The trial was built for 40 per cent; the result stopped at 21 per cent

The gap between the planned and observed effects

EFFECT randomised 604 adults living with HIV who had cryptococcal antigen in blood but no symptoms of meningitis across 11 sites in South Africa and Tanzania. In the arm receiving flucytosine added to fluconazole for 14 days, 40 of 304 participants died, or 13 per cent. In the fluconazole-only arm, 50 of 300 died, or 17 per cent. The absolute difference was 4 percentage points and the relative reduction in mortality was 21 per cent. The protocol had sized 600 participants to detect a 40 per cent relative reduction with 91 per cent power. The trial encountered an effect smaller than the one it was built for, but not obviously trivial, and did not reach statistical significance on its primary superiority analysis.^[1]

That result cannot be translated into 'no difference'. When a power calculation targets a large effect, a smaller real effect may not separate from random variation inside the sample. The 4-point absolute gap in EFFECT is compatible with that account. The same data are also compatible with a genuinely inert treatment; one trial cannot distinguish the two. The correct verdict is that benefit was not demonstrated, not that no benefit exists. The absence of a difference in serious adverse events leading to hospitalisation or death suggests no obvious safety penalty, but it does not repair incomplete efficacy evidence. The next study must carry the uncertainty around the estimate, not simply the 21 per cent headline, into its sample-size calculation.^[1]

Two subgroups, two different evidential weights

A relative mortality reduction of 47 per cent was reported in the 134 participants with high antigen levels. That question was not wholly discovered after the fact: efficacy by baseline antigen titre appears among the protocol's secondary endpoints. The result can therefore serve as a defensible starting point for a new trial focused on people at higher risk. The 29 per cent reduction reported among adherent participants does not carry the same weight. Adherence analysis is absent from the protocol's listed secondary endpoints, and people able to adhere may differ in baseline risk. Prespecification does not confirm a subgroup result; it makes the reason for examining it transparent. A denominator of 134 still leaves wide uncertainty and room for multiplicity.^[1]

The available results come from a conference presentation; no peer-reviewed results paper is reported. The published protocol lets readers verify the design, but the confidence interval for the primary superiority analysis, handling of missing data and the full set of subgroup estimates still belong in the results paper. If a peer-reviewed article published by mid-2027 reports those details under the protocol's definitions, the 47 per cent estimate in the high-antigen group can become an assumption for the next trial. If not, it remains an interesting conference number. The methodological gain is already clear: a missed threshold is not the end of inquiry. It turns the mismatch between the planned effect and the observed difference into the starting point for the next design.^[1]