On a wooden bench outside a clinic in Harare, a grandmother with a few weeks of training sits with a patient and helps her work through a problem: a debt, a husband who has left, a child out of school. They meet six times. There is no psychiatrist present and no medication involved.
In 2016, a randomised trial published in JAMA tested that model, called the Friendship Bench, across 24 clinics in Harare. At six months, 13.7% of people in the intervention group still had depressive symptoms, compared with 49.9% in the enhanced usual care group.
That is one of the strongest results in global mental health. It is also a useful embarrassment for wealthy health systems, because it shows that the main barrier to mental health care in primary care is not knowledge, technology or even specialist supply. It is how we organise and pay for the work.
The gap is enormous everywhere
According to the WHO’s 2022 World Mental Health Report, only 3% of people with depression in low- and lower-middle-income countries receive minimally adequate treatment, compared with 23% in high-income countries. Even in wealthy countries, then, most people with depression are not getting adequate care.
The conventional answer is more specialists: train more psychiatrists, hire more psychologists, open more clinics. That will never close a gap this size. There are not enough specialists in any country to see everyone with common mental disorders, and there never will be.
Primary care is where most people already go. The question is how to make it capable of treating depression and anxiety, not just noticing them.
The evidence has been in for years
Collaborative care, a model in which a primary care clinician works with a care manager and a consulting psychiatrist, tracking patients against a measure such as a depression score and adjusting treatment when they don’t improve, has been tested extensively. A 2012 Cochrane review of randomised trials found it improved depression and anxiety outcomes compared with usual care.
The Friendship Bench makes a related point from a different setting: task-sharing works. Trained, supervised non-specialists can deliver structured psychological treatment with real effect. The WHO now recommends exactly this approach to scale up care for common mental disorders.
So the model exists, and the evidence exists. What’s missing?
The US tried paying for it. Almost nobody billed.
The United States offers a revealing experiment. In 2017, Medicare introduced billing codes that let primary care practices get paid for collaborative care and other forms of behavioural health integration.
The uptake was tiny. A study in the Journal of General Internal Medicine found that in 2017 and 2018, just 0.1% of Medicare beneficiaries with behavioural health conditions received services billed under these codes.
Why would a proven model, finally paid for, go unused? Because the codes paid for the activity without paying for the infrastructure. To bill, a practice has to hire a care manager, contract a psychiatric consultant, set up a patient registry, track symptom scores monthly and document time carefully. All of that has to exist before the first claim is paid. For a small practice, the upfront investment and administrative burden outweigh what the codes return.
This is the trade-off hidden in most “integration” policy. Payers prefer to pay for discrete, countable services. Integration is not a service. It is a way of working, involving teams, registries, follow-up calls and case reviews, and fee-for-service payment is structurally bad at buying ways of working.
Why the Friendship Bench is cheaper to scale
Here is where the low-income example becomes instructive for everyone. The Friendship Bench works partly because it asks for so little of the expensive parts of the system. The counsellors are lay health workers. The intervention is structured and brief. Supervision is built in, rather than added as a billing requirement.
Wealthy systems have tended to do the opposite: take an evidence-based model and wrap it in credential requirements, documentation rules and billing codes that make it expensive to start. The result is a proven intervention available mostly in large health systems that can afford the overhead.
That is not an argument for copying the Friendship Bench wholesale into a US or European clinic. It is an argument that the design principle, simple roles, protocols and supervision, matters more than the credentials of the person on the bench.
What would actually move the needle
- Pay for the team, not the minutes. A per-patient monthly payment for a panel under measurement-based care is easier to operate than time-based codes.
- Fund the start-up costs. Care managers and registries need to exist before revenue appears. Grants or advance payments can bridge that gap.
- Let non-specialists do more. Trained lay counsellors and community health workers can deliver structured therapies under supervision, as the Zimbabwe trial showed.
- Measure and act. The core of collaborative care is not the team chart. It is tracking a symptom score and changing treatment when it doesn’t improve.
What integration looks like on a Tuesday afternoon
“Integration” is one of those words that sounds like progress and often means little. It helps to see what the working version involves.
A primary care practice screens adults for depression using a short questionnaire during routine visits. A patient scores in the moderate range. The GP talks with her, agrees on a plan, perhaps starts an antidepressant, and introduces her to the practice’s care manager before she leaves the building.
The care manager, often a nurse or social worker, adds her to a registry. Over the next weeks, he calls her to check on side effects, repeats the questionnaire, and offers brief structured therapy sessions such as problem-solving or behavioural activation. Once a week, he sits down with a consulting psychiatrist, who may never see the patient, and reviews every patient on the registry who is not improving. The psychiatrist suggests a dose change, a different medication or referral for specialist care. The GP acts on the advice.
