The Same Antibiotic, Unequal Results: Reading the Trial Evidence
Now that the timing of doxy-PEP is clear, the focus shifts to what the trials actually show. The question is not simply whether doxycycline lowers bacterial STI incidence, but how the size and certainty of that reduction compare for syphilis, chlamydia and gonorrhoea. Reading the evidence this way gives us a more useful judgment: which result is strong and consistent, which is smaller or less certain, and why a pooled number cannot be detached from the population and setting that produced it.
What the pooled numbers actually say
The most compact summary comes from WHO’s pooled analysis of randomized controlled trials among men who have sex with men and transgender women. In these studies, participants were compared according to whether they used the event-driven 200 mg doxy-PEP regimen. The outcome was the incidence of bacterial STIs, not a laboratory measurement of how much doxycycline reached a particular tissue or a guarantee about what would happen after one individual exposure.
| Infection or outcome | Pooled relative risk | Certainty | What the estimate means |
|---|---|---|---|
| Syphilis | 0.20 (95% CI 0.11–0.37) | High | About 80% lower relative risk under the studied conditions |
| Chlamydia | 0.15 (95% CI 0.10–0.23) | High | About 85% lower relative risk under the studied conditions |
| Gonorrhoea | 0.52 (95% CI 0.40–0.68) | Moderate | About 48% lower relative risk, with more setting dependence |
| Any of the three STIs | 0.27 (95% CI 0.15–0.47) | Moderate | A combined outcome that conceals the pathogen-specific differences |
A relative risk below 1 means that the outcome occurred less often in the doxy-PEP group than in the comparison group. A relative risk of 0.20 does not mean that 20% of every person is protected, nor that four out of five infections are prevented after any single exposure. It describes the ratio of incidence between groups over the studied follow-up. The percentage translation is useful for seeing the gradient, but it should not be mistaken for an individual forecast.
The table also separates magnitude from certainty. Syphilis and chlamydia have the largest reductions in the pooled analysis and high-certainty evidence. Gonorrhoea still shows a meaningful reduction, but its estimated effect is smaller and the certainty is moderate. Those labels are not grades for the moral importance of the infections, nor are they declarations that one result is real and another is false. They express how confident the guideline is in the evidence supporting the estimate after considering study limitations and variation.
Confidence intervals help show why a point estimate should not stand alone. The syphilis estimate is 0.20, but the interval extends from 0.11 to 0.37; the evidence is compatible with a range of substantial reductions. The gonorrhoea estimate is 0.52, with an interval from 0.40 to 0.68. That interval still lies below 1, supporting a reduction in the pooled analysis, but it does not tell us that every setting will produce a relative risk near 0.52. A pooled estimate summarizes the evidence available; it does not remove the conditions under which the evidence was generated.
The combined outcome of “any STI” illustrates the danger of aggregation. An RR of 0.27 is important for understanding the overall burden of the three infections in the trials, but it cannot answer the pathogen-specific question. A program concerned primarily with syphilis may see a different expected benefit from one concerned primarily with gonorrhoea. The same person can also face different probabilities of acquiring different infections. The aggregate is therefore an additional outcome, not a replacement for the three separate ones.
There is a practical reason to keep this distinction visible. If the headline is that doxy-PEP cuts bacterial STIs by roughly three quarters, a reader may reasonably infer that the protection is broadly uniform. The pathogen-specific estimates contradict that inference. Chlamydia and syphilis show strong pooled effects with high certainty; gonorrhoea shows a smaller effect whose interpretation depends more heavily on setting. The overall conclusion is positive, but it is not flat.
The pooled result is useful only when its pathogen-specific branches remain visible.
Why the trials do not line up perfectly
The pooled numbers are not the first evidence readers encounter. They are built from trials whose populations, locations and observed outcomes were not identical. Looking at those trials side by side does not weaken the overall finding; it tells us what kind of finding it is.
The IPERGAY substudy was conducted among 232 HIV-negative men who have sex with men and transgender women taking HIV PrEP. Participants were assigned either a single 200 mg doxycycline dose after condomless anal or oral sex, ideally within 24 hours and no later than 72 hours, or no medication prophylaxis. Over the follow-up period, doxy-PEP reduced chlamydia risk by 70%, with a hazard ratio of 0.30 and a 95% confidence interval from 0.13 to 0.70. Syphilis risk fell by 73%, with a hazard ratio of 0.27 and a confidence interval from 0.07 to 0.98. Gonorrhoea did not show a statistically significant difference: the hazard ratio was 0.83, with a confidence interval from 0.47 to 1.47.
The phrase “not statistically significant” needs careful handling. It does not mean that the IPERGAY study proved doxy-PEP has no effect on gonorrhoea. Its interval includes values compatible with a reduction as well as values compatible with little or no reduction, so the study did not resolve the question on its own. It did, however, show a clear contrast between the observed chlamydia and syphilis results and the uncertain gonorrhoea result in that setting.
The U.S. DoxyPEP trial supplied a different pattern. It included 501 men who have sex with men and transgender women with HIV infection or taking HIV PrEP, all with a history of condomless sex and at least one STI during the previous 12 months. Participants self-administered 200 mg after condomless anal, oral or vaginal sex. Gonorrhoea incidence was reduced among both the HIV-PrEP group and the group living with HIV, with relative risks of 0.45 and 0.43, respectively. Chlamydia relative risks were 0.12 and 0.26, while early syphilis relative risks were 0.13 and 0.23. The direction of effect was consistent across these groups, but the size was not identical.
