Incident use of loop diuretics and incident heart failure in type-2 diabetes mellitus: a precision medicine approach

Introduction Heart Failure (HF) Guidelines strongly recommend (Class I) loop diuretics for managing symptoms and signs of congestion due to HF. However, many patients without a diagnosis of HF are prescribed loop diuretics. Such patients have a poor prognosis. Many may have a missed diagnosis of HF. Aim To identify characteristics that predict initiation of loop diuretics and investigate whether they are similar to those that predict incident HF. Methods We obtained electronic medical records (EMRs) of patients with type-2 diabetes mellitus (T2DM) from two populations (Hong Kong, China & Scotland), including demographics, laboratory measurements, medications and prior comorbidities. New prescriptions of loop diuretics were extracted from an electronic prescribing system (EPS). Prevalent loop diuretic prescriptions were excluded, with a look-back period of 5 years. Diagnosis of HF was determined based on International Classification of Diseases (ICD) 9th revision codes. We implemented time-dependent machine-learning models for incident loop diuretics with and without heart failure. We integrated state-of-the-art artificial intelligence interpretability with clinical expertise to compare the clinical characteristics in patients with and without heart failure. External validation was carried out using the Scotland cohort. Results Of 210,066 patients with T2DM aged >50 years, 3,008 (1.4%) patients were newly prescribed loop diuretics and 3,658 had incident heart failure between 2009 – 2019. The models showed strong predictive performance (loop diuretic without HF: C-index: 0.87, Brier: 0.03) and (loop diuretic with HF: C-Index: 0.88, Brier 0.03). Table 1 shows the key predictors including age, eGFR, serum albumin and cardiovascular comorbidities. The model predicting incident loop diuretics with HF showed improved performance indicating that HF status is closely aligned with factors predicting loop diuretic initiation. Table 2 shows validation in an independent, diverse population demonstrated consistent predictive performance and similar key risk factors, further supporting the robustness of loop diuretic prescriptions as an early indicator of HF development. Conclusion Patients newly prescribed loop diuretics have clinical characteristics consistent with HF.