VITAL SIGNS COLUMN: Eighteen months into the dismantling of the global AIDS response, how many have died? The honest answer is that nobody knows. Rio’s numbers are inferences, its assumptions and projections disputable. But 30 years and tens of billions of dollars into the HIV/AIDS pandemic, we should not still be guessing.

The week’s Vital Signs came from Rio de Janeiro, where the 26th International AIDS Conference convened last week under the banner “Rethink. Rebuild. Rise.” A stirring slogan that would surely awaken even the dead? Therein lies the question: 18 months after Washington started dismantling the architecture of the global HIV response, how many people are dying as a result?

Nobody knows, because no one has counted the dead. Every figure in circulation is a projection against a hypothetical world without aid cuts. Mourning the unknown dead dampened the vibrant Rio buzz that I have so enjoyed in the past.

A third kind of half-truth

The assassination of evidence by advocacy should worry a global health enterprise proud of its underlying scientific base. That base is now under assault from multiple directions, with “mal-information” – distinct from misinformation or disinformation – being the latest threat.

Mal-information concerns data analysis that could be taken out of context and manipulated to make misleading connections, which damage or advance particular causes. Nowadays, that is facilitated by generally reduced population numeracy, which undermines trust in public policy and institutions and fuels weird theories and conspiracies.

Coming to HIV/AIDS, the message from Rio is that foreign aid cuts cost lives. The arithmetic of this is somewhat awkward. Policymakers scanning the headlines of well-presented UNAIDS updates may not understand, or overlook, that these are centred on epidemiological modelling. And so we rely on estimations. 

However, as we learnt from disputed mortality projections when COVID-19 was advancing relentlessly, it is the assumptions underlying mathematical models that have a huge bearing on the numbers churned out. These, in turn, take on a life of their own to support or refute whatever case protagonists want to push.

Meanwhile, it is doubtful whether the Rio delegates had time to study the UNAIDS modelling methodology note, whose entrails require forensic drilling with advanced statistical tools.

This is what is revealed. As UNAIDS does not count AIDS deaths, it estimates them in worst-affected countries via a model fed by prevalence tests at antenatal clinics, household surveys and clinic treatment registers. The model already contains assumptions about how many people die when treatment is interrupted, and so it cannot independently prove that the aid cuts killed anyone.

Worse, when clinics stop reporting – for example, if funding and staffing shrink – the model reads the silence as falling treatment and over-estimates deaths. Or, when registers go stale, it under-estimates them. 

Furthermore, the published margin of error covers only the model’s own arithmetic, not mistakes in the country data collection process. The historical series must also be revised annually, shifting past figures and making trend comparisons unstable. 

Where countries do not produce data, UNAIDS makes estimates that reviewers cannot inspect. The share filled in by such statistical guesswork is rising, as surveillance and reporting systems fray under resource constraints.

While UNAIDS documents these challenges honestly, the obvious question is whether future global HIV policy can be formulated on this basis. Even stock market indices are easier to understand.

Reduced confidence

With this caveat, the UNAIDS special report for Rio models AIDS-related deaths at 570,000 in 2025 with a range (430,000 to 780,000) so wide as to greatly reduce confidence in the central estimate. The spread is wider still around the number of people living with HIV, put at 41 million within bounds of 35.3 and 47.5 million.

The same modelling offers a mirror image. If the United Nations HIV/AIDS Political Declaration’s targets were met in full, it says, 3.2 million infections and 1.3 million deaths would be averted by 2030. That projects what success would buy, not what failure will cost –  an important distinction lost at the conference.

Confusion comes from other projections that disagree, partly because they model different worlds. One gives 74,000 excess deaths across seven African countries by 2030, assuming a 90-day funding freeze followed by near-total collapse. Another gives 770,000 to 2.9 million across all low- and middle-income countries.

Thirteen months ago, UNAIDS itself feared four million deaths by 2029, on the assumption that American-supported services collapsed entirely. The spread reflects assumptions, not data.

