Hospitalization-Associated Disability After Cardiac Surgery in Elderly Patients - Exploring the Risk Factors Using Machine Learning Algorithms

Abstract

Background: Hospitalization-associated disability (HAD) is associated with prolonged functional decline and increased mortality after discharge. Therefore, we examined the incidence and risk factors associated with HAD in elderly patients undergoing cardiac surgery in Japan. Methods and Results: We retrospectively examined 2,262 elderly patients who underwent elective cardiac surgery at Sakakibara Heart Institute. HAD was defined as a functional decline between time of admission and discharge measured by the Barthel Index. We analyzed clinical characteristics using machine learning algorithms to identify the risk factors associated with HAD. After excluding 203 patients, 2,059 patients remained, of whom 108 (5.2%) developed HAD after cardiac surgery. The risk factors identified were age, serum albumin concentration, estimated glomerular filtration rate, Revised Hasegawa's Dementia Scale, N-terminal pro B-type natriuretic peptide, vital capacity, preoperative Short Physical Performance Battery (SPPB) score, operation times, cardiopulmonary bypass times, ventilator times, length of postoperative intensive care unit stay, and postoperative ambulation start day. The highest incidence of HAD was found in patients with an SPPB score ≤9 and in those who started ambulation >6 days after surgery (76.9%). Conclusions: Several risk factors for HAD are components of frailty, suggesting that preoperative rehabilitation to reduce the risk of HAD is feasible. Furthermore, the association between HAD and a delayed start of ambulation reaffirms the importance of early mobilization and rehabilitation.

Keywords: Cardiac surgery; Elderly patients; Hospitalization-associated disability; Machine learning algorithms; Risk factor.

Figure 1.

Flow diagram of the patient selection process in this study. ARDS, acute respiratory distress syndrome; BI, Barthel Index; HAD, hospitalization-associated disability.

Figure 2.

Feature selection using the Boruta algorithm. Box plots depicting the importance of confounding variables on hospitalization-associated disability (HAD). “Confirmed” variables are those that were found to contribute significantly to HAD. *The cardiopulmonary bypass (CPB) time was considered to be 0 for those who did not undergo CPB. BI, Barthel Index; BMI, body mass index; CABG, coronary artery bypass grafting; CHF, chronic heart failure; COPD, chronic obstructive pulmonary disease; CRRT, continuous renal replacement therapy; CVA, cerebral vascular accident; eGFR, estimated glomerular filtration rate; FEV1.0%, forced expiratory volume % in one second; HAD, hospitalization-associated disability; HDS-R, Revised Hasegawa’s Dementia Scale; ICU, intensive care unit; LVEF, left ventricular ejection fraction; NT-proBNP, N-terminal pro B-type natriuretic peptide; NYHA, New York Heart Association; SPPB, Short Physical Performance Battery; VC, vital capacity.

Figure 3.

Exploration of the hospitalization-associated disability (HAD) risk profiles using the conditional inference tree. The boxes represent binary splits at the cut-off point (corresponding levels are specified) that maximized the discrepancy in the risk of the 2 subsamples of HAD. SPPB, Short Physical Performance Battery.