Validation of an algorithm for identifying patients with febrile neutropenia: an epidemiological study using real-world data in Japan.

Abstract

The epidemiology of febrile neutropenia (FN) remains poorly understood. Although the Diagnosis Procedure Combination (DPC) database offers valuable clinical information, it lacks FN-specific diagnostic codes. This study aimed to develop an algorithm to identify patients with FN using DPC data and to investigate the epidemiology of FN. Data from St. Luke's International Hospital were used to identify DPC-based FN cases in patients with hematological malignancies who underwent chemotherapy, blood culture collection, and antipseudomonal antibiotic therapy. True FN cases were confirmed through chart review, incorporating neutrophil counts and body temperature. The algorithm demonstrated a sensitivity of 79.6%, specificity of 94.1%, PPV of 81.9%, NPV of 93.2%, and an F-measure of 80.6%. A sensitivity analysis for 2015-2020 showed improved performance: sensitivity 98.9%, specificity 99.6%, PPV 82.4%, NPV 99.6%, and F-measure 89.9%. Among the 144,009 patients with hematological malignancies in the DPC database, 50,558 received chemotherapy, and 13,720 (27.1%) developed FN as indicated by the algorithm, with in-hospital mortality rates of 15.0%. Although 76.7% of patients with FN received guideline-recommended first-line antibiotics, 17.0% received carbapenem, suggesting a potential need for antimicrobial stewardship interventions. The algorithm represents a valuable tool for large-scale epidemiological studies and could help inform strategies for FN management in Japan.