Comparing the performances of comorbidity adjustments with and without inclusion of ‘prior hospitalizations’ data
Mansour Taghavi Azar Sharabiani, School of Public Health, Imperial College London
22 June 2011
Based on the C-statistics summarised in Table 3, it seems that the inclusion of 'prior hospitalizations' in the 'index hospitalisation' data results in significant improvements to the predictability of the models that are adjusted for comorbidities (i.e. Baseline model + Charlson/Deyo, Charlson/Romano, or Elixhauser). However, it is possible that these improvements are confounded by the potential improvements to the predictability of the baseline model itself. Thus, the improvements to the predictability could be independent of the performances of the comorbidity adjustment techniques. This issue cannot be settled based on the current results as it seems that the authors have provided only the C-statstics that are related to the predictability of the baseline model for the ‘Index hospitalization only’ (Table 3). Therefore, the C-statistics of the baseline model for ‘Index and prior hospitalisations’ data and also C-statistics of Charlson/Deyo, Charlson/Romano, and Elixhauser wherein baseline model is not included will confirm or rule out the existence of potential confounding effect of the baseline model.
Comparing the performances of comorbidity adjustments with and without inclusion of ‘prior hospitalizations’ data
22 June 2011
Based on the C-statistics summarised in Table 3, it seems that the inclusion of 'prior hospitalizations' in the 'index hospitalisation' data results in significant improvements to the predictability of the models that are adjusted for comorbidities (i.e. Baseline model + Charlson/Deyo, Charlson/Romano, or Elixhauser). However, it is possible that these improvements are confounded by the potential improvements to the predictability of the baseline model itself. Thus, the improvements to the predictability could be independent of the performances of the comorbidity adjustment techniques. This issue cannot be settled based on the current results as it seems that the authors have provided only the C-statstics that are related to the predictability of the baseline model for the ‘Index hospitalization only’ (Table 3).
Therefore, the C-statistics of the baseline model for ‘Index and prior hospitalisations’ data and also C-statistics of Charlson/Deyo, Charlson/Romano, and Elixhauser wherein baseline model is not included will confirm or rule out the existence of potential confounding effect of the baseline model.
Competing interests
None declared