Accuracy of diagnosis codes to identify febrile young infants using administrative data

Abstract

Background Administrative data can be used to determine optimal management of febrile infants and aid clinical practice guideline development. Objective Determine the most accurate International Classification of Diseases, 9th revision (ICD-9) diagnosis coding strategies for identification of febrile infants. Design Retrospective cross-sectional study. Setting Eight emergency departments in the Pediatric Health Information System. Patients Infants age < 90 days evaluated between July 1, 2012 and June 30, 2013 were randomly selected for medical record review from one of four ICD-9 diagnosis code groups: 1) discharge diagnosis of fever, 2) admission diagnosis of fever without discharge diagnosis of fever, 3) discharge diagnosis of serious infection without diagnosis of fever, and 4) no diagnosis of fever or serious infection. Exposure The ICD-9 diagnosis code groups were compared in four case-identification algorithms to a reference standard of fever ≥ 100.4°F documented in the medical record. Measurements Algorithm predictive accuracy was measured using sensitivity, specificity, negative and positive predictive values. Results Among 1790 medical records reviewed, 766 (42.8%) infants had fever. Discharge diagnosis of fever demonstrated high specificity (98.2%, 95% confidence interval [CI]: 97.8-98.6) but low sensitivity (53.2%, 95% CI: 50.0-56.4). A case-identification algorithm of admission or discharge diagnosis of fever exhibited higher sensitivity (71.1%, 95% CI: 68.2-74.0), similar specificity (97.7%, 95% CI: 97.3-98.1), and the highest positive predictive value (86.9%, 95% CI: 84.5-89.3). Conclusions A case-identification strategy that includes admission or discharge diagnosis of fever should be considered for febrile infant studies using administrative data, though under-classification of patients is a potential limitation.

Are we missing respiratory viral infections in infants and children? Comparison of a hospital-based quality management system with standard of care

“…Literature [4,5] shows that disease burden estimates of influenza and other respiratory viruses are largely based on laboratory reports in electronic health records. The following key conclusions could be drawn from this project.…”

“…An ICD code for 'bronchitis due to influenza infection' does not exist, whereas 'influenza pneumonia' (J10.1, J11.1) is catalogued [6]. ICD coding may also be influenced by seasonality [4]. Expanded use of rapid diagnostics would likely increase the yield of virus-specific ICD codes [22].…”

“…Hospital-based estimates usually rely on diagnostic coding or laboratory reports [4,5]. The International Classification of Diseases (ICD) [6,7] provides a broad range of options, but use of ICD codes may vary among healthcare providers.…”

“…The International Classification of Diseases (ICD) [6,7] provides a broad range of options, but use of ICD codes may vary among healthcare providers. Some codes are used to describe a confirmed diagnosis, others are used to indicate a clinical presentation or suspicion [4,6,8]. In children, clinical symptoms may be subtle and atypical, resulting in ICD codes consistent with non-respiratory complaints, such as gastroenteritis or febrile seizures [9,10].…”

Application of the Rochester Criteria to Identify Febrile Infants With Bacteremia and Meningitis

“…To further identify infants with bacteremia or meningitis who only had a diagnosis code assigned of fever, we additionally identified an equal number of randomly selected infants from the PHIS database over the same 2-year period (using a software-based randomizer) who had an ICD-9 admission or discharge diagnosis code for fever (780.6, 778.4, 780.60, or 780.61) 27 without associated ICD-9 diagnosis code for UTI, bacteremia, or meningitis, and who had culture of urine, blood or CSF obtained during the ED visit or hospitalization. We excluded infants with a complex chronic condition, 28 as these infants may undergo a non-standard evaluation, as well as infants transferred from another hospital as the infant may have undergone testing or treatment prior to transfer. A total of 1057 potential cases were identified using the described search strategy.…”