Underhydration Is Associated with Obesity, Chronic Diseases, and Death Within 3 to 6 Years in the U.S. Population Aged 51-70 Years.

Nationally representative data from the National Health and Nutrition Examination Survey (NHANES) indicate that over 65% of adults aged 51-70 years in the U.S. do not meet hydration criteria. They have hyponatremia (serum sodium < 135 mmol/L) and/or underhydration (serum sodium >145 mmol/L, spot urine volume <50 mL, and/or spot urine osmolality ≥500 mmol/kg). To explore potential public health implications of not meeting hydration criteria, data from the NHANES 2009-2012 and National Center for Health Statistics Linked Mortality Files for fasting adults aged 51-70 years (sample n = 1200) were used to determine if hyponatremia and/or underhydration were cross-sectionally associated with chronic health conditions and/or longitudinally associated with chronic disease mortality. Underhydration accounted for 97% of the population group not meeting hydration criteria. In weighted multivariable adjusted Poisson models, underhydration was significantly associated with increased prevalence of obesity, high waist circumference, insulin resistance, diabetes, low HDL, hypertension, and metabolic syndrome. Over 3-6 years of follow-up, 33 chronic disease deaths occurred in the sample, representing an estimated 1,084,144 deaths in the U.S. Alongside chronic health conditions, underhydration was a risk factor for an estimated 863,305 deaths. Independent of the chronic health conditions evaluated, underhydration was a risk factor for 128,107 deaths. In weighted multivariable Cox models, underhydration was associated with 4.21 times greater chronic disease mortality (95% CI: 1.29-13.78, p = 0.019). Zero chronic disease deaths were observed for people who met the hydration criteria and did not already have a chronic condition in 2009-2012. Further work should consider effects of underhydration on population health.


Introduction
Data from the 2009-2012 National Health and Nutrition Examination Survey (NHANES) indicate that the majority of adults ages 51-70 years in the U.S. do not meet hydration criteria [1]. An estimated 72.1% (68.4%-75.5%) of men and 65.9% (61.0%-70.6%) of women who are free of acute illness have a serum sodium below 135 mmol/L, a serum sodium of 145 mmol/L or higher, a spot urine volume less than 50 mL, or a urine osmolality of 500 mmol/kg or higher. Individuals who do not meet these hydration criteria have hyponatremia and/or a condition described as "underhydration" by Kavouras [2], which is characterized by elevated vasopressin and renal conservation of water despite normal plasma osmolality. The 2009-2012 NHANES analysis [1] has implications for national water intake recommendations, if hyponatremia and/or underhydration signal risk for poor health outcomes. years who had valid information about water intake, body weight, and serum sodium. The present study population excluded individuals with extreme serum creatinine (<0.5 or >1.5 mg/dL; n = 75) or albuminuria (urine albumin/creatinine ratio ≥30 mg/g creatinine; n = 424), for whom hydration status might be misclassified. The study population excluded 1335 individuals who were not 8 h fasted at the time of the 2009-2012 examination, as fasting laboratory test results were used to classify study participants with respect to chronic conditions. Participants missing information about chronic disease conditions (n = 41) and vital status through 2015 (n = 2) were excluded. The analysis sample included 1200 individuals aged 51-70 years.
The NHANES 2009-2010 and 2011-2012 protocols were reviewed and approved by the Centers for Disease Control and Prevention (CDC) National Center for Health Statistics (NCHS) Institutional Review Board (IRB) / Ethics Review Board (ERB) (https://www.cdc.gov/nchs/nhanes/irba98.htm). All study participants provided informed consent. The study complied with the Declaration of Helsinki for human subject involved medical research.

Hydration
A fasting blood and spot urine sample were collected and processed at the mobile examination center (MEC) following protocol described in detail elsewhere [1,48,49]. Serum sodium was determined as part of a DxC800 serum biochemistry profile [50,51]. Urine osmolality was determined by freezing point depression osmometer (Osmette II, Model 5005 Automatic Osmometer, Precision Systems Inc., Natick, MA, USA) [52].
Fasting serum sodium, spot urine volume, and spot urine osmolality were used to classify study participants according to previously defined hydration criteria [1,53,54], which are consistent with zero electrolyte free water clearance [55], proposed cutoffs for urine osmolality [56,57], and possible chronic health benefits [58][59][60]. For urine osmolality, the cutoff of 500 mmol/kg was selected because of its sensitivity to cell swelling, suppressed vasopressin release, and urine dilution. Due to the tight relationship between cell shrinkage and vasopressin release [61], higher urine osmolality cutoffs distinguish cell shrinkage and urine concentration.
Participants who met all criteria were identified as euhydrated. Participants who did not meet any one of the criteria were classified as not meeting the hydration criteria. Participants who were unable to void, a sign of vasopressin activation and action, were classified with participants with urine volume below 50 mL, as not meeting the hydration criteria. Participants that did not meet criteria and did not have hyponatremia were classified as underhydrated.

Chronic Conditions
Measured body weight, height, and waist circumference as well as CDC cutoffs (body mass index (BMI) ≥30; waist > 40 inches for men, waist > 35 inches for women) were used to classify the weight status of participants [62].
Consistent with previous NHANES analyses [1,63], fasting insulin determined by enzyme-linked immunosorbent assay (ELISA), fasting glucose determined by enzymatic method, and the Homeostatic Model Assessment for Insulin Resistance (HOMA-IR) equation (with ≥2.5 cutoff) were used to classify participants with respect to insulin resistance. Fasting glucose and glycosylated hemoglobin (HbA1c) were classified as elevated following American Diabetes Association cutoffs (glucose ≥126 mg/dL; HbA1c ≥6.5%) [64]. HbA1c was determined by a laboratory affiliated with the National Glycohemoglobin Standardization Program using a Tosoh A1C G7 system [64,65]. Participants who had an elevated fasting glucose, elevated HbA1c or an oral glucose tolerance test result at or above 200 mg/dL, or who reported ever being told that they had diabetes were classified as having diabetes.
Participants were classified as having hypertension on the basis of blood pressure measurements taken at the MEC visit and American Heart Association cutoffs (systolic blood pressure (SBP) ≥120, diastolic blood pressure (DBP) ≥80) [68] or ever diagnosis of high blood pressure [69].
National Cholesterol Education Program, Adult Treatment Panel III (NCEP/ATPIII) criteria were used to classify participants with respect to metabolic syndrome (three or more of the following conditions: high waist circumference, elevated triglycerides, blood pressure ≥130/ ≥85 mmHg, low HDL (<40 mg/dL for men and <50 mg/dL for women), and fasting glucose (≥110 mg/dL) [70]. Participants with none of the above conditions were identified (BMI, waist circumference, HOMA-IR, fasting glucose, HbA1c and fasting triglyceride below cutoffs, no diabetes, no hypertension and no metabolic syndrome).

