Improved Algorithms Needed to Identify Intravenous Drug Use-Associated Endocarditis

syringe sharing, drug abuse
A team of investigators reviewed data from institutions within the VA healthcare system to assess the ability of 2 algorithms to identify patients with intravenous drug use-associated infective endocarditis.

A team of investigators reported the poor performance of case-finding algorithms that used International Classification of Disease (ICD) diagnosis codes to identify patients with infective endocarditis (IE) associated with intravenous drug use (IVDU). The results of their study were published in Open Forum Infectious Diseases.

The aim of this research was to assess the accuracy of 2 previously described algorithms designed to identify patients with potential IVDU-IE admitted to 125 VA hospitals between January 2010 and December 2018. Algorithm A identified patients assigned concurrent ICD-9/10 codes for IE and drug use during the same admission. Algorithm B identified patients with drug use coded either during the IE admission or during outpatient or other visits within 6 months of admission. The investigators reviewed 400 randomly selected patient charts and determined the positive predictive value (PPV) of each algorithm for clinical documentation of IE, any drug use, IVDU, and IVDU-IE.

Algorithm B identified 1314 patients, a 68% increase compared with the 788 patients identified by algorithm A. For clinical documentation of diagnoses of IE, PPVs were high at 86.5% for algorithm A and at 82.6% for algorithm B. For any drug use, PPVs were also high at 99.0% for algorithm A and at 96.3% for algorithm B. However, for documented IVDU, PPVs were lower at 74.5% for algorithm A and 64.1% for algorithm B.  For combined diagnoses of IVDU-IE, PPVs were also lower at 65.0% for algorithm A and at 55.2% for algorithm B. The lower PPVs were due in part to the lack of ICD codes specific for IVDU. Among patients identified by algorithm B but not A, 72% had clinical documentation of drug use during the IE admission. The investigators concluded that this indicates a failure of algorithm A to capture cases due to incomplete recording of inpatient ICD codes for drug use.

According to the investigators, the use of VA administrative and electronic record data “allowed us to link outpatient and inpatient care episodes for individual patients, and to conduct chart reviews to confirm clinical diagnoses of IE and drug use in samples of patients [from] over 100 VA hospitals.” However, the study still lacked a gold standard for diagnoses of IE, drug use, IVDU, and combined IVDU-IE as there is no curated registry of confirmed IE cases in the national VA system. This prevented the calculation of true sensitivities and specificities of the algorithms against an external gold standard. The results might also lack generalizability outside of VA settings, but the study did have “unique strengths that produced findings that complemented prior studies.”

Based on these results, the investigators concluded that there is a need to develop more accurate algorithms for finding IVDU-IE cases that do not rely entirely on currently available ICD codes. Suggestions for improvement include the creation of ICD codes specific for IVDU coupled with efforts to improve the clinical documentation of route of drug use and natural language processing applied to clinical notes. However, the investigators acknowledge that these strategies may be difficult to implement.


Kobayashi T, Beck B, Miller A, Polgreen P, O’Shea A, Ohl ME. Positive predictive values of two algorithms for identifying patients with intravenous drug use associated endocarditis using administrative data [published online June 1, 2020]. Open Forum Infect Dis. doi: