U.S. Adults With Pain, a Group Increasingly Vulnerable to Nonmedical Cannabis Use and Cannabis Use Disorder: 2001–2002 and 2012–2013

While many individuals can use cannabis without harm, evidence indicates that regular or heavy users are at increased risk for health consequences, including vehicle crashes, respiratory symptoms, emergency department visits, psychiatric symptoms, withdrawal, and cannabis use disorder (1). Despite this evidence, however, U.S. adults have become increasingly likely to perceive cannabis use as harmless (2), and nonmedical use of cannabis, including daily or near-daily use, has increased among U.S. adults since the early 2000s (2–4). The prevalence of adult cannabis use disorder has also increased, including among hospital inpatients (5), Veterans Health Administration patients (6), and in one general population study (4) (although not in another [2]). Given the evidence for an increase in the prevalence of cannabis use disorder, identifying characteristics that increase the risk of frequent nonmedical cannabis use and cannabis use disorder is an important public health issue. Pain may be one such characteristic.

Pain, an unpleasant sensory-emotional experience associated with actual or potential tissue damage (7), is common in U.S. adults (8), is a leading cause of disability (9), and is associated with substance use disorders (10, 11). Since 1996, 34 states have passed laws authorizing cannabis use for various medical conditions, including pain, and 11 states have legalized recreational cannabis use. Although meta-analyses of the effectiveness of cannabis to treat pain are inconsistent (12–14), 66% of adults now view marijuana as beneficial for pain management (15). Professional concerns about cannabis use for pain include the risk of unintended consequences, including frequent nonmedical use and cannabis use disorder (16, 17). Because of the many changes in the marijuana landscape (e.g., more permissive marijuana laws, more favorable public attitudes toward marijuana [2, 15], and greater nonmedical use [2–6]), people with pain may be a group at particular and growing risk for adverse cannabis outcomes, such as frequent nonmedical use and cannabis use disorder. If so, greater clinical and public health efforts to treat or prevent adverse cannabis outcomes among those with pain would be warranted.

To our knowledge, no studies have compared risk for nonmedical cannabis use and cannabis use disorder between those with and without pain in the adult general population, or ascertained whether these risks have changed over time. We used data from two nationally representative surveys of U.S. adults, the 2001–2002 National Epidemiologic Survey on Alcohol and Related Conditions (NESARC) (18) and the 2012–2013 National Epidemiologic Survey on Alcohol and Related Conditions–III (NESARC-III) (19), to examine two issues. First, within each survey, we examined whether the prevalence of any nonmedical cannabis use, frequent nonmedical use, and cannabis use disorder differed between respondents with and without pain. Second, given the many changes in the marijuana landscape, we examined whether these differences widened over time, that is, between the 2001–2002 and 2012–2013 surveys.

Methods

Samples and Procedures

The NESARC (18) and NESARC-III surveys (19) included adults (age ≥18) in households and group quarters sampled in multistage designs. Sample weights adjusted for selection probabilities and nonresponse. The total sample analyzed was 79,402 (43,093 in NESARC, 36,309 in NESARC-III). Rigorous field procedures were similar across surveys (3, 4), including interviewer training (structured home-study and in-class training), supervision by trained supervisors, and random respondent callbacks to verify interview data. Both surveys were sponsored by the National Institute on Alcohol Abuse and Alcoholism (NIAAA), which contracted the field work to large survey organizations (NESARC: U.S. Census Bureau; NESARC-III: Westat), and participants in both surveys were informed that the survey was sponsored by NIAAA. The methodological similarities of the two surveys have facilitated examination of change over time in numerous important health outcomes (4, 20–22). The NESARC response rate was 81.0%. The protocol and written consent procedures were approved by U.S. Bureau of the Census and Office of Management and Budget institutional review boards. The NESARC-III response rate was 60.1%, similar to other contemporaneous nationally representative surveys (23, 24). The protocols and consent procedures (verbal, recorded electronically) were approved by institutional review boards at the National Institutes of Health and Westat.

Measures

The Alcohol Use Disorder and Associated Disabilities Interview Schedule (AUDADIS), a structured computer-assisted diagnostic interview, was used to assess substance use and substance use disorders in both surveys. Identical questions assessed nonmedical substance use, including cannabis.

