Lathan Liou

BACKGROUND: Genetic differences may contribute to interindividual variability in LDL-C (low-density lipoprotein cholesterol) lowering with statin therapy. Polygenic risk scores may help identify individuals unlikely to achieve guideline-concordant LDL-C targets on statins, enabling earlier therapy intensification. METHODS: We developed a multiancestry polygenic risk score for statin non-goal response (PRS-NGR) and evaluated its association with failure to achieve an on-statin LDL-C level of ≤70 mg/dL and with percent LDL-C reduction. This longitudinal cohort study used genotyping and electronic health record-linked data from the All of Us Research Program (2018–2025), the UK Biobank (2014–2023), and the Biobank Japan (2003–2008). Participants were statin users with at least 1 prestatin and 1 on-statin LDL-C measurement. Associations were assessed overall and by genetic ancestry, with replication in the UK Biobank and Biobank Japan. RESULTS: The study included 46 564 participants from All of Us, 37 009 from the UK Biobank, and 3613 from Biobank Japan. In All of Us, higher PRS-NGR was associated with increased odds of non-goal response (odds ratio per SD, 1.43 [95% CI, 1.37–1.49]). Compared with the middle quintile, individuals in the top 1% had a higher risk, whereas those in the bottom 1% had a lower risk of non-goal response. Associations of PRS-NGR were consistent across African, European, and Latin American ancestry groups. Each SD increase in PRS-NGR corresponded to a 1.2-percentage-point smaller LDL-C reduction. Findings were replicated in the UK Biobank (odds ratio per SD, 2.39 [95% CI, 2.23–2.57]) and in Biobank Japan (odds ratio per SD, 1.31 [95% CI, 1.17–1.46]). Integration of PRS-NGR with guideline-based criteria identified individuals who derived a higher LDL-C% change and increased identification of statin-eligible individuals. CONCLUSIONS: We developed and validated a multiancestry polygenic risk score that estimates the risk of non-goal LDL-C response to statin therapy. Incorporation of polygenic risk into lipid-lowering treatment paradigms may improve risk stratification and support more tailored therapy intensification strategies.

Background: As artificial intelligence and machine learning become increasingly influential in clinical practice, it is critical for future physicians to understand how such novel technologies will impact the delivery of patient care. Objective: We describe 2 trainee-led, multi-institutional datathons as an effective means of teaching key data science and machine learning skills to medical trainees. We offer key insights on the practical implementation of such datathons and analyze experiences gained and lessons learned for future datathon initiatives. Methods: We detail 2 recent datathons organized by MDplus, a national trainee-led nonprofit organization. To assess the efficacy of the datathon as an educational experience, an opt-in postdatathon survey was sent to all registered participants. Survey responses were deidentified and anonymized before downstream analysis to assess the quality of datathon experiences and areas for future work. Results: Our digital datathons between 2023 and 2024 were attended by approximately 200 medical trainees across the United States. A diverse array of medical specialty interests was represented among participants, with 43% (21/49) of survey participants expressing an interest in internal medicine, 35% (17/49) in surgery, and 22% (11/49) in radiology. Participant skills in leveraging Python for analyzing medical datasets improved after the datathon, and survey respondents enjoyed participating in the datathon. Conclusions: The datathon proved to be an effective and cost-effective means of providing medical trainees the opportunity to collaborate on data-driven projects in health care. Participants agreed that datathons improved their ability to generate clinically meaningful insights from data. Our results suggest that datathons can serve as valuable and effective educational experiences for medical trainees to become better skilled in leveraging data science and artificial intelligence for patient care.

