Measuring the Relative Contribution of Social Risk Factors in an Enhanced Risk‐Adjustment Model for Unplanned Hospital Readmissions
Year: 2020
Type: article
Abstract: Research Objective Recent research provides evidence that social risk factors should be accounted for when measuring hospital performance. Since 2015, the Massachusetts Center for Health Information and Analysis (CHIA) has publicly reported on all‐payer, unplanned hospital readmissions using an adapted version of the Yale/CMS hospital‐wide readmission measure. According to expert recommendations of a statewide workgroup, CHIA expanded its existing risk‐adjustment model to incorporate patient‐ and community‐level social risk factors. This study aims to establish a method to measure the relative contributions of social risk factors in the enhanced risk‐adjustment model for unplanned hospital readmissions. Study Design The Yale/CMS readmission measure runs a separate risk‐adjustment model for five clinically relevant cohorts—medicine, surgical, neurology, cardiovascular, and cardiorespiratory—each adjusting for patient case mix and hospital service mix. CHIA expanded each model by adding patient‐level factors including race, homeless status, payer type, and dual‐eligibility for Medicaid and Medicare, as well as community‐level factors such as income, wealth, housing, education, and employment. A component analysis was conducted to better understand the relative contributions of these factors in risk adjustment. Factors were first grouped into three domains: the existing model (ie, patient demographics, comorbidities, and condition), patient‐level social risk factors, and community‐level social risk factors. The linear predictors from each logistic regression were normalized to create a domain score. The three domain scores were used as predictors in a hierarchical logistic regression model. For each hospital, the product of its average domain score and the respective domain coefficient was calculated. These amounts were converted to odds ratios to derive percentage change to illustrate the magnitude and direction of domain‐specific adjustments. This process was run for each of the five clinical cohorts. Population Studied Using the Massachusetts Hospital Inpatient Discharge Database (HIDD), a three‐year dataset of Massachusetts residents was created, spanning from July 1, 2016, to June 30, 2018 (ie, state fiscal year 2016‐2018). Three years of HIDD were utilized to increase statistical reliability. Community‐level data from the American Community Survey (2013‐2017) and the US Census (2010) were linked to the HIDD by patient’s zip code. Principal Findings The relative contributions of the domains to the risk‐adjusted readmission rate varied by clinical cohort and by hospital. For example, the amount of adjustment in the medicine cohort by hospital ranged from −0.38% to 0.80% for the community‐level domain and from −4.58% to 4.68% for the patient‐level domain. However, within the surgical cohort, the amount of adjustment for each hospital was slightly differently, ranging from −6.70% to 3.81% for the community‐level domain and from −3.36% to 3.87% for the patient‐level domain. Conclusions The component analysis showed the relative importance of patient‐ and community‐level social risk factors in an enhanced risk‐adjustment of readmissions. Implications for Policy or Practice Component analysis is a helpful way to summarize the results of complex modeling and provides information that would be otherwise masked in the final readmission rate. CHIA plans to use this method in a pilot program with select hospitals to help explain how their rates would be effected by the expanded risk‐adjustment model. Primary Funding Source Commonwealth of Massachusetts.
Source: Health Services Research
Institution: University of Massachusetts Chan Medical School
Language: English
Subfield: Health Information Management
Field: Health Professions
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W3090805274