Joint Exposure
Given potential for a feedback system between hydration and metabolic dysregulation, participants were divided into six groups defined by euhydration, hyponatremia, or underhydration and whether or not they had any of the chronic health conditions (yes/no) at baseline. Participants with underhydration AND any of the chronic conditions at baseline were considered further progressed along the feedback system of metabolic dysregulation than participants with underhydration OR any of the chronic conditions at baseline. The latter groups, in turn, were considered further progressed along the feedback system than participants who were euhydrated and did not have any of the chronic conditions in 2009-2012.

Mortality
Survival time was expressed as the number of months elapsed between the MEC examination in 2009-2012 and December 31, 2015. Vital status was expressed as deceased or assumed alive through December 31, 2015. The underlying cause of death was classified as heart disease (I00-I09, I11, I13, I20-I51), malignant neoplasm (C00-C97), or all other causes, on the basis of the 10th International Statistical Classification of Diseases and Related Health Problems (ICD-10) codes. Deaths with hypertension and/or diabetes listed as a contributing cause were flagged. Deaths with heart disease, malignant neoplasms, hypertension, and/or diabetes listed as an underlying cause or multiple cause were grouped.

Covariates
Age in years, sex, and height measured at the MEC (as proxy for body size) were considered as determinants of individual water intake requirements [71]. Socio-economic covariates included self-reported race/ethnicity (non-Hispanic black, Hispanic, white and other/multi-racial), education (less than high school diploma, high school or General Educational Development (GED), college or postgraduate), and poverty (percent of Federal Poverty Level (FPL)). Season of the MEC examination (winter or spring, summer or fall) was considered as proxy for ambient temperature and environmental determinant of both water intake and water intake requirements. The total length of food fast was measured as the number of hours between the last food or drink consumed, other than water, and the venipuncture.
Behavioral covariates included self-reported prescription medication use in the past month [72,73], lifetime smoking experience [74,75], physical activity in the past 7 days, and dietary intake in the 24 h before the morning blood and urine sample collection.
To index physical activity in the past 7 days, a trained interviewer administered a Global Physical Activity Questionnaire (GPAQ) using the computer-assisted personal interviewing (CAPI) system [76,77]. Participants were asked to self-report how much time they spent walking, biking, or Nutrients 2020, 12, 905 6 of 29 doing moderate or vigorous activities at work and/or for recreation on each day in the previous week. The time estimates were converted to metabolic equivalents (MET) minutes per week, assuming a MET score of 4.0 for moderate physical activity and MET score of 8.0 for vigorous activities.
The previous day total intakes of energy, carbohydrate, alcohol, and solute load were estimated from an in-person 24 h dietary recall, administered by a trained interviewer using the United States Department of Agriculture (USDA) Automated Multiple Pass software and portion size reference guides [78,79]. Interviewers probed for snacks and drinks between meals and food preparation details, including addition of salt and sugar during cooking or at the table. Total carbohydrate intake was expressed as a continuous variable relative to total energy intake. Alcohol was expressed in categorical terms, as no drinks, 1-2 drinks, or 3 or more drinks per day. Total daily intakes of protein, sodium, and potassium were combined to index dietary solute load: (gm protein x 5.8) + mEq (Na + K). The calculated index underestimates the total solute load, because solute, such as chloride, were not available in the dataset.

Descriptive Analysis
The bivariate relationship between serum sodium and urine osmolality in the study sample was described by scatterplot. The estimated population-representative distributions of serum sodium, urine osmolality, and urine volume in the study population were described using weighted geometric means and 95% confidence intervals. The estimated population-representative prevalence of euhydration and not meeting the hydration criteria in the study population were described using weighted proportion and 95% confidence intervals.
The study population was described in terms of risk factors that might confound effects of hydration on chronic disease risk, namely, socio-economic, environmental, and behavioral determinants of water intake requirements and water intake. The weighted proportion of the study population with each risk factor was estimated by hydration classification. Unadjusted weighted Poisson models with robust error variances tested for association between each risk factor and hydration classification.

Cross-Sectional Relationship between Chronic Disease and Hydration Classification
Assuming that chronic health conditions may increase risk of not meeting the hydration classification and, conversely, that not meeting the hydration classification may increase risk of having a chronic condition, effects in both directions were estimated.
The weighted prevalence of not meeting the hydration criteria was estimated, stratified by chronic health condition. Unadjusted and multi-variable adjusted weighted Poisson models with robust error variances were used to test if the prevalence ratio of not meeting the hydration criteria in 2009-2012 varied by health condition in 2009-2012. A first multivariable model (Model 2) controlled for age, sex, height, race-ethnicity, education, poverty level, physical activity, total energy intake, total carbohydrate intake, alcoholic drinks consumed, dietary solute load, past or current cigarette smoking, prescription medication use, season of MEC measurement, and length of fast prior to venipuncture. A second multivariable model (Model 3) added control for the other chronic conditions (obesity, insulin resistance, diabetes, dyslipidemia, and hypertension) to check if metabolic dysregulation related to concurrent chronic conditions might mediate or explain the association. To explore for effects specific to underhydration, a set of models with the same specification were fit to test if the prevalence ratio of underhydration varied by chronic health condition.
The weighted prevalence of each chronic condition was next estimated, stratified by hydration classification. Unadjusted and multi-variable adjusted weighted Poisson models with robust error variances were used to test if the prevalence ratio of having each chronic condition in 2009-2012 varied by hydration classification in 2009-2012 (not meeting the hydration criteria vs. euhydration as well as underhydration vs. euhydration). Multivariable models adjusted for covariates, as described above.

