Abstract CN19
Background
Both Charlson's comorbidity index (CCI) and Elixhauser's comorbidity index (ECI) have been used to adjust comorbidity in individual patients using administrative databases in health care quality. The Korean Health Insurance Review and Assessment Service (HIRA) database were used to compare the performance of CCI and ECI in predicting Nurse-Sensitive Outcomes (NSOs) with gastric resections for gastric cancer.
Methods
Multiple logistic regression was used to compare CCI and ECI for NSOs adjusted with patients and hospital characteristics. The base model included the route of admission, surgical operation type, lymph node dissection, hospital location, and the type of hospital. Three modeling ways of including factors were used: The item model is a model in which disease items in CCI or ECI are added as separate factors. The Categorical model is a model that categorizes CCI (Charlson's score < 3 or ≥ 3) or ECI (Elixhauser's score < 0, 0, 1-4, ≥ 5) are used. The Continuous model is a model in which CCI or ECI is added as a quantitative variable.
Results
The total number of patients with gastric resections for gastric cancer in Korea for two years was 33,323. The number of patients with at least one NSO was 3,545 (10.64%). Elixhauser method using item was the best model with the highest C-statistic and lowest AIC for total NSO (C-statistics 0.729), deep vein thrombosis (C-statistics 0.741), and physiologic/metabolic derangement (C-statistics 0.733). A model with Charlson's items was the best for upper gastrointestinal tract bleeding (C-statistics 0.793). A model with Elixhauser items was the best for in-hospital mortality (C-statistics 0.863), urinary tract infection (C-statistics 0.746), shock/cardiac arrest (C-statistics 0.779), CNS complication (C-statistics 0.804), wound infection (C-statistics 0.799) and pulmonary failure (C-statistics 0.785). A model with Charlson's items was the best for hospital-acquired pneumonia (C-statistics 0.747) and pressure ulcer (C-statistics 0.759).
Conclusions
The ECI performed significantly better than the CCI in predicting 8 of 11 NSOs. It is necessary to choose ECI or CCI for each nursing indicator according to the predictive performance of models.
Clinical trial identification
Editorial acknowledgement
Legal entity responsible for the study
The authors.
Funding
Has not received any funding.
Disclosure
All authors have declared no conflicts of interest.