Improving statistical inference when interest focuses on the identification of extreme random effects in clustered data

点击翻译按钮获取中文摘要

PROJECT ABSTRACT Statistical models that generate predicted random effects are widely used to evaluate the status of and rank patients, physicians, hospitals and health plans from longitudinal and clustered data. Predicted random effects have been proven to outperform simpler approaches such as standard regression models, on average. These predicted random effects are often used to identify extreme or outlying values, such as elderly patients with rapid declines in their health or poorly performing hospitals. When interest focuses on the extremes rather than performance on average, there has been no systematic investigation of best approaches. We propose to develop novel methods for prediction of extreme or outlying values and systematically evaluate their performance using theoretical calculations, simulations and examples. Merely predicting extreme or outlying values is rarely sufficient and decision rules for identifying extreme values in a statistically rigorous manner are also needed. We will develop such approaches and provide easy- to-use software to implement the recommended methods.