Assessment of the Accuracy of Using ICD-9 Diagnosis Codes to Identify Pneumonia Etiology in Patients Hospitalized With Pneumonia

Abstract

Importance: Administrative databases may offer efficient clinical data collection for studying epidemiology, outcomes, and temporal trends in health care delivery. However, such data have seldom been validated against microbiological laboratory results.

Objective: To assess the validity of International Classification of Diseases, Ninth Revision (ICD-9) organism-specific administrative codes for pneumonia using microbiological data (test results for blood or respiratory culture, urinary antigen, or polymerase chain reaction) as the criterion standard.

Design, setting, and participants: Cross-sectional diagnostic accuracy study conducted between February 2017 and June 2019 using data from 178 US hospitals in the Premier Healthcare Database. Patients were aged 18 years or older admitted with pneumonia and discharged between July 1, 2010, and June 30, 2015. Data were analyzed from February 14, 2017, to June 27, 2019.

Exposures: Organism-specific pneumonia identified from ICD-9 codes.

Main outcomes and measures: Sensitivity, specificity, positive predictive value, and negative predictive value of ICD-9 codes using microbiological data as the criterion standard.

Results: Of 161 529 patients meeting inclusion criteria (mean [SD] age, 69.5 [16.2] years; 51.2% women), 35 759 (22.1%) had an identified pathogen. ICD-9-coded organisms and laboratory findings differed notably: for example, ICD-9 codes identified only 14.2% and 17.3% of patients with laboratory-detected methicillin-sensitive Staphylococcus aureus and Escherichia coli, respectively. Although specificities and negative predictive values exceeded 95% for all codes, sensitivities ranged downward from 95.9% (95% CI, 95.3%-96.5%) for influenza virus to 14.0% (95% CI, 8.8%-20.8%) for parainfluenza virus, and positive predictive values ranged downward from 91.1% (95% CI, 89.5%-92.6%) for Staphylococcus aureus to 57.1% (95% CI, 39.4%-73.7%) for parainfluenza virus.

Conclusions and relevance: In this study, ICD-9 codes did not reliably capture pneumonia etiology identified by laboratory testing; because of the high specificities of ICD-9 codes, however, administrative data may be useful in identifying risk factors for resistant organisms. The low sensitivities of the diagnosis codes may limit the validity of organism-specific pneumonia prevalence estimates derived from administrative data.

Conflict of interest statement

Conflict of Interest Disclosures: Dr Deshpande reported receiving grants from the Agency for Healthcare Research and Quality (AHRQ) during the conduct of the study; grants and nonfinancial support from Clorox Company and other support from Ferring Pharmaceuticals outside the submitted work. Dr Zilberberg reported receiving personal fees from Cleveland Clinic during the conduct of the study; grants from Spero, grants from Merck, personal fees from Nabriva, personal fees from Melinta, grants from Lungpacer, grants from Astellas, other support from JNJ, grants from Tetraphase, and grants from The Medicines Company outside the submitted work. Dr Imrey reported receiving grants from the AHRQ during the conduct of the study. Dr Haessler reported receiving grants from the AHRQ during the conduct of the study. Dr Richter reported receiving grants from the AHRQ during the conduct of the study; grants and other support from bioMerieux, grants from BD Diagnostics, grants from Hologic, grants from Diasorin, grants from Lifescale Affinity, grants from Roche, and grants from ARLG outside the submitted work. Dr Rothberg reported receiving grants from the AHRQ during the conduct of the study. No other disclosures were reported.

Figures

Figure.. Participant Selection Flow Diagram

Abbreviations: CT, computed tomography; LOS, length of stay (in days); PDX, principal diagnosis; POA, present on admission; SDX, secondary diagnosis. aPatients with viral pneumonia as principal diagnosis but without initial antibiotic treatments were included. bOn a 1-patient 1-admission basis, with a single eligible admission randomly selected from each patient's eligible admissions (178 hospitals).