Back

Where the penicillin gap for rheumatic heart disease is widest: a country triage, a single supply chokepoint, and the case for not suspending prophylaxis

Jun 26, 2026 · 8:30 AM

Abstract

Rheumatic heart disease (RHD) kills more than 373,000 people a year, almost all in low- and middle-income countries, and the prophylaxis that prevents it is one injection of benzathine penicillin G (BPG) every three to four weeks at a cost of cents per dose. We asked where a marginal dollar against RHD buys the most averted death, built a 20-country burden-coverage-supply scorecard, mapped the BPG supply chain, modeled the deaths and dollars of closing the gap, and wrote the dossier a ministry needs to keep injecting. The naive expectation is that coverage and stockout data point to the worst-hit countries. They do the opposite: the four largest-burden countries (India, China, Pakistan, Indonesia) have no national register and no documented stockout, while the only citable coverage numbers come from low-burden settings. Closing the gap in five high-burden countries alone could avert roughly 234,000 deaths a year for a drug bill on the order of $18M to $168M. We recommend funding the API supply neck and the missing registers, in that order.

Background

RHD begins as untreated childhood strep throat that triggers an autoimmune attack on the heart valves. It affects over 40 million people and causes more than 300,000 deaths a year, almost entirely in poorer countries, and it kills the young: one-year mortality reaches 17% among children and adolescents in endemic regions [1]. The cure is old and off-patent. The 2021 NEJM GOAL trial gave the first randomized proof that BPG works in latent disease: among 818 children, progression occurred in 0.8% on penicillin versus 8.3% with no treatment over two years, a tenfold reduction [2]. WHO put RHD on the agenda with resolution WHA71.14 in 2018 and issued new guidelines in 2024 [3]. The precedent that the problem is fixable is the rich world itself, which eliminated RHD decades ago with the same molecule. What does not exist is the named, sourced map that tells a funder or a ministry exactly where prophylaxis fails and why. That map is the artifact here.

Method

Four stages, each pure desk synthesis over published data, each verified by an independent evaluator with fresh eyes against quoted done-criteria. M1 cross-references IHME Global Burden of Disease (GBD) prevalence and mortality, national-register and survey coverage figures, and documented BPG stockouts into a ranked 20-country scorecard [4]. M2 maps the supply chain from active-pharmaceutical-ingredient (API) makers down to finished-dose products and lays the shortage events on a dated timeline [5]. M3 applies the GOAL relative risk reduction (derived explicitly as 90.4% from 0.8% versus 8.3%) to GBD burden at a stated $/dose to estimate avertable deaths and drug cost per country, with a low/base/high sensitivity band [6]. M4 sets the rare fatal injection reaction from the 2022 AHA advisory against that benefit, with number-needed-to-treat against number-needed-to-harm [7]. All four milestones passed on the first attempt. The evaluator spot-checked citations against source text in each pass (for example, confirming the WHO survey's 39 of 95 countries, the GOAL 0.8% versus 8.3%, and the Birru meta-analysis 6 fatal of 11,587). Artifacts and verdicts are linked throughout and listed under Provenance.

Findings

1. The data are brightest where the burden is dimmest, and dark where it is worst.

The four largest-burden countries have no national register and no representative coverage figure. India holds about 10.5 million prevalent cases and roughly 166,000 deaths a year (GBD 2021); China, Pakistan, and Indonesia round out the top four [4]. None of them carries a documented BPG stockout in the sources located, and none has a national adherence number. Meanwhile the only citable coverage figures come from settings where the absolute burden is far smaller. The map is blank exactly where a funder most needs it. The "none documented" stockout cells for India, China, and Pakistan are a statement about the evidence base, not about supply security: absence of a published shortage report is not proof of a secure supply.

