Electronic Health Record Data May Lead to Underestimates of Cannabis Use—Especially Among Older Populations

Javanbakht et al. published an important paper in this issue of the Journal of the American Geriatrics Society (JAGS) that estimated prevalence of current cannabis use among older adults within a large healthcare system in Los Angeles, California.¹

The authors should be commended for appropriately noting many of the limitations of utilizing a data source derived from self-report. We hope to expand this discussion by further considering the potential limitations associated with utilizing data from electronic health records.

In this study, patients were asked about substance use via face-to-face interviews during their annual physical exam visits with medical staff. They were first asked if they had used any substance from a list of over thirty drugs that included cannabis; those who answered affirmatively regarding any use were subsequently queried about “current” use of said drugs in an open-ended manner. While validated drug screenings are most ideal,² asking about drug use in any systematic manner is more advantageous than not. However, it is worth noting that asking one overall question about drug use that then leads to specific follow-up questions if answered affirmatively (i.e., gate questions) tends to underestimate the specific behaviors of concern.^3,4

An additional consideration that can lead to underestimation of prevalence is the manner through which self-reported responses are obtained. It has been demonstrated that participants are more likely to underreport drug use during face-to-face interviews (as compared to surveys),⁵ especially if not conducted in a private setting (e.g., away from others, such as family members).⁶ While the authors reasonably put forth that the legalization of recreational cannabis in their study location may have been associated with destigmatized reporting of its use, we would implore caution—the direct probing of use between provider and patient in this study may in fact offset any potential reduction in stigma surrounding use owing to the recent legality of use. Please note that considerations surrounding stigma are further discussed below.

Current use in this study was commonly defined by medical staff as use within the past 1–3 months. The varying time frames adopted by staff and patients alike is problematic. In the existing literature, current use is typically defined as past-month use.⁷ The authors correctly note that this is a limitation that may, in part, contribute to underestimation of prevalence of use among patients who use infrequently, but most of their discussion then focuses on limited recall of infrequent events. Indeed, this is a valid point, though we would like to add that less recent use is associated with infrequent use. Specifically, as shown in Figure 1, among noninstitutionalized adults in the United States (US) in 2019 (data from the National Survey on Drug Use and Health [NSDUH]⁷), past-month cannabis users were much more likely to engage in higher frequency use than those engaging in past-year but not past-month use (p<.001). As such, we believe that the small window period adopted in this study is more likely to detect past-month users (who are more likely to be frequent users) as opposed to less frequent users (who are more likely to underreport use).

Figure 1.

Frequency of cannabis use in relation to recency of use among noninstitutionalized individuals in the United States, 2019

The authors also mention that self-reported cannabis use typically has a high concordance with biospecimen test results, but it should be noted that in two of the studies cited by the authors,^8,9 older patients were more likely to test positive after having not reported cannabis use. In addition, it has been shown that the prevalence of recanting, which is defined as the subsequent denial of previously self-reported use of a drug on a follow-up survey,¹⁰ can increase with age. In fact, adults of ages 45–64 at baseline in one study were found to be at over twice the odds of recanting cannabis and “hard” drug use compared to those of ages 25–44, with prevalence of recanting reaching 30% at 10 years following the initial survey.¹¹ This further suggests that self-report tends to be less reliable in older populations.

As noted by the authors, other limitations include a reluctance to report or ask about cannabis use due to perceived stigma associated with use. As they thoughtfully point out, liberalized cannabis laws in California have indeed led to lower levels of disapproval toward use.¹² However, we feel it is imperative to note that overall disapproval remains high among older populations in the US. As shown in Figure 2, among noninstitutionalized adults in the US in 2019 (data from the NSDUH⁷), disapproval (and strong disapproval) significantly increase with age, with over half (52.7%) of adults ages ≥65 expressing disapproval towards adults trying cannabis (p<.001). This is important to consider because people who stigmatize drug use are more likely to recant use,¹³ as earlier discussed, and those who are more disapproving of cannabis use are less likely to use,¹⁴ or perhaps less likely to report use.

Figure 2.

Disapproval towards cannabis use according to age among noninstitutionalized adults in the United States, 2019

Finally, we must recognize that utilizing electronic health record databases can have its own limitations. Selection bias is a particular issue when there are factors that lead to systematic under-enrollment according to specific patient characteristics.¹⁵ For example, if people who use cannabis are more or less likely to have annual physical examinations, then electronic medical record data may not be representative of the overall population. Furthermore, not only are people of certain demographics more likely to visit the hospital,¹⁶ but people with health conditions may be more likely to visit hospitals more often or have more complete records (e.g., regarding prescription orders).¹⁷ Not only can this bias results, but more visits is likely associated with increases in the likelihood of reporting recent drug use as well. Even when the majority of physicians probe for and record drug use during medical visits,^18,19 a sizeable portion of patients may still be unaccounted for. Physicians may also be less likely to ask older patients in particular about drug use given a relatively low prevalence of use.

Ultimately, we would like to commend the authors for presenting their findings in this issue of JAGS. Health implications of cannabis use is a topical issue and use among older populations is certainly an important one to consider in public health. Bringing attention towards the need for routine cannabis-related screening among older adults in primary care settings appears beneficial, but interpretation of these findings should take into account a more expansive discussion of the study limitations.