Longitudinal Effect of Hydration Classification on Death
The weighted number of deaths occurring to the study population within 3 to 6 years of the MEC examination was estimated, stratified by hydration classification in 2009-2012, health condition in 2009-2012, and all-cause or chronic disease-associated death. Health condition was defined as any or none of the following health conditions: obesity, high waist circumference, insulin resistance, diabetes, elevated glucose, HgA1c or triglycerides, low HDL, hypertension, or metabolic syndrome. All-cause and chronic disease-associated death rates were estimated with jackknife confidence intervals. The stset command available in stata was used for survival analyses with a time-on-study scale [81] and failure defined as 'all-cause death' or 'chronic-disease associated death'. Kaplan-Meier survival curves were graphed by hydration classification as well as by the joint exposure of hydration classification and chronic health condition at baseline.
Weighted Cox proportional hazards models were used to test for differences in the hazard rate by hydration classification. All multivariable models controlled for all covariates described above for Model 2. The aim of this descriptive analysis was to determine if euhydration, as defined previously [1], signals significantly lower mortality or not. The aim was not to estimate an independent causal effect of hydration, as some confounding and time-varying variables were not available in NHANES (see Figure 1). A first set of unadjusted and multivariable-adjusted models estimated the mortality risk associated with euhydration vs. not meeting hydration criteria [1]. A second set of models estimated the mortality risk associated with euhydration vs. underhydration, more specifically.

Longitudinal Effect of Hydration Classification on Death
The weighted number of deaths occurring to the study population within 3 to 6 years of the MEC examination was estimated, stratified by hydration classification in 2009-2012, health condition in 2009-2012, and all-cause or chronic disease-associated death. Health condition was defined as any or none of the following health conditions: obesity, high waist circumference, insulin resistance, diabetes, elevated glucose, HgA1c or triglycerides, low HDL, hypertension, or metabolic syndrome. All-cause and chronic disease-associated death rates were estimated with jackknife confidence intervals. The stset command available in stata was used for survival analyses with a time-on-study scale [81] and failure defined as 'all-cause death' or 'chronic-disease associated death'. Kaplan-Meier survival curves were graphed by hydration classification as well as by the joint exposure of hydration classification and chronic health condition at baseline.
Weighted Cox proportional hazards models were used to test for differences in the hazard rate by hydration classification. All multivariable models controlled for all covariates described above for Model 2. The aim of this descriptive analysis was to determine if euhydration, as defined previously [1], signals significantly lower mortality or not. The aim was not to estimate an independent causal effect of hydration, as some confounding and time-varying variables were not available in NHANES (see Figure 1). A first set of unadjusted and multivariable-adjusted models estimated the mortality risk associated with euhydration vs. not meeting hydration criteria [1]. A second set of models estimated the mortality risk associated with euhydration vs. underhydration, more specifically. The blue circles represent the outcome. The pink and grey circles represent covariates that were observed and unobserved by National Health and Nutrition Examination Survey (NHANES), respectively. The DAG assumes that the hydration classification is a function of water intake requirements, water intake, and health condition. Many unobserved covariates and interrelationships between covariates are not shown. Figure 1 was generated using DAGgity.net v2.3.
If hydration and chronic health condition(s) interact in a negative feedback system, then effects of health condition(s) on mortality would be expected to be worse if hydration criteria are not met. To check for this possibility, unadjusted and multivariable adjusted weighted Cox proportional The yellow circle highlights the joint exposure. The blue circles represent the outcome. The pink and grey circles represent covariates that were observed and unobserved by National Health and Nutrition Examination Survey (NHANES), respectively. The DAG assumes that the hydration classification is a function of water intake requirements, water intake, and health condition. Many unobserved covariates and interrelationships between covariates are not shown. Figure 1 was generated using DAGgity.net v2.3.
If hydration and chronic health condition(s) interact in a negative feedback system, then effects of health condition(s) on mortality would be expected to be worse if hydration criteria are not met. To check for this possibility, unadjusted and multivariable adjusted weighted Cox proportional hazards models tested for greater risk of mortality associated with not meeting the hydration criteria among individuals with one or more chronic health conditions in 2009-2012.

Sensitivity Analyses
The hydration classification exposure selected for this analysis was defined to account for U-shaped hydration effects and distinguish euhydration from hyponatremia as well as underhydration. To illustrate need to account for U-shaped hydration effects, Poisson models, specified as described above, tested for a significant U-shaped relationship between serum osmolality and mortality. Serum osmolality was calculated from serum chemistry results as 1.86 × (Na+ + K+) + 1.15 × glucose + urea + 14 [82]. The models included a main effect for serum osmolality and serum osmolality*serum osmolality interaction term. The sample for this sensitivity analysis included adults aged 71-80 years.
The specific aims of this paper were focused on the 51-70-year-old age group to align with the previous analysis [1] as well as account for potential effect modification by older age related to factors such as decreased urine concentrating ability and/or increased risk of isotonic hypovolemia with age [1]. Unadjusted and multivariable-adjusted weighted models tested whether age group (71-80 vs. 51-70 years) modified the effect of underhydration on chronic disease mortality. The sample for this sensitivity analysis included adults aged 71-80 years.
Beyond accounting for U-shape, the hydration classification exposure selected for this analysis was assumed to reflect biologically meaningful variability that serum osmolality masks. Serum osmolality may be normal at the expense of vasopressin-mediated conservation of body water. Vasopressin-mediated pathways are associated with increased chronic disease mortality [19,34,37]. To check this working assumption, additional multivariable adjusted weighted Poisson models tested for an effect of underhydration on chronic disease mortality for persons aged 51-70 years with a calculated serum osmolality in the normal range, between 285 and 295 mmol/kg [82].
Finally, to consider if underhydration effects on chronic disease mortality might conceivably be attributable to vasopressin-mediated conservation of body water, as opposed to underlying diabetes by itself (elevated urine osmolality due to glycosuria), a last set of unadjusted and multivariable weighted adjusted Poisson models tested for a difference in the hazard between low urine volume and euhydration for individuals with urine osmolality below 500 mmol/kg. Figure 2 illustrates the distribution of serum sodium and urine osmolality for the study sample. Serum sodium ranged from 129 to 145 mmol/L, with a mean (standard error (SE)) of 139 (0.1) mmol/l. Urine osmolality ranged from 66 to 1180 mmol/kg, with a mean (SE) of 537 (11) mmol/kg. Urine volume ranged from 1 to 466 mL, with a mean (SE) of 77 (3) mL.

