The question: how many people will be impacted by the the DOGE interruption of PEPFAR PrEP (Pre-Exposure Prophylaxis) programs under USAID, what number of those impacts will involve HIV infection and/or death, and can we be confident in these numbers?
Some caveats, first. I'm not analyzing ART expenditures under PEPFAR, which is the bigger part of the program but out of scope for the discussion, here. I'm not going to try to evaluate moral or political considerations, because I don't think either of us are interested in that question. There's a plausible asterisk on any upper bound because UNAID was trying to upscale every program simultaneously, and because programs have second- and third-order effects, but these are also spaces where even the most unlikely number is nonfalsifiable. Almost all of these numbers depend (heavily) on data from a very small number of sources; what level of trust you give them will likely vary a lot, but they have gaps. Several of those gaps will favor undercounts (non-reporting by an org, large year-over-year changes since last submission, survey failures), but others will favor overcounts (id or org boundary issues can lead to multiple initiations for one person, Goodhart'd data logging, survey failures).
I think one or both of us has some confusion about PrEP vs. ART for pregnant women. From what documentation I can find, PrEP is recommended for HIV-negative pregnant women to avoid HIV transmission from outside sources to the pregnant parent. Where a pregnant or breastfeeding person is HIV-positive, all the literature I can find points to ART being recommended to prevent vertical transmission to the child. If that mostly wasn't supposed to be impacted and a number of clinics stopped PrEP but didn't stop ART, and ART was the recommended course for HIV-positive pregnant people, it's probably not going to be persuasive to argument vertical transmission -- but in turn, I don't know either the practice of medicine here, nor can I find enough I trust about any ART numbers.
((Oh, boy, Rubio said something nice. Time to update the exact other direction!))
Officially, PEPFAR is federally authorized by Congress at a specific dollar amount that has to be spent, and the Trump administration claims that it's merely moving how that spending from USAID to the Department of States while tots unrelated resisting reauthorization, but even assuming that's true there have and will continue to be unavoidable disruption. Optimistically, this analysis gives us a higher-end (though not quite a highest plausible) bound for how many people-PrEP-courses were immediately disrupted, realistically there's near-certainly more people who would have been initiated but weren't, pessimistically it tells us what realistic numbers will need to be waved around during various legal challenges if/when the Department of State turns out to have been lying again. I'm not going to try to evaluate either the political or legal odds of what happens in any of these cases.
Okay, caveats done with.
Layer One:
The first problem is figuring out how many people PEPFAR provided PrEP in January 2025. You'd expect this to be an easy question. It's not; there's a bunch of different possible metrics we could even start looking at. Some of them are into the millions, such as if you use all new initiations into PrEP programs over a year. But most numbers that give anything close to 700k people are around what data.pepfar.gov (cw: zip file with tab-separated-values) puts under the PrEP_CT statistic, which measures people who return in a quarter after their first initiation into PrEP (the PrEP_NEW stat). The Q4 2024 numbers for PrEP_CT (people continuing PrEP treatment from a previous visit) from PEPFAR gives 809,921, with the quarter before that at 761,119 continuing treatments. That's actually a higher number than the original 700k!
Though I will caveat that it's people. I can't say whether "Most recipients of preventative HIV medication were women whose husbands had HIV but who hadn't contracted it yet, or pregnant women with HIV trying to prevent their babies from contracting it." is true or not (though see discussion re: vertical tranmission and ART above), but it's at minimum a claim I'd be very careful before making. See this study for a period where it was very close to half "adolescent girl/young woman" and half "key populations" (msm, trans, sex workers, iv drug use, prisoners) who were uniquely vulnerable. The prevention tab here that should break this out is completely useless and the data.pepfar.gov spreadsheet doesn't give me any more fidelity than a 2-2.5:1 female:male ratio for FY2024 or individual quarters, so I don't have any level of confidence just at the level of programs. None of these stats measure marital status or pregnant-HIV-negative from pregnant-HIV-positive, many don't really separate out pregnant from non-pregnant; PEPFAR itself may not know these values. And I suspect that the gender-coding for PrEP programs is not 100% accurate.
But there's no way to get 700k women out of the PrEP_CT statistic honestly, and I don't think there's any combination of PrEP_CT and PrEP_NEW that gets you within 100k of that. The Q4 2024 PrEP_CT "female" number is 516,069.
USAID plays a large part of these programs, but it's not the only part, or even the only part of US financing. This study estimates that 25% of PrEP initiations came through the CDC. There's less information on the Department of Defense side of the program (DHAPP). It looks to have been included in the data.pepfar.com numbers at least some of the time, given Burundi, but because it's about interacting with military organizations it's very unlikely that their impact will relate much, if at all, to the direct money in side of things. If I had to give a guess, I'd say probably closer to 5% than 10% of PrEP_CT numbers? There's probably some cross-program impact, since the CDC and USAID weren't exactly mortal enemies, but because so much of the USAID programs worked through grant projects that have to be kept separate from the CDC spend I don't think it's very high either.
Taking the higher number of 809921 and subtracting 30%, we're at 566945.
Layer Two:
But not everyone PEPFAR reached, PEPFAR saved. Even in higher-income countries, PrEP is not perfect at preventing HIV transmission, and compliance with PrEP is far from perfect even in the best of circumstances, nevermind spaces with severe poverty, weather, or civil disruption. In an ideal world, we'd have either large-scale longitudinal studies from PrEP programs, but these are few, far between, and not very representative.
