AUDIT, CAGE and ASSIST: screening as pre-test probability

Two people answer the same questionnaire in the same week. The first is in the waiting room of a primary care unit, has come to renew a prescription for an antihypertensive, and marks two positive answers on the CAGE. The second is in the emergency department, with acute pancreatitis, and marks the same two answers. The score is identical. The probability of alcohol use disorder, after the score, is very different in the two rooms, because it was already very different before any question. Anyone who reads the result without taking that into account treats a number as a diagnosis.

NEXO is a daily medical diagnosis game: one clinical case a day, up to six progressive clues, and ICD-10 proximity feedback across five levels.

This text uses screening instruments for alcohol and other substance use to train a central idea of clinical reasoning: every test changes a probability that already existed. The CAGE, the AUDIT, the AUDIT-C, the single question and the ASSIST are the setting. The subject is how you interpret a positive or negative result, which error each instrument tends to produce, and how to turn the score into a conversation that helps the person.

Key points

• Screening instruments change the probability of a problem; the clinical assessment that follows is what closes the diagnosis.
• The same score has different predictive values in populations with different prevalences.
• The CAGE detects more severe presentations better and detects risky use poorly; the AUDIT and the AUDIT-C were designed for a broader spectrum.
• A well-worded single question already works as an initial screen for alcohol and for other substances in primary care.
• The time frame of each instrument (lifetime or past year) changes what a positive means.
• The score opens a conversation; how the result is communicated influences whether the person comes back.

Screening is not diagnosis: what the result means

Screening means asking a simple question of many people without a specific complaint, to separate those who need a more careful assessment. By definition, screening trades precision for reach. It accepts false positives, which will be clarified in the following assessment, and tries to miss few relevant cases. The diagnosis of a substance use disorder, in turn, depends on clinical criteria, a detailed history and context.

That difference seems obvious, but it often disappears in the medical record. "CAGE positive" becomes "alcoholic" on the problem list, and the label then influences future decisions. The correct reasoning is different: a positive result raises the probability that an alcohol-related problem exists, and that probability must be confirmed or ruled out by a clinical conversation. A negative result likewise lowers the probability, but does not reduce it to zero, especially when the prior suspicion was high.

The US Preventive Services Task Force recommendation for screening for unhealthy alcohol use in adults in primary care illustrates the design well: screening comes with a brief intervention for those with risky use and with referral for assessment when there are signs of a disorder. The instrument is the first step in a pathway, not the endpoint.

Sensitivity, specificity and predictive values

It is worth recalling the four concepts in the language of screening. Sensitivity is the ability to give a positive result in those who have the problem; specificity is the ability to give a negative result in those who do not. These two numbers describe the instrument and change little from one room to another. Positive predictive value and negative predictive value describe what a result means for the person in front of you, and these change a great deal with prevalence.

The confusion between the two pairs is the source of much of the error. An article may report that a questionnaire has good sensitivity in a specialty service sample, where almost everyone has the problem. Taken to a general population, the same questionnaire produces many false positives. The reverse also happens: an instrument calibrated for the general population may look poor in a specialty service because its cutoffs do not separate moderate from severe cases well. Reading the validation article while asking "who was this tested on?" is part of critical reading of articles.

The post on pretest probability details the general logic. Here we will apply it to instruments that, because they are questionnaires, are often treated with less rigor than a blood test.

CAGE, AUDIT, AUDIT-C and single question: what each one captures

Each instrument was built for a different target, and the target determines the type of error it makes.

CAGE

Proposed by Ewing in 1984, CAGE has four questions about feeling the need to cut down, being annoyed by criticism, feeling guilty and drinking first thing in the morning. The framing is lifetime, and the questions explore consequences and perception of one's own consumption, not quantity. The review by Dhalla and Kopec concludes that CAGE works reasonably to identify more severe cases but performs poorly at detecting risky use, precisely the group in which brief intervention makes the most difference. Studies in primary care, such as the one by Aertgeerts and colleagues, reinforce this limitation.

AUDIT

The Alcohol Use Disorders Identification Test was developed in a collaborative project of the World Health Organization, described by Saunders and colleagues. It has ten questions covering quantity and frequency, signs of dependence and consequences, with reference to the last year. The WHO manual describes score ranges linked to risk levels and suggested interventions. The review by Reinert and Allen shows consistent performance across different settings and populations.

AUDIT-C

The first three AUDIT questions, about consumption, form AUDIT-C, validated by Bush and colleagues. It is short enough to fit into a consultation routine and captures risky use well. Bradley and colleagues compared CAGE and AUDIT in the same population and showed that the choice of instrument changes who is identified. Performance also varies between women and men and between population groups, which requires thinking about cutoffs suited to the population served.

