Comparison of Medical Cannabis Use Reported on a Confidential Survey vs Documented in the Electronic Health Record Among Primary Care Patients

This survey study compares the use of cannabis for medical purposes as reported in electronic health records (EHRs) with use reported in a confidential survey.


Introduction
Cannabis and cannabinoid use in the US is prevalent and increasing. 1,2 A majority of states have legalized medical cannabis use, and among these, 18 have legalized nonmedical use. 3,4 A recent study found the prevalence of past-year cannabis use among primary care patients routinely screened for cannabis use in a state with legal nonmedical use was greater than 20%. 5 Documentation of patients' medical cannabis use in the electronic health record (EHR) can support patient-clinician discussions of the risks of cannabis use and exploration of treatment alternatives. Patients use cannabis for a variety of health conditions, [6][7][8][9][10] and although evidence suggests potential benefit for neuropathic pain, appetite, nausea and vomiting, spasticity, and shortterm sleep outcomes, most health conditions for which patients use cannabis have insufficient or nonexistent evidence of benefit, potential contraindications, and more effective first-line treatment options. [11][12][13] Moreover, cannabis use has known risks, including increased risk of cannabis and other substance use disorders, mental health disorders, acute care utilization, and withdrawal. [13][14][15][16][17][18][19][20][21] The prevalence of EHR-documented medical cannabis use may be low in comparison to selfreported prevalence. 22,23 The recent study of patients routinely screened for past-year cannabis use 5 also found that only 2% of patients had documentation of medical cannabis use in their EHR over a 1-year period, including documentation of explicit (ie, medical use) and implicit (ie, use to selfmanage a health condition or symptom) medical use.
To understand how EHR documentation of medical use compares with patient report, we used a confidential patient survey to (1) estimate the prevalence of explicit and implicit medical cannabis use among primary care patients in a state with legal nonmedical cannabis use, and (2) compare the performance of EHR-documented medical cannabis use with patient-reported medical cannabis use on the survey as the reference standard.
of KPWA primary care patients are screened annually for marijuana use, without reference to medical or nonmedical use. 26,30 Among 108 950 primary care patients eligible from March 28, 2019, to September 12, 2019, 5000 (4.6% of eligible), including patients who reported no past-year cannabis use, were randomly sampled within 60 days of their index cannabis screen date to ensure proximity of survey to index screen ( Figure). Patients were oversampled for higher frequency of past-year cannabis use and for members of racial and ethnic minoritized groups, including Hispanic/Latinx, to ensure adequate representation of racial and ethnic minoritized groups and those who use cannabis. Sampled patients received a mailed invitation, information about confidentiality, request to access patient EHR data, web-survey URL, unique study identifier, a $2 incentive, and notification of receipt of compensation ($20) for survey completion. Follow-up reminder calls offered telephone completion, prompts to complete online, and/or an offer to email the survey link. Patients acknowledged informed consent before completing the survey.  Diseases and Related Health Problems, Tenth Revision (ICD-10) codes for diagnoses in the year before the index screen date were also obtained for respondents.

Measures Patient Report of Medical Cannabis Use on a Confidential Survey
The survey first asked 2 questions about the frequency and recency of past-year cannabis use (the only 2 required for survey completion). Those with a response other than never for past-year use (ie, indicating patient report of past-year cannabis use) were asked additional questions, including 2 about past-year medical use. The primary survey measure of interest, which was used previously, 22 other symptoms (write-in option), and none. A binary indicator of patient report of implicit medical use was created for patient report of any of the 11 reasons, including other symptoms.

EHR-Documented Medical Cannabis Use
Patients were categorized as having EHR-documented medical cannabis use if more than 1 EHR note and/or an ICD-10 diagnosis indicated medical cannabis use. To identify medical cannabis use in respondent EHRs, a binary indicator of past-year medical cannabis use was created from EHR text records using methods described previously. 5,32 Medical use was defined by clinician recommendation or characterized by clinician or patient as use to manage a health condition or symptom, explicitly (eg, medical marijuana most days) or implicitly (eg, cannabis for low back pain).
Although the EHR did not prompt clinicians to document medical use, clinicians could document reasons for use in notes. In brief, all patient EHR notes, within the year before the index cannabis screen date, were evaluated using an automated machine-learned natural language processing (NLP) algorithm applied to EHR notes to (1)

Statistical Analysis
All analyses were weighted to account for stratified random survey sampling and nonresponse, unless noted otherwise, so that prevalence estimates were representative of the eligible primary care sample ( Figure). To account for survey sampling, weights were created for the inverse proportion of

JAMA Network Open | Substance Use and Addiction
Prevalence of Medical Cannabis Use Reported by Primary Care Patients vs EHR Documentation eligible patients randomly sampled within each of 10 strata resulting from 5 cannabis screen responses and the indicator for patients from racial and ethnic minoritized groups (eTable 1 in the Supplement). 33,34 Inverse probability weights were estimated using logistic regression to account for differences between respondents and nonrespondents according to demographic characteristics available at sampling. The 2 weights, multiplied, were applied to survey respondent data to obtain estimates representative of the eligible primary care sample (eTable 2 in the Supplement). 35 Unweighted characteristics of survey respondents and nonrespondents were compared using 2-sided χ 2 tests of independence with significance set at P < .05. Demographic and clinical characteristics of the primary care sample (ie, survey respondents weighted to eligible primary care sample) were described on the basis of survey and EHR data. Main analyses estimated the weighted prevalence of medical cannabis use based on patient report and EHR documentation, with 95% CIs to convey precision of estimates. 36

JAMA Network Open | Substance Use and Addiction
Prevalence of Medical Cannabis Use Reported by Primary Care Patients vs EHR Documentation

JAMA Network Open | Substance Use and Addiction
Prevalence of Medical Cannabis Use Reported by Primary Care Patients vs EHR Documentation   Asking patients about use of cannabis to manage health conditions alongside routine cannabis screening may improve recognition and documentation of medical cannabis use and the management of health conditions for which cannabis is being used.