A 2024 cross-sectional study of nearly 1,600 primary care patients found that cannabis use disorder prevalence climbed from roughly 13% among less-than-monthly users to 45% among daily users on a routine one-question clinical screen. Even infrequent cannabis use was associated with meaningful rates of disorder, suggesting that all patients who disclose use may warrant a follow-up conversation about risk.

Cannabis use screening in primary care has expanded rapidly in states with legal recreational access, but clinicians have lacked concrete data on what each frequency category actually means for patient risk. Without frequency-stratified prevalence estimates, a positive screen tells a physician very little about whether the patient sitting in front of them is likely to meet criteria for a clinically significant disorder. This study provides those numbers for the first time within a routine screening program, transforming a vague signal into a calibrated risk conversation.

As cannabis legalization expands, primary care clinicians face a growing need to identify patients at risk for cannabis use disorder from among the many who disclose recreational or medical use. The Single-Item Cannabis Screener (SIS-C), a one-question frequency-of-use screen embedded in routine care at Kaiser Permanente Washington, provides a practical entry point but has lacked evidence connecting its response categories to CUD prevalence. This study sought to determine whether the four frequency strata of the SIS-C (less than monthly, monthly, weekly, and daily) correspond to meaningfully different rates of CUD as assessed by the Composite International Diagnostic Interview Substance Abuse Module (CIDI-SAM), a validated instrument aligned with DSM-5 diagnostic criteria.

In a weighted analysis projecting to the broader primary care population, the prevalence of any CUD rose monotonically from 12.7% among less-than-monthly users to 25.3% for monthly, 33.3% for weekly, and 44.6% for daily users. Moderate-to-severe CUD, a threshold with clearer treatment implications, rose from 0.9% to 20.3% across the same gradient. All linear trends were highly significant (p less than 0.001). The frequency categories also aligned with corroborating measures of cannabis intensity, including days per week, episodes per day, and route of administration. However, the study’s 34% survey response rate, single-site design in a legal-cannabis state, and average 78-day lag between screening and survey represent important constraints on generalizability. The authors emphasize the need for multi-site replication and longitudinal validation.

No Safe Harbor: Cannabis Use Disorder Is Common Even Among Infrequent Users in Primary Care

When a patient tells their primary care physician they use cannabis less than once a month, the clinical instinct is often to move on. This study suggests that instinct may be leaving roughly 1 in 8 patients with an unrecognized cannabis use disorder. What makes this paper valuable is not that it discovers a novel biological mechanism or proposes a new therapy. Its contribution is more practical than that. The researchers asked exactly the right operational question: given that we already have a one-question screen embedded in routine care, what does each answer category actually tell us about a patient’s probability of meeting diagnostic criteria for CUD? The answer is striking in its clarity. The dose-response gradient from 13% to 45% across frequency categories is both statistically robust and clinically actionable. Think of it the way we think about alcohol screening: asking how often someone drinks tells you something real about their risk of alcohol use disorder. Not everything, but enough to know whether a deeper conversation is warranted. The SIS-C performs exactly that function for cannabis.

That said, I read this paper with clear-eyed caution about its limitations. The most consequential vulnerability is the 34% survey response rate. Imagine polling a neighborhood about whether they lock their doors, but only 1 in 3 households answers. You can weight the results by what you know about who responded, such as age and income, but you cannot know whether the people who stayed silent are the ones most likely to be burgled. The same logic applies here: inverse probability weighting adjusts for observable demographic differences between respondents and nonrespondents, but it cannot account for unmeasured behavioral differences in cannabis use or CUD status. This means the specific prevalence numbers, while directionally credible, carry more uncertainty than their confidence intervals alone would suggest. The single-site design in Washington state, where recreational cannabis has been legal since 2012, further limits how confidently we can project these findings to other health systems, other patient demographics, or jurisdictions where cannabis remains illegal.

What I would tell a patient is this: even if you use cannabis rarely, it is worth a brief conversation about whether it is causing problems in your life, because the data show that risk does not begin only at high frequency. What I would tell a colleague is that we finally have the frequency-stratified CUD prevalence data needed to move beyond a simple yes-or-no interpretation of cannabis screening. And what I would tell a policymaker is that this study justifies investing in cannabis screening infrastructure, but the numbers require replication across diverse settings before they should anchor clinical thresholds or resource allocation decisions. Ultimately, a single-item screen’s value is not in its diagnostic precision but in its ability to open a calibrated clinical conversation. This study shows that even the coarsest frequency categorization carries meaningful signal about a patient’s risk profile.

This study sits at a transitional point in the cannabis screening research arc. Prior work from the same group at Kaiser Permanente Washington established that roughly 22% of primary care patients report past-year cannabis use and that about 21% of those users meet CUD criteria. What was missing was the granular link between the screen’s frequency categories and actual CUD probability. This paper fills that gap, providing the first evidence that a clinician can use a patient’s SIS-C answer to estimate their approximate CUD risk tier. The monotonic gradient across four categories is internally consistent and supported by multiple corroborating exposure measures.

