Abstract:
|
Diagnostic error is a major public health problem and affects more than 12 million Americans per year. It is imperative to identify sources of diagnostic errors and monitor misdiagnosis-related harm in health institutes to improve diagnostic accuracy and patient care. Traditionally however, this effort often relies on asertaining errors through labor-intensive manual review of medical records and are constrained by reviewer bias and poor documentation. In addition, such reviews are time-consuming and almost impossible to carry out in large scale, limiting our ability to attain timely feedback on the evolving medical practice and to inform policy making. In this talk, we discuss a mixture survival model that can identify misdiagnosis related harm based on electronic medical record without labor intensive manual reviews. We also propose harm functions accompanied by profile analysis measures that can be used to evaluate and compare institution-level diagnostic performances. We illustrate proposed methods using stroke occurrence data in the Taiwan National Health Insurance Research Database.
|