic Architectures of Medical Images Revealed by Registration of Multiple Modalities

The advent of biobanks with vast quantities of medical imaging and paired genetic measurements creates huge opportunities for a new generation of genotype-phenotype association studies. However, disentangling biological signals from the many sources of bias and artifacts remains difficult. Using diverse medical images and time-series (ie, magnetic resonance imagings [MRIs], electrocardiograms [ECGs], and dual-energy X-ray absorptiometries [DXAs]), we show how registration, both spatial and temporal, guided by domain knowledge or learned de novo, helps uncover biological information. A multimodal autoencoder comparison framework quantifies and characterizes how registration affects the representations that unsupervised and self-supervised encoders learn. In this study we (1) train autoencoders before and after registration with nine diverse types of medical image, (2) demonstrate how neural network-based methods (VoxelMorph, DeepCycle, and DropFuse) can effectively learn registrations allowing for more flexible and efficient processing than is possible with hand-crafted registration techniques, and (3) conduct exhaustive phenotypic screening, comprised of millions of statistical tests, to quantify how registration affects the generalizability of learned representations. Genome- and phenome-wide association studies (GWAS and PheWAS) uncover significantly more associations with registered modality representations than with equivalently trained and sized representations learned from native coordinate spaces. Specifically, registered PheWAS yielded 61 more disease associations for ECGs, 53 more disease associations for cardiac MRIs, and 10 more disease associations for brain MRIs. Registration also yields significant increases in the coefficient of determination when regressing continuous phenotypes (eg, 0.36 ± 0.01 with ECGs and 0.11 ± 0.02 for DXA scans). Our findings reveal the crucial role registration plays in enhancing the characterization of physiological states across a broad range of medical imaging data types. Importantly, this finding extends to more flexible types of registration, such as the cross-modal and the circular mapping methods presented here.

Proteogenomic analysis integrated with electronic health records data reveals disease-associated variants in Black Americans

BACKGROUNDMost GWAS of plasma proteomics have focused on White individuals of European ancestry, limiting biological insight from other ancestry-enriched protein quantitative loci (pQTL).METHODSWe conducted a discovery GWAS of approximately 3,000 plasma proteins measured by the antibody-based Olink platform in 1,054 Black adults from the Jackson Heart Study (JHS) and validated our findings in the Multi-Ethnic Study of Atherosclerosis (MESA). The genetic architecture of identified pQTLs was further explored through fine mapping and admixture association analysis. Finally, using our pQTL findings, we performed a phenome-wide association study (PheWAS) across 2 large multiethnic electronic health record (EHR) systems in All of Us and BioMe.RESULTSWe identified 1,002 pQTLs for 925 protein assays. Fine mapping and admixture analyses suggested allelic heterogeneity of the plasma proteome across diverse populations. We identified associations for variants enriched in African ancestry, many in diseases that lack precise biomarkers, including cis-pQTLs for cathepsin L (CTSL) and Siglec-9, which were linked with sarcoidosis and non-Hodgkin's lymphoma, respectively. We found concordant associations across clinical diagnoses and laboratory measurements, elucidating disease pathways, including a cis-pQTL associated with circulating CD58, WBC count, and multiple sclerosis.CONCLUSIONSOur findings emphasize the value of leveraging diverse populations to enhance biological insights from proteomics GWAS, and we have made this resource readily available as an interactive web portal.FUNDINGNIH K08 HL161445-01A1; 5T32HL160522-03; HHSN268201600034I; HL133870.

Language-model-based patient embedding using electronic health records facilitates phenotyping, disease forecasting, and progression analysis

Current studies regarding the secondary use of electronic health records (EHR) predominantly rely on domain expertise and existing medical knowledge. Though significant efforts have been devoted to investigating the application of machine learning algorithms in the EHR, efficient and powerful representation of patients is needed to unleash the potential of discovering new medical patterns underlying the EHR. Here, we present an unsupervised method for embedding high-dimensional EHR data at the patient level, aimed at characterizing patient heterogeneity in complex diseases and identifying new disease patterns associated with clinical outcome disparities. Inspired by the architecture of modern language models-specifically transformers with attention mechanisms, we use patient diagnosis and procedure codes as vocabularies and treat each patient as a sentence to perform the patient embedding. We applied this approach to 34,851 unique medical codes across 1,046,649 longitudinal patient events, including 102,739 patients from the electronic Medical Records and GEnomics (eMERGE) Network. The resulting patient vectors demonstrated excellent performance in predicting future disease events (median AUROC = 0.87 within one year) and bulk phenotyping (median AUROC = 0.84). We then illustrated the utility of these patient vectors in revealing heterogeneous comorbidity patterns, exemplified by disease subtypes in colorectal cancer and systemic lupus erythematosus, and capturing distinct longitudinal disease trajectories. External validation using EHR data from the University of Washington confirmed robust model performance, with median AUROCs of 0.83 and 0.84 for bulk phenotyping tasks and disease onset prediction, respectively. Importantly, the model reproduced the clustering results of disease subtypes identified in the eMERGE cohort and uncovered variations in overall mortality among these subtypes. Together, these results underscore the potential of representation learning in EHRs to enhance patient characterization and associated clinical outcomes, thereby advancing disease forecasting and facilitating personalized medicine.

Phenotypic Findings Associated with Variation in Elastin

Variation in the elastin gene ( ELN ) may contribute to connective tissue disease beyond the known disease associations of Supravalvar Aortic Stenosis and Cutis Laxa. Exome data from MyCode Community Health Initiative participants were analyzed for ELN rare variants (mean allele frequency <1%, not currently annotated as benign). Participants with variants of interest underwent phenotyping by dual chart review using a standardized abstraction tool. Additionally, all rare variants that met inclusion criteria were collapsed into an ELN gene burden score to perform a Phenome-wide Association Study (PheWAS). Two hundred and ninety-six eligible participants with relevant ELN variants were identified from 184,293 MyCode participants. One hundred and three of 254 living participants (41%) met phenotypic criteria, most commonly aortic hypoplasia, arterial dilation, aneurysm, and dissection, and connective tissue abnormalities. ELN variation was significantly (P <2.8×10 -5 ) associated with "arterial dissection" in the PheWAS and two connective tissue Phecodes approached significance. Variation in ELN is associated with connective tissue pathology beyond classic phenotypes.

eTOC Blurb

Carriers of variants of interest in the elastin gene ( ELN ) were evaluated for presence of findings that could be associated with the variation. Chart review and Phenome-wide Association Studies were used. Results are consistent with variation in ELN being associated with findings affecting elastic tissues beyond classic phenotypes.

Genetic association studies using disease liabilities from deep neural networks

The case-control study is a widely used method for investigating the genetic underpinnings of binary traits. However, long-term, prospective cohort studies often grapple with absent or evolving health-related outcomes. Here, we propose two methods, liability and meta, for conducting genome-wide association study (GWAS) that leverage disease liabilities calculated from deep patient phenotyping. Analyzing 38 common traits in ~300,000 UK Biobank participants, we identified an increased number of loci compared to the conventional case-control approach, with high replication rates in larger external GWAS. Further analyses confirmed the disease-specificity of the genetic architecture with the meta method demonstrating higher robustness when phenotypes were imputed with low accuracy. Additionally, polygenic risk scores based on disease liabilities more effectively predicted newly diagnosed cases in the 2022 dataset, which were controls in the earlier 2019 dataset. Our findings demonstrate that integrating high-dimensional phenotypic data into deep neural networks enhances genetic association studies while capturing disease-relevant genetic architecture.