Study outcomes included three cannabis variables: any nonmedical use, frequent nonmedical use, and DSM-IV cannabis use disorder, with the time frame specified as the past 12 months. Nonmedical use was defined to respondents as use without a prescription or other than prescribed, for example, to get high (3). Any use was defined as at least one time, and frequent use as at least three times a week. Cannabis use disorder was defined as meeting criteria for DSM-IV cannabis dependence or abuse, combined because their criteria reflect a single condition (25). Most of the 22 cannabis use disorder symptom items used in the two surveys were identical; examination of the few slight differences showed that they could not account for the large differences in prevalence in the two surveys (3, 4).

Pain was measured in NESARC and NESARC-III with an item from the Medical Outcomes Study 12-Item Short Form Health Survey, Version 2 (SF-12) (26), a valid measure used widely in clinical (27) and general population surveys (28). This pain item uses a 5-point scale (not at all, a little bit, moderately, quite a bit, extremely) to measure how much pain interfered with “your normal work, including both work outside the home and housework” during the past 4 weeks. Consistent with a previous study (29), responses were dichotomized, coding moderate to extreme interference as positive and other responses as negative.

Control covariates included gender, age group (18–29, 30–44, 45–64, ≥65 years), race/ethnicity (Hispanic, non-Hispanic white, non-Hispanic black, Native American, Asian/Pacific Islander), education (less than high school, high school graduate or GED, at least some college), marital status (married or living together, previously married, never married), and family income ($0–$19,999, $20,000–$34,999, $35,000–$69,999, ≥$70,000). In sensitivity analyses, we used a covariate indicating whether respondents’ states had a medical marijuana law, as evaluated by legal and economic experts and used in previous studies (3, 30). At the time of NESARC, California, Colorado, Hawaii, Maine, Nevada, Oregon, and Washington had medical marijuana laws. At the time of NESARC-III, these and nine additional states—Arizona, Connecticut, Maryland, Massachusetts, Michigan, Montana, New Jersey, New Mexico, and Vermont—had medical marijuana laws.

Statistical Analysis

As in other studies utilizing the two surveys (3, 4, 20, 21), we concatenated the NESARC and NESARC-III data sets, adding a variable indicating survey. Using logistic regression, we modeled cannabis outcomes as a function of pain, survey, and pain-by-survey interactions, with control covariates including sociodemographic characteristics (age, race/ethnicity, gender, education, marital status, income) and covariate-by-pain interactions, allowing covariate effects to differ between participants with and without pain. For each outcome, predictions from the logistic models were used to generate predicted marginal prevalences (i.e., prevalences standardized to the distribution of sociodemographic characteristics of the sample pooled across the two surveys) in each survey by pain status (yes or no). Group comparisons within each survey (i.e., participants with and without pain) were conducted with contrasts between those with and without pain, with risk differences indicating group differences in absolute risk and risk ratios indicating group differences in relative risk. Between-survey comparisons (2001–2002 compared with 2012–2013) were tested with pain-by-time interactions on the additive scale for risk differences and the multiplicative scale for risk ratios. Interaction on the additive scale tested whether the difference in absolute risk of cannabis outcomes between those with and without pain had changed over time (i.e., difference in risk difference), and interaction on the multiplicative scale tested whether the ratios of relative risk for cannabis outcomes between those with and without pain had changed over time (i.e., ratio of risk ratios). Estimates for differences in risk differences and 95% confidence intervals were obtained using the PRED_EFF option in SUDAAN. Estimates for ratios of risk ratios and 95% confidence intervals were obtained using Fieller’s method incorporating predicted marginal prevalences and standard errors from SUDAAN. Interactions on the additive scale are of particular interest from a public health perspective, as they point to groups that experience the greatest population-level risk (or shifts in risk) and thus may represent important targets for prevention and intervention (see the Supplemental Material section and Table S6 in the online supplement) (31–45). To yield estimates representative of the U.S. adult population, we incorporated the survey weights, and used SUDAAN 11.0.1 (46) to adjust for the complex sampling designs. All tests were two-tailed, with the significance threshold set at <0.05.

Note that in comparing groups by their pain status, interpretation of the 95% confidence intervals differs depending on whether differences in absolute risk (risk difference) or relative risk are considered. When evaluating risk differences (risk of the outcome in those with pain minus risk in those without pain), a risk difference of 0.0 indicates no group difference by pain status. Accordingly, a risk difference whose 95% confidence interval does not include 0.0 is statistically significant at p<0.05, as are differences in risk difference estimates whose 95% confidence intervals do not include 0.0. When evaluating group differences using risk ratios (the probability of the outcome in those with pain divided by the probability of the outcome in those without pain), a risk ratio of 1.0 indicates no difference by pain status. Accordingly, risk ratios and ratios of risk ratio whose 95% confidence intervals do not include 1.0 are statistically significant at p<0.05.