Although prediction models for heart transplantation outcomes have been developed previously, a comprehensive benchmarking of survival machine learning methods for mortality prognosis in the most contemporary era of heart transplants following the 2018 donor heart allocation policy change is warranted. This study assessed seven statistical and machine learning algorithms–Lasso, Ridge, Elastic Net, Cox Gradient Boost, Extreme Gradient Boost Linear, Extreme Gradient Boost Tree, and Random Survival Forests in a post-policy cohort of 7,160 adult heart-only transplant recipients in the Scientific Registry of Transplant Recipients (SRTR) database who received their first transplant on or after October 18, 2018. A cross-validation framework was designed in mlr. Model performance was also compared in a seasonally-matched pre-policy cohort. In the post-policy cohort, Random Survival Forests and Cox Gradient Boost had the highest performances with C-indices of 0.628 and 0.627. The relative importance of some predictive variables differed between the pre-policy and post-policy cohorts, such as the absence of ECMO in the post-policy cohort. Survival machine learning models provide reasonable prediction of 1-year posttransplant mortality outcomes and continual updating of prediction models is warranted in the contemporary era.

Despite the widespread inclusion of statistics in medical school curricula as per the Liaison Committee on Medical Education requirements, the statistical competency among medical students and clinicians remains low. A 2007 study of 277 medical residents revealed only 41.1% scored correctly on a statistical knowledge survey, with minimal understanding of key concepts such as confidence intervals and adjusted odds ratios. A more recent 2023 study of 898 clinicians showed similar deficiencies in understanding efficacy, p-values, and discrimination metrics despite high confidence. This perspective argues for a paradigm shift from teaching statistical applications to focusing on statistical communication. We believe current statistics instruction lacks emphasis on communicating statistical results to patients. Teaching statistical concepts as tools for patient communication, rather than extensions of mathematics, can enhance understanding and ensure patients make informed decisions. Reframing statistical education to focus on communication could potentially address traditionally perceived learning barriers, improve understanding, and foster confidence. In this article, we outline several example reframings of teaching classical statistical concepts emphasizing interpretation and communication. Future strategies such as aligning statistics education closer to residency, revising exam content, updating accreditation requirements, and developing standardized communication primers can help ensure future clinicians are well-equipped to practice evidence-based medicine and effectively communicate statistical information in our increasingly data-driven world.

BACKGROUND: Coronary artery disease (CAD) is a complex, heterogeneous disease with distinct etiological mechanisms. These different etiologies may give rise to multiple subtypes of CAD that could benefit from alternative preventions and treatments. However, so far, there have been no systematic efforts to predict CAD subtypes using clinical and genetic factors. METHODS: Here, we trained and applied statistical models incorporating clinical and genetic factors to predict CAD subtypes in 26 036 patients with CAD in the UK Biobank. We performed external validation of the UK Biobank models in the US-based All of Us cohort (8598 patients with CAD). Subtypes were defined as high versus normal LDL (low-density lipoprotein) levels, high versus normal Lpa (lipoprotein A) levels, ST-segment–elevation myocardial infarction versus non–ST-segment–elevation myocardial infarction, occlusive versus nonocclusive CAD, and stable versus unstable CAD. Clinical predictors included levels of ApoA, ApoB, HDL (high-density lipoprotein), triglycerides, and CRP (C-reactive protein). Genetic predictors were genome-wide and pathway-based polygenic risk scores (PRSs). RESULTS: Results showed that both clinical-only and genetic-only models can predict CAD subtypes, while combining clinical and genetic factors leads to greater predictive accuracy. Pathway-based PRSs had higher discriminatory power than genome-wide PRSs for the Lpa and LDL subtypes and provided insights into their etiologies. The 10-pathway PRS most predictive of the LDL subtype involved cholesterol metabolism. Pathway PRS models had poor generalizability to the All of Us cohort. CONCLUSIONS: In summary, we present the first systematic demonstration that CAD subtypes can be distinguished by clinical and genomic risk factors, which could have important implications for stratified cardiovascular medicine.

