How to compare and report department specific mortality rates for peer review using the Perinatal Database of The Netherlands

Abstract

Since characteristics of patient populations of obstetric departments vary substantially with respect to the pathology presented, (perinatal) mortality rates as such can not be used for a fair peer review without adjustment for those differences. Using the Perinatal Database of The Netherlands (LVR), data on approximately 80 000 newborns annually from 1985 to 1991 inclusive were used in statistical models to predict the perinatal mortality risks of four subpopulations of different gestational age in about 125 obstetric departments. As predictors for perinatal mortality we used only those risk factors which were judged to reflect the ‘pathology’ of the patient; risk factors associated with or resulting from hospital care and/or policy are (by definition) excluded. The statistical technique of logistic regression was used. The expected and observed mortality within each of these four subpopulations (for 1985–1991) was sent anonymously to each department as a specially designed graphical overview. This method of (anonymous) peer review was met favourably. In peer review the difference between observed and expected mortality should be used, rather than mortality per se, to adjust for differences in basic risk in the intake populations on which the departmental policy has no influence.