Sample Characteristics
Of 1200 study participants, 881 did not meet one or more hydration criteria. A total of 22 participants were hyponatremic, 10 of which had a urine osmolality below 500 mmol/kg. A total of 859 participants were classified as underhydrated, including 322 participants with zero urine or a urine volume below 50 mL. Urine osmolality was 500 mmol/kg or higher for 796 participants, including 537 participants with urine volume of 50 mL or more.
Applying sample weights, an estimated 71.4% (67.0%-75.4%) of fasting U.S. adults aged 51-70 years in 2009-2012 did not meet one or more hydration criteria. Among those who did not meet the hydration criteria, 2.7% (1.3%-5.5%) had hyponatremia and 97.3% (94.5%-98.7%) were underhydrated. An estimated 69.4% (65.5%-73.1%) of the study population was underhydrated. Table 1 describes the hydration classification by risk factors for chronic health conditions and/or mortality. Compared to euhydration, the prevalence of not meeting the hydration criteria varied Nutrients 2020, 12, 905 9 of 29 significantly by race/ethnicity, cigarette smoking, total daily dietary solute load, total energy, and carbohydrate intake.
weighted adjusted Poisson models tested for a difference in the hazard between low urine volume and euhydration for individuals with urine osmolality below 500 mmol/kg. Figure 2 illustrates the distribution of serum sodium and urine osmolality for the study sample. Serum sodium ranged from 129 to 145 mmol/L, with a mean (standard error (SE)) of 139 (0.1) mmol/l. Urine osmolality ranged from 66 to 1180 mmol/kg, with a mean (SE) of 537 (11) mmol/kg. Urine volume ranged from 1 to 466 mL, with a mean (SE) of 77 (3) mL.    Table 2A describes the number of NHANES participants and weighted proportion of the study population who did not meet one or more hydration criteria, stratified by chronic condition. The table aligns with groups defined in the prior NHANES analysis [1]. Independent of the covariates controlled in Model 2, the prevalence of not meeting hydration criteria was significantly higher among individuals with obesity, high waist circumference, insulin resistance, low HDL, and metabolic syndrome than among individuals without these conditions. Except for the relationship between obesity and not meeting hydration criteria, the observed associations were explained away by control for other chronic conditions (Model 3).

Prevalence of Not Meeting Hydration Criteria by Chronic Health Condition
Whereas Table 2A describes the prevalence of not meeting one or more hydration criteria, Table 2B focuses on the prevalence of underhydration, more specifically, excluding hyponatremia. The same pattern of results was observed for underhydration. Independent of the covariates controlled in Model 2, the prevalence of underhydration was significantly higher among individuals with obesity, high waist circumference, insulin resistance, low HDL, and metabolic syndrome than among individuals without these conditions. Table 3A,B describes the number of NHANES participants and weighted prevalence of each chronic health condition, stratified by hydration classification. As above, Table 3A describes associations for not meeting one or more of the hydration criteria, whereas Table 3B focuses on underhydration specifically. In unadjusted weighted Poisson models, obesity, high waist circumference, insulin resistance, diabetes, low HDL, hypertension, and metabolic syndrome were significantly more prevalent among individuals who did not meet hydration criteria. The proportion of people without any chronic condition was two times greater for people who were euhydrated. The relationships remained statistically significant after adjustment for all covariates in Model 2. The multivariable-adjusted prevalence ratio for each condition (Model 2) was reduced by more than 10% after further adjustment for all other chronic conditions (Model 3). Table 4 summarizes the number of deaths that occurred within 3 to 6 years of the 2009-2012 MEC examination. A total of 52 deaths in the 2009-2012 NHANES study sample represented 1,724,051 deaths to persons ages 51-70 years, nationally. The observed deaths were concentrated among persons who were underhydrated in 2009-2012. Figure 3a illustrates the difference in survival for people who were hyponatremic or underhydrated vs. euhydrated. The unadjusted all-cause death rate was over two times higher for people who did not meet the hydration criteria in 2009-2012 (weighted hazard ratio (HR): 2.37, 95% confidence interval (CI): 1.02-5.50, p = 0.044), either because of hyponatremia (weighted HR: 2.32, 95% CI: 0.32-16.89, p = 0.396) or underhydration (weighted HR: 2.37, 95% CI: 1.03-5.49, p = 0.044). The estimated increase in all-cause mortality associated with not meeting hydration criteria remained over two times higher after adjustment for covariates (adjusted weighted HR: 2.32, 95% CI: 0.96-5.63, p = 0.062). The estimated increase in all-cause mortality associated with underhydration, more specifically, also remained over two times higher (adjusted weighted HR: 2.33, 95% CI: 0.97-5.60, p = 0.059).     National Health and Nutrition Examination Survey (NHANES) participants were followed through December 31, 2015 for all-cause and chronic disease-associated mortality. With survey weights applied, estimates are representative of fasting U.S. adults aged 51-70 years in 2009-2012. Euhydrated: participants had a serum sodium of 135-144 mmol/L, a urine volume greater than or equal to 50 mL, and a urine osmolality below 500 mmol/kg. Did not meet hydration criteria: participants who did not meet one or more criteria for euhydration were classified as not meeting hydration criteria. Hyponatremic: participants had a serum sodium below 135 mmol/L. Underhydrated: participants had a serum sodium of 145 mmol/L or higher, a urine volume below 50 mL, or a urine osmolality of 500 mmol/kg or higher. One or more: participants were classified as having one or more chronic health conditions if they had obesity, high waist circumference, insulin resistance, diabetes, elevated glucose, elevated HgA1c, elevated triglycerides, low HDL, hypertension, or metabolic syndrome in 2009-2012. None: participants were classified as having none of the chronic health conditions if they did not have any of the listed conditions in 2009-2012. The number of deaths and death rates with jackknife confidence intervals were estimated with NHANES survey weights applied. Hyponatremia: serum sodium below 135 mmol/L; euhydration: serum sodium 135-144 mmol/L, urine volume 50 mL or higher, and urine osmolality below 500 mmol/kg; underhydration: serum sodium 145 mmol/L or higher, urine volume below 50 mL, or urine osmolality 500 mmol/kg or higher.

