No new HIV infections among 2,134 given a twice yearly injection

Two randomised trials published in the New England Journal of Medicine, one in 2024 and one in 2025, tested the same idea: an injection given under the skin twice a year to keep people from acquiring HIV. The first reported no infections at all in the injection group. The second reported two. Numbers that low are rare enough in prevention research that the design underneath them is worth reading carefully [1][2].

What the two trials tested

The drug is lenacapavir, given as an injection under the skin every 26 weeks. Both trials compared it against tablets that people already take daily for the same purpose, an approach called preexposure prophylaxis, meaning medicine taken before any contact with the virus rather than after [1][2].

The first trial, PURPOSE 1, enrolled adolescent girls and young women in South Africa and Uganda and assigned them by chance, in a two to two to one split, to the injection, to one daily tablet combination, or to a second daily tablet combination used as the active comparison. Everyone also received a matching inactive injection or tablet so that nobody knew which group they were in [1].

The second trial, PURPOSE 2, ran the same comparison in cisgender men, transgender women, transgender men and gender nonbinary people, at sites across North America, South America, Africa and Asia, assigning two people to the injection for every one assigned to a daily tablet [2].

What they found

In the first trial, among 5,338 participants who were HIV negative at the start, 55 people acquired the virus during the trial. None of them were in the injection group, which held 2,134 people. Thirty nine infections occurred among the 2,136 people on one tablet combination and 16 among the 1,068 on the other [1].

In the second trial, among 3,265 people in the main analysis, 2 infections occurred in the injection group and 9 in the daily tablet group. Expressed as a rate, that is 0.10 infections per 100 person years against 0.93. A person year is one person followed for one year, so a rate per 100 person years is the count expected if 100 people were watched for a year each [2].

The comparison that carries the result

Neither trial included a group given nothing. Withholding a prevention method that already works would not be acceptable, so the trials did something else: they estimated how much infection would have occurred in the same population without any prevention, by measuring the rate among everyone screened for entry. That estimate was 2.41 infections per 100 person years in the first trial and 2.37 in the second [1][2].

The headline efficacy figures are the injection rate set against that estimate. This is a reasonable way to run such a trial and it is also the softest joint in the argument, because the screened population is not the enrolled population and the estimate is built rather than observed. Both papers report the comparison against the daily tablets as well, which is the harder test, and the injection was lower on that comparison too [1][2].

What the trials do not settle

The first trial states plainly that adherence to both tablet combinations was low. That matters for how the result should be read. Part of what separated the groups is chemistry and part is the difference between an injection given twice a year at a clinic and a tablet somebody has to remember every morning. The trial cannot tell those two apart, and it was not built to [1].

Side effects were not absent. In the first trial, reactions at the injection site occurred in 68.8 percent of the injection group against 34.9 percent of those receiving the inactive injection, and 4 people, 0.2 percent, stopped the trial regimen because of them. In the second, 26 of 2,183 people in the injection group, 1.2 percent, stopped for the same reason, against 3 of 1,088 on tablets [1][2].

Both trials also ran in populations chosen for high background risk. That is what makes a prevention trial readable in a reasonable time, and it is also why the rates in either group cannot be carried over to a different population.

Between a trial and a clinic

A separate report, published in Clinical Infectious Diseases in 2026, followed the first 501 people to receive the injection outside a trial, across 17 sites in the United States. Starting it was often delayed by insurance processes. Early stopping was uncommon at 3.6 percent, but that is higher than the rate at which people left the trials, and it was linked to injections in the abdomen and to reactions that needed a clinical visit [3].

None of that contradicts the trials. It describes the gap that always exists between a result and its use, and it is the part of the record that a striking efficacy figure tends to crowd out.

Who ran them

Both trials were funded by Gilead Sciences, the company that makes the drug, and the author lists of both include company employees alongside academic investigators. That is stated in the papers themselves. It does not make a result wrong, and a trial that says who paid for it is behaving as it should; it sets how much independent replication a finding needs before it is treated as settled [1][2].

What would settle it

Longer follow up in populations that were not selected for high background risk, records of what happens when the injection is used in ordinary services rather than trial sites, and evidence on whether people return every 26 weeks over years rather than months. Two trials with almost no infections in one group is an unusually clean result. It is not the same as knowing what the same medicine does in a clinic that no trial is watching [1][2][3].