2. Even the best-resourced registers report failure.

Setting Coverage (≥80% of scheduled doses) Source year
Uganda (retained in care) ~91% adherent 2017 [4]
Ethiopia (Jimma clinic zone) ~63% adherent 2019 [4]
Uganda (Mulago cohort) ~58% adherent earlier [4]
Australia, First Nations (AIHW) ~32% received 80–100% of doses 2024 [4]
Fiji (screen-detected cohort) ~7% adequate adherence; 41% had no injections recorded 2016 [4]

These numbers are not directly comparable: they use different denominators and definitions, and Uganda's ~91% is conditioned on retention in care, so the binding constraint there is enrollment, not adherence. The point stands that a register is necessary but nowhere near sufficient. Australia, a high-income country, reaches only about 32% of First Nations patients on the dose threshold.

3. The whole world's BPG rests on three plants in China.

The peer-reviewed CHAI analysis states that by 2016 only three Chinese API manufacturers remained, supplying the global market, with production concentrated in two regions of China [5]. The finished-dose layer looks broader (CHAI counted at least 30 finished-product makers, and at least seven branded products reach market), but every one of those products is reconstituted from API that traces to those three plants. The chain is an hourglass with a thin neck. It has failed at both ends within the decade: upstream in the 2014–2016 shortage (39 of 95 responding countries reported a BPG stockout) and downstream in 2023–2025, when a single US supplier (Pfizer's Bicillin L-A) recalled lots in July 2025 and tipped the country into shortage during a syphilis epidemic [4][5]. CHAI estimated global demand at 74–100 million doses against a need of about 200 million for RHD alone [5].

4. The drug is so cheap that the lives, not the budget, are the constraint.

Quantity Value Basis
Annual deaths, 5 counted countries (India, China, Pakistan, Bangladesh, Nepal) 287,500 GBD via M1 [6]
Deaths avertable/yr (base case) ~234,000 287,500 × 0.90 × 0.904 [6]
Annual doses to close the gap (4 with prevalence counts) ~168M prevalence × 0.90 × 13 [6]
Annual drug cost @ $0.11 / $0.30 / $1.00 per dose $18.5M / $50.4M / $168M [6]
Drug-only cost per death averted ~$215 to ~$718 [6]

Globally, applying the same base formula to the GBD 2021 anchor (373,345 deaths) gives roughly 304,000 avertable deaths a year (sensitivity range ~118,000 to 321,000), against a global RHD drug bill of tens of millions to about $200M a year on CHAI's 200-million-dose need [6]. Even the worst corner of the sensitivity space (high price, low efficacy) costs about $1,700 per death averted, cheap by any global-health benchmark.

5. The safety scare does not justify suspending prophylaxis.

The 2022 AHA presidential advisory is narrowly about patients with severe or advanced valvular RHD, not penicillin in general [7]. It reframed the rare fatal reactions as cardiovascular collapse in already-failing hearts rather than anaphylaxis, and it recommended continuing intramuscular BPG for the low-risk majority while routing the highest-risk minority to oral penicillin. The numbers: number-needed-to-treat is about 13 patients to prevent one progression event (GOAL); number-needed-to-harm is about 2,000 patients, or about 32,000 injections, to cause one fatal reaction, and that reaction is almost always confined to severe-RHD patients who can be flagged and rerouted [7]. The documented fatal harm in the entire pooled literature is single digits across about 155,000 injections (6 fatal of 11,587 patients, all in severe RHD) [7].

Recommendations

Item Executor Cost Anchor
Volume guarantee to keep the 3 Chinese API lines at capacity, or qualify one non-Chinese API line CHAI / Gates / WHO not separately costed here 3 API plants, two Chinese regions [5]
Build national RHD registers in India, China, Pakistan, Indonesia Health ministries drug is $18M–$168M/yr for 5-country gap; register/delivery cost not included no register in the top-4 burden countries [4]
Issue the benefit-harm dossier so programs keep injecting under the AHA advisory Reach / World Heart Federation / ministries essentially free NNH/NNT ≈ 150 [7]
Fund supervised-administration and oral-route stratification for severe RHD Programs low-cost steps AHA 2022 advisory [7]

Sequence the supply neck first, because no register fixes a stockout and the failure mode is shared by all 30-plus finished-dose makers. Build the registers second, because the model is silent exactly where the burden is largest and a funder cannot target what is unmeasured. Run the dossier in parallel at near-zero cost, because a ministry that suspends a program over six documented deaths forfeits a benefit measured in the hundreds of thousands. The drug budget is a rounding error; the binding constraints are supply security and delivery.

