Risk of Mortality in Adult Cancer Febrile Neutropenia Patients with a Machine Learning Approach.
Chintan Shah,Xinsong Du,Rohit Bishnoi,Jiang Bian
DOI: https://doi.org/10.1200/jco.2018.36.15_suppl.e13562
IF: 45.3
2018-01-01
Journal of Clinical Oncology
Abstract:e13562 Background: Febrile neutropenia (FN) is a major dose-limiting toxicity of myelosuppressive therapy in cancer patients and impose substantial morbidity and mortality risks. We analyzed National Inpatient Sample (NIS) database to understand the trend in FN and predict mortality in cancer patients who experienced FN using a machine learning (ML) approach. Methods: We used the Healthcare Cost and Utilization Project (HCUP) Nationwide Inpatient Sample (NIS) data. Cancer discharges were identified using the HCUP Clinical Classifications Software (CCS) categorizations (11-45), within which FN patients were identified with International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes (ICD-9-CM: 7806 and 288, 2841 or 1125). We examined trends in FN discharges and mortality rates using during 2003 – 2015, and then built various models on NIS 2013 data to predict mortality. Results: The percentage of discharges secondary to cancer-related FN kept increasing from 0.2% in 2003 and stabilized around 2% since 2010. However, associated cost has continued to rise (average cost of $36,853 vs $105,912 respectively, in 2003 and 2015). The unadjusted mortality rate also kept increasing from 2.2% in 2003 and has stabilized around ~ 4.2% since 2010. The logistic regression model showed significantly higher odds of mortality with advancing age (2.8%, 3.8%, 7%, 8.3% respectively, for age group 18-44, 45-64, 65-84, 85+). Presence of bacteremia compared to no bacteremia (4.5 vs, 3.3, P < 0.01), non-teaching hospitals compared to teaching hospital (4.9% vs, 4.3%, P < 0.01), and comorbidities such as hypertension, paralysis, hypothyroidism, depression showed higher risk of mortality (P < 0.05). Patients admitted over weekend and transferred from another institute had higher odds of mortality (P < 0.01). Conclusions: Even in this era of widespread use of granulocyte colony-stimulating factor for primary prophylaxis and less cytotoxic immunotherapy, mortality from FN remains significantly higher and continues to impose significant financial burden. Identification of risk factors and predicting mortality risk will help clinicians take better therapeutic decisions and thus patient survival.