A Likely Double Count in a Published Water–Mortality Study

Independent computational audit of source-water classification in Zhou et al. (2022), based on US NHANES 1999–2014 public-use dietary files. Beginning in 2005–2006, NHANES includes plain drinking water in the total dietary moisture variable. Literal reconstruction of the source-water equations described in the paper risks adding that plain water again, including it within both “food water” and total water. On a matched reconstruction sample (N=35,163), median food-water estimates are 949 g/day under the literal calculation and 556 g/day after subtracting plain water once; the published study reports 924 g/day. In exploratory models linking to mortality through 2019 (N=31,046, deaths=5,898), the Q4-vs-Q1 food-water HR changes from 0.910 (95% CI 0.805–1.029) to 0.979 (0.883–1.085). This is not an exact replication of the published 2015 mortality follow-up or the original covariate/survey-weighted models, nor direct evidence of the authors' source code. The deposit includes a technical report, independent XPT-to-derived-data scripts, participant-level derived public-use analytical data, full archived model-specification ledger, plain-water code validation tables, and reproducibility instructions. Remaining discrepancies include 94 person-days in the 2009–2010 plain-water code validation and a different final sample from the original paper. The first public description appeared at https://nullinstitute.org/likely-double-count-water-mortality-study/