Evaluation of quality of care using registry data: the interrelationship between length-of-stay, readmission and mortality and impact on hospital outcomes.
BMJ Qual Saf. 2014 Apr;23(4):350-1
Authors: Marang-van de Mheen PJ, Lingsma HF, Middleton S, Kievit J, Steyerberg EW
INTRODUCTION: Hospital mortality, readmission and length of stay (LOS) are commonly used measures for quality of care, given availability in administrative data. However, these measures are interrelated. For example, a short LOS due to patient's death should be interpreted differently than short LOS in survivors. And patients who died cannot be readmitted. In this study we aim to disentangle the relationship between mortality, readmission and LOS and propose a way to jointly report the three figures to facilitate insight and evaluation of quality of care.
METHODS: Data from the Global Comparators Project were used, in which 22 hospitals from 5 countries have reconciliated the different coding systems of their administrative admission data to obtain risk-adjusted hospital outcomes. Patients discharged between 2007-2011 were included. Three outcomes were considered: mortality, readmission, and prolonged LOS (>75 percentile). We analyzed all patients, stroke patients and colorectal patients as we expected these conditions to vary in short-term mortality and readmission/long LOS. We assessed the correlations between the three standardized outcomes: mortality versus readmission (survivors), mortality versus long LOS, readmission (survivors) versus long LOS (survivors) and long LOS (deaths) versus long LOS (survivors). Second we constructed a composite measure with 5 levels: survivors no readmission normal LOS (best), survivors no readmission long LOS, survivors readmission normal LOS, survivors readmission long LOS, deaths (worst). This composite measure was analyzed using ordinal regression, to obtain a single standardized rate to compare hospitals.
RESULTS: A total of 4,134,359 admissions were included in the analysis, with 76,517 for stroke and 31,736 for colorectal patients. The overall mortality rate was 3.1%, the readmission rate (in survivors) was 7.4% and 20.5% of the admissions had a long LOS (for stroke: 13.9%, 7.1% and 23.0%; for colorectal: 5.0%, 10.4% and 45.7%). The median number of admission per hospital was 170,497 (range 9,294 to 430,731). Standardized (risk-adjusted) outcome rates varied largely between hospitals: 55-140 (mortality), 58-116 (readmission), 50-165 (long LOS). No correlation was found between standardized mortality and readmission rates, and between readmission and long LOS rates (survivors). However, standardized mortality and long LOS rates were positively correlated (r=0.73, p=0.0001), indicating longer hospital stay in patients who died. Long LOS (survivors) was highly correlated with long LOS (deaths) (r=0.74 p<0.01), indicating that some hospital had a long LOS regardless of their mortality rates. The figure shows the variation in the composite outcome measure, consistent with a variation in standardized rates between 43 and 171 (for stroke: 34-162; for colorectal: 33-1.9). This composite measure correlated well with all individual measures, except readmission (r=0.06 p=0.79) caused by the smaller variation between hospitals in readmission rates, therefore weighted less.
DISCUSSION: The three outcome measures were highly related. Disentangling the interrelations in outcomes facilitates insight so that hospitals get better directions for quality improvement. We propose to summarize the three outcomes into a single composite measure. The variation between hospitals in this composite measure is larger than for the individual measures, indicating a more accurate (detailed) representation of quality of care.
DECLARATION OF COMPETING INTERESTS: None.
PMID: 24639573 [PubMed - in process]