An elastic net machine learning model developed using routinely collected laboratory indicators identified clinically documented comorbid heart failure in hospitalised patients with chronic obstructive pulmonary disease, maintaining discrimination and calibration across a temporal external validation cohort.
Heart failure represents a frequent and prognostically important comorbidity among individuals with chronic obstructive pulmonary disease, yet the cardiac condition is often documented late during routine medical care. To address this delay, investigators developed and temporally validated an interpretable machine learning framework designed to detect clinically documented comorbid heart failure solely from routinely ordered blood tests.
The retrospective study established a development cohort comprising 1,087 patients with chronic obstructive pulmonary disease hospitalised between January 2023 and December 2025, which included 292 individuals with clinically documented heart failure. For the temporal external validation cohort, researchers evaluated 268 patients hospitalised between January and September 2026, of whom 63 had clinically documented heart failure.
Methodologically, twenty-five candidate predictors were screened using least absolute shrinkage and selection operator regression together with the Boruta algorithm. Nine candidate algorithms were trained without class-imbalance correction, and the final model architecture was selected based on cross-validated discrimination and out-of-fold calibration. Internal validity was established through bootstrap optimism correction with the complete modeling pipeline nested across 1,000 resamples, and the framework was benchmarked against NT-proBNP.
Constructed around 12 routine predictors, the final elastic net model yielded an apparent development area under the receiver operating characteristic curve of 0.817, an optimism-corrected value of 0.799, and an external validation area under the curve of 0.759 with a 95 percent confidence interval spanning 0.683 to 0.835. External calibration proved near-ideal, displaying an intercept of -0.133 and a slope of 0.958, while SHAP variable importance rankings were preserved across cohorts. Applying a development-derived threshold targeting 90 percent sensitivity classified 98 of the 268 temporal validation patients (36.6 percent) as negative, indicating potential echocardiography savings at the expense of missing 11 of the 63 patients with heart failure (17.5 percent).
Benchmarking demonstrated that NT-proBNP alone achieved an external validation area under the curve of 0.983, confirming that the algorithmic model provides no added discrimination in settings where natriuretic peptide testing is already routinely available. Accordingly, the authors proposed positioning the model as a zero-marginal-cost pre-screening mechanism applied to existing laboratory panels to prioritise patients requiring downstream natriuretic peptide evaluation and echocardiography, noting that external validation across independent healthcare centres is required prior to clinical implementation.
Subject of Research: Machine learning identification of heart failure in patients with chronic obstructive pulmonary disease using routine laboratory indicators
Article Title: A machine learning model for identifying heart failure in patients with chronic obstructive pulmonary disease based on routine laboratory indicators
Article References: Liu, Z., Ren, Z., Xu, L., Gao, C., Liu, J., & Wang, C. (2026). A machine learning model for identifying heart failure in patients with chronic obstructive pulmonary disease based on routine laboratory indicators. BMC Cardiovascular Disorders. https://doi.org/10.1186/s12872-026-06966-4
Image Credits: AI Generated
DOI: 10.1186/s12872-026-06966-4
Keywords: chronic obstructive pulmonary disease, heart failure, routine laboratory indicators, machine learning, elastic net, LASSO regression, Boruta algorithm, temporal external validation, NT-proBNP, echocardiography, discrimination, calibration, SHAP values, retrospective cohort study, bootstrap optimism correction, biomarker pre-screening, clinical decision support
News Source: Ophelia Keating. (November 11, 2026). Routine Blood Test Model Identifies Heart Failure in Patients With COPD. Scienmag.



