To compare Jarman-derived hospital standardised mortality ratios (HSMR) and Linkage-derived cumulative mortality ratios (CMR).

HSMR and CMR values for four groups of hospitals were derived from four single-year cohorts of linked patient admissions and deaths, and compared; differences were explored and reasons for non-matching and discordance were suggested.

For the group of metropolitan teaching hospitals the Jarman-derived HSMR value of 0.95 (95% CI 0.93–0.97) was significantly lower than the Linkage-derived CMR value of 0.99 (95% CI 0.97–1.01). The opposite result was seen for the group of metropolitan non-teaching public hospitals: the Linkage-derived CMR of 0.81 (0.77–0.85) was significantly lower than the Jarman method HSMR of 1.03 (0.98–1.07). Unmatched deaths in the study resulted from a condition unique to the Jarman method. Besides there being fewer deaths based only on the hospital separation records, 593 of these records suggested wrongly diagnosed deaths, possibly due to erroneous coding. Also, 467 hospital separation records deemed not to have involved transfer in the Jarman method database were detected by the Linkage method. However, excluding these unmatched deaths did not result in total concordance for two-method hospital-related deaths.

Incorrect deaths in the Jarman method can be overcome by using the Linked method. The Jarman method, unable to adjust for the contiguous transfers related to the death, apportioned excess deaths unfairly to the teaching hospitals group.What is known about the topic? HSMR based on hospital separation record information can reflect hospital performance if monitored over a regular period. Despite considerable variability, inter-hospital comparison league tables of hospitals based on such ratios have been published.What does this paper add? This study demonstrated that the Linkage-derived CMR, utilising valid details from the state death registry, more accurately ascertains number of deaths than does the Jarman method-derived HSMR.What are the implications for practitioners? Where data linkages are possible, dual death derivations by the Jarman method and the Linked method can identify any unmatched or discordant deaths. Detailed exploration may help identify any differing hospital discharge practices.