No dose-response curve for aid

The principal villain portrayed as responsible for future projected deaths is the aid cuts. How justified is that?

Donor government funding for HIV/AIDS dropped 25% in a year – from $8.3 billion in 2024 to $6.2 billion in 2025, the lowest level since 2007.  

It was entirely American: US disbursements fell $2.09 billion against a global fall of $2.08 billion. Other donors had already cut earlier, and now their combined total held level.

That could potentially translate into raised AIDS mortality, but by how much is uncertain. There is no “dose-response curve” for foreign aid akin to that for a drug, or even for a public health intervention like clean air.

Meanwhile, the overall financing picture is complex. Foreign aid cuts for HIV have been partially offset by a 4% increase in domestic funding in 2025, with more than 55 countries now committing to raise their own contributions. Domestic resources, public and private, carry 59% of total HIV financing in low- and middle-income countries. Total resources therefore fell by only 6%, to $17.6 billion.

The models don’t tell us how many lives are thereby saved.

What was measured, not modelled?

Meanwhile, and remembering the pathology of HIV progression, it is not the magnitude of aid cuts that matters but where they fall. What we do know, because it has been measured rather than estimated, comes from the US programme’s own returns. 

Between 2024 and 2025, prevention spending fell 51%, testing by 17% and pre-exposure prophylaxis (PrEP) initiations by a third.

Direct service staff fell by 62,541, a quarter of that cadre. Some 77,000 fewer children received treatment through those programmes. Facilities that had treated ALMOST 442,000 people simply stopped reporting, their status now unknown.

That last figure is the modelling problem made flesh. Silence from a clinic is read by the model as treatment failing and converted into deaths – whether or not anybody died. How that is folded into a mortality calculation is not clear.

Meanwhile, the foreign aid picture is not static. Could US money be spluttering back? The US Congress appropriated $4.633 billion for bilateral HIV assistance this fiscal year, against the administration’s request of $2.9 billion. But the administration is delaying disbursements as far as possible.

The evidence sits in the gap between the two verbs. US HIV appropriations were flat across 2023, 2024 and 2025, yet disbursements fell 31% last year. Our AIDS mortality modelling is too simple to accommodate such shenanigans by factoring in actual aid flows.

A further twist is the America First Global Health Strategy, issued by the State Department last September, in which the US is seeking bilateral agreements with countries.

Ageing multilateralists like me should, in principle, be dismayed by such blatant subjugation of the collective global good to any country’s national interest.

But before automatically condemning the new US approach, is it not worth keeping an open mind on whether this is good or bad for long-term HIV control?

Time will tell. The 34 bilateral compacts signed since December 2025 commit some $24 billion to 2030, of which around $14 billion is from the US and close to $10 billion is required from recipients. 

Sovereignty-sensitive people will not like aid conditionalities. But this is not new, even in multilateral approaches where the World Bank is the master. And if this incentivises stronger national ownership and sustainability, it is a useful counter to the dependency that traditional aid has created.

Conditionality concerns

There are also concerns about sharing data and biomaterials, or about leveraging health assistance in business bargaining.  

A Kenyan court suspended implementation days after the country’s leadership signed an MOU with the US. Zambia and Zimbabwe backed away, the former having found its health agreement entangled with American access to copper and cobalt.

More serious for public health, the agreement texts examined so far are largely silent on “key populations”. These are people most at risk of HIV, including men who have sex with men, sex workers, transgender people, and people who inject drugs. 

Where preventive medication (PrEP) appears at all, it is as a commodity line item, with no commitment to reach the people most exposed.

Key populations are essential because this is where HIV/AIDS is increasingly concentrated, as we embark on the remaining journey toward an “AIDS-free world”. 

The critical question is whether the US will allow counterpart recipient funds to target key groups as part of jointly funded national programmes, even if American funding may not be used for them for ideological reasons.

