NLM Gateway
A service of the U.S. National Institutes of Health
Your Entrance to
Resources from the
National Library of Medicine
    Home      Term Finder      Limits/Settings      Search Details      History      My Locker        About      Help      FAQ    
Skip Navigation Side Barintended for web crawlers only

Application of a Propensity Score for Risk Adjustment in Physician Group Profiling for Asthma Care.

Huang C, Frangakis C, Dominici F, Diette G, Damberg C, Wu A; Academy for Health Services Research and Health Policy. Meeting.

Abstr Acad Health Serv Res Health Policy Meet. 2002; 19: 22.

Johns Hopkins University, School of Public Health, 1620 McElderry St, Suite 11C4, Baltimore, MD 21205; Tel: (410) 502-6384; E-mail: ichuang@jhsph.edu

RESEARCH OBJECTIVE: Patient selection of providers may bias comparisons of quality of care among different health care organizations. Propensity scores have been used in the past to balance a large number of covariates between two groups to reduce this bias, and have been shown to be more effective than conventional risk-adjustment models. We developed a propensity score-based risk-adjustment model to compare satisfaction with asthma care among 20 physician groups, and compared its performance to that of conventional regression-based models. STUDY DESIGN: This was a cross-sectional study using patient surveys. Patient satisfaction with asthma care was the indicator of physician group performance. We developed a propensity score approach based on individual patient characteristics, using a multinomial logit model to estimate the probability of a patient being cared from each one of 20 physician groups. Based on this propensity score, patients in each physician group were assigned to one of five strata. The overall adjusted group performance is calculated as odds ratio of excellent vs. less than excellent satisfaction with asthma care for each physician group related to a reference group, based on a weighted probability of satisfaction in each of these strata. The criteria for assessing different risk adjustment strategies were the discriminative power of the model, the number of physician groups that changed in ranking, and the number of physician groups that moved into a different quintile of ranking. We compared the propensity score-based model to conventional risk-adjustment models that included (1) no risk-adjustment, (2) the Consumer Assessment of Health Plans Study's (CAHPS) risk-adjustment model, and (3) a model adjusting for socio-demographic, clinical, and health status (SF-36) dimensions. Ranking changes were tested using the Spearman rank test. POPULATION STUDIED: Data were collected by the Pacific Business Group on Health (PBGH) from patients with asthma selected at random from each of 20 physician groups in California. A total of 2,515 patients completed the mailed survey (response rate of 32.3%). PRINCIPAL FINDINGS: The propensity score approach was able to balance the distribution of covariates between 20 physician groups. Compared to the unadjusted model, both conventional and propensity score risk-adjustment models result in changes in absolute and quintile rankings (35%-65% of plans changed in absolute ranking and 20%-30% changed in quintile ranking; weighted kappa = 0.81-0.88 vs. 80% changed in absolute ranking and 45% changed in quintile ranking; weighted kappa of 0.69, respectively). Comparisons of the propensity score model to the conventional risk-adjustment models also showed a greater shift in absolute and quintile rankings (70%-75% changed in absolute ranking and 30%-40% changed in quintile ranking; weighted kappa = 0.75-0.81). CONCLUSIONS: The propensity score model is designed theoretically to balance covariates more effectively than other approaches. Changes in ranking therefore suggest that the propensity score approach provides practically better adjustment for selection bias than conventional risk-adjustment models. IMPLICATIONS FOR POLICY, DELIVERY OR PRACTICE: This is among the first studies to demonstrate the advantage of a propensity score approach to risk adjustment in provider profiling. Use of propensity score approach may help to improve fairness in comparisons among physician groups and health plans. PRIMARY FUNDING SOURCE: Pacific Business Group on Health

Publication Types:
  • Meeting Abstracts
Keywords:
  • Bias (Epidemiology)
  • California
  • Cross-Sectional Studies
  • Health Status
  • Humans
  • Insurance, Health
  • Logistic Models
  • Patient Satisfaction
  • Physicians
  • Physicians, Family
  • Probability
  • Risk Adjustment
  • economics
  • hsrmtgs
Other ID:
  • GWHSR0002545
UI: 102274221

From Meeting Abstracts




Contact Us
U.S. National Library of Medicine |  National Institutes of Health |  Health & Human Services
Privacy |  Copyright |  Accessibility |  Freedom of Information Act |  USA.gov