Sensitivity, specificity and predictive values of ICD-10 substance use codes in a cohort of substance use-related endocarditis patients

Background Healthcare databases have the potential to become efficient tools for epidemiological research in People Who Inject Drugs (PWID). The validity of ICD-10 codes for specific substances in this population has not been assessed. Objectives Validate ICD-10 diagnosis codes relating to the use of specific substance classes in a cohort of endocarditis patients. Methods Our study sample consisted of 379 first-episode infective endocarditis patients (Male: 208, Female: 171), aged 18-55, admitted to any of three hospitals in London, Ontario from 2007 to 2018. Of these, 287 used drugs. We validated ICD-10 substance use codes for opioids (F11), stimulants (F15), cocaine (F14) and multiple substances (F19). Sensitivity, specificity, Positive Predictive Value (PPV) and Negative Predictive Value (NPV) were calculated for each code, using self-reported substance use documented on medical record review as a gold standard. We conducted a comparative analysis between code-negative users and code-positive users for each substance. Results All substance use codes shared the same pattern: high specificity, high PPV and low sensitivity, with code F11 yielding the highest PPV (96.3%; 95% C.I.: 90.8-98.6) and sensitivity (42.6%; 95% C.I. 36.3-49.1). The code-positives and code-negatives for each substance did not differ significantly in any characteristics compared. Conclusion Our results suggest that the individual ICD-10 codes analyzed should not be used for research without adjustment for low sensitivity. However, due to high PPV and specificity, these codes may still have potential for research use. Because code-negative patients did not differ from code-positive patients, their data may be extrapolated to the overall group of substance users.