Clinical and genetic contributions to medical comorbidity in bipolar disorder: a study using electronic health records-linked biobank data

Bipolar disorder is a chronic and complex polygenic disease with high rates of comorbidity. However, the independent contribution of either diagnosis or genetic risk of bipolar disorder to the medical comorbidity profile of individuals with the disease remains unresolved. Here, we conducted a multi-step phenome-wide association study (PheWAS) of bipolar disorder using phenomes derived from the electronic health records of participants enrolled in the Mayo Clinic Biobank and the Mayo Clinic Bipolar Disorder Biobank. First, we explored the conditions associated with a diagnosis of bipolar disorder by conducting a phenotype-based PheWAS followed by LASSO-penalized regression to account for correlations within the phenome. Then, we explored the conditions associated with bipolar disorder polygenic risk score (BD-PRS) using a PRS-based PheWAS with a sequential exclusion approach to account for the possibility that diagnosis, instead of genetic risk, may drive such associations. 53,386 participants (58.7% women) with a mean age at analysis of 67.8 years (SD = 15.6) were included. A bipolar disorder diagnosis (n = 1479) was associated with higher rates of psychiatric conditions, injuries and poisonings, endocrine/metabolic and neurological conditions, viral hepatitis C, and asthma. BD-PRS was associated with psychiatric comorbidities but, in contrast, had no positive associations with general medical conditions. While our findings warrant confirmation with longitudinal-prospective studies, the limited associations between bipolar disorder genetics and medical conditions suggest that shared environmental effects or environmental consequences of diagnosis may have a greater impact on the general medical comorbidity profile of individuals with bipolar disorder than its genetic risk.

Toward Reliable Symptom Coding in Electronic Health Records for Symptom Assessment and Research: Identification and Categorization of International Classification of Diseases, Ninth Revision, Clinical Modification Symptom Codes

To date, symptom documentation has mostly relied on clinical notes in electronic health records or patient-reported outcomes using disease-specific symptom inventories. To provide a common and precise language for symptom recording, assessment, and research, a comprehensive list of symptom codes is needed. The International Classification of Diseases, Ninth Revision or its clinical modification ( International Classification of Diseases, Ninth Revision, Clinical Modification ) has a range of codes designated for symptoms, but it does not contain codes for all possible symptoms, and not all codes in that range are symptom related. This study aimed to identify and categorize the first list of International Classification of Diseases, Ninth Revision, Clinical Modification symptom codes for a general population and demonstrate their use to characterize symptoms of patients with type 2 diabetes mellitus in the Cerner database. A list of potential symptom codes was automatically extracted from the Unified Medical Language System Metathesaurus. Two clinical experts in symptom science and diabetes manually reviewed this list to identify and categorize codes as symptoms. A total of 1888 International Classification of Diseases, Ninth Revision, Clinical Modification symptom codes were identified and categorized into 65 categories. The symptom characterization using the newly obtained symptom codes and categories was found to be more reasonable than that using the previous symptom codes and categories on the same Cerner diabetes cohort.

A comprehensive analysis of the health effects associated with smoking in the largest population using UK Biobank genotypic and phenotypic data

Background

Smoking is a widespread behavior, while the relationship between smoking and various diseases remains a topic of debate.

Objective

We conducted analysis to further examine the identified associations and assess potential causal relationships.

Methods

We utilized seven single nucleotide polymorphisms (SNPs) known to be linked to smoking extracting genotype data from the UK Biobank, a large-scale biomedical repository encompassing comprehensive health-related and genetic information of European descent. Phenome-wide association study (PheWAS) analysis was conducted to map the association of genetically predicted smoking status with 1,549 phenotypes. The associations identified in the PheWAS were then meticulously examined through two-sample Mendelian randomization (MR) analysis, utilizing data from the UK Biobank (n = 487,365) and the Sequencing Consortium of Alcohol and Nicotine Use (GSCAN) (n = 337,334). This approach allowed us to comprehensively characterize the links between smoking and disease patterns.

