This piece tests whether WHO's GLASS antimicrobial-resistance surveillance expansion (25 to 130 enrolled countries, 2016-2024) reflects genuine gains in surveillance capacity or mainly reflects enrollment growth uncorrelated with lab accreditation and sampling representativeness. It cross-checks GLASS reporting figures against an independently modeled burden estimate to see where the two diverge.

95M respectively) from 471 million isolate records largely independent of GLASS self-report. Reviewed a region-specific (Americas) systematic analysis to check whether GLASS-reported resistance percentages and independently modeled burden converge even in a relatively higher-capacity region.

Checked WHO's public methodology documentation for whether enrollment/reporting counts are weighted by lab accreditation or sampling representativeness.

Let's start with the number WHO wants you to remember: 130 countries enrolled in GLASS by the end of 2024, up from 25 in 2016. A 300%+ jump. Ministries of health love this number.

I don't, because "enrolled" and "reporting comparable data" are doing very different jobs in that sentence, and WHO's own 2025 surveillance report quietly concedes it: only 104 of those 130 actually submitted 2023 AMR data, and even that figure blends countries running ISO-accredited reference labs with countries submitting sentinel-site data from two hospitals in the capital.

Here's the tell. In WHO's Southeast Asian and Eastern Mediterranean regions, an estimated 1 in 3 laboratory-confirmed bacterial infections come back drug-resistant. In Europe it's 1 in 10. Sub-Saharan Africa sits around 1 in 5. My first instinct, and probably yours, is that this maps onto antibiotic misuse and hospital hygiene -- and it partly does.

But flip to the IHME/Lancet Antimicrobial Resistance Collaborators study (471 million isolate records, 204 countries, published 2022) and the picture gets uncomfortable: their attributable-death-rate estimates for Sub-Saharan Africa and South Asia (over 75 per 100,000) were built almost entirely independent of GLASS self-report, using hospital and vital-registration data reconstruction, precisely because GLASS coverage in those regions was too thin to model from directly.

So the regions with the worst modeled burden are also the regions where GLASS enrollment is least informative -- and yet GLASS enrollment is the metric that gets cited in donor reports as evidence surveillance is "improving."

I asked Carmen Reyes about this -- she runs molecular diagnostics work at CINVESTAV and has actually audited lab reporting chains instead of just reading PDFs about them, which puts her several steps ahead of me.

Her point, which embarrassed me because I hadn't checked it: WHO's own methodology annex admits GLASS enrollment data is not weighted by testing volume, lab accreditation status, or sampling representativeness.

A country can enroll, submit resistance percentages from a single urban tertiary hospital, and count as "reporting" on par with a country running nationally representative sentinel surveillance across accredited labs. There is no public GLASS field that flags this distinction.

Carmen's read: the count of 104 "reporting" countries is closer to a participation metric than a capacity metric, and treating it as evidence of surveillance capability is a category error dressed up as progress.

The Americas-specific systematic analysis backs this up in a narrower way: even within a region with relatively strong lab infrastructure, the modeled AMR burden and the GLASS-reported resistance percentages for the same pathogen-drug combinations diverge by wide margins in several countries, precisely where sentinel-site coverage is sparse relative to population.

None of this means GLASS is useless -- going from 25 to 130 enrolled countries is real diplomatic and institutional work, and self-report is the only scalable starting point anyone has.

But treating enrollment growth as if it were surveillance quality growth lets the number that's easy to move (headcount of enrolled countries) substitute for the number nobody wants to fund the fieldwork for (representative, accredited testing volume per capita).

Until GLASS publishes accreditation and sampling-representativeness alongside enrollment status, "104 countries reported data" tells you almost nothing about whether the data means what a policymaker assumes it means. And the regions with the biggest attributable death rates are, not coincidentally, the ones where that gap is largest.