Cluster analysis of multimorbidity and healthcare burden based on machine learning: results from CHARLS
Keywords: multimorbidity cluster analysis healthcare burden machine learning
TL;DR: Our findings provide important evidence to guide clinical intervention and health care, as well as to optimize outcomes for patients with multimorbidity.
Abstract: Background: A comprehensive system for managing multiple concurrent conditions and guidelines is absent. we aimed to pinpoint and validate clusters of individuals with multimorbidity on different Chinese regions, as well as the connection to healthcare burden and mortality, and provide tailored clinical strategies and preventative actions to support clinicians and health care.
Methods: T-distributed Stochastic Neighbor Embedding (tSNE) was performed to reduce the dimensionality of data through machine learning and k-means were used to identify clusters of multimorbidities. The Cox proportional hazard models were used to estimate hazard ratios (HRs) and 95% CIs for the associations between distinct cluster groups and the risks of death. To evaluate the link between hospital admissions and various clusters and stratified groups, negative binomial regression models were employed, addressing the disproportionate spread of hospitalizations. Odds ratios were calculated using logistic regression to examine the cluster differences in mortality rates at the end of the second, fourth and seventh years.
Results: A four-group solution was pinpointed to describe comorbidity patterns and were named according to their main features: cancer, respiratory and digestion, hypertension and heart and digestion. Within these cluster groups, those with heart and disgestive group exhibited the most comorbidities, averaging 5.32 diseases per individuals, in contrast to the cancer group, which had the fewest, averageing 2.73 diseases per person. Using the cancer group as a benchmark, with HR observed in groups with respiratory and digestion, hypertension, and heart and digestion group were 1.260 (1.137-1.397), 1.358 (1.240-1.489) and 1.258 (1.125-1.407), respectively. Subsequently, after adjusting for confounders, the risk was slightly decreased. During the 7-year follow-up, 300 deaths (31.064 per 1000 person-years) occurred in cancer cohort and 377 deaths (39.790 per 1000 person-years) occurred in respiratory and digestion cohort. And the hypertension group had the highest RD mortality rate of 1.381 per 1000 person-years (95% CI, 1.260, 1.514). In heart and digestion group, the frequency of one-year hospital stays, the latest hospitalization days, personal hospital stays, and one-year hospital stays topped the list, followed closely by the hypertension group in term of out-of-pocket expenses and total one-year hospitalization costs.
Conclusion: Four multimorbidity clusters were identified and associated with distinct clinical characteristics, healthcare use and mortality rates, primarily targeting middle-aged and elderly. Our findings provide important evidence to guide clinical intervention and health care, as well as to optimize outcomes for patients with multimorbidity.
Submission Number: 33
Loading