This study attempted to determine the effects of recency, duration and severity of comorbidity on model performance and risk estimation of death within one year of hospital admission, readmission within 30 days of hospital separation and hospital length of stay (LOS). Medical patients (n=326,456) with a hospital admission in WA from 1990-1996 were randomly selected. Linked hospital morbidity data were extracted for 102 comorbidities, categorised into 16 diagnostic chapters. Statistical modelling was performed, incorporating the specified comorbidities, for the three outcomes. Variables were developed for the presence of comorbidity, as well as for measures of recency, duration and severity of the secondary conditions. Cox regression was used to analyse mortality and readmission outcomes while multiple linear regression was used for LOS. Models were fitted and compared for comorbidity presence-only and for the four condition variables. Risk of post-hospitalisation mortality and readmission was calculated from both models. Model-fit was significantly greater (P<0.0001) with the four variable model for both post-hospitalisation outcomes, compared with modelling comorbidity presence only. For index LOS, adjusted r-square values were improved for the four variable model (0.312) compared with the presence only model (0.096). Nine and 13 comorbidity diagnostic chapters were associated with an increased risk of death or readmission respectively. Adjustment for condition recency, severity and duration resulted in modified risk estimation for all diagnostic chapters. The results indicate that incorporating measures of comorbidity recency, duration and severity strengthens model-fit above that achieved with analysing the presence/absence only for mortality, readmission and LOS outcomes. Further, it is possible to create adjustment factors for these variables to more accurately estimate outcome risk. Failure to consider these aspects of comorbidity may lead to poor risk estimation.
|Publication status||Published - Apr 2004|