摘要：When comparing the performance of health care providers, it isimportant that the e ect of such factors that have an unwanted e ect onthe performance indicator (eg. mortality) is ruled out. In register basedstudies randomization is out of question. We develop a risk adjustmentmodel for hip fracture mortality in Finland by using logistic regression. Themodel is used to study the impact of the length of the register follow-upperiod on adjusting the performance indicator for a set of comorbidities.The comorbidities are congestive heart failure, cancer and diabetes. Wealso introduce an implementation of the minimum description length (MDL)principle for model selection in logistic regression. This is done by usingthe normalized maximum likelihood (NML) technique. The computationalburden becomes too heavy to apply the usual NML criterion and therefore atechnique based on the idea of sequentially normalized maximum likelihood(sNML) is introduced. The sNML criterion can be evaluated eciently alsofor large models with large amounts of data. The results given by sNML arethen compared to the corresponding results given by the traditional AICand BIC model selection criteria. All three comorbidities have clearly ane ect on hip fracture mortality. The results indicate that for congestiveheart failure all available medical history should be used, while for cancerit is enough to use only records from half a year before the fracture. Fordiabetes the choice of time period is not as clear, but using records fromthree years before the fracture seems to be a reasonable choice.
关键词：Code length; hip fracture; logistic regression; maximum likelihood.
