(P-130) Prognostic Impact of Three Comorbidity Indices on 1 Year Mortality Among Patients with Hip Fracture

Background: Comorbidity adjustment is an important part of epidemiological modelling. The availability of large administrative databases has supported the construction of different comorbidity indices.
Three established indices are diagnoses-based Charlson Comorbidity Index and Elixhauser Comorbidity Index, and prescription-based Rx-Risk Index.
The relative strength of these three indices as well as the effect of combining multiple indices in predicting mortality is not well studied in patients with hip-fracture.

Objectives: To evaluate the 1 year mortality prognostic value of Charlson, Elixhauser and Rx-Risk indices individually and combined among patients with hip fracture.

Methods: Using the Danish Multidisciplinary Hip Fracture Registry directly linked to population based Danish health registries, we identified 33,172 hip fracture surgery patients aged ≥65 years (2014-2018). We calculated Charlson and Elixhauser indices based on all inpatient and outpatient visits discharge diagnoses within a 10 year lookback period, and Rx-Risk index based on all dispensing within a one year lookback period.
We examined the comorbidity indices individually and pairwise as additions to a reference model including only age and sex.
We used logistic regression to assess discrimination by calculating c-index and the contrast in c-index to the reference model with 95% confidence intervals (CI). Besides, we evaluated the calibration of each model graphically.

Results: In total, 27% patients died within one year of hip fracture. The reference model obtained c-index of 0.678 (CI 0.671-0.684).
Adding Charlson, Elixhauser and Rx-Risk individually to the reference model resulted in c-indices of 0.713, 0.708 and 0.714, respectively. This corresponds to c-index improvements of 0.035 (CI 0.031–0.039), 0.030 (CI 0.026-0.033) and 0.036 (CI 0.032–0.040).
Pairwise combinations of Charlson + Elixhauser, Charlson + Rx-Risk, Elixhauser + Rx-Risk improved the c-index with 0.039 (CI 0.035-0.043), 0.050 (CI 0.046-0.055), and 0.053 (CI 0.49-0.058), respectively compared to the reference model.
Calibration was acceptable for all models.

Conclusions: The reference model with sex and age contributed most to the predicted risk for one year mortality. Adding indices individually improved the model further - Charlson and Rx-Risk slightly more than Elixhauser.
Combining the two diagnoses-based indices resulted in negligible additional improvement, whereas the information captured by the Rx-Risk was partly complemented by each of the diagnosis-based indices resulting in a further improvement of the C-index.