Prevalence and prediction of masked uncontrolled hypertension in patients recently hospitalised for myocardial infarction

Aims To study the prevalence of masked uncontrolled hypertension (MUCH) in patients recently hospitalised for myocardial infarction, and to develop machine learning-based prediction models identifying MUCH. Methods Ambulatory blood pressure measurement (ABPM) was performed in 99 patients following hospitalisation for a myocardial infarction. Sixty-two clinical variables were eligible for machine learning. Variable importance for the prediction of MUCH (office blood pressure <140/90 mm Hg at ABPM start but mean 24-h blood pressure ≥130/80 mm Hg) was assessed using the least absolute shrinkage and selection operator (LASSO) and the Boruta algorithms. Logistic regression, LASSO and random forest models using the top variables were evaluated by receiver operating characteristic area under the curve (AUC) in repeated cross-validation. Results Mean age was 62.1±8.2 years, 73 (74%) were males. The ABPM was performed at a median of 11 weeks after discharge. Among 96 patients with valid 24-h ABPM recordings, 32 (33%) had 24-h mean blood pressure ≥130/80 mm Hg and 17 (18%) were identified with MUCH. Machine learning identified discharge diagnoses of diabetes and hypertension, and kidney dysfunction as most important predictors of MUCH. The best random forest, logistic regression and LASSO models showed mean AUC 0.82, 0.80, and 0.80, respectively, for prediction of MUCH. Conclusion One in five patients at follow-up after a myocardial infarction have MUCH. The readily available variables diabetes, hypertension, and kidney dysfunction were identified as the most important predictors of MUCH, which may be implemented in a prediction model for identifying this clinically challenging blood pressure phenotype.