Single question

Smith and colleagues validated, in primary care, a single question about how many times in the last year the person drank above a limit of drinks in one day. The same group validated a single question for drug use, including nonmedical use of prescription medicines. The advantage is speed; the limitation is the absence of information on severity, which must come next.

In practice, this means that choosing the instrument is a choice about which error you prefer to make:

• If the goal is to find risky use for brief intervention, an instrument centered on quantity, such as AUDIT-C, misses fewer cases.
• If the goal is to estimate severity, the full AUDIT brings more information.
• If time is very short, the single question serves as an entry point, with AUDIT reserved for those who answer positive.
• If the question is about other substances, ASSIST or an instrument such as TAPS covers a broader spectrum.

Pretest probability: the same score in two rooms

Go back to the two people from the opening. What changes between the rooms is the base rate: the proportion of people with the problem in that population before any question. In a primary care unit, this proportion is lower than in an emergency department that receives pancreatitis, trauma and seizures. The post on base rate in diagnosis shows how forgetting this number is one of the most common sources of interpretation error.

A didactic example, with hypothetical round numbers, makes the logic visible. Imagine a questionnaire that correctly identifies eight out of every ten people with the problem and that gives a negative result for nine out of every ten people without it.

• Room A, with low prevalence: among a thousand people, a hundred have the problem. The questionnaire finds eighty of them and marks ninety of the nine hundred without the problem as positive. Of the hundred and seventy positives, fewer than half have the problem.
• Room B, with high prevalence: among a thousand people, four hundred have the problem. The questionnaire finds three hundred and twenty and marks sixty of the six hundred without the problem as positive. Of the three hundred and eighty positives, the great majority have the problem.

The instrument is the same, with the same performance. What changed was the population. In room A, a positive calls for a careful conversation before any conclusion. In room B, a negative deserves distrust, because the chance of a false negative has grown. This reasoning in natural frequencies is easier to teach and to explain to the person being seen than formulas with likelihood ratios.

There is also a second level. Within the same room, the clinical history changes the individual probability. A rise in liver enzymes, repeated falls, absences from work or a worried family member raise the suspicion before the questionnaire. The instrument comes in afterwards, as one more piece of data, and what you do with it depends on the starting point.

The negative under high suspicion

The most dangerous error tends to be the reassuring negative. Imagine a person with recurrent pancreatitis, altered liver enzymes and worried family members, who answers negative to every question on the CAGE. The temptation is to record "negative screening" and move on. But the probability before the question was high, and an instrument with moderate sensitivity lets some cases slip through. In this situation, the negative calls for another approach: ask about quantity and frequency instead of consequences, talk alone, bring the subject up again at another time.

The same holds for the opposite direction. A young person with no clinical sign at all, who marks one positive item on a lifetime framing question, probably does not need an extensive investigation. Adjusting the intensity of the approach to the post-test probability, and not to the binary result, is what separates reasoning from protocol.

When the person knows what the question is looking for

Consumption questionnaires are transparent: whoever answers knows what a "yes" will set off. This adds a layer of uncertainty that laboratory tests do not have. The report may be reduced by fear of judgment or enlarged by someone seeking help who wants to be taken seriously. Treating the answer as data, and not as absolute truth, includes considering the relationship, the setting and the reason for the visit.

One last note on the post-test. After the score, the revised probability has to fit into a decision. If it fell below the threshold at which you would act, record it and move on. If it fell above the threshold for conversation but below the threshold for diagnosis, the next step is a detailed clinical assessment. Thinking in thresholds makes it explicit why two professionals with the same score can make different decisions, and why that is not always an error.

Table: instrument, target and reading trap

The table summarizes the design of each instrument and the most common reading error. Specific cut-off points are in the manuals and original articles and should be adjusted to the population being seen.

Instrument	What it asks	Time window	Best for	Reading trap
CAGE	Perception and consequences of consumption	Lifetime	Suggesting more severe conditions	Treating a negative as absence of risky use; treating an old positive as a current problem
AUDIT	Consumption, signs of dependence and consequences	Past year	Estimating level of risk and guiding the intervention	Using the same cut-off point in any population
AUDIT-C	Quantity and frequency	Past year	Quick screening for risky use	Confusing risky use with an established disorder
Single question (alcohol)	Days of consumption above a limit	Past year	Entry point in short visits	Stopping at the positive without assessing severity
Single question (drugs)	Drug use or non-medical use of medications	Past year	Open the conversation about other substances	Ignore non-medical use of prescribed medications
ASSIST	Use, problems and concern by substance	Lifetime and recent months	Risk score by substance	Adding different substances together as if they were a single risk

Look at the time window column. A positive CAGE may reflect a problem the person resolved years ago. Recording "positive CAGE" without asking when the episodes happened creates an outdated label that follows the person from visit to visit.