From a pharmacological standpoint, the finding that 86.9% of daily users reported inhalation as their primary route is noteworthy, as inhalation delivers rapid THC peaks associated with higher reinforcement potential and greater respiratory risk. Clinicians should be aware that daily use combined with inhalation represents a compounded risk profile for both CUD and pulmonary harm. The study does not address cannabis product potency, which is an increasingly important variable as high-THC concentrates gain market share. The single most actionable recommendation from this evidence is straightforward: treat every positive SIS-C response, regardless of frequency category, as an invitation for a brief CUD risk conversation rather than a clinical non-event.

This is a cross-sectional survey study nested within a routine primary care screening program, using stratified random sampling and inverse probability weighting. It occupies a mid-tier position in the evidence hierarchy: above case series and expert opinion, but below cohort studies and randomized trials. The single most important inference constraint is that the cross-sectional design cannot establish whether high-frequency cannabis use preceded CUD or whether emerging CUD drove increases in use frequency.

The monotonic relationship between cannabis use frequency and CUD prevalence is consistent with a well-established body of epidemiological research identifying frequency of use as the strongest predictor of cannabis use disorder development. Prior work from the same Kaiser Permanente Washington research group reported an overall 21% CUD prevalence among past-year users in primary care, and the weighted average across frequency strata in the present study aligns with that estimate. What this paper adds is the clinically actionable stratification: it confirms that frequency-based screening captures meaningful risk heterogeneity, extending findings from population-level surveys into the operational context of clinical screening. No prior study had mapped CUD prevalence directly onto the SIS-C frequency categories in routine care.

The most consequential analytic choice was the use of inverse probability weighting to project survey results onto the broader primary care population. An alternative approach, such as reporting only unweighted survey-respondent estimates, would have produced different point estimates and arguably more defensible confidence intervals, since the weighting necessarily rests on assumptions about nonrespondent similarity. Additionally, had the investigators performed sensitivity analyses stratifying by the time lag between SIS-C completion and survey response, or had they included patients who did not disclose cannabis use on the SIS-C as a comparison group, the picture of CUD prevalence and screening performance could have shifted meaningfully. The core dose-response gradient would likely persist under any reasonable analytic alternative, but the absolute prevalence estimates at each tier should be treated as approximate.

The most likely overinterpretation is treating these prevalence estimates as universal benchmarks applicable to all primary care populations. These data are specific to patients who disclosed past-year cannabis use on a non-confidential clinical screen within a single integrated health system in a legal-cannabis state. Patients who chose not to disclose cannabis use, and patients in other healthcare systems or jurisdictions, may have very different CUD prevalence profiles. A related misreading is the assumption that daily cannabis use always equals CUD: in fact, more than 55% of daily users in this sample did not meet criteria for any CUD. Conversely, dismissing less-than-monthly users as risk-free ignores the 12.7% prevalence of any CUD in that group.

This study contributes the first frequency-stratified CUD prevalence estimates for a real-world clinical cannabis screen in primary care, demonstrating a robust gradient from 13% to 45% across four use frequency categories. It does not establish causation, does not validate the SIS-C as a diagnostic tool, and does not speak to populations outside one integrated health system in a legal-cannabis state. For practice today, the message is clear: every positive cannabis screen, at any frequency, warrants a brief, evidence-informed conversation about risk.

Does this study prove that using cannabis causes cannabis use disorder?

No. This was a cross-sectional study, which means it captured a snapshot at one point in time. It shows that people who use cannabis more frequently are more likely to meet criteria for CUD, but it cannot tell us whether frequent use caused the disorder or whether the disorder drove more frequent use. Longitudinal studies following patients over time would be needed to untangle that question.

If I use cannabis only occasionally, should I be worried about cannabis use disorder?

The study found that about 1 in 8 people who reported using cannabis less than once a month still met criteria for some level of CUD. This does not mean occasional use is dangerous for everyone, but it does suggest that even infrequent users may benefit from an honest conversation with their physician about how cannabis fits into their life and whether it is causing any difficulties.

Do these numbers apply to everyone who uses cannabis?

Not necessarily. These estimates come from one health system in Washington state, where recreational cannabis has been legal since 2012. The patients studied were insured, receiving regular primary care, and willing to disclose cannabis use on a clinical screen. People in different healthcare settings, different states, or those who do not disclose their cannabis use may have different risk profiles. Replication in other populations is needed before these numbers can be applied broadly.

What is the SIS-C, and is it a diagnostic test for cannabis use disorder?

The SIS-C (Single-Item Cannabis Screener) is a simple one-question tool that asks patients how often they have used cannabis in the past year. It is a screening tool, not a diagnostic test. It helps identify which patients might benefit from a fuller assessment, much like a blood pressure reading identifies patients who may need further cardiovascular evaluation. The actual CUD diagnoses in this study were made using a separate, more comprehensive instrument called the CIDI-SAM.

Have thoughts on this? Share it: