Using ICD-10-CM codes to detect illicit substance use: A comparison with retrospective self-report

•ICD-10 codes had high specificity and low sensitivity for detecting substance use.•Sensitivity was higher for detecting more frequent use.•Sensitivity was highest for outpatient visits and lowest for ED visits. Understanding whether International Classification of Disease, 10th Revision, Clinical Modification (ICD-10-CM) codes can be used to accurately detect substance use can inform their use in future surveillance and research efforts. Using 2015–2018 data from a retrospective cohort study of 602 safety-net patients prescribed opioids for chronic non-cancer pain, we calculated the sensitivity and specificity of using ICD-10-CM codes to detect illicit substance use compared to retrospective self-report by substance (methamphetamine, cocaine, opioids [heroin or non-prescribed opioid analgesics]), self-reported use frequency, and type of healthcare encounter. Sensitivity of ICD-10-CM codes for detecting self-reported substance use was highest for methamphetamine (49.5 % [95 % confidence interval: 39.6–59.5 %]), followed by cocaine (44.4 % [35.8–53.2 %]) and opioids (36.3 % [28.8–44.2 %]); higher for participants who reported more frequent methamphetamine (intermittent use: 27.7 % [14.6–42.6 %]; ≥weekly use: 67.2 % [53.7–79.0 %]) and opioid use (intermittent use: 21.4 % [13.2–31.7 %]; ≥weekly use: 52.6 % [40.8–64.2 %]); highest for outpatient visits (methamphetamine: 43.8 % [34.1–53.8 %]; cocaine: 36.8 % [28.6−45.6 %]; opioids: 33.1 % [25.9–41.0 %]) and lowest for emergency department visits (methamphetamine: 8.6 % [4.0−15.6 %]; cocaine: 5.3 % [2.1−10.5 %]; opioids: 6.3 % [3.0−11.2 %]). Specificity was highest for methamphetamine (96.4 % [94.3–97.8 %]), followed by cocaine (94.0 % [91.5–96.0 %]) and opioids (85.0 % [81.3–88.2 %]). ICD-10-CM codes had high specificity and low sensitivity for detecting self-reported substance use but were substantially more sensitive in detecting frequent use. ICD-10-CM codes to detect substance use, particularly those from emergency department visits, should be used with caution, but may be useful as a lower-bound population measure of substance use or for capturing frequent use among certain patient populations.