Introduction: Multiple chronic conditions (MCC) diminish the pre-stroke reserve that aids post-stroke adaptation and recovery. Through machine learning, we developed a MCC index that integrates pre-stroke comorbid conditions, functional and cognitive factors, as well as their interactions, to predict post-stroke functional outcome (FO) in a bi-ethnic, population-based cohort study. Methods: Ischemic stroke patients (2008-2017) were interviewed at baseline and 90 days. FO score (range 1-4, higher scores worse) at 90 days was measured by averaging 22 activities of daily living (ADL)/instrumental activities of daily living (IADL) and dichotomizing the score into favorable (1-3) and unfavorable FO (>3, a lot of difficulty with ADL/IADLs). Multiple linear regression was fit with a Lasso penalty to select predictors among 22 chronic conditions from ICD codes and medical records, pre-stroke function, cognitive impairment, social support, marital status, depression, age, initial stroke severity (NIHSS) and all pairwise interactions. We developed an MCC index by weighting selected predictors using β-coefficients. Adjusted R 2 , discrimination and calibration of the model were assessed. Results: Among 1,035 stroke survivors, 69% were Mexican American, 51% were female, mean age was 68 (SD=12), and median initial NIHSS was 4 (IQR:2-8). Median FO score was 2.36 (IQR:1.55-3.41); 32% had unfavorable FO. The final model contained the pre-stroke modified Rankin Score (mRS), initial NIHSS, age, congestive heart failure (CHF), weight loss, diabetes, other neurological disease, initial NIHSS х pre-stroke mRS, dementia х age, CHF х renal failure and pre-stroke mRS х history of stroke/TIA, which explained 44% of variability in FO score. The MCC index was well calibrated (p=0.28) and predicted unfavorable FO well (c-statistic, 0.84) in the internal validation dataset. Conclusion: A new MCC assessment tool was developed and validated to improve the prediction of post-stroke FO. Weight loss, other neurological disease and interactions between MCC were discovered as novel predictors. Efforts to improve stroke prognosis may benefit from a better understanding, prevention and management of MCC in population at high risk for stroke.
Read full abstract