Osundiji, Mayowa

Abstract Patients with rare genetic disorders often wait years for a correct diagnosis, highlighting the urgent need for early and efficient triage for those at risk. Electronic health records (EHRs) are rich in useful information, yet the extent to which this data is sufficient to generate prediagnostic signals is uncertain. We trained machine learning models on billing codes and phenotypes extracted from clinical notes and applied these across eight conditions in a longitudinal cohort of roughly 3 million patient records from the Mayo Clinic, seeking early identification of conditions. The number of patients identified early varied by condition, ranging from 10% to 89% at 99% specificity. Median lead times exceeded one year prior to the first diagnostic billing code for most conditions, including Fabry disease (1,408 days), hereditary angioedema (2,755 days), neurofibromatosis type 1 (1,067 days), and hereditary hemorrhagic telangiectasia (2,320 days). We show that ML models trained on EHR data can prioritize patients for clinician review and confirmatory genetic testing and that the integration of diverse phenotypic data sources provides superior predictive value.

Background. Rare diseases collectively affect an estimated 300 million people worldwide, yet patients wait an average of 4-8 years for a correct diagnosis after visiting multiple physicians, accumulating unexplained findings, and suffering preventable disease progression before a unifying diagnosis is reached. Electronic health records (EHRs) capture this pre-diagnostic trajectory in rich detail, but conventional methods typically collapse years of longitudinal phenotypic signals into a single static snapshot; our approach instead models this temporal structure directly. Methods. We conducted a retrospective cohort study using longitudinal EHR data from approximately 3 million patients at Mayo Clinic Platform_Accelerate. Cases were patients with confirmed diagnoses of 10 rare diseases; controls (~5,200 per disease) had no record of any rare disease. HPO phenotype terms were extracted from clinical notes and laboratory results using a medspaCy NLP pipeline; ICD codes were grouped at three-character level. We developed and trained two temporal models, GRU-Attn and Conformer, that encode patient histories as quarterly time-binned sequences spanning up to 30 years, and compared these against three static baselines (CatBoost, XGBoost, logistic regression). Models were evaluated systematically at prediction horizons h = 1-10 years before diagnosis. Results. In the full held-out cohort benchmark, temporal models outperformed static baselines across the large majority of diseases and prediction horizons, with the advantage widening substantially as the horizon lengthened. At 1-2 years prior to diagnosis, mean AUROC was 0.928 (GRU-Attn) versus 0.871 (CatBoost); at 6-10 years prior to diagnosis, 0.726 versus 0.693, with static models degrading to near-chance for some diseases. Combining HPO and ICD features yielded the strongest performance, with a mean gain of +0.063 AUROC over the best single source at 1-2 years prior to diagnosis. Separately, in a manually curated cohort of 143 confirmed-undiagnosed cases with extended EHR histories, GRU-Attn identified 138 cases before first disease mention and all 143 cases before formal diagnosis at the prespecified 99th-percentile specificity threshold, with median lead times of 1.1-8.3 years and 1.9-21.7 years, respectively. Conclusions. By learning from the temporal trajectory of phenotypic signals rather than their static aggregate, temporal EHR models improved retrospective early detection performance across several rare diseases, most powerfully for conditions where pre-diagnostic findings accrue gradually and heterogeneously over years. These results show that longitudinal EHR trajectories carry rare-disease signal beyond static phenotypic burden and support prospective evaluation of temporal EHR modeling for rare-disease risk stratification.

Background: Stroke is a frequent manifestation of non-bacterial thrombotic endocarditis (NBTE), yet current knowledge largely stems from case reports and small series. This systematic review aimed to comprehensively describe stroke presentation, imaging features, management, and outcomes in NBTE patients. Methods: A comprehensive search of EMBASE, PubMed, MEDLINE, Web of Science and PsycINFO databases was conducted through September 2023. Studies reporting individual cases or case series of NBTE-related stroke were included. Extracted data included demographics, vascular risk factors, stroke presentation, imaging features, NBTE etiology, treatment modalities (unfractionated heparin/low molecular weight heparin (UFH/LMWH), warfarin, direct oral anticoagulants (DOACs), and valvular intervention), stroke recurrence, and mortality. Multivariate regression was performed to identify predictors of recurrent stroke and mortality. Results: A total of 129 patients from 118 studies were included (median age 56 [IQR:44-64], 62% females). Right (28%) and left (26%) hemispheric stroke syndromes accounted for the majority of clinical presentation, and multivessel ischemic infarcts (>2 territories) were observed in 66%. Conventional vascular risk factors were infrequent (1%-21%), while deep vein thrombosis and systemic embolism occurred in 59% and 38% of patients, respectively, either preceding or coinciding with stroke onset. Malignancy and autoimmune diseases were identified in 72% and 22%, respectively. Anticoagulation was initiated in 66% and 21% underwent valvular treatment. Over a median follow-up of 1.3 months, stroke recurrence and death occurred in 48% and 50%, respectively. Malignancy significantly increased mortality risk (OR 16.8, P<0.001), while autoimmune-related NBTE (OR 0.2, P=0.029), and anticoagulation with UFH/LMWH (OR 0.3, P=0.011) and VKA (OR 0.09, P=0.002) reduced mortality. Recurrent stroke was more likely in patients with malignancy (OR 16.8, P=0.001) and those receiving DOACs (OR 6.2, P=0.057) Conclusion: NBTE should be considered in stroke patients lacking conventional risk factors, particularly with underlying malignancy or autoimmune disease. Malignancy and DOAC use may increase the risk of recurrent strokes in this patient population.