Results

The PheWAS analysis produced 34 phenotypes that demonstrated significant associations with smoking (P = 0.05/1460). Importantly, sickle cell anemia and type 2 diabetes exhibited the most significant SNPs (both 85.71% significant SNPs). Furthermore, the MR analyses provided compelling evidence supporting causal associations between smoking and the risk of following diseases: obstructive chronic bronchitis (IVW: Beta = 0.48, 95% confidence interval (CI) 0.36-0.61, P = 1.62×10-13), cancer of the bronchus (IVW: Beta = 0.92, 95% CI 0.68-1.17, P = 2.02×10-13), peripheral vascular disease (IVW: Beta = 1.09, 95% CI 0.71-1.46, P = 1.63×10-8), emphysema (IVW: Beta = 1.63, 95% CI 0.90-2.36, P = 1.29×10-5), pneumococcal pneumonia (IVW: Beta = 0.30, 95% CI 0.11-0.49, P = 1.60×10-3), chronic airway obstruction (IVW: Beta = 0.83, 95% CI 0.30-1.36, P = 2.00×10-3) and type 2 diabetes (IVW: Beta = 0.53, 95% CI 0.16-0.90, P = 5.08×10-3).

Conclusion

This study affirms causal relationships between smoking and obstructive chronic bronchitis, cancer of the bronchus, peripheral vascular disease, emphysema, pneumococcal pneumonia, chronic airway obstruction, type 2 diabetes, in the European population. These findings highlight the broad health impacts of smoking and support smoking cessation efforts.

Principled distillation of UK Biobank phenotype data reveals underlying structure in human variation

Data within biobanks capture broad yet detailed indices of human variation, but biobank-wide insights can be difficult to extract due to complexity and scale. Here, using large-scale factor analysis, we distill hundreds of variables (diagnoses, assessments and survey items) into 35 latent constructs, using data from unrelated individuals with predominantly estimated European genetic ancestry in UK Biobank. These factors recapitulate known disease classifications, disentangle elements of socioeconomic status, highlight the relevance of psychiatric constructs to health and improve measurement of pro-health behaviours. We go on to demonstrate the power of this approach to clarify genetic signal, enhance discovery and identify associations between underlying phenotypic structure and health outcomes. In building a deeper understanding of ways in which constructs such as socioeconomic status, trauma, or physical activity are structured in the dataset, we emphasize the importance of considering the interwoven nature of the human phenome when evaluating public health patterns.

Complex patterns of multimorbidity associated with severe COVID-19 and long COVID

BACKGROUND

Early evidence that patients with (multiple) pre-existing diseases are at highest risk for severe COVID-19 has been instrumental in the pandemic to allocate critical care resources and later vaccination schemes. However, systematic studies exploring the breadth of medical diagnoses are scarce but may help to understand severe COVID-19 among patients at supposedly low risk.

METHODS

We systematically harmonized >12 million primary care and hospitalisation health records from ~500,000 UK Biobank participants into 1448 collated disease terms to systematically identify diseases predisposing to severe COVID-19 (requiring hospitalisation or death) and its post-acute sequalae, Long COVID.

RESULTS

Here we identify 679 diseases associated with an increased risk for severe COVID-19 (n = 672) and/or Long COVID (n = 72) that span almost all clinical specialties and are strongly enriched in clusters of cardio-respiratory and endocrine-renal diseases. For 57 diseases, we establish consistent evidence to predispose to severe COVID-19 based on survival and genetic susceptibility analyses. This includes a possible role of symptoms of malaise and fatigue as a so far largely overlooked risk factor for severe COVID-19. We finally observe partially opposing risk estimates at known risk loci for severe COVID-19 for etiologically related diseases, such as post-inflammatory pulmonary fibrosis or rheumatoid arthritis, possibly indicating a segregation of disease mechanisms.

CONCLUSIONS

Our results provide a unique reference that demonstrates how 1) complex co-occurrence of multiple - including non-fatal - conditions predispose to increased COVID-19 severity and 2) how incorporating the whole breadth of medical diagnosis can guide the interpretation of genetic risk loci.

