Heat and hearts: An exposure-anchored computational phenotyping framework for assessing cardiovascular vulnerability during extreme heat.

OBJECTIVE

Extreme heat is associated with increased cardiovascular vulnerability. We developed and validated a heat exposure-anchored cardiovascular vulnerability (HECV) computational phenotype for outpatient visits using longitudinal EHR data linked to high-resolution temperature.

MATERIALS AND

METHODS

We assembled a loyalty cohort of adult patients from a large Chicago-based health system (2017-2023) and linked all encounters to Daymet daily temperature data for acute heat exposure at a 1 km2 resolution. HECV cases were identified and matched to non-HECV controls. Structured EHR data from the 12 months prior to diagnosis informed a phenome-wide association study (PheWAS) and penalized conditional logistic regression. The model was evaluated in a held-out test set using a January 1, 2018 landmark, with calculated time-to-event outcomes and time-dependent area-under-the-curve.

RESULTS

Among 104,439 loyalty cohort patients (62.7% female; mean age 43.6 years), key predictors included low free thyroxine, reduced kidney function, cardiovascular medications, and prior echocardiography. PheWAS showed associations with chronic kidney disease, mitral valve disorders, and venous thromboembolism (all p < 0.001). Our model for HECV risk discrimination had an AUC = 0.85 at 2 years, 0.82 at 3 years, and 0.81 at 5 years. Risk stratification showed clear separation: 2-year HECV incidence ranged from 20.4% (lowest tertile) to 50.0% (highest); 5-year risk from 87.0% to 98.1%.

CONCLUSION

Routinely collected EHR features, combined with external heat metrics, can identify patients with elevated HECV and support proactive clinical or public-health responses. Computational phenotyping of HECV produced a discriminative, scalable risk tool with strong time-dependent AUCs, enabling identification of heat-vulnerable patients prior to event onset.