Background – Currently, no predictive models exist for developing LV systolic dysfunction in patients with isolated LV dilation (ILVD). We developed a predictive model using EHR data from patients with ILVD.

Methods – De-identified patient records in the temporal database of a tertiary care center were analyzed between 2020-2023. (follow-up as of 9/2024). Inclusion criteria are modeling details in Fig. A/B. The endpoint was LVEF < 50% or a future echo (“progression”). Five ML models were trained with data parsed from transthoracic echos and comorbidities extracted from clinic notes. A unanimous voting ensemble model was created based on operating points for individual models.

Results – Of 15,042 patients with at least mild LV dilation, 4,230 had a follow up echo (mean age 75 years, 61% female, 63% white). Progression was observed in 6.5% of patients, with a 450-day median time to progression. Observation time was between 80 and 1,693 days. A higher LVMI, systolic heart rate, age, BNP, and male gender/black race were associated with higher model output predicting progression (Fig. C). A unanimous voting ensemble model had the highest precision (0.30), specificity (0.96) and accuracy (0.91). Patients predicted to progress had a 7.97x higher risk of progressing within the observation period (p<0.001, Fig. E).

Conclusion – Our preliminary work sheds light on predictors/ability to predict progression in patients with ILVD. Further work is needed to prospectively validate performance in distinct populations.

VIEW THE PUBLICATION

VIEW THE POSTER