Phenome-wide association study identifies new clinical phenotypes associated with Staphylococcus aureus infections

BACKGROUND

Phenome-Wide Association study (PheWAS) is a powerful tool designed to systematically screen clinical observations derived from medical records (phenotypes) for association with a variable of interest. Despite their usefulness, no systematic screening of phenotypes associated with Staphylococcus aureus infections (SAIs) has been done leaving potential novel risk factors or complications undiscovered.

METHOD AND COHORTS

We tailored the PheWAS approach into a two-stage screening procedure to identify novel phenotypes correlating with SAIs. The first stage screened for co-occurrence of SAIs with other phenotypes within medical records. In the second stage, significant findings were examined for the correlations between their age of onset with that of SAIs. The PheWAS was implemented using the medical records of 754,401 patients from the Marshfield Clinic Health System. Any novel associations discovered were subsequently validated using datasets from TriNetX and All of Us, encompassing 109,884,571 and 118,538 patients respectively.

RESULTS

Forty-one phenotypes met the significance criteria of a p-value < 3.64e-5 and odds ratios of > 5. Out of these, we classified 23 associations either as risk factors or as complications of SAIs. Three novel associations were discovered and classified either as a risk (long-term use of aspirin) or complications (iron deficiency anemia and anemia of chronic disease). All novel associations were replicated in the TriNetX cohort. In the All of Us cohort, anemia of chronic disease was replicated according to our significance criteria.

CONCLUSIONS

The PheWAS of SAIs expands our understanding of SAIs interacting phenotypes. Additionally, the novel two-stage PheWAS approach developed in this study can be applied to examine other disease-disease interactions of interest. Due to the possibility of bias inherent in observational data, the findings of this study require further investigation.

Genetic Architecture and Clinical Outcomes of Combined Lipid Disturbances

BACKGROUND

Dyslipoproteinemia often involves simultaneous derangements of multiple lipid traits. We aimed to evaluate the phenotypic and genetic characteristics of combined lipid disturbances in a general population-based cohort.

METHODS

Among UK Biobank participants without prevalent coronary artery disease, we used blood lipid and apolipoprotein B concentrations to ascribe individuals into 1 of 6 reproducible and mutually exclusive dyslipoproteinemia subtypes. Incident coronary artery disease risk was estimated for each subtype using Cox proportional hazards models. Phenome-wide analyses and genome-wide association studies were performed for each subtype, followed by in silico causal gene prioritization and heritability analyses. Additionally, the prevalence of disruptive variants in causal genes for Mendelian lipid disorders was assessed using whole-exome sequence data.

RESULTS

Among 450 636 UK Biobank participants: 63 (0.01%) had chylomicronemia; 40 005 (8.9%) had hypercholesterolemia; 94 785 (21.0%) had combined hyperlipidemia; 13 998 (3.1%) had remnant hypercholesterolemia; 110 389 (24.5%) had hypertriglyceridemia; and 49 (0.01%) had mixed hypertriglyceridemia and hypercholesterolemia. Over a median (interquartile range) follow-up of 11.1 (10.4-11.8) years, incident coronary artery disease risk varied across subtypes, with combined hyperlipidemia exhibiting the largest hazard (hazard ratio, 1.92 [95% CI, 1.84-2.01]; P=2×10-16), even when accounting for non-HDL-C (hazard ratio, 1.45 [95% CI, 1.30-1.60]; P=2.6×10-12). Genome-wide association studies revealed 250 loci significantly associated with dyslipoproteinemia subtypes, of which 72 (28.8%) were not detected in prior single lipid trait genome-wide association studies. Mendelian lipid variant carriers were rare (2.0%) among individuals with dyslipoproteinemia, but polygenic heritability was high, ranging from 23% for remnant hypercholesterolemia to 54% for combined hyperlipidemia.

CONCLUSIONS

Simultaneous assessment of multiple lipid derangements revealed nuanced differences in coronary artery disease risk and genetic architectures across dyslipoproteinemia subtypes. These findings highlight the importance of looking beyond single lipid traits to better understand combined lipid and lipoprotein phenotypes and implications for disease risk.