ASSIST and other substances: one score per substance

The Alcohol, Smoking and Substance Involvement Screening Test, developed by the WHO and validated in several countries by Humeniuk and colleagues, asks separately about tobacco, alcohol, cannabis, cocaine, stimulants, sedatives, hallucinogens, inhalants, opioids and other substances. For each one, it produces an involvement score that guides the level of intervention. The WHO manual for the ASSIST describes the application and the associated brief intervention.

For clinical reasoning, the interesting point is that the ASSIST forces a reading by substance. A person may have low risk for alcohol and moderate risk for prescribed sedatives, and these two pieces of information lead to different conversations. Adding everything into a general impression of "drug user" erases exactly what the instrument was made to separate.

There is also a selection trap. Applying the ASSIST only to those who "seem" to use substances reproduces the bias that universal screening tries to avoid. The decision of whom to screen is itself a decision subject to stereotype. The post about overdiagnosis discusses the opposite side: screening without a strategy for the result also creates harm, with labels and referrals that help no one.

Populations where the reading changes

Some groups require special attention in interpretation:

• Adolescents: patterns of episodic use and lower total volume change the meaning of each question; instruments specific to the age group are usually preferred.
• Older adults: smaller amounts already carry risk from drug interactions, falls and metabolic changes, and cutoffs for young adults may underestimate the problem.
• Pregnant people: any use has clinical relevance, and the goal of screening shifts from "risky use" to "any use".
• People with another language or culture: the concept of "dose" and the way of drinking vary, and a literal translation of the questions may distort the answers.

In all these groups, the principle is the same: the instrument was validated in someone, and the question is whether that person resembles those who took part in the validation.

Finally, remember that questionnaires depend on self-report. Fear of judgment, legal or employment consequences and the way the question is asked all influence the answer. A confidential setting and a neutral question improve data quality as much as the choice of instrument.

From score to conversation: communication and brief intervention

The screening result only helps if it becomes a conversation. The review by Moyer and colleagues on brief interventions in people who were not seeking treatment shows benefit in reducing use, and that is the reason to screen in primary care. The way the result is communicated decides whether the person listens.

A few practical principles help:

1. Ask permission. "May I comment on what your answers suggest?" opens more than "your test was positive".
2. Give the result as information, not as judgment. Show where the score falls and what that usually means in terms of risk.
3. Ask what the person thinks. Their perception of their own use guides the next step better than any cutoff.
4. Agree on one small next step. A concrete goal, a new conversation or an additional assessment.
5. Record it in neutral language. "Risky alcohol use by AUDIT-C, conversation started" informs more and stigmatizes less than a label.

These steps depend on listening, and listening is also part of reasoning. The post about empathy and clinical reasoning in the consultation shows how the quality of the relationship improves the quality of the history. The text about narrative medicine reminds us that an account of consumption is a story with context, not just a quantity. To organize what the person brings, it is worth reviewing how to interpret signs, symptoms and history.

It is also worth thinking about what happens afterwards. Screening on its own, applied once and forgotten, loses almost all of its value. Reapplying the same instrument in future consultations makes changes visible, and a change in the same person is more informative than any comparison with the average. A person who moves from a low score to an intermediate one within a year is telling a story, and that story may be linked to grief, unemployment, chronic pain or insomnia. The score points to the question; the context answers it.

Finally, the conversation needs time, and time is short. That is why choosing a quick instrument and preparing neutral opening sentences are practical decisions. Rehearsing these sentences in simulation, before using them in a consultation, reduces the discomfort of the person asking and increases the chance of an honest answer from the person answering.

When the score suggests a substance use disorder, referral for specialist assessment follows the local network. If the text touched on something personal, seek the help you need from a trusted health service.

How NEXO trains this reasoning

In NEXO, each fictional case with a real clinical presentation arrives in up to six clues. The first clue works as the context of the room: age, setting, complaint. It sets the pre-test probability of each hypothesis before any specific data. The following clues work as tests, and the exercise is to notice how much each one should move your list.

The five-level proximity feedback, anchored in ICD-10, shows whether your hypothesis is in the right chapter, group or category. In a case where alcohol use appears as a clue, this helps you notice when the screening data pulled the hypothesis away from the diagnosis the other clues supported. The Collection page shows the scope of the cases, and the limits of use are at /economia.

