# Frozen analysis protocol ## Project **Working title:** Does NHANES-Defined Underhydration Predict Long-Term Mortality? A Replication and Extended Follow-up of U.S. Adults Aged 51–70 Years **Protocol version:** 1.0 **Frozen:** 2026-09-29 **Status at freeze:** No 2019-linked outcome analysis for the target cohort has been run or inspected in this project. Earlier exploratory work had examined spot urine osmolality alone in NHANES 2011–2012 using the 2015 mortality vintage; those results are considered prior exploratory knowledge and are not a test of the prespecified composite exposure below. ## Objective Replicate the published 2009–2012 NHANES analysis of the association between a prespecified hydration classification and mortality through 2015, then test whether the same baseline classification predicts mortality with follow-up extended through 2019. The target of inference is **prognostic association**, not the causal effect of drinking water. ## Source study to replicate Stookey JD, Kavouras SA, Suh H-G, Lang F. *Underhydration Is Associated with Obesity, Chronic Diseases, and Death Within 3 to 6 Years in the U.S. Population Aged 51–70 Years.* Nutrients. 2020;12(4):905. doi:10.3390/nu12040905. Published benchmarks used only for the reproduction gate: - analytic N = 1,200 - 319 met all euhydration criteria - 881 failed at least one hydration criterion - 22 were hyponatremic - 859 were classified as underhydrated - 52 all-cause deaths through 2015 - 33 chronic-disease-associated deaths through 2015 - adjusted weighted HR, underhydration vs euhydration: - all-cause mortality: 2.33 (95% CI 0.97–5.60) - chronic-disease-associated mortality: 4.21 (95% CI 1.29–13.78) ## Data Baseline: - NHANES 2009–2010 - NHANES 2011–2012 Mortality: - 2015-vintage NCHS Public-Use Linked Mortality Files for the reproduction gate - 2019-vintage NCHS Public-Use Linked Mortality Files for the prospective extension Records are joined by SEQN. ## Target population and reproduction sample Attempt to reproduce the source study sample exactly before any 2019 result is examined. Core restrictions reported by the source study: - age 51–70 years - non-acutely ill baseline population matching the prior NHANES hydration study - valid information required by that source population, including water intake, body weight and serum sodium - serum creatinine 0.5–1.5 mg/dL inclusive - UACR <30 mg/g - at least 8 hours fasting - complete information required for the source study's chronic-condition classification - eligible linked mortality status Any ambiguity encountered while reproducing the original sample must be documented. No ambiguous rule may be chosen by looking at the 2019 mortality association. ## Exposure classification Use fasting serum sodium, first spot urine volume and spot urine osmolality. **Euhydrated** All of: - serum sodium >=135 and <145 mmol/L - spot urine volume >=50 mL - spot urine osmolality <500 mOsm/kg **Hyponatremic** - serum sodium <135 mmol/L **Underhydrated** Does not meet all euhydration criteria, is not hyponatremic, and has at least one of: - serum sodium >=145 mmol/L - spot urine volume <50 mL - spot urine osmolality >=500 mOsm/kg Participants unable to void are to be handled as in the source study, i.e. with the low-urine-volume category, if the underlying NHANES coding permits faithful reconstruction. Primary comparison: **underhydrated vs euhydrated**. Hyponatremic participants are excluded from this two-category primary contrast. ## Primary outcome All-cause mortality through 31 December 2019. Rationale: in the 2019 public-use LMF, final vital status is not altered for disclosure protection, whereas follow-up time and underlying cause of death can be replaced with synthetic values for some records. ## Secondary outcome Chronic-disease-associated mortality, reconstructed as closely as public-use variables permit from: - heart disease - malignant neoplasm - diabetes flag - hypertension flag This is secondary because public-use cause-of-death information may be synthetic for some records. ## Survey design For pooled 2009–2010 and 2011–2012: - fasting subsample weight = 0.5 * WTSAF2YR - use SDMVSTRA - use SDMVPSU ## Primary model Survey-weighted Cox proportional hazards regression with time-on-study scale and underhydration vs euhydration as the exposure. Adjustment set follows the source study's Model 2: - age - sex - height - race/ethnicity - education - poverty level - physical activity - total energy intake - carbohydrate intake relative to total energy - alcohol category: 0, 1–2, >=3 drinks/day - dietary solute-load index: protein(g)*5.8 + mEq(sodium + potassium) - past/current cigarette smoking - prescription medication use - season of MEC examination - fasting duration No variable may be added to or removed from the primary model based on the observed 2019 exposure coefficient. ## Reproduction gate Before interpreting the 2019 extension: 1. Recreate baseline sample and hydration groups using the 2015-vintage files. 2. Compare N, hydration counts, death counts and weighted adjusted HRs with the published benchmarks. 3. If material disagreement remains, stop and resolve the discrepancy before treating the 2019 result as a replication. A small mismatch caused by documented mortality-vintage linkage revisions may be reported, but the source of the mismatch must be identified rather than silently accepted. ## Prespecified 2019 analyses ### A. Primary extension Same baseline cohort, same exposure definition and same primary adjustment set; mortality follow-up through 2019. ### B. Minimal-adjustment model Age + sex, for comparison with the fully adjusted model. ### C. Late-death falsification analysis Among participants classified alive in the 2015-vintage LMF, test whether baseline underhydration predicts becoming deceased in the 2019-vintage LMF. This analysis uses the difference between mortality vintages rather than attempting to reconstruct an exact 1 January 2016 landmark time from public-use follow-up months. ### D. Component decomposition Estimate associations separately for: - serum sodium >=145 mmol/L - urine volume <50 mL - Uosm >=500 mOsm/kg The decomposition is explanatory/sensitivity analysis, not a replacement primary endpoint. ### E. Renal-function sensitivity Repeat after excluding or adjusting for indicators that can alter urine concentrating ability, where available without breaking the reproduction sample: - eGFR <60 mL/min/1.73m² - albuminuria - diuretic use Because the source reproduction sample already excludes UACR >=30 mg/g and extreme creatinine, these checks are secondary. ### F. Uosm sensitivity Model spot Uosm as a continuous exposure using a prespecified flexible form (restricted cubic spline), with the same warning that spot Uosm is not equivalent to habitual water intake. ### G. Cycle consistency Report 2009–2010 and 2011–2012 estimates separately when event counts are sufficient for interpretable estimates. ## Diagnostic checks - proportional-hazards diagnostic - event counts by exposure group - effective sample size / sparse-cell warnings - missingness by variable - weight distribution - exposure prevalence by cycle - compare 2015 and 2019 vital status for the same SEQN - identify cases whose 2015/2019 linkage classification differs ## Interpretation rules frozen before 2019 analysis The result will be considered supportive of persistent prognostic association if: - the 2019 HR remains materially above 1, - uncertainty is narrower than in the original short follow-up, - direction is not explained by one exposure component alone, - and the late-death analysis is directionally consistent. The hypothesis will be considered weakened or falsified if: - the extended estimate moves near 1 or reverses, - the association is concentrated only in early deaths, - it is driven almost entirely by one unstable component, - or reasonable renal-function / linkage sensitivities remove it. Statistical significance alone is not the decision rule. ## Causal language Permitted: - associated with - predicted - prognostic association - persisted / attenuated / did not replicate Not permitted: - underhydration caused mortality - drinking more water would reduce mortality - the analysis identifies an optimal water intake ## Reporting Primary guideline: STROBE. ## Change control After this protocol is frozen, any deviation must be logged with: - date - exact change - reason - whether the 2019 outcome result was already known at the time of the change Exploratory analyses added after outcome inspection must be labelled exploratory.