The French ANRS DOXYVAC study adds a third context. It enrolled men who have sex with men who had been taking HIV PrEP for at least six months and had experienced an STI in the previous year. In the doxy-PEP comparison, the adjusted hazard ratio was 0.49 for gonorrhoea, 0.11 for chlamydia and 0.21 for syphilis. Once again, the three infections did not move together: chlamydia and syphilis showed larger reductions, while gonorrhoea showed a smaller but statistically supported reduction.
The metrics are not identical, and that matters when the trials are placed in one table. IPERGAY reports hazard ratios, the DoxyPEP results above report relative risks, and DOXYVAC reports adjusted hazard ratios. Each is a relative comparison, but a hazard ratio incorporates the timing of events during follow-up, whereas a relative risk compares incidence between groups. The numbers can show a common direction and a broad pattern, but they should not be treated as if they were measurements from one common instrument. Even the word “reduced” needs its comparator and follow-up attached.
Randomization strengthens the within-study comparison because the groups are formed to make systematic differences less likely to explain the result. It does not make the study population interchangeable with every other population, and it does not make a result from one setting immune to differences in circulating bacteria, exposure networks or service delivery. A randomized trial can therefore provide strong evidence for what happened under its conditions while leaving a genuine question about how the estimate will travel.
These results support two conclusions at once. First, the intervention is not resting on one isolated successful trial. Across major studies, syphilis and chlamydia show a repeated pattern of substantial benefit, and gonorrhoea shows benefit in some settings even though it was not significant in IPERGAY. Second, the pattern is not perfectly uniform. The gonorrhoea estimate is sensitive enough to setting and study context that a single pooled value should not be used as a local guarantee.
Several explanations can coexist for differences among trials. The prevalence of tetracycline resistance may differ between the bacterial populations circulating in each setting. Participant characteristics and sexual-exposure networks may change the chance and site of exposure. Adherence may vary. Outcome ascertainment and statistical precision may differ, and chance can affect estimates even in well-designed randomized trials. The available evidence supports these as possible contributors; it does not establish that resistance alone explains every cross-trial difference.
A disciplined reading keeps three layers separate. The first is the within-trial observation: compared with no doxy-PEP, the intervention was associated with fewer infections in a defined group and follow-up period. The second is the cross-trial pattern: syphilis and chlamydia reductions recur more consistently, while gonorrhoea moves from an uncertain result in IPERGAY to reductions in DoxyPEP and DOXYVAC. The third is applicability: whether the observed pattern is likely to hold in a new setting depends on how similar the target population, bacterial strains, exposure sites, adherence and care system are. Moving from the first layer to the third is a reasoned judgment, not an automatic consequence of statistical significance.
This approach also prevents the null result in one trial from being overread in either direction. A non-significant result is not evidence that the intervention has no effect; it may reflect a smaller sample, a wider interval or a setting in which the effect is genuinely smaller. Conversely, a significant result in another setting is not proof that the same effect will be reproduced everywhere. The evidence becomes more informative when the trials are treated as related observations whose differences need explanation, rather than as competing verdicts.
This is the point at which an evidence summary must stop short of a mechanistic overclaim. The trials demonstrate differences in incidence between doxy-PEP and comparison groups. They do not, by themselves, identify every biological step that produced those differences. They show that a common event-driven regimen can lead to unequal outcomes; explaining why requires attention to the characteristics of the pathogen and the setting, not just more repetition of the percentages.
What the comparison can—and cannot—establish
The comparison can establish a strong, high-certainty preventive signal for syphilis and chlamydia in the studied populations, and a meaningful but more moderately certain gonorrhoea signal. It can establish that gonorrhoea results vary across trials, including one study without a statistically significant difference and later studies with reductions. It can establish that the combined “any STI” result is useful for overall prevention but insufficient for understanding any one pathogen.
There is no contradiction between saying that doxy-PEP works and saying that its effect is conditional. “Works” is a claim about the average comparison observed in a defined study, while “conditional” identifies the circumstances that give the average meaning. The two statements become contradictory only if the first is silently expanded into “works equally for every pathogen, every person and every setting.” The trials do not support that expansion.
It cannot establish that a person who takes doxy-PEP will receive an 80%, 85% or 48% reduction in their own risk. It cannot establish that the same relative effect will occur where the circulating gonococcal population has a different tetracycline-resistance profile. It cannot establish that the intervention works equally in populations not represented by the main evidence base. Those are not evasive caveats; they are the conditions needed to apply a group-level estimate responsibly.
The evidence also does not tell us that the smaller gonorrhoea effect is an accidental statistical nuisance. WHO explicitly links that effect to the prevalence of circulating tetracycline resistance in N. gonorrhoeae. The trial comparison reveals the pattern; the next question is the biological and public-health relationship that can make that pattern change from one place to another.
The right way to carry the results forward is therefore conditional. Ask which pathogen is the target, how closely the population and exposure context resemble the trials, and what local information could change the estimate. For syphilis and chlamydia, the evidence gives a relatively firm basis for expecting substantial benefit in the studied setting. For gonorrhoea, the evidence supports benefit but demands more attention to local susceptibility. That distinction is the foundation for understanding why antimicrobial resistance is not a side issue in doxy-PEP policy.