Background It remains unknown whether estimated glomerular filtration rate (eGFR) using the refit Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation without a term for race is associated with mortality and the need for kidney replacement therapy (KRT) differentially between Black and White heart transplant recipients. Methods We studied 25,900 adults included in the Scientific Registry of Transplant Recipients. We classified recipients into six categories of eGFR (< 30, 30 to < 45, 45 to < 60, 60 to < 90, 90 to < 120, ≥ 120 ml/min/1.73 m²) using the race-neutral CKD-EPI refit equation, and assessed survival with multivariable adjusted Cox proportional hazards regression. Results The association between pre-transplant race-neutral eGFR and mortality varied by race (P_interaction = 0.006). Compared to White patients with an eGFR of 90–120 ml/min/1.73 m², the mortality rates were 57% (95% CI 1.25, 1.98), 29% (95% CI 1.11, 1.51), 34% (95% CI 1.19, 1.52), and 19% (95% CI 1.06, 1.33) higher in Black patients with an eGFR less than 30, 30–45, 45–60, and 60–90 ml/min/1.73m², respectively; and 53% (95% CI 1.28, 1.82), 49% (95% CI 1.33, 1.66), and 23% (95% CI 1.11, 1.35) higher among White patients with an eGFR less than 30, 30–45, and 45–60 ml/min/1.73 m², respectively. The association between pre-transplant eGFR and the need for KRT during follow-up was similar between Black and White patients (P_interaction = 0.57). Conclusions Worsening pre-transplant eGFR using the new race-neutral CKD-EPI refit equation was associated with a higher rate of post-heart transplant mortality and KRT in Black and White recipients. The racial disparity in post-heart transplant mortality was narrower in the setting of severe kidney dysfunction. Graphical abstract

BACKGROUND As artificial intelligence and machine learning become increasingly influential in clinical practice, it is critical for future physicians to understand how such novel technologies will impact the delivery of patient care. OBJECTIVE We describe a trainee-led, multi-institutional datathon as an effective means of teaching key data science and machine learning skills to medical trainees. We offer key insights on the practical implementation of such datathons and analyze experiences gained and lessons learned for future datathon iterations. METHODS We detail a recent datathon organized by MDplus, a national student-run nonprofit consisting of over 3,000 medical trainee members and physician-innovators. To assess the efficacy of the datathon as an educational experience, a short opt-in post-datathon survey was sent to all registered participants. Survey responses were de-identified and anonymized before downstream analysis to assess the quality of datathon experiences and areas for future work. RESULTS Our virtual datathon was attended by approximately 200 medical trainees across the United States. A diverse array of medical specialty interests were represented amongst participants, with 44% of survey participants expressing an interest in Internal Medicine, 33% in Surgery, and 19% in Radiology. Participant skills in leveraging Python and R for analyzing medical datasets improved after the datathon, and survey respondents enjoyed participating in the datathon (average score: 4.23 / 5). CONCLUSIONS The datathon proved to be an effective and cost-effective means of providing medical trainees the opportunity to collaborate on data-driven projects in healthcare. Participants agreed that the datathon improved their ability to generate clinically meaningful insights from data. Our results suggest that datathons can serve as valuable and effective educational experiences for medical trainees to become better skilled in leveraging data science and artificial intelligence for patient care.

Fine-mapping and gene-prioritisation techniques applied to the latest Genome-Wide Association Study (GWAS) results have prioritised hundreds of genes as causally associated with disease. Here we leverage these recently compiled lists of high-confidence causal genes to interrogate where in the body disease genes operate, which, in previous studies, has mostly been investigated by testing for enrichment of GWAS signal among genes with cell/tissue specific expression. By integrating GWAS summary statistics, gene prioritisation results, and RNA-seq data from 46 tissues and 204 cell types, we directly analyse the gene expression of putative disease genes across the body in relation to 11 major diseases and cancers. In tissues and cell types with established disease relevance, disease genes show higher and more specific gene expression compared to control genes. However, we also detect elevated expression in tissues and cell types without previous links to the corresponding disease. While some of these results may be explained by cell types that span multiple tissues, such as macrophages in brain, blood, lung and spleen in relation to Alzheimer’s disease ( P -values < 10 ^-3 ), the cause for others is unclear and warrants further investigation. To support functional follow-up studies of disease genes, we identify technical and biological factors influencing their expression, and highlight tissues in which higher expression is associated with increased odds of inclusion in drug development programs. We provide our systematic testing framework as an open-source, publicly available tool that can be utilised to offer novel insights into the genes, tissues and cell types involved in any disease, with the potential for informing drug development and delivery strategies.