Performance Profiling in Primary Care: Does the Choice of Statistical Model Matter?

摘要: >> Background. Profiling is increasingly being used to generate input for improvement efforts in health care. For these be successful, profiles must reflect true provider performance, requiring an appropriate statistical model. Sophisticated models are available account the specific features of performance data, but they may difficult use and explain providers. Objective. To assess influence model on primary care Data Source. Administrative data (2006–2008) 2.8 million members a Dutch insurer who were registered with 1 4396 general practitioners. Methods. Profiles constructed 6 quality measures 5 resource measures, controlling differences case mix. Models include ordinary least squares, generalized linear models, multilevel models. Separately each model, providers ranked z scores classified as outlier if belonging 10% worst or best performance. The impact evaluated using weighted kappa rankings overall, percentage agreement designation, changes over time. Results. Agreement among was relatively high overall (kappa typically .0.85). designation more variable often below 80%, especially outliers. Rankings similar processes than outcomes expenses. annual per low all Conclusions. Differences small, choice did affect rankings. In addition, most appear driven largely by chance, regardless that used. Profilers should pay careful attention both measures.