Complex patterns of multimorbidity associated with severe COVID-19 and Long COVID

Early evidence that patients with (multiple) pre-existing diseases are at highest risk for severe COVID-19 has been instrumental in the pandemic to allocate critical care resources and later vaccination schemes. However, systematic studies exploring the breadth of medical diagnoses, including common, but non-fatal diseases are scarce, but may help to understand severe COVID-19 among patients at supposedly low risk. Here, we systematically harmonized >12 million primary care and hospitalisation health records from ~500,000 UK Biobank participants into 1448 collated disease terms to systematically identify diseases predisposing to severe COVID-19 (requiring hospitalisation or death) and its post-acute sequalae, Long COVID. We identified a total of 679 diseases associated with an increased risk for severe COVID-19 (n=672) and/or Long COVID (n=72) that spanned almost all clinical specialties and were strongly enriched in clusters of cardio-respiratory and endocrine-renal diseases. For 57 diseases, we established consistent evidence to predispose to severe COVID-19 based on survival and genetic susceptibility analyses. This included a possible role of symptoms of malaise and fatigue as a so far largely overlooked risk factor for severe COVID-19. We finally observed partially opposing risk estimates at known risk loci for severe COVID-19 for etiologically related diseases, such as post-inflammatory pulmonary fibrosis (e.g., MUC5B, NPNT, and PSMD3) or rheumatoid arthritis (e.g., TYK2), possibly indicating a segregation of disease mechanisms. Our results provide a unique reference that demonstrates how 1) complex co-occurrence of multiple - including non-fatal - conditions predispose to increased COVID-19 severity and 2) how incorporating the whole breadth of medical diagnosis can guide the interpretation of genetic risk loci.

Distinguishing vulnerability and resilience to posttraumatic stress disorder evaluating traumatic experiences, genetic risk and electronic health records

What distinguishes vulnerability and resilience to posttraumatic stress disorder (PTSD) remains unclear. Levering traumatic experiences reporting, genetic data, and electronic health records (EHR), we investigated and predicted the clinical comorbidities (co-phenome) of PTSD vulnerability and resilience in the UK Biobank (UKB) and All of Us Research Program (AoU), respectively. In 60,354 trauma-exposed UKB participants, we defined PTSD vulnerability and resilience considering PTSD symptoms, trauma burden, and polygenic risk scores. EHR-based phenome-wide association studies (PheWAS) were conducted to dissect the co-phenomes of PTSD vulnerability and resilience. Significant diagnostic endpoints were applied as weights, yielding a phenotypic risk score (PheRS) to conduct PheWAS of PTSD vulnerability and resilience PheRS in up to 95,761 AoU participants. EHR-based PheWAS revealed three significant phenotypes positively associated with PTSD vulnerability (top association "Sleep disorders") and five outcomes inversely associated with PTSD resilience (top association "Irritable Bowel Syndrome"). In the AoU cohort, PheRS analysis showed a partial inverse relationship between vulnerability and resilience with distinct comorbid associations. While PheRSvulnerability associations were linked to multiple phenotypes, PheRSresilience showed inverse relationships with eye conditions. Our study unveils phenotypic differences in PTSD vulnerability and resilience, highlighting that these concepts are not simply the absence and presence of PTSD.

Medical history predicts phenome-wide disease onset and enables the rapid response to emerging health threats

The COVID-19 pandemic exposed a global deficiency of systematic, data-driven guidance to identify high-risk individuals. Here, we illustrate the utility of routinely recorded medical history to predict the risk for 1883 diseases across clinical specialties and support the rapid response to emerging health threats such as COVID-19. We developed a neural network to learn from health records of 502,460 UK Biobank. Importantly, we observed discriminative improvements over basic demographic predictors for 1774 (94.3%) endpoints. After transferring the unmodified risk models to the All of US cohort, we replicated these improvements for 1347 (89.8%) of 1500 investigated endpoints, demonstrating generalizability across healthcare systems and historically underrepresented groups. Ultimately, we showed how this approach could have been used to identify individuals vulnerable to severe COVID-19. Our study demonstrates the potential of medical history to support guidance for emerging pandemics by systematically estimating risk for thousands of diseases at once at minimal cost.

