Prediction of in-hospital mortality among pancreatic cancer patients using modified Cancer Frailty Assessment Score Tool: A National Inpatient Sample database study from 2016-2022.

Abstract

789 Background: Pancreatic cancer impose major burden on healthcare resource utilization in oncology care. We employed the national inpatient sample database within United States to further validate a risk scoring model for in-hospital mortality. We hypothesized that a broad range of features with equivalency scoring would further improve the already validated Cancer Frailty Assessment Score Tool (C-FAST) predictive performance over conventional hospital frailty risk index tools. Methods: We extracted around 245 variables, including demographics, admission details (elective vs. nonelective), and ICD-10 codes capturing co-morbidities and treatments. Codes included broad categories ranging from cancer treatments, infections, hematologic, endocrine, cardiac, respiratory, gastrointestinal, musculoskeletal, and other systems. Predictive score was assessed for diagnostic accuracy with area under the ROC curve (AUC), and the receiver operating characteristic curve was used to evaluate the performance of modified C-FAST (mC-FAST) against the conventional scores like hospital frailty risk (HFR) index, Charslon comorbidity index (CCI) and Elixhauser comorbidity index (ECI). STATA version 17.0 was used for statistical analysis. Results: A total of 155,102 patients with a diagnosis of pancreatic cancer were included in the final analysis from the year of admission 2016 till 2022. These patients had a mean age of 68.35+-11.39 years among whom 52.31% were males and in-hospital crude mortality rate was found to be 7.53%. mC-FAST performed better than other conventional scores with an AUC of 0.87 to predict in-hospital mortality as opposed to HFR (AUC: 0.69), ECI (AUC: 0.62), and CCI (AUC: 0.61). The mC-FAST prediction model works slightly better in females, Hispanic and Asian population, patients with payer Medicaid, higher socioeconomic status, urban teaching location, and private (not-for-profit) hospitals. Inflation adjusted hospitalization cost (+221.54$ [214.9-228.2$]) and length of hospital stay (+0.08 days) were also significantly associated with mC-FAST (P<0.001). Conclusions: C-FAST conventionally utilizes ICD-10 codes to include clinical variables and demographic data available within the database to stratify the risk scoring model. It is considered a high-dimensional predictive model that is used to predict in-hospital mortality and the predictability of home discharge in cancer patients. We validated a robust score using mC-FAST that predicted in-hospital mortality among pancreatic cancer patients. AUC of frailty indices for predicting in-hospital mortality with pancreatic cancer. Frailty indices AUC 95% confidence interval Arithmetic mean CCI 0.612 0.607-0.617 5.67 +- 2.57 ECI 0.620 0.615-0.625 5.14 +- 2.06 HFR 0.687 0.682-0.692 5.58 +- 4.26 mC-FAST 0.870 0.865-0.875 35.04 +- 17.55