Oxford/AstraZeneca COVID-19 vaccine trial
Two valid lab methods counted the same batch differently; around 1,300 participants received a lower first dose.
Voysey et al., The Lancet, 2021 (opens in a new tab)Clean Data isn't always Comparable Data
From the literature
Vitamin D · Germany, 1998–2011
Estimated deficiency moved from above 20% to below 20% once samples were standardised.
Rabenberg et al., BMC Public Health, 2018 (opens in a new tab)Vitamin D · USA, 1988–2006
An apparent decline across four national survey periods disappeared after standardisation.
Schleicher et al., Am J Clin Nutr, 2016 (opens in a new tab)Creatinine calibration
Assay-calibration bias could shift estimated kidney function by 10–20% between survey cycles.
Selvin et al., Am J Kidney Dis, 2007 (opens in a new tab)ICD-10 coding switch · USA, 2015
About one in six detailed diagnostic categories shifted by 20% or more overnight.
Ellis et al., JAMA Network Open, 2020 (opens in a new tab)Query burden in late-stage trials
An average trial sends 96,980 queries back to sites, yet only 1.7% of values change as a result.
Stokman et al., J Soc Clin Data Manag, 2021 (opens in a new tab)FDA & EMA joint AI principles
Provenance must be traceable, and monitoring is required because data drifts.
FDA / EMA, January 2026 (opens in a new tab)Watch what we do
Request access
New here? Request access first. Once we approve your email, enter it here to watch.
News & events
Who we are
Join us
No open roles right now. But if you're interested in what we do and have something to offer, send us your CV anyway at ayeshasanober@
Curious? So are we.
Request sent
Thank you. We'll reply once we've read it.