Multiomic foundations of human prefrontal cortex tissue function

The prefrontal cortex (PFC) is a region of the brain that in humans is involved in the production of higher-order functions such as cognition, emotion, perception, and behavior. Neurotransmission in the PFC produces higher-order functions by integrating information from other areas of the brain. At the foundation of neurotransmission, and by extension at the foundation of higher-order brain functions, are an untold number of coordinated molecular processes involving the DNA sequence variants in the genome, RNA transcripts in the transcriptome, and proteins in the proteome. These "multiomic" foundations are poorly understood in humans, perhaps in part because most modern studies that characterize the molecular state of the human PFC use tissue obtained when neurotransmission and higher-order brain functions have ceased (i.e., the postmortem state). Here, analyses are presented on data generated for the Living Brain Project (LBP) to investigate whether PFC tissue from individuals with intact higher-order brain function has characteristic multiomic foundations. Two complementary strategies were employed towards this end. The first strategy was to identify in PFC samples obtained from living study participants a signature of RNA transcript expression associated with neurotransmission measured intracranially at the time of PFC sampling, in some cases while participants performed a task engaging higher-order brain functions. The second strategy was to perform multiomic comparisons between PFC samples obtained from individuals with intact higher-order brain function at the time of sampling (i.e., living study participants) and PFC samples obtained in the postmortem state. RNA transcript expression within multiple PFC cell types was associated with fluctuations of dopaminergic, serotonergic, and/or noradrenergic neurotransmission in the substantia nigra measured while participants played a computer game that engaged higher-order brain functions. A subset of these associations - termed the "transcriptional program associated with neurotransmission" (TPAWN) - were reproduced in analyses of brain RNA transcript expression and intracranial neurotransmission data obtained from a second LBP cohort and from a cohort in an independent study. RNA transcripts involved in TPAWN were found to be (1) enriched for RNA transcripts associated with measures of neurotransmission in rodent and cell models, (2) enriched for RNA transcripts encoded by evolutionarily constrained genes, (3) depleted of RNA transcripts regulated by common DNA sequence variants, and (4) enriched for RNA transcripts implicated in higher-order brain functions by human population genetic studies. In PFC excitatory neurons of living study participants, higher expression of the genes in TPAWN tracked with higher expression of RNA transcripts that in rodent PFC samples are markers of a class of excitatory neurons that connect the PFC to deep brain structures. TPAWN was further reproduced by RNA transcript expression patterns differentiating living PFC samples from postmortem PFC samples, and significant differences between living and postmortem PFC samples were additionally observed with respect to (1) the expression of most primary RNA transcripts, mature RNA transcripts, and proteins, (2) the splicing of most primary RNA transcripts into mature RNA transcripts, (3) the patterns of co-expression between RNA transcripts and proteins, and (4) the effects of some DNA sequence variants on RNA transcript and protein expression. Taken together, this report highlights that studies of brain tissue obtained in a safe and ethical manner from large cohorts of living individuals can help advance understanding of the multiomic foundations of brain function.

Phenotypic Associations With the HMOX1 GT(n) Repeat in European Populations

Heme oxygenase 1 is a key enzyme in the management of heme in humans. A GT(n) repeat length in the heme oxygenase 1 gene (HMOX1) has been widely associated with a variety of phenotypes, including susceptibility to and outcomes in diabetes, cancer, infections, and neonatal jaundice. However, studies have generally been small and results inconsistent. In this study, we imputed the GT(n) repeat length in participants from 2 UK cohort studies (the UK Biobank study (n = 463,005; recruited in 2006-2010) and the Avon Longitudinal Study of Parents and Children (ALSPAC; n = 937; recruited in 1990-1991)), with the reliability of imputation tested in other cohorts (1000 Genomes Project, Human Genome Diversity Project, and Personal Genome Project UK). Subsequently, we measured the relationship between repeat length and previously identified associations (diabetes, chronic obstructive pulmonary disease, pneumonia, and infection-related mortality in the UK Biobank; neonatal jaundice in ALSPAC) and performed a phenomewide association study in the UK Biobank. Despite high-quality imputation (correlation between true repeat length and imputed repeat length > 0.9 in test cohorts), clinical associations were not identified in either the phenomewide association study or specific association studies. These findings were robust to definitions of repeat length and sensitivity analyses. Despite multiple smaller studies identifying associations across a variety of clinical settings, we could not replicate or identify any relevant phenotypic associations with the HMOX1 GT(n) repeat.

