Alcohol Screening: Three Questions, and 1.6 Fewer Drinks a Week

It is eleven at night and someone is still turning over a sentence from that morning's appointment. Somewhere between the blood-pressure cuff and the printout, a nurse asked how many drinks there were in a typical week, and then how often four or five arrived in a single evening. Nothing happened afterwards. The visit moved on. But the question landed the way questions land when they arrive without warning and are written down, and the search that follows it, hours later, is rarely about alcohol at all. It is about what was being looked for, what the answer went into, and what happens next.

The question resurfaced in the news feeds this week as a guideline story, which is the form these things take: a professional-audience item about a task force, a draft, a letter grade. That coverage is written for the clinician holding the questionnaire, not the person answering it. The underlying documents, though, are unusually specific about the machinery, and they are candid about where the evidence for it runs thin.

The standing recommendation is a final statement dated November 13, 2018. It gives a Grade B to screening adults aged 18 and older, including pregnant persons, for unhealthy alcohol use in primary care settings, and to providing people whose drinking is risky or hazardous with brief behavioral counseling. A Grade B is the Task Force's second-strongest category of recommendation. For adolescents aged 12 to 17 it issues an I statement instead, concluding that current evidence is insufficient to weigh the benefits against the harms. The recommendation explicitly does not cover people already diagnosed with alcohol use disorder or already seeking treatment for it, a boundary that matters more than it sounds and is returned to below.

An update has been underway since August 5, 2025, when a draft recommendation statement was posted for comment. It reaches the same two conclusions, a B for adults and an I statement for adolescents, on a larger evidence base: 79 trials covering 40,486 adults. As of this writing the Task Force's recommendation page still serves the 2018 final statement, so the draft is where the newer numbers live and the 2018 statement is what is formally in force. Two evidence reviews of the same question, seven years apart, turn out to be a useful thing to read side by side, because they agree on direction and differ a little on magnitude.

The instruments are short by design. The briefest is a single question: how many times in the past year has the person had five or more drinks in a day, for men, or four or more, for women. One occasion is a positive result. The AUDIT-C is three questions covering frequency of drinking, typical quantity on a drinking day, and how often heavy drinking occasions happen. The full AUDIT is those three plus seven more about dependence and alcohol-related problems, for ten in total. There are US-specific versions, the USAUDIT and USAUDIT-C, rewritten around the 14-gram American standard drink.

What a score does is sort, not diagnose, and the draft review is frank that the sorting is imperfect. Evidence on the accuracy of the US-adapted instruments in adults is limited: the review found only two studies, and both reported a score of 7 or more as the optimal cutoff for the USAUDIT-C. In adolescents, studies of the full AUDIT used its standard cutoff of 8 or more, and the USAUDIT-C's performance at detecting alcohol use disorder in that age group ranged from a sensitivity of 0.61 (95% CI, 0.53 to 0.69) to 0.79 (95% CI, 0.71 to 0.85), with specificity from 0.57 (95% CI, 0.48 to 0.65) to 0.79 (95% CI, 0.73 to 0.83). A test that misses between a fifth and two-fifths of the cases it is aimed at is a conversation starter, and the guidance treats it as one.

The thresholds the questions are built around are population definitions rather than personal verdicts. The CDC's binge-drinking definition, which the single-question screen borrows, is five or more drinks for men or four or more for women during an occasion. Moderate drinking is defined as no more than two drinks in a day for men and no more than one for women, and the guidance states that pregnant women should not consume alcohol. Alcohol use disorder is something else again: a DSM-5 diagnosis describing a maladaptive pattern of use leading to clinically significant impairment or distress, which no three-question screen establishes.

The counseling that follows a positive screen is smaller than the phrase suggests. Most of the interventions studied ran 15 minutes or less and were completed in a single session, generally involving human contact by phone or in person. Many used the SBIRT approach, which pairs screening with a brief intervention and, where warranted, referral to treatment, or the five A's mnemonic: ask, advise, assess, assist, arrange.

