Two trials cut treatment length in high-burden settings without losing ground

Non-inferiority trials rarely make news, because their best possible result is "no worse." In treatments that are long, toxic, expensive and hard to complete, that framing understates what is at stake. Shortening a regimen without losing efficacy is a gain in every dimension except the one being measured.

Two 2026 trials did this in settings where treatment burden decides whether treatment finishes at all.

Six months for rifampicin-resistant tuberculosis

Drug-resistant tuberculosis has historically required regimens lasting well beyond six months, with substantial toxicity and high rates of loss to follow-up.

BEAT Tuberculosis screened 432 people in South Africa and randomised 403 — 203 to the trial strategy and 200 to control [s1]. A successful outcome occurred in 174 of 202 participants (86.1%) in the trial-strategy group and 172 of 200 (86.0%) in the control group, an adjusted risk difference of −0.2 percentage points (95% CI, −6.9 to 6.5; P = 0.001 for non-inferiority) [s1].

Two-tenths of a percentage point apart, on a success rate of about 86% in both arms [s1].

Safety went in the same direction. Adverse events of grade 3 or higher occurred during treatment in 63 of 202 participants (31.2%) on the trial strategy and 74 of 200 (37.0%) on control, with 10 deaths in each group [s1].

That said, the confidence interval on the risk difference runs from −6.9 to 6.5 percentage points [s1]. Non-inferiority was established against the trial's prespecified margin, not against the possibility of a several-point difference in either direction. With about 200 participants per arm, that is what the design can deliver.

The trial was funded by the US Agency for International Development and others [s1] — worth noting given the disruption to US global health funding documented elsewhere.

Four days for severe childhood pneumonia

The second trial addresses a WHO recommendation directly. Standard treatment for African children hospitalised with severe community-acquired pneumonia has been five days of intravenous antibiotics. The trial asked whether children could step down to oral treatment on clinical improvement, and for how long in total.

Between December 2020 and August 2023, 2,248 children were screened and 1,101 randomly allocated — 480 girls (43.6%) and 621 boys (56.4%) — with the primary outcome available in 1,055 (95.8%) [s2].

Both oral step-down strategies were non-inferior to intravenous-only treatment. Primary outcomes occurred in 33 of 475 (6.9%) on co-amoxiclav, 27 of 484 (5.6%) on amoxicillin, and 6 of 96 (6.3%) on intravenous-only, with upper 95% confidence limits of 6.0 for co-amoxiclav and 5.7 for amoxicillin [s2]. There was no evidence that co-amoxiclav beat plain amoxicillin: adjusted risk difference 1.3% (95% CI, −1.8 to 4.4; P = 0.40) [s2].

On duration, every arm was non-inferior to eight days. Primary outcomes occurred in 8 of 194 (4.1%) in the 4-day group, 10 of 190 (5.3%) at 5 days, 16 of 188 (8.5%) at 6 days, 16 of 193 (8.3%) at 7 days, and 10 of 194 (5.2%) at 8 days, with all upper 95% confidence limits at or below 6.0% [s2].

Read that sequence carefully. The 4-day group had the lowest event rate at 4.1%, and the 6- and 7-day groups the highest at 8.5% and 8.3% [s2]. Nobody should conclude that shorter treatment works better; this is chance in arms of under 200. What it does rule out is a dose-response in the expected direction, which is the finding that matters.

Harms did show a gradient. Antibiotic-related or serious adverse events occurred in 1 (1.0%) in the intravenous-only group and 12 (2.5%) on amoxicillin, an adjusted risk difference of −2.3% (95% CI, −4.2 to −0.3; P = 0.025), and 16 (3.4%) on co-amoxiclav (+0.8% [−1.2 to 2.8]; P = 0.42), with rates increasing with randomised duration (slope estimate +0.9% [0.1 to 1.7]; P = 0.032) [s2].

More days of antibiotics produced more adverse events without producing better outcomes [s2]. That is the argument for shortening, and it is stronger than the efficacy comparison.

The authors' conclusion is specific: for sub-Saharan African children hospitalised with severe pneumonia without complicating factors, stepping down on clinical improvement to oral amoxicillin with a total duration of 4–5 days is non-inferior to WHO-recommended 5-day intravenous treatment [s2]. The exclusion of complicating factors is doing real work in that sentence.

Why the adherence arithmetic matters more than the effect size

Both trials measure outcomes among people who received the assigned treatment. Neither can capture the larger effect that shortening produces in practice: more people finishing.

A drug-resistant tuberculosis regimen that runs six months rather than longer is a regimen more patients complete, and completion is what prevents relapse and further resistance. A child who steps down to oral amoxicillin can leave hospital, which frees a bed and removes the cannula.

That is why "no worse" understates these results. The trials establish clinical equivalence; the operational gains sit on top and are not in the confidence intervals.

One caution against over-reading: both trials excluded complicated cases, and both were conducted in research conditions with close monitoring [s1][s2]. Shortening treatment is safe when someone is watching for the patients who need longer.

What to watch

Whether WHO guidance for severe childhood pneumonia moves to oral step-down at 4–5 days, and how quickly national protocols follow [s2]. Whether the BEAT Tuberculosis strategy is adopted where drug-resistant tuberculosis burden is highest, and whether US funding disruption affects that rollout [s1].

This article describes published trial results. It is not medical advice.

Sources

• [s1] A Pragmatic Trial of a 6-Month Strategy for Rifampicin-Resistant Tuberculosis, New England Journal of Medicine, 2026;394(24):2429–2439.
• [s2] Oral step-down, optimal drug, and total duration of antibiotic treatment in African children hospitalised with severe pneumonia, The Lancet, 2026;408(10554):545–557.