MixEHR-SurG: A joint proportional hazard and guided topic model for inferring mortality-associated topics from electronic health records

Survival models can help medical practitioners to evaluate the prognostic importance of clinical variables to patient outcomes such as mortality or hospital readmission and subsequently design personalized treatment regimes. Electronic Health Records (EHRs) hold the promise for large-scale survival analysis based on systematically recorded clinical features for each patient. However, existing survival models either do not scale to high dimensional and multi-modal EHR data or are difficult to interpret. In this study, we present a supervised topic model called MixEHR-SurG to simultaneously integrate heterogeneous EHR data and model survival hazard. Our contributions are three-folds: (1) integrating EHR topic inference with Cox proportional hazards likelihood; (2) integrating patient-specific topic hyperparameters using the PheCode concepts such that each topic can be identified with exactly one PheCode-associated phenotype; (3) multi-modal survival topic inference. This leads to a highly interpretable survival topic model that can infer PheCode-specific phenotype topics associated with patient mortality. We evaluated MixEHR-SurG using a simulated dataset and two real-world EHR datasets: the Quebec Congenital Heart Disease (CHD) data consisting of 8211 subjects with 75,187 outpatient claim records of 1767 unique ICD codes; the MIMIC-III consisting of 1458 subjects with multi-modal EHR records. Compared to the baselines, MixEHR-SurG achieved a superior dynamic AUROC for mortality prediction, with a mean AUROC score of 0.89 in the simulation dataset and a mean AUROC of 0.645 on the CHD dataset. Qualitatively, MixEHR-SurG associates severe cardiac conditions with high mortality risk among the CHD patients after the first heart failure hospitalization and critical brain injuries with increased mortality among the MIMIC-III patients after their ICU discharge. Together, the integration of the Cox proportional hazards model and EHR topic inference in MixEHR-SurG not only leads to competitive mortality prediction but also meaningful phenotype topics for in-depth survival analysis. The software is available at GitHub: https://github.com/li-lab-mcgill/MixEHR-SurG.

Clinical associations with a polygenic predisposition to benign lower white blood cell counts

Polygenic variation unrelated to disease contributes to interindividual variation in baseline white blood cell (WBC) counts, but its clinical significance is uncharacterized. We investigated the clinical consequences of a genetic predisposition toward lower WBC counts among 89,559 biobank participants from tertiary care centers using a polygenic score for WBC count (PGSWBC) comprising single nucleotide polymorphisms not associated with disease. A predisposition to lower WBC counts was associated with a decreased risk of identifying pathology on a bone marrow biopsy performed for a low WBC count (odds-ratio = 0.55 per standard deviation increase in PGSWBC [95%CI, 0.30-0.94], p = 0.04), an increased risk of leukopenia (a low WBC count) when treated with a chemotherapeutic (n = 1724, hazard ratio [HR] = 0.78 [0.69-0.88], p = 4.0 × 10-5) or immunosuppressant (n = 354, HR = 0.61 [0.38-0.99], p = 0.04). A predisposition to benign lower WBC counts was associated with an increased risk of discontinuing azathioprine treatment (n = 1,466, HR = 0.62 [0.44-0.87], p = 0.006). Collectively, these findings suggest that there are genetically predisposed individuals who are susceptible to escalations or alterations in clinical care that may be harmful or of little benefit.