Those who research diagnostic reasoning or teaching with cases can learn about the NEXO Researcher program. Doctors who want to review cases find the path in NEXO Curator, content creators in NEXO Creators and institutions at /closer. The design of the game is explained in what NEXO is.

Frequently asked questions

Does a positive CAGE confirm alcohol use disorder?

CAGE raises the probability, and confirmation depends on clinical assessment with diagnostic criteria. Because the framing is lifetime, a positive result may reflect a past problem. The review by Dhalla and Kopec discusses these limits.

Which instrument should be used in primary care?

Instruments focused on consumption, such as AUDIT-C or the single question, capture at-risk use better, which is the target of brief intervention. The full AUDIT complements it when screening is positive, as described in the WHO manual.

Why does the same score mean different things in different places?

Because predictive value depends on prevalence. In populations with more cases, a positive result is more likely to be true, and a negative result is more likely to be false. It is the logic of pre-test probability applied to questionnaires.

Should drug screening include prescribed medications?

Yes. The single question validated by Smith and colleagues includes non-medical use of prescribed medications, a group that often stays out of the conversation.

How can the result be recorded without stigmatizing?

Describe the instrument, the result and the next step in neutral language, for example "at-risk alcohol use by AUDIT-C, brief intervention performed". Labels such as "alcoholic" carry judgment and influence future conduct.

Does this text teach how to diagnose substance use disorders?

The text is educational and deals with reading instruments as probabilistic reasoning. Diagnosis and treatment follow guidelines and clinical assessment. If you need support, seek the help you need from a health service.

Final checklist

• I estimated the pre-test probability of the population and of the person before looking at the score.
• I chose the instrument based on the target: risky use, severity or other substances.
• I checked the instrument's time window before interpreting a positive result.
• I read the ASSIST by substance, not as an overall impression.
• I gave the result back with permission and neutral language.
• I agreed on a concrete next step and recorded it descriptively.

Play the case of the day: open NEXO, read the first clue and estimate which hypotheses are already likely before seeing the second.

References

1. Ewing JA. Detecting alcoholism: the CAGE questionnaire. JAMA. 1984. PMID 6471323.
2. Dhalla S, Kopec JA. The CAGE questionnaire for alcohol misuse: a review of reliability and validity studies. Clin Invest Med. 2007. PMID 17716538.
3. Aertgeerts B, Buntinx F, Kester A. The value of the CAGE in screening for alcohol abuse and alcohol dependence in general clinical populations: a diagnostic meta-analysis. J Clin Epidemiol. 2004. PMID 15019008.
4. Saunders JB, Aasland OG, Babor TF, de la Fuente JR, Grant M. Development of the Alcohol Use Disorders Identification Test (AUDIT). Addiction. 1993. PMID 8329970.
5. Reinert DF, Allen JP. The alcohol use disorders identification test: an update of research findings. Alcohol Clin Exp Res. 2007. PMID 17250609.
6. Bush K, Kivlahan DR, McDonell MB, Fihn SD, Bradley KA. The AUDIT alcohol consumption questions (AUDIT-C). Arch Intern Med. 1998. PMID 9738608.
7. Bradley KA, Bush KR, McDonell MB, Malone T, Fihn SD. Screening for problem drinking: comparison of CAGE and AUDIT. J Gen Intern Med. 1998. PMID 17551799.
8. Smith PC, Schmidt SM, Allensworth-Davies D, Saitz R. Primary care validation of a single-question alcohol screening test. J Gen Intern Med. 2009. PMID 19247718.
9. Smith PC, Schmidt SM, Allensworth-Davies D, Saitz R. A single-question screening test for drug use in primary care. Arch Intern Med. 2010. PMID 20625025.
10. US Preventive Services Task Force. Screening and behavioral counseling interventions to reduce unhealthy alcohol use in adolescents and adults. JAMA. 2018. PMID 30422199.
11. Humeniuk R, Ali R, Babor TF, et al. Validation of the Alcohol, Smoking And Substance Involvement Screening Test (ASSIST). Addiction. 2008. PMID 18373724.
12. McNeely J, Wu LT, Subramaniam G, et al. Performance of the Tobacco, Alcohol, Prescription Medication, and Other Substance Use (TAPS) Tool. Ann Intern Med. 2016. PMID 27595276.
13. Moyer A, Finney JW, Swearingen CE, Vergun P. Brief interventions for alcohol problems: a meta-analytic review of controlled investigations in treatment-seeking and non-treatment-seeking populations. Addiction. 2002. PMID 11964101.
14. World Health Organization. AUDIT: the Alcohol Use Disorders Identification Test. Guidelines for use in primary care. who.int.
15. World Health Organization. The ASSIST screening test version 3.0 and feedback card. who.int.