Best Practices for Identifying Hospitalized Lower Respiratory Tract Infections Using Administrative Data: A Systematic Literature Review of Validation Studies

• Published in: Infectious diseases and therapy Vol. 13; no. 4; pp. 921 - 940
• Main Authors: Hanquet, Germaine, Theilacker, Christian, Vietri, Jeffrey, Sepúlveda-Pachón, Ingrid, Menon, Sonia, Gessner, Bradford, Begier, Elizabeth
• Format: Journal Article
• Language: English
• Published: Cheshire Springer Healthcare 01.04.2024; Springer; Springer Nature B.V; Adis, Springer Healthcare
• Subjects: Administrative data; Algorithms; Bacterial pneumonia; Best practice; Best practices; Codes; Computational linguistics; Hospitalization; ICD; Incidence; Infections; Infectious Diseases; Internal Medicine; Language processing; Literature reviews; Lower respiratory tract infection; Medicine; Medicine & Public Health; Natural language interfaces; Nosocomial infection; Pneumonia; Respiratory diseases; Respiratory tract; Review; Sensitivity; Systematic review; Validation studies; Validation; Pneumonia; Sensitivity; Specificity; Administrative data; Systematic review; ICD; Incidence; Lower respiratory tract infection
• ISSN: 2193-8229; 2193-6382; 2193-6382
• DOI: 10.1007/s40121-024-00949-8

Introduction Estimating the burden of lower respiratory tract infections (LRTIs) increasingly relies on administrative databases using International Classification of Diseases (ICD) codes, but no standard methodology exists. We defined best practices for ICD-based algorithms that estimate LRTI incidence in adults. Methods We conducted a systematic review of validation studies assessing the use of ICD code-based algorithms to identify hospitalized LRTIs in adults, published in Medline, EMBASE, and LILACS between January 1996 and January 2022, according to PRISMA guidelines. We assessed sensitivity, specificity, and other accuracy measures of different algorithms. Results We included 26 publications that used a variety of ICD code-based algorithms and gold standard criteria, and 18 reported sensitivity and/or specificity. Sensitivity was below 80% in 72% (38/53) of algorithms and specificity exceeded 90% in 77% (37/48). Algorithms for all-cause LRTI ( n  = 18) that included only pneumonia codes in primary position ( n  = 3) had specificity greater than 90% but low sensitivity (55–72%). Sensitivity increased by 5–15%, with minimal loss in specificity, with the addition of primary codes for severe pneumonia (e.g. sepsis) while pneumonia codes were in secondary position, and by 13% with codes from LRTI-related infections (e.g. viral) or other respiratory diseases (e.g. empyema). Sensitivity increased by 8% when pneumonia codes were in any position, but specificity was not reported. In hospital-acquired pneumonia and pneumococcal-specific pneumonia, algorithms containing only nosocomial- or pathogen-specific ICD codes had poor sensitivity, which improved when broader pneumonia codes were added, in particular codes for unspecified organisms. Conclusion Our systematic review highlights that most ICD code-based algorithms are relatively specific, but miss a substantial number of hospitalized LRTI adult cases. Best practices to estimate LRTI incidence in this population include the use of all pneumonia ICD codes for any LRTI outcome and, to a lesser extent, those for other LRTI-related infections or respiratory diseases.