In the last two decades, we have witnessed the widespread use of structural analysis of data. The area, generally called Network Science, concentrates on understanding complex phenomena by looking for properties that emerge from the relationships between the pieces of data instead of the traditional mining of the data itself. A commonly used structural analysis in networks consists of finding subgraphs whose density of connections within the subgraph surpasses that of outside connections; called Community Detection. Many techniques have been proposed to find communities as well as benchmarks to evaluate the algorithms ability to find these substructures. Until recently, the literature has mostly neglected the fact that these communities often represent common characteristic of the elements in the community. For instance, in a social network, communities could represent: people who follow the same particular sport, people from the same classroom, authors working in the same field of study, to name a few. The problem here is one of community detection selection as a function of the ground truth provided by available meta-data. In this work, we propose the use of rank stability (entropy of ranks) to assess communities identified using different techniques from the perspective of meta-data. We validate our approach using a large-scale data set of on-line social interactions across multiple community detection techniques.
Background: Higher-than-expected heart failure (HF) readmissions affect half of US hospitals every year. The Hospital Reduction Readmission Program has reduced risk-adjusted readmissions, but it has also produced unintended consequences. Shared care models have been advocated for HF care, but the association of shared care networks with HF readmissions has never been investigated. Objective: This study aims to evaluate the association of shared care networks with 30-day HF excessive readmission rates using a longitudinal observational study. Methods: We curated publicly available data on hospital discharges and HF excessive readmission ratios from hospitals in California between 2012 and 2017. Shared care areas were delineated as data-driven units of care coordination emerging from discharge networks. The localization index, the proportion of patients who reside in the same shared care area in which they are admitted, was calculated by year. Generalized estimating equations were used to evaluate the association between the localization index and the excessive readmission ratio of hospitals controlling for race/ethnicity and socioeconomic factors. Results: A total of 300 hospitals in California in a 6-year period were included. The HF excessive readmission ratio was negatively associated with the adjusted localization index (β=–.0474, 95% CI –0.082 to –0.013). The percentage of Black residents within the shared care areas was the only statistically significant covariate (β=.4128, 95% CI 0.302 to 0.524). Conclusions: Higher-than-expected HF readmissions were associated with shared care networks. Control mechanisms such as the Hospital Reduction Readmission Program may need to characterize and reward shared care to guide hospitals toward a more organized HF care system.
Objectives This study aimed to evaluate the impact of shared care networks on heart failure readmission rates. Background Higher-than-expected heart failure (HF) readmissions affect half of US hospitals every year. The Hospital Reduction Readmission Program (HRRP) has reduced risk-adjusted readmissions, but it has also produced unintended consequences. Shared care models have been advocated for HF care, but the association of shared care networks with HF readmissions has never been investigated. Methods We curated publicly available data on hospital discharges and HF excessive readmission ratios (ERRs) from hospitals in California between 2012 and 2017. Shared Care Areas (SCAs) were delineated as data-driven units of care coordination emerging from discharge networks. The localization index (LI), the proportion of patients who reside in the same SCA in which they are admitted, were calculated by year. Generalized estimating equations (GEE) were used to evaluate the association between the LI and the ERR of hospitals controlling for race/ethnicity and socioeconomics factors. Results A total of 300 hospitals in California in a 6-yr period were included. The HF excessive readmission ratio (ERR) was negatively associated with the localization index (beta: -0.0474; 95% CI: -0.082 to -0.013). The percentage of Black residents within the SCAs was the only statistically significant covariate (beta: 0.4128; 95% CI: 0.302 to 0.524). Conclusions Higher-than-expected HF readmissions were associated with shared care networks. Control mechanisms such as the HRRP may need to characterize and reward shared care to guide hospitals towards a more organized HF care system.