One of the ways they're not representative is that awkward PrEP_NEW stat. This measures (almost) all new enrollments into PEPFAR PrEP programs. It's also universally been close (80%-100%) to the PrEP_CT stat for the same quarter, and similar (75%-92%) the PrEP_CT stat of the next quarter, since the PrEP_CT stat started being collected in 2022. Now, there are some reasons here, not just direct noncompliance. Unlike ART, PrEP isn't necessarily something you need to take your whole life. Among key populations, you don't want people spending forever in prison or injecting drugs, and not every person who goes into sex work is going to stay in sex work. If it is ever used for pregnancy (and breastfeeding), that's a few years. Even where needed, it's not tolerated especially well, and there's a lot of supply chain or travel disruptions that can just make it hard to achieve even if the user wants to continue.
But this drastically changes a lot of what you'd expect to see out of the program. If someone is married to (and having unprotected sex with) an HIV-positive man, and only takes PrEP for six months, that does still help them... but probably not for a full year.
Almost all PrEP usage under PEPFAR programs has been oral PrEP. There are different regimens, but from the user's perspective, they're a regular pill. While there have been a number of countries that allow up to three-month prescription dispenses for PrEP, and a few pilot programs that allow up to six-month dispenses, there's been a lot of considered hesitance and tradeoffs to consider (most critically, regular followup visits are being used to also test for HIV).
There are a handful of longer-term PrEP approaches: an injectable formulation that's supposed to be taken once every eight weeks, and an insertable ring for cis women (and trans men?). But PrEPwatch puts less than 18k initiations in Africa for these in their current data combined, and they're probably a rounding error.
On the more optimistic side, not everyone that PEPFAR reached needed to be saved. Part of the adolescent girl/young woman target group is just in a monogamous relationship (or poly relationship with, or considering dating) where HIV status is untrusted, unknown, or unknowable. See this (early, qualatative) study for Kenya for some examples within PEPFAR. These programs aren't worthless! There's a Bayesian analysis here where you try to estimate the approximate rate such a woman would be exposed to HIV from a median romantic partner. But since we don't have and probably can't get a numerator here at any level of confidence, it's kinda worthless to try to include as anything but a notch of uncertainty.
PEPFAR doesn't provide this sort of data; we can't tell how many people stayed on PrEP for how long, or whether they left the program by no longer needing PrEP, no longer having the ability to get it, or having contract HIV. Studies working with subsets of PEPFAR PrEP show about 40% discontinue within six months, and over half of those discontinued did not restart within a year of initiation. Whether they stopped taking it for happy reasons or unhappy ones, they stopped taking it. Even where cycling in and out, at best that's going to drastically reduce effectiveness of PrEP.
That's a fuzzy 20% of the population who are dropping from the program on their own.
Layer Three:
PrEP is also not perfectly effective, even for people who are constantly using the medication. Nor, thankfully, is HIV perfectly contagious. This metastudy gives a relative risk ratio of 0.47-0.67 for heterosexual couples using PrEP (for those heterosexuals, from studies generally in Africa). Estimates for yearly risk for seriodiscordant couples range a lot more heavily and especially for hets seems very dependent on local conditions, but one of the trials used in that metastudy had the best results for seriodiscordant heterosexual couples was a subgroup with 0.5 infections in treatment versus 1.99 infections in control per 100 person-years. There's a few studies giving better numbers or (more often) bigger spreads, especially for particularly vulnerable key populations, but I don't think you're going to get a wider spread than 2 treatment : 10 control infections per 100 person-years.
(Caveat: law of small numbers.)
From an epidemiological perspective, that's amazing. And there are, looking forward, a lot of potential for better approaches: event-driven oral tablets seem like they're doing a little bit better, and long-acting injectibles PrEP avoids some of the stigma and social problems oral doses do. But it starts swallowing a lot of the remaining effect size, not just in the relatively small portion of people using PrEP who nevertheless get HIV, but also the larger group who aren't on PrEP and don't get HIV in that time period, either because of other preventative measures or by dumb luck.
And You <Don't> Add Them Up
If you want to throw these numbers to get a final impact, you can. All of them end up lower than 100k deaths or HIV infections across a four-year-period, and many result in less than 100k deaths or HIV infections in a four-year period. But I'm not going to actually advocate that number being any more real. This is, ultimately, an argument by just adding zeros. There are times where that's the only option available, and you have to make a decision, and it's better to make a decision with known-bad data than no data.
((Though I'll caveat that the degrees of freedom here make it so easy to put a thumb on the scales that I'm hesitant to even do that. Layer Three can swing anywhere from 0.08 to 0.015 even while taking these numbers and studies totally honestly. And I'm pretty sure there are some unknowns in there, the confidence intervals could fit a bus, the whole PrEP_NEW vs vs PrEP_CURR vs PrEP_CT makes the entire dataset a mess, and it's quite possible that some of effects are double-counted or undercounted above.))
Regardless of what you're doing, though, if you're doing anything with made-up statistics, you have to remember they're made up. Because if you're making extremely confident claims about a story that's a wild guess, it undermines any opportunity you'll ever have to persuade anyone later.