A Lesotho trial let health workers start diabetes drugs. The result is inconclusive
After 12 months, HbA1c was 6.5% among people managed by community health workers versus 7.1% among those referred to clinics. With 103 participants analysed, the confidence interval crosses zero.
| Group | Value (%) |
|---|---|
| Community health worker care | 6.5 |
| Facility referral | 7.1 |
Type 2 diabetes is growing in low- and middle-income countries while access to chronic care lags behind it, and one proposed answer is to move care out of clinics and into villages. A cluster-randomised trial in rural Lesotho tested a strong version of that idea: community health workers who not only screened and monitored but initiated and titrated first-line drug treatment themselves, guided by a tablet-based clinical decision support system [s1].
The trial's own conclusion is carefully hedged — that this model may improve engagement in care and glycaemic control, and that larger studies are required to confirm it [s1]. The hedging is warranted, and the reasons are instructive about what a trial of this size can and cannot establish.
What was tested
The trial was nested within the Community-Based Chronic Care Lesotho cohort, which spans 103 rural villages managed by trained and supervised community health workers [s1]. Cohort participants aged 40 or older, or with a body mass index of at least 25 kg/m², were screened at home, and all those found to have type 2 diabetes were enrolled in the trial [s1].
Villages, not individuals, were randomised. In intervention villages, community health workers provided the care directly — including initiation and monitoring of metformin, atorvastatin and aspirin — working from a tablet-based clinical decision support system [s1]. In control villages, participants were referred to facility-based care in the usual way [s1].
The primary analysis was restricted to participants with uncomplicated and uncontrolled type 2 diabetes, defined as fasting glucose of at least 7 mmol/l, with glycated haemoglobin at 12 months as the outcome [s1].
The numbers
Between 13 May 2023 and 31 January 2024, 5,785 cohort participants were screened and 252 — 4·4% — were diagnosed with type 2 diabetes [s1]. Of those, 103 met the criteria for the primary analysis: 51 in control villages and 52 in intervention villages [s1].
The analysed group was 73·8% female, with a mean age of 62·3 (SD 12·8) years and a mean baseline HbA1c of 7·2% (SD 1·4) [s1].
At 12 months, mean HbA1c was 7·1% (SD 1·9) in the control arm and 6·5% (SD 1·3) in the intervention arm, giving an adjusted mean difference of −0·46% (95% CI −1·14 to 0·22) [s1].
That interval is the whole story. It spans from a clinically meaningful 1·14-point advantage for community health worker care down through zero to a 0·22-point advantage for clinic referral. The point estimate favours the intervention; the data do not exclude no difference, or a small difference in the other direction.
Two secondary observations are reported: engagement in care was higher in the intervention arm, and no relevant difference in safety outcomes was observed between arms [s1].
Why the trial ended up this small
The trial was not underpowered by design failure so much as by epidemiology. Screening 5,785 people produced 252 diabetes diagnoses [s1], and the primary-analysis criteria — uncomplicated and uncontrolled disease — narrowed that further to 103 [s1]. In a cluster-randomised design, where the effective sample size is reduced further by within-village correlation, 103 participants is a thin base for detecting a sub-1-point HbA1c difference.
The mean baseline HbA1c of 7·2% [s1] compounds the problem. That is only modestly above target, which leaves limited room for improvement in either arm and shrinks the effect size a trial would need to detect. A cohort with worse baseline control would have made the comparison easier — and would also have been a different, sicker population.
What is genuinely novel here
The authors note that randomised trials of community health worker-led models in which the workers themselves initiate, titrate and monitor first-line diabetes treatment are lacking [s1]. That is the part worth separating from the equivocal primary result.
Most task-shifting programmes stop short of prescribing. Screening, adherence support, blood pressure checks and referral are widely delegated; drug initiation generally is not. This trial delegated it, with decision-support software as the scaffolding, and reported no relevant difference in safety outcomes between arms [s1]. In a trial of 103 analysed participants that is a reassuring signal rather than a safety demonstration — small trials cannot rule out uncommon harms — but it is the kind of signal that justifies running the larger study.
What it does not show
It does not show that community health worker-led diabetes care lowers HbA1c more than clinic care. The confidence interval forbids that claim. It does not establish safety at scale. It does not address complicated diabetes, since the primary analysis excluded it. And it says nothing about durability beyond 12 months.
It also cannot be read as a general verdict on task-shifting. This was one model — supervised community health workers embedded in an existing cohort, with digital decision support and a defined three-drug formulary — in one setting.
What to watch next
The cohort is registered as NCT05596773 and the trial as NCT05743387 [s1]. The authors' stated next step is larger studies to confirm the findings [s1]. The useful thing to watch is whether a follow-on trial is powered for HbA1c or, more sensibly given these results, for engagement and retention in care — the outcome on which this trial reported a difference, and the one where the gap between village-based and clinic-based care is largest to begin with.
Sources
Sources
More on
Two GLP-1 pills went head to head. One won on blood sugar and lost on side effects.
ACHIEVE-3 randomised 1,698 people to orforglipron or oral semaglutide for 52 weeks. Orforglipron cut HbA1c more at every dose comparison — and produced more side effects and more dropouts.
Tirzepatide added to basal insulin cut HbA1c by 1.5 points in a China trial
SURPASS-CN-INS randomised 257 adults at 26 Chinese hospitals. The 10 mg and 15 mg doses performed almost identically against placebo, and diarrhoea affected roughly a third of those on the higher doses.
Across 41 trials, tirzepatide cut the most fat, and the most lean mass
Pooling 2,906 participants, researchers ranked how antidiabetic drugs affect fat versus lean body mass. Exercise blunted liraglutide's muscle loss; metformin, insulin, and DPP4 inhibitors barely moved either measure.
Among GLP-1 users, oxycodone carried the highest short-term gut risk
A 411,000-patient claims study found adults on GLP-1 drugs who started oxycodone had more severe constipation and bowel obstruction at 30 days than those starting hydrocodone or tramadol.