Chronic Diseasee Mortality
An estimated 1,084,144 deaths in the study population were attributed to or associated with heart disease, malignant neoplasms, hypertension, and/or diabetes. The unadjusted risk of chronic disease-associated death was about four times higher for underhydrated vs. euhydrated people in 2009-2012 (unadjusted weighted HR: 4.44, 95% CI: 1.57-12.54, p = 0.006). After adjustment for covariates, underhydration remained significantly associated with chronic disease-associated death (adjusted weighted HR: 4.21, 95% CI: 1.29-13.78, p = 0.019). Figure 3 (b) illustrates the difference in survival.

Sensitivity Analyses
For adults aged 51-80 years, a U-shaped relationship was observed between serum osmolality and all-cause as well as chronic disease mortality. For the 51-70-year-old age group, a U-shaped relationship was observed between serum osmolality and all-cause, but not chronic disease mortality (see Table A1). The tails of the serum osmolality distribution, below 285 mmol/kg and above 300 mmol/kg, were associated with the increased risk of chronic disease mortality (see Table A2).
The effect of underhydration on chronic disease associated mortality was significantly modified by age group (adjusted weighted HR (SE) for the interaction term: 0.23 (0.17), p-value = 0.048).

Mortality Associated with the Joint Effect of Hydration Classification and Chronic Health Condition
Across the different chronic health conditions in 2009-2012, deaths were concentrated among those who were underhydrated. Health condition-specific data were not reported because of small numbers. There were fewer than 10 chronic disease-associated deaths in the euhydrated group. Hyponatremia: serum sodium below 135 mmol/L; euhydration: serum sodium 135-144 mmol/L, urine volume 50 mL or higher, and urine osmolality below 500 mmol/kg; underhydration: serum sodium 145 mmol/L or higher, urine volume below 50 mL, or urine osmolality 500 mmol/kg or higher.

Chronic Disease Mortality
An estimated 1,084,144 deaths in the study population were attributed to or associated with heart disease, malignant neoplasms, hypertension, and/or diabetes. The unadjusted risk of chronic disease-associated death was about four times higher for underhydrated vs. euhydrated people in 2009-2012 (unadjusted weighted HR: 4.44, 95% CI: 1.57-12.54, p = 0.006). After adjustment for covariates, underhydration remained significantly associated with chronic disease-associated death (adjusted weighted HR: 4.21, 95% CI: 1.29-13.78, p = 0.019). Figure 3b illustrates the difference in survival.

Sensitivity Analyses
For adults aged 51-80 years, a U-shaped relationship was observed between serum osmolality and all-cause as well as chronic disease mortality. For the 51-70-year-old age group, a U-shaped relationship was observed between serum osmolality and all-cause, but not chronic disease mortality (see Table A1). The tails of the serum osmolality distribution, below 285 mmol/kg and above 300 mmol/kg, were associated with the increased risk of chronic disease mortality (see Table A2).
The effect of underhydration on chronic disease associated mortality was significantly modified by age group (adjusted weighted HR (SE) for the interaction term: 0.23 (0.17), p-value = 0.048).

Mortality Associated with the Joint Effect of Hydration Classification and Chronic Health Condition
Across the different chronic health conditions in 2009-2012, deaths were concentrated among those who were underhydrated. Health condition-specific data were not reported because of small numbers. There were fewer than 10 chronic disease-associated deaths in the euhydrated group. Table 4 summarizes the number of deaths that occurred by presence or absence of chronic health condition in 2009-2012. Among those with one or more chronic health conditions in 2009-2012, compared to euhydration, not meeting hydration criteria was associated with significantly greater risk of all-cause mortality within 3 to 6 years (unadjusted weighted HR: 2.51, 95% CI: 1.14-5.51, p = 0.024; adjusted weighted HR: 2.50, 95% CI: 1.07-5.85, p = 0.036). Compared to euhydration, underhydration was associated with significantly greater all-cause mortality (unadjusted weighted HR: 2.51, 95% CI: 1.13-5.54, p = 0.025; adjusted weighted HR: 2.47, 95% CI: 1.06-5.71, p = 0.036). It was not possible to estimate the effect of underhydration on mortality for participants with none of the chronic conditions at baseline, due to the small number of deaths in this group. No chronic disease-associated deaths were observed during the follow-up period for people who did not have any of the chronic conditions and were, also, euhydrated in 2009-2012. Compared with people who did have one or more chronic condition, but were classified as euhydrated, in 2009-2012, people who were not identified as having any chronic condition at baseline but were underhydrated were significantly more likely to die (unadjusted weighted HR: 5.41, 95% CI: 1.32-22.15, p = 0.020; adjusted weighted HR: 8.43, 95% CI: 1.15-61.75, p = 0.037).
It was not possible to estimate the effect of underhydration on mortality for participants with none of the chronic conditions at baseline, due to the small number of deaths in this group. No chronic disease-associated deaths were observed during the follow-up period for people who did not have any of the chronic conditions and were, also, euhydrated in 2009-2012. Compared with people who did have one or more chronic condition, but were classified as euhydrated, in 2009-2012, people who were not identified as having any chronic condition at baseline but were underhydrated were significantly more likely to die (unadjusted weighted HR: 5.41, 95% CI: 1.32-22.15, p = 0.020; adjusted weighted HR: 8.43, 95% CI: 1.15-61.75, p = 0.037).

