Every combination of tests in the roadside data outperformed every single test. That is the whole argument for what changed. The office cannabis policy and the one-page chairside algorithm now carry 5 more findings than they did: a 25-second eye exam and a cognitive assessment. The 5 are lack of convergence, rebound dilation, nystagmus, finger to nose, and the teach-back. Three from the eyes, two from the cognitive side. Here they are, along with what they can and can’t tell you.
Why more findings, specifically
A chairside judgment is only as good as the number of independent things supporting it. That sounds like a truism until you look at the numbers behind it.
A note before the numbers, because these are 2 different lists. What follows is a sensitivity ladder, not the 5 findings. Nystagmus and the teach-back have no sensitivity figure in the literature, so they do not appear here even though they are 2 of the 5.
From the 2016 drug recognition analysis of 302 cannabis-only cases against 302 controls:
- Rebound dilation on its own: 70.9% sensitivity
- Lack of convergence on its own: 78.8%
- Finger to nose on its own, at 3 or more misses out of 6: 87.1% or better on every diagnostic measure
- Either eye test positive: 92.7%
The combination outperformed every single finding, and the gap is not small. So the useful move is to carry more independent elements, each with a defined normal, each scoreable, each chartable.
That is what version 2 does.
One number I am deliberately leaving out
The paper’s headline result is better than anything above: requiring 2 or more of 4 psychophysical tests produced 96.7% or better on every diagnostic measure. You will see that figure quoted. It does not belong in a dental policy, and it is worth saying why.
Those 4 are finger to nose (3 or more misses out of 6), eyelid tremors during the Modified Romberg Balance test, 2 or more clues on the One Leg Stand, and 2 or more clues on the Walk and Turn.
Read the last three again. Modified Romberg is standing with feet together, head tilted back, eyes closed. One Leg Stand is standing with one foot 6 inches off the floor for 30 seconds. Walk and Turn is 9 heel-to-toe steps along a line, a prescribed turn, and 9 steps back.
Three of the 4 require a patient who is standing and walking. You cannot run them on someone reclined in a dental chair, and asking a patient to heel-toe a line across your operatory is not a clinical exam. Only finger to nose is native to the chair.
So the 96.7% is the roadside ceiling. It is not available to us, and quoting it would be borrowing credibility from a battery we cannot perform. The 92.7% is ours, and it needs a penlight.
The eye exam. 25 seconds, a penlight.
Convergence. Hold a pen tip at arm’s length and bring it slowly toward the bridge of the patient’s nose.
Normal is both eyes tracking inward and crossing, holding the target to about 2 inches. The finding is positive when one or both eyes break outward before that point, at which stage the patient sees 2 tips instead of 1. Score it by watching the eyes rather than by asking the patient, because someone with longstanding suppression will fail the test and report a single image anyway.
It appeared in 78.8% of cannabis cases and 10.9% of controls. Convergence insufficiency is also a real condition in its own right and gets more common with age, so weigh the finding against the patient in front of you.
Rebound dilation. Shine a penlight directly in the eye and watch for 15 seconds.
Normally the pupil constricts and stays constricted for as long as the light is on. The finding is positive when it constricts briefly in the first few seconds, then steadily re-opens and never returns to its constricted size.
It appeared in 70.9% of cannabis cases and 1.0% of controls, and of the 7 drug categories in the roadside protocol, cannabis is the one that most often produces it.
The trap is calling normal pupillary unrest a rebound. Unrest is a continuous, irregular flicker in pupil size under steady light, and it is normal in everyone. Rebound dilation is a one-way drift that opens and stays open. Unrest wanders both ways. That distinction is the whole test, and it is the reason this finding is easy to score loosely.
Nystagmus, which runs the other way.
Cannabis does not cause it. Vertical gaze nystagmus was found in 0 of 302 cannabis cases and 0 of 302 controls. Horizontal gaze nystagmus appeared in 2.65% of cases against 0.33% of controls, a difference that did not reach significance. Nystagmus belongs to alcohol, CNS depressants, dissociative anesthetics, and inhalants, and to anticonvulsants like phenytoin and carbamazepine, to lithium toxicity, and to vestibular and cerebellar disease.
Which is exactly why it belongs in a dental screen. You are about to give this patient nitrous, a benzodiazepine, or an opioid, and additive central nervous system depression is the hazard that actually hurts people. Nystagmus in a patient who smells of cannabis tells you something else is on board. Treat it as a hold on any sedation case until you know what.
One caution on technique. Physiologic gaze-evoked nystagmus is common in healthy people: one study found it in 21% of normal-vision subjects at 10 degrees of gaze and 34% at 20 degrees, and another found it in 71% at 30 degrees, rising to 100% at wider angles. Held at extreme gaze, essentially everyone has it. Judged by eye at the chair, without training and without a way to measure the angle, this test will generate false positives faster than information. If you are not confident in the finding, score it absent.
The cognitive assessment
Finger to nose. Eyes closed, head tipped back, arms out, index fingers extended. Call the hand; the patient touches the tip of the index finger to the tip of the nose and returns. 6 trials. Record the misses and which part of the finger landed.
A 2-part question. One instruction with 2 components. “Take your glasses off and tell me the last thing you ate today.” A patient who does one and drops the other has shown you a divided-attention deficit in real time.
The teach-back. “Tell me back in your own words what we’re doing today, and name one thing that could go wrong.”
