Here is the specific, well-defined problem: there is no public, standardized field anywhere in WHO's Global Antimicrobial Resistance and Use Surveillance System (GLASS) that distinguishes a country submitting nationally representative, accredited-lab resistance data from a country submitting resistance percentages from two urban hospitals.

Both count identically as "reporting."

As of the 2025 WHO report, 130 countries are enrolled and 104 reported 2023 data -- a genuinely large jump from 25 countries in 2016 -- but that headline number is a participation count, not a data-quality count, and it is being treated as the latter by donors, journalists, and national governments citing their own "improved surveillance" status.

Why this matters beyond pedantry: policy and funding decisions get built on top of GLASS resistance percentages.

A country reporting a high resistance rate from one referral hospital's ICU (where resistant infections concentrate because that's where the sickest patients end up) looks, on paper, identical to a country reporting a low rate from a representative, population-weighted sentinel network.

Donor funding for stewardship programs, WHO's own priority-pathogen lists, and pharmaceutical R&D incentive design (which increasingly references GLASS-adjacent resistance trend data) all inherit this ambiguity.

The IHME/Lancet 2019 global burden study had to build an entirely separate modeling pipeline -- 471 million isolate records, hospital and vital-registration reconstruction -- specifically because GLASS coverage in the highest-burden regions (Sub-Saharan Africa, South Asia) was too sparse and too non-representative to use directly.

That's a tell: the people doing the most careful burden estimation work don't trust GLASS's raw numbers enough to use them without a completely independent cross-check.

Root causes, briefly: GLASS was designed as a participation-building tool first, aimed at getting countries with zero surveillance infrastructure to start submitting anything, which was and is a reasonable sequencing choice.

But the metadata needed to later distinguish good data from thin data -- lab accreditation status (ISO 15189 or equivalent), sampling frame (population-representative sentinel sites vs. convenience samples from referral hospitals), and testing volume relative to population -- was never made a mandatory, public field.

Countries have no strong incentive to volunteer information that would make their own data look thinner, and WHO has limited leverage to demand it without risking participation itself (the same trade-off health data systems hit constantly: ask for more rigor and you get less coverage).

27 million people who die annually from drug-resistant infections directly, concentrated overwhelmingly in the regions where this exact data-quality ambiguity is worst -- because those are the same regions where sparse, non-representative sampling is most common.

Fixing the metadata gap doesn't cure a single infection, but it would let anyone -- WHO, a donor, a national ministry -- actually tell the difference between "surveillance is improving here" and "one hospital started filling out a form here," which is a precondition for targeting scarce stewardship and lab-capacity funding at the places that need it rather than the places that already reported.