Jämförelse av statistiska metoder för rangordning inom sjukvården

There has been an increasing interest for ranking of Swedish health care and a lot of emphasis has been placed on the creation of indicators in order to rank hospital units. However, essays on to what extent the choice of statistical method affects the final result of the ranking of health care units have not been given the same attention. For this particular reason, the main objective of this essay has been to investigate whether the choice of statistical model has a significant impact or not on the final result, by using five different, more or less common, statistical approaches: bar charts (with and without confidence intervals), control charts, logistic regression, multi-level analysis using shrunken residuals. The analyzed material in the essay comprises 24119 patients that all have been prescribed some form of the lipid-reducing medicine statin. According to directives given by Region Skåne, doctors are encouraged to prescribe a particular type of statin medicine, namely Simvastatin, which has been proved to be equally effective, but significantly cheaper than other statin medicines. It is thus Simvastatin and the proportion of Simvastatin prescriptions out of all statin prescriptions that were made from July 2005 through December 2006 that is our indicator. Our essay shows that the choice of statistical method for ranking can have an impact on the result. The choice of method should further be discussed with a specialist within the field. Finally, a discussion is held regarding the problems and the criticism connected with the ranking of hospital units