In the draft review, that conversation was associated with drinking about 1.6 fewer drinks per week than controls (mean difference -1.6; 95% CI, -2.2 to -1.0; 38 studies; n=17,816). The medians underneath the pooled figure are worth seeing: intervention groups reduced by 3.6 drinks a week and control groups by 2.3, which is to say both arms improved and the counseling accounted for the gap between them. The odds of exceeding recommended drinking limits fell 35% (OR 0.65; 95% CI, 0.55 to 0.76) and the odds of any heavy episodic drinking fell 26% (OR 0.74; 95% CI, 0.64 to 0.85). Among pregnant women, the odds of abstinence more than doubled (OR 2.26; 95% CI, 1.25 to 4.07), with a number needed to treat of 6.

The 2018 review reached the same place from a smaller base and, in places, reported larger effects: 37 adult trials and 15,974 participants, a weighted mean difference of -1.59 (95% CI, -2.15 to -1.03) with an I-squared of 63%, an odds ratio of 0.60 (95% CI, 0.53 to 0.67) for exceeding recommended limits, 0.67 (95% CI, 0.58 to 0.77) for heavy-use episodes, and the same 2.26 for abstinence in pregnancy on a tighter interval (95% CI, 1.43 to 3.56). That I-squared of 63% is the honest caveat sitting under every pooled number here: the trials disagreed with each other substantially, and a single averaged figure smooths over settings, populations and counselors that were not alike.

The adolescent gap is the largest one, and the Task Force does not paper over it. Five trials covering 2,964 adolescents were identified; four were conducted in US primary care and three of those found no statistically significant effect. The exception was a single trial of 1,871 adolescents reporting fewer alcohol-related diagnoses in the intervention group, 4.8% against 7.8% (OR 0.69; 95% CI, 0.51 to 0.94). The screening instruments available for this age group were limited in number and carried lower sensitivity and specificity, alcohol use disorder is uncommon in adolescents to begin with, and several studies drew on populations with a high prevalence of alcohol conditions, which limits how far their results travel. The I statement is not a verdict against screening teenagers; it is a statement that the studies needed to judge it have not been done.

Two other gaps apply to adults as well. There is no evidence indicating an optimal screening interval, so how often the question should be asked is unsettled. And no identified studies reported on the harms of alcohol screening in either adults or adolescents, which means the harm side of the ledger is unmeasured rather than measured and found to be zero.

The boundary the recommendation draws around itself is also a clinical one. This guidance addresses risky or hazardous drinking short of a treatment-seeking diagnosis; it excludes people already diagnosed with alcohol use disorder or already in care. That exclusion exists partly because the situations diverge medically. Abrupt cessation after heavy daily drinking can produce a withdrawal syndrome that is dangerous without medical supervision, which is why detoxification sits in clinical hands and not in the fifteen-minute conversation these trials measured.

The number that will get quoted is 1.6 drinks a week, and the fair thing to say is that it is small and real at the same time. Set against the fifteen minutes it costs, a reliable one-drink-plus shift and a 35% cut in the odds of exceeding recommended limits make this one of the better-evidenced short conversations in primary care, which is a lower bar than it sounds; most things asked in a checkup have nothing like 79 trials behind them. What deserves more weight than it gets is the shape of the uncertainty rather than its size. An I-squared of 63% says the trials genuinely disagreed, so the pooled figure describes a literature, not a person. The absent harms literature is a real absence, and a screen with no harms studies is not a screen shown to be harmless. And the adolescent I statement should be read as the Task Force wrote it: the research has not been done, which is a different and more fixable problem than an intervention that failed. The most useful thing in these documents is not the effect size at all. It is the clarification of what the question at the desk is doing, which is sorting a population into a conversation, on an instrument that misses a meaningful fraction of what it looks for.