Abstract
Background: Mortality is widely used to assess quality of hospital care. Inpatient mortality is easily available in administrative data. The use of periods other than length of stay is questionable. We compared different overlapping and disjunctive periods for the calculation of mortality associated with hospital care.
Subjects and Methods: Information from public quality reports covering insured from local sickness funds were retrospectively recorded. Nineteen thousand eight hundred thirteen patients from 69 hospitals were included for 5 tracers. The relationship between different periods, or time spans from admission to death or discharge, was assessed calculating the nonparametric correlation coefficient for mortality rates and standardized mortality ratios. The periods were hospital stay, 30, 90, and 365 days after admission, 31 to 90 days, and 91 to 365 days. The variation within each period was assessed with the coefficient of variation.
Results: There is a strong relationship between overlapping periods for nonsurgical tracers with r > 0.559 (P < .001). The surgical tracers present a reverse relationship between 30-day and 91-day to 365-day mortality with r = -0.724 (P = .012) for colorectal carcinoma with operation and r = -0.490 (P = .028) for total hip replacement in case of femur fracture. Nonsurgical tracers show a decreasing variation of mortality rates with extending periods, whereas colorectal carcinoma shows a stable and small variation.
Conclusions: For nonsurgical tracers, hospital stay is the best period to assess mortality associated with hospital stay. The courses (sequences of 4 mortality rates for one tracer in a hospital) for surgical tracers, in particular colorectal carcinoma, appear as a harvesting effect with an association of high in-hospital mortality with low mortality in the medium term.