Someone finally did the study.
A paper just published in JAMA Health Forum said something many of us who work with Global Burden of Disease (GBD) data have quietly wondered: how stable are these estimates, really?
Zavalis et al. (2026) compared GBD risk factor estimates across 8 iterations from 2010 to 2023 and found substantial instability (particularly for behavioral and dietary risks). Half of the death estimates had a coefficient of variation exceeding 0.2, and for some dietary risks, 70–96% of GBD 2021 point estimates fell outside the GBD 2019 uncertainty intervals. That's not noise, that's a signal worth taking seriously.
As someone who has published multiple Quality of Care Index (QCI) papers using GBD data (covering oral disorders, lip and oral cavity cancer, and orofacial clefts), I think this paper deserves an honest read from anyone in our field.
To be clear: GBD remains one of the most ambitious and valuable epidemiological projects ever undertaken. The team should be credited for continuously updating and improving their methods, and for making the data publicly available in a way that enables studies like this one.
But this analysis is a useful reminder that GBD-derived indices, including QCI, reflect the data and methodology of a specific iteration. When the underlying estimates shift substantially between versions, our conclusions may shift too, even without a single new patient being diagnosed or treated.
What does this mean in practice?
— Interpret QCI trends with appropriate caution
— Be explicit about which GBD iteration you used
— Treat uncertainty intervals as a floor, not a ceiling
— Consider sensitivity analyses across iterations where feasible
I'm glad someone finally did this study systematically. It doesn't invalidate GBD-based research, but it sharpens the conversation about what we can and cannot confidently conclude from it.
jamanetwork.com/journals/jama-…
English