Limitations

  1. The headline death figures rest on an extrapolation a critic can reject. GOAL measured echocardiographic progression of latent disease over two years, not mortality, and not in established symptomatic RHD. Applying its 90.4% relative risk reduction to established-RHD deaths crosses both outcome and population; a published critique of exactly this transfer exists, and the low-sensitivity case halves the efficacy to show how soft the number is [6].
  2. GBD burden is modeled, not measured, with wide uncertainty intervals (India ~8.2M–13.2M prevalent), and several high-burden countries (Indonesia, DRC, Nigeria, Tanzania) had no extractable absolute count, so the 234,000 figure is a floor over five countries, not a global total [4][6].
  3. Coverage figures use different denominators and definitions and are not directly comparable; they describe register cohorts, not the share of the GBD-prevalent population protected [4].
  4. The three API manufacturers are counted authoritatively by CHAI but named only inferentially from a 2017 investigation and trade databases; whether a fourth non-Chinese source (Sandoz) sells open-market API is unresolved [5].
  5. The cost figures are drug-only. Delivery, registers, cold chain, and supervised injection dwarf the molecule cost and are excluded; $60M buys the drug, not the program [6].
  6. One citation error survives in M1: a Northern Territory (Australia) register study is mislabeled as a New Zealand study in the New Zealand row. The cell remains non-blank and is also backed by a genuine New Zealand source, so no finding depends on the mislabel [4].

References

  1. Rheumatic heart disease burden and one-year mortality. https://pmc.ncbi.nlm.nih.gov/articles/PMC12925792/ ; https://www.acc.org/latest-in-cardiology/journal-scans/2021/11/14/00/07/secondary-antibiotic-prophylaxis-aha-2021
  2. Beaton A, et al. Secondary Antibiotic Prophylaxis for Latent Rheumatic Heart Disease (GOAL). NEJM 2021;385:2211–2221. https://www.nejm.org/doi/full/10.1056/NEJMoa2102074 ; plain-language figures https://scienceblog.cincinnatichildrens.org/in-uganda-rheumatic-heart-disease-research-reaches-its-goal/
  3. WHO, rheumatic heart disease (WHA71.14, 2024 guidelines). https://www.who.int/news-room/fact-sheets/detail/rheumatic-heart-disease
  4. M1 scorecard. artifacts/2026-06-25-m1-coverage-stockout-scorecard.md . Key sources: GBD 2021 South Asia https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11212786/ ; WHO 2014–2016 shortage survey https://pmc.ncbi.nlm.nih.gov/articles/PMC5744908/ ; AIHW 2024 https://www.aihw.gov.au/reports/indigenous-australians/arf-rhd/contents/secondary-prophylaxis/delivery-of-bpg-to-first-nations-people ; Fiji https://pubmed.ncbi.nlm.nih.gov/27730711/
  5. M2 supply-chain map. artifacts/2026-06-26-m2-bpg-supply-chain-map.md . Anchor: CHAI/Oxford, International Health 2024;16(3):279. https://pmc.ncbi.nlm.nih.gov/articles/PMC10987389/
  6. M3 cost-of-inaction model. artifacts/2026-06-26-m3-cost-of-inaction-model.md . Latent-to-established critique https://pmc.ncbi.nlm.nih.gov/articles/PMC10593280/
  7. M4 benefit-harm dossier. artifacts/2026-06-26-m4-benefit-harm-dossier.md . AHA advisory: Sanyahumbi A, et al. JAHA 2022;11(5):e024517. https://www.ahajournals.org/doi/10.1161/JAHA.121.024517 ; harm incidence https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12057857/ ; https://pubmed.ncbi.nlm.nih.gov/1674296/

Provenance

This report was produced by an autonomous research loop that ran four milestones, each generated and then checked by an independent evaluator against quoted done-criteria with citation spot-checks; all four passed on the first attempt. The artifacts and the four PASS verdicts live in artifacts/ and verdicts/ in this problem directory.