We conducted several sensitivity analyses of the risk difference results. First, we reran the model for cannabis use disorder after adding cannabis withdrawal to the dependence criteria (assessed identically in NESARC and NESARC-III) and requiring at least three of seven of the cannabis dependence criteria to score dependence as positive. This was done to reflect post-NESARC evidence on the validity and coherence of cannabis withdrawal with the other cannabis use disorder criteria (25). Second, we omitted the covariate-by-pain interactions to examine whether differential covariate effects on pain drove the results. Third, we ran models with age in three-way interactions with pain and survey to explore whether difference-in-difference results varied by age. Fourth, we ran models with state medical marijuana law status at the time of each survey as a covariate. In this analysis, following previous methods (3), only participants from the 42 states included in both surveys were included (41,706 from NESARC and 36,309 from NESARC-III). Finally, to explore whether state medical marijuana law status modified the results, we ran models with medical marijuana law status in three-way interaction models similar to those for age, defining the three-level medical marijuana law variable as never medical marijuana law, medical marijuana law enacted by 2001, and medical marijuana law enacted between 2002 and 2012.

Results

The prevalence of pain among U.S. adults was 19.3% (SE=0.31) in the 2001–2002 survey and 20.0% (SE=0.44) in the 2012–2013 survey. Between the two surveys, overall increases occurred in the prevalence of any nonmedical cannabis use (from 4.1% [SE=0.15] to 9.5% [SE=0.27]), frequent nonmedical use (from 1.2% [SE=0.08] to 3.7% [SE=0.15]), and cannabis use disorder (from 1.5% [SE=0.8] to 2.9% [SE=0.13]).

Pain and Any Past-Year Nonmedical Cannabis Use

The predicted prevalences of any nonmedical cannabis use among respondents with and without pain are listed in Table 1 for the 2001–2002 survey (5.15% and 3.74%) and the 2012–2013 survey (12.42% and 9.02%). As shown in Table 2, pain was associated with significantly greater absolute risk (risk difference) and relative risk (risk ratio) for any nonmedical cannabis use in both surveys. The difference-in-risk-differences test (Table 3) indicated that the risk difference for any nonmedical cannabis use between those with and without pain was greater in the 2012–2013 survey than in the 2001–2002 survey (difference in risk difference=1.99, 95% CI=0.69, 3.29). The 95% confidence interval did not include zero, indicating that the difference in risk differences between the two surveys was statistically significant. The risk ratios reflecting greater risk for any nonmedical cannabis use among those with than those without pain did not differ significantly over time (risk ratio=1.38 in both surveys [Table 2]; ratio of risk ratios=1.00, 95% CI=0.82, 1.24 [Table 3]).

Frequent Nonmedical Cannabis Use

The predicted prevalences of frequent nonmedical cannabis use among respondents with and without pain are listed in Table 1 for the 2001–2002 survey (1.35% and 1.11%) and the 2012–2013 survey (5.03% and 3.45%). In the 2001–2002 data, the predicted prevalence of frequent nonmedical cannabis use did not differ between those with and without pain on the risk difference or risk ratio scale (Table 2). However, in the 2012–2013 data, the predicted prevalence of frequent nonmedical use was significantly greater among respondents with pain than those without pain on both the risk difference and risk ratio scales (Table 2). As shown in Table 3, while the ratio of risk ratios between the two time periods did not reach significance (ratio of risk ratio=1.20, 95% CI=0.85, 1.83), the difference in risk differences indicated that the absolute risk difference between those with and without pain was significantly greater in the 2012–2013 survey than in the 2001–2002 survey (difference in risk difference=1.35, 95% CI=0.47, 2.23).

DSM-IV Cannabis Use Disorder

The predicted prevalences of DSM-IV cannabis use disorder in respondents with and those without pain (Table 1) were 1.77% and 1.35% in the 2001–2002 data and 4.18% and 2.74% in the 2012–2013 data. In the 2001–2002 data, the predicted prevalence of cannabis use disorder differed between those with and without pain on the risk difference scale but not on the risk ratio scale (Table 2). However, in the 2012–2013 survey, the predicted prevalence of cannabis use disorder was significantly greater among those with than those without pain on both the risk difference and risk ratio scales (Table 2). While the ratio of risk ratios did not reach significance between the two time periods (Table 3) (ratio of risk ratio=1.16, 95% CI=0.80, 1.83), the difference in risk differences test indicated that the absolute risk difference between those with and without pain was significantly greater in the 2012–2013 survey than in the 2001–2002 survey (difference in risk difference=1.02, 95% CI=0.18, 1.86).

