Fever of Unknown Origin: A Validation Study of Danish ICD-10 Diagnosis Codes

Real-world data in form of routinely collected clinical data are a valuable resource for epidemiological research in infectious disease. We examined the validity of a discharge diagnosis of fever of unknown origin from hospital discharge registries. We identified patients with a first in- or outpatient diagnosis (primary or secondary) of fever of unknown origin (ICD-10 code R50.0; R50.8, R50.9) recorded in the Danish National Patient Registry (DNPR) between 2010 and 2017 in the North Denmark Region. We based the validation cohort on a mix of patients diagnosed at a highly specialized university department of infectious diseases (n=100), other internal medicine departments (n=50), and patients diagnosed at a regional non-university hospital (n=50). We estimate positive predictive value (PPV) of diagnosis for fever of unknown origin using medical records as reference. The PPV of a diagnosis of fever of unknown origin for patients diagnosed at the infectious disease department was 61% (95% CI: 51-71%). For other internal medicine departments, it was 14% (95% CI: 6-27%), and for the non-university hospital it was 16% (95% CI: 7-29%). To achieve higher PPVs, we excluded immunocompromised patients, patients who were diagnosed with infection, cancer or rheumatic disease within 7 days after admission, and/or patients with a short hospital stay (maximum 3 days) and no subsequent hospital contact within 1 month. The PPV for diagnoses from the Department of Infectious Diseases improved to 82% (95% CI: 68-91%) for other internal medicine departments it improved to 31% (95% CI: 11-59%), and for the non-university hospital it improved to 36% (95% CI: 13-65%). We found that only diagnoses made in the Department of Infectious Diseases accurately identified fever of unknown origin, whereas diagnoses made in other units mainly covered infection-related fever, cancer-related fever, or short unspecific fever without further diagnostic work-up.