Statistical linkage of treatment to diagnosis for research and monitoring of practice patterns.
OBJECTIVES: At each patient contact general practitioners enter information about the diagnoses and the interventions in the electronic medical record (EMR) system. If there is only one diagnosis during a single patient-physician contact, then a causal connection between the diagnosis and the intervention is established. Otherwise it is uncertain what may have been the cause of the intervention. METHODS: Ideally the general practitioners would record a match between each intervention and the diagnosis that justifies it. However, most EMR software is not capable of recording explanatory matches. Furthermore, supplying the matches is a resource-demanding task. In this study the general practitioners were supplied with a matching module for the EMR, so data with full matching between intervention and diagnosis was collected (our "gold standard"). The study models how close the full matching can be recreated by model linkage using different kinds of simple assumptions. RESULTS: The modeling demonstrates a raise in the measure of prediction (sensitivity) from 41.9 percent for a completely random linkage to 90.9 percent based on simple assumptions in the model. The small substantial potential further gain makes it less attractive to apply more intricate assumptions or use more complex modeling (including neural networks). CONCLUSIONS: The perspective of the study lies in the support of the general practitioners with software for comparison of their decisions with those of their peers and also with guidelines. Thus a system for simple quality assurance and awareness to untypical decisions could be incorporated into the electronic medical record systems.