Discussion
The results of this study suggest that over 95% of U.S. adults aged 51-70 years who do not meet hydration criteria are underhydrated. The results further suggest that underhydration signals a statistically significant health risk. Underhydration, that is, serum sodium above the 135-144 mmol/L range, spot urine volume less than 50 mL, or urine osmolality of 500 mmol/kg or higher, was cross-sectionally associated with increased prevalence of multiple chronic health conditions. Underhydration was longitudinally associated with increased risk of all-cause and chronic disease associated mortality within 3-6 years.

Hydration Classification
Researchers debate which biomarker to use for epidemiologic analysis, dietary recommendations, and public health policy. Various hydration biomarkers have been developed to index a variety of body fluid disorders [7,83,84]. Any one biomarker may not be optimally sensitive to all types of body fluid disorders [9]. This analysis explored the public health significance of not meeting hydration criteria specified in a previous NHANES analysis [1], testing for health effects of hyponatremia and/or underhydration, conditions that would be flagged by the hydration criteria.
Researchers agree that serum osmolality and its main component, serum sodium, are sensitive biomarkers of total body water deficit. With every 2% body weight loss via sweat during exercise, plasma osmolality increases about 5 mmol/kg [85]. Researchers further agree that serum osmolality and sodium have limited variability under conditions of daily life due to tight homeostatic regulation [55,86]. Researchers disagree, however, about relying on a biomarker that has limited variability under conditions of daily life to index hydration in people who are not hospitalized or not performing extreme exercise [86]. By itself, a normal serum osmolality or sodium can mask variability in metabolic and/or physiologic compensation. It does not discriminate individuals who have normal serum osmolality and/or normal serum sodium, thanks to urine concentration, from individuals who have normal values with no compensation required. Beyond the potential for unmeasured heterogeneity due to homeostatic compensation, use of serum osmolality or serum sodium to index hydration is complicated by the narrow 10 mmol/L normal range, variation in the set-point for homeostasis, and U-shaped relationship with health outcomes. A one-unit difference in serum osmolality or sodium may be clinically insignificant or only reflect measurement error. Cutoffs for defining the optimal normal range for these biomarkers may differ within and between individuals because the homeostatic set-point for serum osmolality or sodium can vary [61,87]. As hypo-and hypernatremia are both associated with adverse health outcomes, including death [10,88], use of continuous serum osmolality and serum sodium in epidemiologic analysis inappropriately masks U-shaped effects. Indeed, continuous serum osmolality is not linearly associated with mortality in NHANES data [89].
Researchers agree that spot urine osmolality is sensitive to acute and mild changes in fluid intake in the short-term [55,86], but disagree about using this biomarker to index 'usual' status in epidemiologic hydration studies. Despite disagreement, spot urine osmolality at a given time of day may be relatively stable among free-living individuals, as free-living individuals have stable within-person 24 h urine volume and osmolality, as well as intransigent fluid intake patterns [90]. Community-dwelling individuals who report lower fluid intake have elevated antidiuretic hormone (arginine vasopressin, AVP) and higher urine osmolality [57]. In non-acutely ill adults, following a sustained increase in total daily water intake (i.e., a change in the 'usual' level of water intake), spot urine osmolality decreases for all urine collections across the day, including the morning fasting collection [90,91]. The hydration classification used in this analysis was selected to match that defined for a previous NHANES analysis [1]. The hydration classification represents a compromise between only using serum osmolality or serum sodium and only using urine osmolality. It also represents an alignment of epidemiological research on free-living non-acutely ill populations with clinical differential diagnosis of fluid disorders in hospitalized patients [92].

Cross-Sectional Association between Hydration Classification and Chronic Health Conditions
The relationship between hydration classification and chronic health condition was described from two points of view, with hydration classification treated as either the dependent or independent variable. One or more chronic health condition was cross-sectionally associated with not meeting the hydration criteria and underhydration, more specifically. Conversely, not meeting hydration criteria and underhydration were respectively associated with chronic health conditions. Although a statistically significant association in both directions is anticipated when a statistically significant association is observed in one direction, the study provides weighted estimates of the proportion exposed and magnitudes of association for each direction of effect for U.S. adults aged 51-70 years.
Population-representative estimates of each effect may be useful for future meta-analysis, causal inference do-calculus, program planning, or policy making.
Independent of covariates, obesity, high waist circumference, and insulin resistance each respectively increased the prevalence of not meeting the hydration criteria by 20%. Low HDL cholesterol increased the prevalence of not meeting the hydration criteria by 15%. The prevalence ratio for not meeting the hydration criteria was 15% higher for individuals with metabolic syndrome compared to individuals without metabolic syndrome. The results suggest need or opportunity for clinicians to systematically assess hydration for individuals with conditions of the metabolic syndrome, and chronic disease researchers to consider hydration as a potential confounder and/or effect modifier for chronic disease morbidity and mortality. The National Academy of Medicine report recognizes that illness increases water requirements [71]. Further research is needed to determine what level of water intake reduces the risk of underhydration for individuals with metabolic syndrome.
With hydration classification treated as the independent variable, the results suggest that not meeting the hydration criteria and underhydration, more specifically, have public health significance. Not meeting the hydration criteria and underhydration were associated with significantly increased prevalence of obesity, insulin resistance, diabetes, dyslipidemia, hypertension, and metabolic syndrome. Independent of age, sex, race/ethnicity, height, education, poverty level, physical activity, dietary solute load, total energy intake, carbohydrate intake, past or current cigarette smoking, prescription medication use, season, and length of fast, the proportion of people without any of the aforementioned conditions was 39% lower for individuals who did not meet hydration criteria compared to individuals who did meet the hydration criteria.