Such twists and turns are difficult to include in current modelling, which does not reveal the weightings given to multiple factors at play. But new AI capabilities make that possible, whenever there is a resurgence of evidential rigour over convenient advocacy.

The harms donors are not causing

The consequences of donor leverage are one thing, but domestic policies have their own impacts. In 2026, 168 countries criminalised sex work, 152 the possession of small amounts of drugs, and 66 same-sex relations. 

For the first time since UNAIDS began tracking these trends, criminalisation of marginalised populations increased.

The Sahel led the way. Burkina Faso and Niger newly criminalised same-sex activity, and Senegal raised its penalties this year. Visits to Senegal’s treatment centres then fell by a quarter between January and February, as patients stayed away for fear of arrest.

This drives the condition underground. It would show up as falling rates in the data inputs of conventional models, which would then underestimate mortality. Meanwhile, no country can end AIDS while criminalising the people most at risk of it.

Aid cuts did not write those laws. That raises the question of whether foreign aid should be used to compensate for irresponsible domestic policies.

Nor did aid cuts set the price of the drug that could end HIV transmission. This was a lively topic in Rio. Two injections a year of Lenacapavir prevented every infection among women in its South African and Ugandan trial and cut infections by 96 % in a second trial. With 22 % of people living with HIV still not on treatment, this is game-changing.

But it carries a United States list price of $28,218 a year as a patented product, while analysts estimate it could be made generically for $25 to $40.

Pending generic availability, the manufacturer has offered no-profit supply through the US President’s Emergency Plan for AIDS Relief (PEPFAR) and the Global Fund across the 120 countries of its voluntary licence. This is enough for up to two million people over three years. 

Compared to the past, this is a lightning-speed transition from costly patented to more affordable generic medicines.

However, 17 middle-income countries, including Brazil, sit outside that licence and account for 19% of new HIV cases. This exposes an uncomfortable truth: solidarity in global health does not come free. Even among developing countries, the richer ones need to carry a bigger financial burden for the sake of the really poor.  Will they?

How this prevention breakthrough is integrated into mortality modelling is unclear and depends on the currently unknown rate of scale-up.

One detail is informative of the state of the world. Americans will benefit enormously and should thank South Africans for the trials that proved the drug. Yet Washington began a phased drawdown of its HIV programme in South Africa in June, citing the country’s failure to meet its policy demands.

A legacy worth leaving

Tracking HIV/AIDS through mathematical modelling using outdated assumptions is increasingly questionable. Especially in an era of rapid policy and pharmaceutical innovations, tightening resources, and geopolitical and social shifts that are not always benign. 

This is not a sound basis for the smart national and global strategies necessary to achieve the AIDS-free shared goal.

There is something indictable here. Thirty years into the pandemic, we still base many of our actions on inferences, deductions and sometimes, frankly, guesswork. Despite expending tens of billions of dollars, including creating two dedicated international bodies – UNAIDS and the Global Fund to Fight AIDS, Tuberculosis and Malaria – and an extensive ecosystem of national bodies and numerous NGOs.

UNAIDS is meanwhile contemplating its own end. The UN80 review proposed closing it by the end of this year. UNAIDS has countered with a phased plan, and its board expects recommendations in October. 

Whatever is decided, it should consider its legacy. That legacy cannot be advocacy or therapeutic advances, because those are mostly the push of courageous people who have themselves endured HIV and those who work directly with them. 

As a Joint Programme of the biggest and most influential United Nations agencies, UNAIDS should leave behind something more systematic and tangible. How about a robust global system for measuring – not estimating – actual AIDS-related mortality?

 

Mukesh Kapila is professor emeritus of global health and humanitarian affairs at the University of Manchester, and a board member of Health Policy Watch. Vital Signs in Global Health is a regular column, and the opinions expressed are solely those of the author.  

Image Credits: Felipe Varanda/ IAS, Peter Tatchwell Foundation.