That last one earns its own line in the chart, and here is why.
Every other finding in this policy tells you a drug is probably on board. Useful, and still one step away from the decision. What you actually have to determine is whether this patient can understand what you are proposing, appreciate how it applies to them, reason through the options, and communicate a choice. That is the accepted clinical standard for capacity, and the teach-back measures it directly, in the patient’s own language, on the record.
A patient who can do it has demonstrated understanding and appreciation. A patient who cannot has a documented capacity problem, and you never had to use the word impaired.
What none of this establishes
There is no score and no cutoff. These are data points that converge or they do not, and the weight of them drives the decision.
A positive eye finding suggests a drug is on board. It does not establish impairment and it does not establish inability to consent. You can have lack of convergence and completely intact capacity, and plenty of people do. Treating a physical finding as proof of a cognitive one is the exact failure this policy exists to prevent.
More elements make the pattern defensible. They do not make any single element decisive.
Where all of it comes from, and what that costs
None of it came out of dentistry.
It comes from the Drug Evaluation and Classification Program, the 12-step protocol law enforcement runs at the roadside, and specifically from a 2016 analysis in Accident Analysis and Prevention. Dentistry never ran the study. So we are borrowing from a discipline that did, and borrowing carries a cost you have to name out loud.
Those 302 controls were police officers and academy students. Sober, young, fit, and already trained on the very tests they were being scored on. Every specificity figure in the paper is flattered by that design.
The harder version arrived in 2023, in a randomized, placebo-controlled trial published in JAMA Psychiatry. Certified drug recognition instructors evaluated 184 cannabis users. Instructors sit above the 72 hours of classroom work and 32 hours of field certification that make someone a drug recognition expert to begin with, so this was the ceiling of available expertise. They classified 81.0% of the THC group as impaired. They also classified 49.2% of the placebo group as impaired, and believed THC was responsible in 99.2% of everyone they flagged.
Convergence tells the same story twice. In the roadside data, 10.9% of controls failed it. In the placebo-controlled trial, 49.2% of the placebo group failed it. Same test, real people instead of police cadets, and the false-positive rate quadrupled.
The Modified Romberg eyelid-tremor criterion did not replicate at all: 79.3% in the THC arm against 79.4% in placebo. That is one of the 4 components in the roadside protocol’s best-performing combination, which is a second reason not to lean on that result.
Roughly half of sober people fail this battery when it is scored by the best-trained examiners available. Our patients are older, more anxious, and more medicated than a research volunteer. Ours will do worse.
And on older specifically, there is a hard limit worth knowing. The 2016 study excluded everyone aged 60 and over, from both the cases and the controls, citing age limitations documented in the original field sobriety test validation work. So none of these numbers have been shown to hold in a patient over 60. In a general dental practice that is a large share of the schedule, and it means the figures above describe a population your operatory only partly resembles.
Which is why it runs as a capacity screen
A patient who has used cannabis raises 2 questions, and only 1 of them belongs to a dentist.
Whether someone is legally impaired is a determination law enforcement makes, using specially trained examiners and a protocol built for the roadside. Whether this patient can give valid consent, and whether it is safe to layer nitrous or a benzodiazepine or an opioid on top of what is already on board, is a clinical judgment a dental office is licensed and obligated to make.
The screen informs the second question. It never settles the first.
That framing also changed how odor is handled. Odor used to sit at the center of the workflow. It is now the trigger and nothing more, because it tells you the route and rough recency and nothing about dose, timing, or fitness to treat. An odor-triggered workflow also fails silently, since vape pens and edibles produce no smell at all and a patient can arrive with a climbing dose without tripping it. The screen now has 3 entry points instead of 1: odor, disclosure on the health history, or an unexplained finding like resting tachycardia in an otherwise well patient.
What to do with a positive
One positive finding raises vigilance. Recheck vitals, slow the visit down, reassess consent before anything irreversible.
Several converging findings with a failed teach-back is a capacity problem, and elective care waits.
A clean screen in a patient who disclosed use is a reason to proceed with ordinary caution, documented as such. Disclosure is not a finding against the patient.
Nystagmus at any point changes the question from cannabis to what else, and it holds any sedation case until you have the answer.
Then chart the findings and not the conclusion. Record the negatives too, because “converges to 2 inches, no rebound dilation, no nystagmus, 0 of 6 finger-to-nose misses, teach-back intact” is a stronger entry than silence, and it is what makes a decision to proceed defensible a year later.
The updated downloads
Both files carry all of this now.
The one-page chairside decision algorithm runs 7 steps, with the eye exam and the teach-back built in. Print it and pin it in the operatory.
The office cannabis policy template carries the full reasoning, the technique descriptions with normal and abnormal spelled out for each, and the sources, so you can see where the 2 studies disagree rather than inherit one side of it. That link is the editable Word version so you can fill in your practice name and adapt the language. There is also a read-only PDF if you just want to look it over first.
One caveat worth repeating. Whether you can refuse a lawful cardholder, discipline an employee, or write a blanket rule is state-specific and unsettled. Know your own state’s program, and run the finished policy past your own attorney before adopting it.
Dentistry has no map for the cannabis patient. The ADA encouraged best practices in 2021 and then declined to write them, citing thin data, and 5 years on that is still the posture. So anyone with an office policy is drawing their own, and the drawing gets better as the data elements accumulate. If you downloaded the earlier files, replace them.