Potential for Endogenous Effects of Hydration and Chronic Disease
The results of this study were consistent with a potential negative feedback mechanism. Excepting the association between obesity and hydration classification, control for all other chronic health conditions explained away each observed association, signaling that the relationship between each health condition and hydration could conceivably be mediated by metabolic dysregulation due to other chronic conditions. Alternatively, each association might be confounded by one or more other health condition(s). If metabolic dysregulation mediates the relationship between each health condition and hydration, then metabolic dysregulation and the health conditions that are upstream to the metabolic dysregulation should not be controlled in multivariable analysis [93], but rather treated as a joint exposure. If, on the other hand, one or more health conditions confounds the association, explaining the relationship by a pathway that does not involve the index health condition, then stratified analyses would be needed to test for each association among those with no other condition. Longitudinal data with carefully selected study populations are needed to determine if each chronic health condition prospectively predicts underhydration.
The finding that obesity remained cross-sectionally associated with hydration classification after control for all other health conditions might suggest that this relationship preceded associations between underhydration and the other health conditions assessed in this analysis. Obesity might be a cause or consequence of underhydration. Large body size increases water intake requirements and likelihood of inadequate water intake. On the other hand, lower absolute and relative intake of drinking water have been observed to reduce energy expenditure, reduce fat oxidation, and induce insulin resistance, which contribute to positive fat balance [94]. Obesity reflects a sustained period of positive fat balance, due primarily to reduced oxidation of fat consumed [95,96]. The residual association that remained after control for other health conditions might also reflect measurement or misclassification error related to the method and cutoffs used to assess each health condition. Individuals with insulin resistance detectable by a HOMA-IR value of 2.5 may be more likely to already be obese than individuals with a milder HOMA-IR of 1.5. Longitudinal data and carefully defined health conditions would be required to determine the order of associations.

Potential Causal Mechanisms Linking Hydration with Metabolic Syndrome
Underhydration, that is, having a serum sodium above the normal range, low urine volume and/or concentrated urine, implies osmotic stress on cells. Osmotic stress on cells is a plausible causal factor for the metabolic syndrome, which is characterized by multiple simultaneous health conditions, as osmotic cell shrinkage acts as a metabolic switch that simultaneously alters multiple biochemical pathways, cellular processes, and physiological systems [29].
Longitudinal data suggest parallel relationships between various correlates of osmotic stress on cells and risk of the metabolic syndrome. Behavioral determinants of hyperosmotic stress on cells, such as salt intake, lower absolute water intake, and intake of hypertonic beverages such as sugar-sweetened beverages instead of drinking water [97][98][99][100][101]; blood biomarker measures of hyperosmotic stress on cells, such as serum hypernatremia, hypertonicity, or hyperosmolality [26,32,33,38,88,102]; intracellular measures of biochemical response to hyperosmotic cell shrinkage such as upregulation of the serumand glucocorticoid-inducible kinase 1 (SGK1) [103,104]; and physiologic measures of response to hyperosmotic cell shrinkage, such as increases in plasma copeptin [18,19,[105][106][107][108], are independently associated with incident metabolic dysregulation and/or chronic disease risk.
The parallel results suggest that osmotic stress on cells or some correlate(s) of osmotic stress on cells may be causally related to metabolic dysregulation and chronic disease risk. Any one or more measure might conceivably be the necessary and sufficient etiologic factor(s) for metabolic syndrome. Given that osmotic cell shrinkage simultaneously impacts multiple biochemical pathways, multiple cellular mechanisms may mediate effects of suboptimal fluid intake on pathophysiology and life span. Additional experimental effort is needed to unravel the molecular pathophysiology of the relationship between not meeting hydration criteria, underhydration, and chronic health conditions. The success of disease prevention and treatment efforts depends on identification of the necessary and sufficient causal factor(s).

Longitudinal Association between Hydration Classification and Mortality
The above paragraphs consider the potential for endogenous, interactive effects of hydration and health conditions on disease progression, as well as the potential for health conditions to simply confound effects of hydration on disease progression. If the underlying mechanism(s) involve(s) feedback of endogenous effects, then an individual's stage in the feedback cycle at baseline would be an appropriate exposure for longitudinal mortality analysis. The effect to be estimated would be the joint effect of the hydration classification and chronic condition on mortality risk. If, on the other hand, baseline chronic health condition confounds the effect of hydration on disease progression, then the longitudinal effect to estimate would be the effect of hydration among individuals without chronic health conditions at baseline.
In the present analysis, the combination of one or more chronic health conditions in 2009-2012 and underhydration in 2009-2012 was associated with significantly greater all-cause and chronic disease mortality compared to the combination of one or more chronic conditions and euhydration in 2009-2012. The data are consistent with sub-optimal hydration exacerbating metabolic dysregulation and accelerating disease progression. It was not possible to estimate the effect of underhydration on mortality among those without a chronic health condition at baseline, as only five deaths occurred in this group. Zero chronic disease associated deaths were observed for people who did not have any of the chronic conditions and, also, met the hydration criteria in 2009-2012.
The results of the present NHANES analysis are consistent with the results of Mohan et al. [10] and inconsistent with those reported by Kant and Graubard [89]. Mohan et al. [10] report significantly greater mortality associated with hypernatremia vs. normonatremia in NHANES participants followed from 1999-2004 through December 31, 2006. They also report significantly greater mortality associated with hyponatremia vs. normonatremia. The U-shaped relationship between serum sodium and mortality may explain why Kant and Graubard [89] observed no linear association between continuous serum osmolality and all-cause mortality with up to 13 years of follow-up. The present sensitivity analyses confirmed a significant U-shaped relationship between serum osmolality and mortality in the 2009-2012 NHANES data (see Appendix A). Null effects of individual hydration parameters on mortality may also reflect the fact that no single hydration measure isolates optimal hydration for non-acutely ill individuals. Normal serum osmolality may mask heterogeneity due to homeostatic compensation. Low urine osmolality may mask hyponatremia. The present sensitivity analyses confirmed significant variability in mortality associated with underhydration (low urine volume or urine osmolality above 500 mmol/kg) for individuals with a serum osmolality in the normal range, between 285-295 mmol/kg.

