ICD-9 Codes for Identifying Influenza Hospitalizations in Children

To the Editor: The effect of influenza on young children is substantial , but most infections are clinically unrecognized (1). As a result, without routine laboratory confirmation of influenza infection in patients admitted to the hospital with influen-zalike illness, accurate estimates of influenza-related hospitalization rates are difficult to obtain. Several statistical models have been developed to generate estimates of excess or influenza attributable hospitalizations, all of which calculate the rate of hospitalization above baseline during periods in which influenza is circulating (2–8). However, their accuracy is limited when viruses such as respiratory syncytial virus (RSV) and parainfluen-za are cocirculating with influenza. International Classification of Diseases, 9th revision (ICD-9) diagnostic codes specific to influenza (487.0, 487.1, and 487.8) are easily retrieved from hospital discharge records. However, researchers and public health officials have rarely used them for influenza hospitalization surveillance, presumably because they lack sensitivity for identifying true influenza infections, although this assumption has never been tested. To determine the sensitivity and positive predictive value of influenza-specific ICD-9 admission or discharge codes (487.0, 487.1, and 487.8), we conducted a retrospective cohort study of all patients <21 years of age hospitalized at the Children's Hospital of Philadelphia with laboratory-confirmed influenza during 3 consecutive influenza seasons (July 2001 through June 2004) (9). We compared admission and discharge ICD-9 codes with influenza laboratory results. All specimens were initially tested by rapid solid-phase immunoassay for RSV (Binax; Portland, ME, USA) and influenza (Binax). Direct fluorescent antibody testing for adenovirus, influenza A and B, parainfluenza virus types 1, 2, and 3, and RSV was performed on specimens negative by solid-phase immunoassay for RSV or influenza. Comprehensive viral culture was established for all specimens negative for respiratory viruses by direct fluorescent antibody test. Of 715 cases of laboratory-confirmed influenza identified (Table), 617 (86%) were identified by rapid testing and 98 (14%) by viral culture after rapid test results were negative. A total of 529 patients had influenza-specific admission or discharge ICD-9 codes. The sensitivity of influenza-specific ICD-9 codes was 65% (95% confidence interval [CI] 61%–68%), and the positive predictive value was 88% (95% CI 84%–90%) (Table). Of 66 patients who had influenza-specific admission or discharge ICD-9 codes but negative influenza laboratory results, laboratory tests confirmed parainfluenza (n = 42), Haemophilus influenzae (n = 6; 1 with a positive blood culture and 5 with positive respiratory cultures), H. parainfluenzae (n = 1 wound infection), adenovirus (n = 1), and …