Measuring standardised mortality ratios of hospitals: Challenges and recommendations

Abstract

Objectives of this thesis This thesis has the following objectives: - To study the effects of ‘referral bias’ and ‘casemix and coding issues’ on the current Dutch HSMR calculation. - To identify potential adjustments in the estimation of the HSMR to improve its validity as a performance indicator. Outline of this thesis The thesis starts with investigating the theoretical method underlying the calculation of the HSMR, the so-called indirect standardisation method. In chapter 2, the indirect standardisation method is compared with the direct standardisation method. Also, pitfalls of HSMR resulting from the indirect standardisation method are discussed, and recommendations are given to reduce the shortcomings of this method. Subsequently, the thesis investigates potential modifications of the currently used model for HSMR calculation. To adjust for casemix differences between hospitals, parameters of comorbidities are included in the model underlying the HSMR calculation. In chapter 3, the commonly used Charlson comorbidity measure is compared with the Elixhauser comorbidity measure. Discriminative performance of the casemix correction models based on these two comorbidity measures is compared and their effects on the HSMRs of individual hospitals are explored. The Dutch HSMR is currently based on in-hospital mortality. However, discharge patterns, average length of hospital stay, and transfers all affect inhospital mortality. In chapter 4, effects of the inclusion of post-discharge mortality on HSMRs are compared with those of in-hospital mortality. In the final part of the thesis we zoom in onto the mortality ratios of specific patient populations, rather than that of an entire hospital population. In chapter 5, the focus is on SMRs of specific diagnosis groups requiring specialised care offered by specialised hospitals. The SMRs of specialised and nonspecialised hospitals are compared and the influence of referral patterns on SMRs is investigated. Current HSMR calculation is based on administrative databases and said to lack important clinical predictors. In chapter 6, the casemix adjustment model for cardiac surgery patients, based on an administrative database, is compared with the validated clinical EuroSCORE prediction model, based on a clinical database. Also influences of the two models on eventual SMRs are compared. Finally, in chapter 7, the results and implications of this thesis are summarised and discussed together with insights and recommendations to improve the validity and utility of HSMRs.