If the patient’s score hasn’t dropped after eight to twelve weeks, the plan changes. That is the core of the model: measure, and act when measurement shows no improvement.
None of this is technologically complex. What it requires is a specific role, the care manager, whose job is to follow people up; a registry that shows who is not getting better; and protected time for the psychiatrist’s review. In most practices, none of those exist by default.
The mechanism behind the billing failure
The US experience with collaborative care codes shows why payment design matters so much.
Fee-for-service payment works best for activities that are discrete, visible and done by one person: a visit, a procedure, a test. Collaborative care is none of those. It is a set of small activities spread across a team and across weeks: calls, registry reviews, case discussions, brief therapy sessions. To bill for it, practices have to track minutes across staff members, document the psychiatric consultation, and ensure the patient has consented to the care. The administrative burden falls on exactly the small practices least able to absorb it.
There is also a timing mismatch. A practice has to hire the care manager and contract the psychiatrist before it can bill anything. The revenue comes later, uncertainly, and depends on enrolling enough patients. For many practices, that risk is not worth taking, however good the evidence.
The result is predictable. The model is adopted mainly by large health systems with the resources to build the infrastructure, and the patients who could benefit most, those served by small, stretched practices, rarely see it.
The strongest counterargument
The strongest objection to expanding mental health care in primary care is about capacity and quality.
Primary care is already overloaded. GPs in many countries have short appointments, long lists and rising demand. Asking them to take on depression and anxiety management could reduce the quality of everything else they do, or result in superficial care: a questionnaire, a prescription and little follow-up. Critics also worry that task-sharing to lay counsellors, while effective in trials, may lose fidelity at scale if training and supervision weaken.
These concerns are valid. Primary care cannot absorb mental health care without additional staff and time, and scaling task-shared therapies does require continuing investment in supervision. The Friendship Bench works partly because supervision is built in, not bolted on.
But the alternative is not specialist care for everyone. It is no care for most. The WHO figure that only 3% of people with depression in low- and lower-middle-income countries receive minimally adequate treatment makes that clear. The real choice is between imperfect care in primary settings and no care at all. The task is to make the imperfect care good enough, through measurement, supervision and adequate staffing, not to reject it.
Second-order effects
Specialist services can change shape. When primary care handles more common mental disorders, psychiatrists can focus on complex cases, and much of their influence can come through consultation rather than direct care. That multiplies their reach.
Physical health outcomes may improve. Depression often accompanies chronic physical conditions such as diabetes and heart disease, and can make them harder to manage. Treating it in the same setting creates the opportunity to address both together.
Stigma may soften. Receiving mental health care from the same clinic that treats blood pressure may feel less stigmatising than attending a separate mental health service. That could bring in people who would never seek specialist care.
Risk of medicalisation. On the other side, routine screening without follow-up capacity could lead to more prescribing without improvement. Screening only makes sense when there is a plan for those who screen positive.
A note on digital tools
Digital mental health tools are often proposed as the way to close the treatment gap cheaply. They may help, particularly for people comfortable with apps and for mild symptoms. But the lesson of collaborative care applies to them too. The active ingredient is not the content alone. It is measurement, follow-up and a plan that changes when a person isn’t improving. An app that delivers therapy modules without anyone noticing when a user stops, or gets worse, reproduces the weakest version of primary care mental health: something offered, nothing tracked. Digital tools work best as part of a team-based model, not as a substitute for one.
What this means for decision-makers
For ministries and payers, it means paying for teams and outcomes rather than minutes. A monthly per-patient payment for people enrolled in measurement-based care, with a bonus for improvement, would be simpler for practices to operate than time-based codes.
For health system leaders, it means funding care managers as a core role, not a project, and building simple registries into existing records.
For primary care clinicians, it means adopting measurement: using a standard symptom score at every contact for patients being treated for depression, and changing course when it doesn’t improve.
For funders in low-income settings, it means backing task-sharing models with long-term investment in training and supervision, not just pilot grants.
What to watch next
Watch whether payers move from time-based collaborative care codes toward simpler per-patient payments. Watch whether task-sharing programmes can hold their results as they scale beyond research settings. And watch whether countries begin to report treatment coverage for depression, not just prevalence, as a routine health system indicator.
The takeaway
Integrating mental health into primary care is not an unsolved problem. It is a solved problem that most health systems have not been willing to fund in a way that works.
A grandmother on a bench in Harare, with training and supervision, outperformed usual care by a wide margin. The lesson is not about benches. It is that the evidence-based version of care can be simple, and our payment systems keep making it complicated.
What’s stopping your primary care teams from treating depression, not just referring it? Tell us in the comments.
