JMIR Publications

ABSTRACT

Background:

Enhanced management of multimorbidity constitutes a major clinical challenge. Multimorbidity shows well-established causal relationships with high use of healthcare resources and, specifically, with unplanned hospital admissions. Enhanced patients’ stratification is key for achieving effectiveness through personalised post-discharge service selection.

Objective:

The study has a two-fold aim: i) Generation and assessment of predictive models of mortality and readmission at 90-days after discharge; and ii) Characterization of patients’ profiles for personalised service selection purposes.

Methods:

Gradient Boosting techniques were used to generate predictive models based on multisource data (registries, clinical/functional and social support) from 761 non-surgical patients admitted in a tertiary hospital over a 12-month period (Oct 2017 to Nov 2018). K-means clustering was used to characterize patient’s profiles.

Results:

Predictive models performance (AUC [sensitivity/specificity]) was: 0.82 [0.78/0.70] and 0.72 [0.70/0.63] for mortality and readmissions respectively. Four patients’ profiles were identified. Briefly, the Reference patients (cluster #1, n=281, 37%), 54% men and 71±16 yrs. of age, showed 4% mortality and 16% readmissions at 90 days post-discharge. The Unhealthy lifestyle profile (cluster #2, n=179, 23.5%) was predominantly composed of males (76.5%) with similar age, 70±13 yrs., but showed slightly higher mortality (6%) and markedly higher re-admission rate (27.4%). Frailty patients (cluster #3, n= 152, 20%) were older (81±13 yrs.) and predominantly female (41.5% males). They showed medical complexity with high level of social vulnerability and the highest mortality (15%), but with a similar hospitalization rate (25%) than #2. Finally, the Medical complexity profile (cluster #4, n=149, 19.6%), 83±9 yrs., 56% males, showed the highest clinical complexity resulting in 13% mortality and the highest re-admission rate, 38%.

Conclusions:

The results indicated potential to predict mortality and morbidity-related adverse events leading to unplanned hospital readmissions. The resulting patient’s profiles fostered recommendations for personalised service selection with potential for value-generation.