Confounding
Effects of hydration on the metabolic syndrome and chronic disease risk may be confounded by determinants of water requirements, determinants of water intake, and determinants of water handling in the body. Time invariant and time-varying environmental, socio-economic, behavioral, psychological, and biological factors may determine an individual's water requirements and/or water intake at a given point in time. Environmental factors may include ambient temperature; altitude; the availability, accessibility, and quality or palatability of water; and place-based conditions, such as school or workplace policy (e.g., vending machine rules, recess or toilet breaks). Socio-economic factors may include cultural norms or ethnicity, education, and poverty. Behavioral factors include physical activity, solute intake, sleep, smoking, alcohol intake, and medication use. Psychological factors include stress and pain. Biological factors include age, sex, body size, race, metabolic adaptation (e.g., to habitual caffeine intake or heat stress), renal concentrating/diluting capacity, disease related factors (e.g., insult accumulation/wear and tear on organ systems, glycosuria), and genetic factors.
The present cross-sectional analysis was able to control for some environmental, socio-economic, behavioral, and genetic factors, but data were not available to also control for psychological factors. Stress and pain may confound the observed associations because stress and pain can trigger vasopressin and cortisol release [109,110], which stimulate urine concentration. At the same time, stress and pain induce metabolic shifts that favor the flight or fight response, a profile of metabolic dysregulation that is characteristic of the metabolic syndrome. In addition to better control for environmental and psychological factors, future prospective studies of hydration effects on chronic disease risk should address time-varying factors.
Variables which might confound the observed cross-sectional associations between hydration and chronic health conditions may also confound the observed longitudinal relationships between hydration and mortality. In the present study, longitudinal models predicting death within 3-6 years controlled for the same covariates addressed in the cross-sectional analysis. On the basis of the hypothesized causal mechanism illustrated in Figure 1, the analysis did not control for all possible confounding variables.
The present study observed that spot urine volume below 50 mL was associated with significantly increased mortality risk. This finding, considered with the association between serum hypernatremia and increased mortality reported by Mohan et al. [10], suggests that the observed association between underhydration and increased mortality is not solely attributable to urine osmolality of 500 mmol/kg or more due to uncontrolled diabetes-related glycosuria. Urine glucose data were not available to rule out a confounding by glycosuria. Future work to confirm and characterize an independent effect of underhydration on mortality should consider potential confounding by glycosuria.
Acknowledging that observational data might be informative regarding causal effects, even when confounding variables are unmeasured [93], the present analysis reported weighted probabilities of having health conditions and estimated number of deaths, in addition to relative hazard ratios, in case readers wish to apply front door or do-calculus techniques to estimate the probability of death caused by underhydration. Front door adjustment may not be applicable to the present estimation problem, however, if the unmeasured confounders (e.g., environmental factors, pain, psychological stress) have a strong effect on the mediating metabolic dysregulation. Independence of the intermediate from the unobserved confounder is a prerequisite for using front door adjustment [93]. The lack of data regarding metabolic dysregulation from an intermediate time point limits the approach. Information or assumptions about a weak, negligible effect of environmental and psychological factors on metabolic dysregulation, which remains consistent over time, in the population of interest might enable front door adjustment.

Analysis Limitations
The present analysis was limited by the non-randomized, observational design, lack of serial measures of hydration biomarkers, chronic health conditions, relatively short 3 to 6 year follow-up period and small number of deaths. Although chronic health condition was determined from laboratory tests and multivariable models were used to control for confounding, the results are vulnerable to measurement or misclassification error and residual confounding. Interpretation of the results is limited to descriptive purposes and hypothesis generation about underlying causal mechanisms.
The analysis defined hydration in terms of biomarkers, which are sensitive to hypertonic dehydration but may be relatively insensitive to isotonic dehydration and/or hypohydration for persons with deficits in urine concentrating ability. The hydration criteria may have limited interpretation for individuals with advanced chronic disease. Serum sodium and urine osmolality are influenced by chronic health conditions. Hyperglycemia can distort serum sodium downwards. Urine glucose may distort urine osmolality upwards. Medications prescribed for chronic disease, such as diuretics, may disassociate serum sodium and urine osmolality from dietary intake. Renal urine concentration and/or urine dilution capacity may influence urine osmolality. The present sensitivity analyses confirmed that older age significantly modifies the effect of underhydration on chronic disease mortality.

Future Directions
In addition to the question of whether or not underhydration truly causes an increased risk of chronic disease and mortality, the present results raise questions about what type and amount of water intake could reduce said risk. In the preceding NHANES analysis, the prevalence of not meeting hydration criteria was significantly higher for non-acutely ill men and women aged 51-70 years with daily total water intake (TWI) below 45 mL/kg or plain drinking water intake (PWI) below 20 mL/kg [1]. The discussion of that paper calls for careful sensitivity and specificity analyses to identify the optimal ml/kg TWI and PWI for each population sub-group and condition. Group-and condition-specific randomized trials are needed to test effects of different levels of water intake, defined relative to optimal ml/kg TWI or PWI cutoffs, on chronic disease incidence and progression.

Conclusions
The present study extends inference regarding the reported high prevalence of not meeting hydration criteria among U.S. adults aged 51-70 years [1]. The study provides nationally representative estimates of association between not meeting the hydration criteria and the prevalence of chronic health conditions. Despite the sample size limitations, the analysis reveals a statistically significant decrease in survival, within 3 to 6 years, for adults aged 51-70 years with a pre-existing chronic health condition who did not meet the hydration criteria, and were underhydrated, more specifically. The results draw attention to the public health impact of not meeting hydration criteria.    [110]. The multivariable adjusted weighted models controlled for age in years, sex, height, race/ethnicity, poverty level, education level, physical activity level, total energy intake, dietary solute load, percent of total dietary energy from carbohydrate, number of alcoholic drinks, smoking, medication use, season, and hours of fast before measurement.