A randomised trial of 558 people compared three ways of supporting people with alcohol use disorder through the same digital platform. The cheapest model, where people used it on their own, cost $162 per patient and cut about 42 heavy drinking days a year. The most expensive, built into clinical care, cost $802 and cut about 48.
The analysis was published on 3 November 2026 in the peer-reviewed, open access journal npj Digital Medicine, as an early accepted version. It is worth reading because it asks a question every health system faces: how much support is worth paying for?
The study at a glance
| Journal | npj Digital Medicine (peer-reviewed, open access) |
| Published | 3 November 2026 (early accepted version) |
| Who and where | People with alcohol use disorder. The abstract does not name the trial sites; the authors work at universities in Wisconsin and Kentucky, USA |
| Type of study | Cost-effectiveness analysis of a registered randomised clinical trial (NCT04011644) |
| Size | 558 participants |
| Main finding | 42.4 fewer heavy drinking days a year at $162 per patient (self-monitored), 40.2 at $524 (peer-supported), 48.3 at $802 (clinically integrated) |
| Read the paper | doi:10.1038/s41746-026-03422-7 |
What the researchers did
Alcohol use disorder is a medical condition in which a person has trouble controlling their drinking. The authors call it a major public health issue that is often under-addressed in primary care, the everyday clinic where most people see a doctor.
Digital tools can reach many people, but a health system also needs to know what they cost before it adopts one. So the team took a trial in which everyone used a shared digital platform, an app, and compared three kinds of support around it:
- Self-monitored: people used the platform on their own.
- Peer-supported: use of the platform came with support from peers.
- Clinically integrated: use of the platform was tied into clinical care.
The researchers estimated what each model cost to deliver per patient. They measured benefit as the drop in heavy drinking days, scaled up to a full year.
What they found
The clinically integrated model cut the most heavy drinking days, and it also cost the most. Here is what each one delivered:
- Self-monitored: 42.4 fewer heavy drinking days a year, at $162 per patient. That works out to $3.80 for each day avoided.
- Peer-supported: 40.2 fewer days, at $524. That is $13.00 per day avoided.
- Clinically integrated: 48.3 fewer days, at $802. That is $16.60 per day avoided.
Put another way, the most intensive model cost about five times as much as the cheapest and cut about 6 more heavy drinking days a year. The authors calculated that each extra day avoided by moving from self-monitored to clinically integrated support cost $107.90.
In the basic comparison, peer support came last. Economists call this “dominated”: it cost more than self-monitoring and cut fewer days.
The picture changed in two further analyses. When the authors added later use of health care, the results suggested that more intensive support may pay off economically. And when they modelled running the programme at scale, peer support became cost-efficient for large patient populations.
Why it matters
The authors’ conclusion is that no single model is best everywhere. They suggest health systems may benefit from offering several levels of support, matched to what patients prefer and what the organisation can manage.
That is a practical message for clinics with tight budgets. The cheapest option here was not an empty one, and the costliest may still be worth it for a system that also counts later care. The choice depends on what is being counted. For more on technology in primary care, see our explainer on AI that flags missed specialist referrals.
The catch
- All three groups had the platform. The abstract describes no group without it, so these numbers compare support models. They do not show how much of the drop came from the app itself.
- Only the abstract is online so far. This early version gives no margins of error or statistical tests, so we cannot say how certain the 6-day gap is. The text may also be edited before the final version.
- The figures are annualised. The reductions are expressed as a yearly rate, and the abstract does not give the length of follow-up.
- The scale result comes from a model. It is a calculation, not something observed in real clinics.
- One disclosure. One author holds shares in CHESS Health, an addiction software company that the paper says did not develop this app. He is also an associate editor at the journal and was not involved in its review. The trial was funded by the US National Institute on Alcohol Abuse and Alcoholism, which had no role in the study.
What it could mean for Ethiopia and similar settings
This study did not test anything in Ethiopia, and the abstract does not say where participants were treated. Its dollar costs cannot simply be carried over to a health system with different salaries, phone access and clinic staffing.
The question it asks does carry over. Where clinicians are few, it helps to know how much benefit a lower-touch model can deliver before paying for a more intensive one. Any local programme would need its own trial and its own costing.
Questions people ask
What is a heavy drinking day?
It is a day on which someone drinks above a set amount. Definitions vary, and the abstract does not give the trial’s exact cut-off.
Does this mean the cheapest option is the best one?
Not by itself. It had the lowest cost per day avoided, but the clinically integrated model cut more days, and the authors found no single model that suits everyone.
Can an app replace treatment for alcohol use disorder?
This study does not answer that. It compared support models around a digital platform, not an app against care such as counselling or medicines like disulfiram. If this affects you, talk to your doctor before changing anything.
Who paid for the research?
The US National Institute on Alcohol Abuse and Alcoholism, grant R01-AA024150.
Source: Li X et al. Cost-effectiveness analysis of digital medicine support models for alcohol use disorder. npj Digital Medicine, 3 November 2026. doi:10.1038/s41746-026-03422-7
Written with AI assistance from the published paper and reviewed by Dagim Amare, M.D., before publication. This article explains research and is not medical advice.