Sensitivity Analyses

The results were virtually unchanged after cannabis withdrawal was added to the criteria (see Table S1 in the online supplement) and after the covariate-by-pain interaction terms were omitted (see Table S2 in the online supplement). Age did not significantly modify the results for any or frequent nonmedical use (see Table 3A in the online supplement). For cannabis use disorder, using the ≥65-year age group as the reference group, the difference-in-difference test was significantly stronger for respondents in the 18- to 29-year age group (group contrast=3.82, 95% CI=0.57, 7.07) (see Table 3B in the online supplement). The results were similar after medical marijuana law status was added to the models (see Table S4 in the online supplement), the only difference being that the risk difference for DSM-IV cannabis use disorder in those with and without pain became nonsignificant in the 2001–2002 data. No test exploring modification of change in the pain-cannabis associations over time by state medical marijuana law status was significant (see Table S5 in the online supplement).

Discussion

To our knowledge, this is the first study to examine the differences in prevalence of nonmedical cannabis use and cannabis use disorder among U.S. adults with and without pain, and whether the risk differences changed over a period when the prevalence of moderate to severe pain remained similar. In the 2001–2002 survey, any nonmedical cannabis use and cannabis use disorder were more prevalent among respondents with pain than those without pain. In the 2012–2013 survey, the prevalences of all three cannabis outcomes were greater among those with pain than those without pain. Furthermore, when considered in absolute risk terms, the differences between those with and without pain increased over time. Extrapolating to the number of U.S. adults potentially affected, approximately 1.5 million more adults with pain were frequent nonmedical cannabis users in the 2012–2013 period than in the 2001–2002 period, and approximately 0.9 million more adults with pain had past-year cannabis use disorder in the 2012–2013 period than in the 2001–2002 period. These results suggest that among adults with pain, frequent nonmedical cannabis use and cannabis use disorder are growing problems.

Although the risk ratios (multiplicative scale) were unable to capture a significant pain-by-time period interaction for any of the cannabis outcomes, pain-by-time period interactions were found to be significant for all three outcomes using the risk differences (additive scale). Ours is not the first study to find a difference in interaction results depending on the scale used (44). As discussed further in the Supplemental Material section and Table S6 in the online supplement, and in a large methodological literature (31–45), for a study such as ours, results on the additive scale were the ones of interest, since we were interested in determining whether, during a period of striking changes in the U.S. marijuana landscape, adults in a particular group (i.e., those with pain) were more affected than others. The study findings suggest that in the context of the changing landscape, a shift occurred in the distribution of individuals among whom the cannabis outcomes occurred, and that those with pain are now a group particularly vulnerable to frequent nonmedical use and cannabis use disorder, thus warranting special attention.

While some may question inclusion of a variable representing “any” nonmedical cannabis use as an adverse outcome in a health study, this variable ensures consistency with many other epidemiologic studies of cannabis use that include such a variable (2, 47), and it is also analogous to the “any” drinking variable in the literature on alcohol (21, 48). At the same time, frequent nonmedical cannabis use exposes users to risk for adverse medical sequelae such as respiratory symptoms, injury due to intoxication-related impairments, onset of psychiatric disorders including psychotic disorders (49), and onset of cannabis use disorder (3, 50). Thus, frequent nonmedical cannabis use and cannabis use disorder are clearly public health concerns. With rates of frequent nonmedical use and cannabis use disorder increasing among U.S. adults (4–6, 51), identifying subgroups at particular risk for these conditions is necessary to inform clinical and public health intervention. Our results suggest that adults in pain are one such group.

Given the current U.S. opioid crisis, ecological studies of national U.S. data have examined the relationship of medical marijuana laws to state-level rates of opioid outcomes, finding decreased rates of opioid prescriptions (52–56), overdoses (57, 58), and opioid-related hospitalizations (59) after medical marijuana law enactment. These studies were interpreted as indicating that increased cannabis availability might help mitigate the opioid crisis. However, ecological studies are a weak design for investigating the causes of individual behaviors (43, 60–63), and none of these studies took individual pain or pain-related conditions into account. A large study of U.S. adults showed that among those with pain, cannabis use predicted higher incident nonmedical opioid use and opioid use disorder (63), and another national study showed that presence of a medical marijuana law predicted increased rates of nonmedical opioid use among 12th graders (64). These studies, along with the present findings, suggest the need for caution in advocating cannabis as a widespread substitute for opioids in treating chronic noncancer pain.

Sensitivity analyses showed that the results were robust to the influence of several factors, including medical marijuana law status. Results for any and frequent nonmedical cannabis use were consistent across age groups. However, for cannabis use disorder, differences between those with and without pain changed more over time among young adults (the 18- to 29-year age group) than among older groups. Replication of this finding would be useful, as would a study examining whether the result is an age, period, or cohort effect.

Popular beliefs hold that cannabis is an effective treatment for pain (15, 65), although meta-analyses of cannabis for treating pain (12, 13) show mixed efficacy, particularly for cannabis consumed in plant form (13). Further research with stronger designs may support public beliefs about cannabis or its constituent components as treatment for pain. Meanwhile, the need remains for other interventions to manage pain that do not incur the risk of another substance use disorder (i.e., cannabis use disorder) as an adverse treatment outcome.

Limitations of the study should be noted. Pain was indicated by a single variable rather than by physician assessment. However, NESARC and NESARC-III estimates of pain prevalence were nearly identical to National Health Interview Survey estimates (20.4%) (66), which are based on a well-validated pain scale (67). Future national surveys of substance use should measure pain and painful medical conditions more extensively. Cannabis variables were based on self-report and thus subject to social desirability bias. The NESARC did not ask about cannabis use exclusively for medical purposes (i.e., no nonmedical use). NESARC-III asked about this, finding that of the 36,309 participants, 82 used cannabis exclusively for medical purposes (68). These individuals were more likely to have pain than those whose used cannabis solely for nonmedical purposes and those who used it medically and nonmedically (68). We did not include exclusively medical users in our analyses because they were not asked frequency of use or cannabis use disorder items; future studies should do so. Given the cross-sectional nature of the surveys, we could not address the possibility that cannabis use caused pain, although this appears unlikely, or that initial cannabis use for pain eventually led to cannabis use disorder, which should also be examined in the future. Future studies should also incorporate information on state prescription opioid policies and use of opioids as prescribed and nonmedically. Also, as household surveys, NESARC and NESARC-III did not include participants who were medically institutionalized (a group possibly less likely to be nonmedical marijuana users) or incarcerated (a group possibly more likely to be users), and our findings cannot be generalized to these special populations. Also, NESARC-III compensated participants for participation, whereas NESARC did not. How compensation might have affected responses is unknown, although survey methodology studies suggest little effect of compensation on data accuracy (69). Additionally, NESARC interviewers were U.S. Census Bureau employees, whereas NESARC-III interviewers were Westat employees. Some have speculated on whether participant willingness to report cannabis use varied by whether their survey interviewer was directly employed by the federal government (70). However, the fact that participants in both surveys were informed of the governmental sponsor (NIAAA) may have mitigated this possibility.

This study did not address potential mechanisms of the results, including increasingly positive attitudes toward cannabis use, increasing public beliefs that cannabis effectively treats pain, changes in state marijuana laws, and increasing prevalence of conditions that may covary with pain and cannabis use, for example, psychiatric disorders. All of these should be examined in future studies. Finally, a higher response rate in NESARC-III would have been preferable, as survey participants may be healthier than nonparticipants (71). However, the NESARC-III rate resembled other contemporaneous national surveys, and sample weights adjusted the findings to represent the overall general population.

Conclusions

This study suggests that adults with pain are a group with particular and growing vulnerability to frequent nonmedical cannabis use or cannabis use disorder, for which prevention and clinical intervention efforts are warranted. Future studies in larger samples should examine many additional questions about pain and nonmedical cannabis use, including whether pain and additional risk factors (e.g., psychiatric disorders) contribute to the risk differences in the cannabis outcomes. Meanwhile, although marijuana legalization has social justice and tax revenue advantages, state policy makers and the health departments that implement state marijuana laws should also consider and plan for potential increases in health problems if new laws and policies increase the prevalence of adult cannabis use disorder (3, 72), especially among those with pain. Also, media reports and public education programs should provide credible public information about the risks of cannabis use, including for people with pain. A recent analysis of media reports on marijuana showed that few addressed health consequences, and of those that did, an increasing proportion focused only on favorable aspects (11% in 2012 compared with 65% in 2016) (73). Greater balance is needed in reporting marijuana issues. Finally, those developing public education programs should not assume that simple information is effective, but instead work with constituent groups to determine messaging that conveys balanced, credible information about the nature and magnitude of risks from cannabis use, including among the large group of U.S. adults with pain. Psychiatrists and other mental health professionals treating patients with moderate to severe pain should be informed about the potential risks of cannabis, including cannabis use disorder, provide information about these risks to their patients, and monitor patients for signs and symptoms of cannabis use disorder.