Exposure concentrations assessed from spot urine samples are typically adjusted to a constant creatinine concentration to correct for dilution of the samples due to the variable hydration status of the participants. However, recent studies have suggested that this commonly used approach of correcting urinary analyte concentrations by creatinine could result in biased estimates in some instances (Christensen et al. 2014; Hoet et al. 2016; Middleton et al. 2016). Urine specimens are susceptible to inter- and intra-individual variations in dilution due to a variety of factors such as water intake, circadian fluctuations, physical activity, temperature, humidity, age, and disease status (Aylward et al. 2014; Hoet et al. 2016). As a result, measurement error can bias analyte concentrations in the form of under- or overestimations if adjustments for dilution are not made. Creatinine is a byproduct of muscle metabolism excreted from the body primarily through glomerular filtration, in addition to active secretion by peritubular capillaries of the kidney (Barr et al. 2005). In healthy individuals, urinary creatinine concentrations are correlated with anthropometric measurements including body mass and body mass index (BMI); lean body mass, rather than adiposity, likely explains these associations (Baxmann et al. 2008; Gerchman et al. 2009; Yeh et al. 2015). These relationships have implications for epidemiologic analyses of biomarkers and obesity, in which creatinine correction may be inappropriate.

With more than one-third of the U.S. adult population obese, interest in understanding how exposure to environmental chemicals may contribute is increasing (Maull et al. 2012; Ogden et al. 2014; Thayer et al. 2012). However, a limited number of studies have focused on the association of arsenic exposure with obesity. These studies have been heterogeneous in terms of study populations, biomarkers of exposure, and results (Ettinger et al. 2014; Grashow et al. 2014; Lin et al. 2014; Ronco et al. 2010). In an occupational study of welders in the U.S. (n=74), toenail arsenic concentrations were inversely associated with BMI (β=−0.03, p=0.006) after adjusting for age, caloric intake, alcohol intake, smoking status, and season of toenail clipping (Grashow et al. 2014). A cross-sectional analysis of young adults of African descent (n=500) across five countries (Ghana, South Africa, Seychelles, Jamaica, and the United States) found blood arsenic concentrations greater than the median (8.48μg/L) were associated with significantly lower odds of being overweight [odds ratio (OR)=0.27, 95% CI: 0.09–0.81], but not with obesity (OR=0.86; 95% CI: 0.24–3.07), after adjustment for age, sex, site location, marital status, education, paid employment, smoking, alcohol use, and fish intake (Ettinger et al. 2014). A cross-sectional study of Chilean women of childbearing age (n=107) found no crude associations between urinary arsenic concentrations and BMI or fat mass percentage (p-value=0.79) (Ronco et al. 2010). Average urinary arsenic concentrations when expressed as per gram of creatinine were 11.3μg/g (95% CI: 8.9–20.3) among women with a BMI<18.5 kg/m2, 11.9μg/g (95% CI: 6.2–16.1) among women with a BMI between 19 and 24.9 kg/m2, and 12.3μg/g (95% CI: 6.9–16.5) among women with a BMI>25 kg/m2 (Ronco et al. 2010). Last, a cross-sectional study of Taiwanese adolescents (n=303) found a significant age- and sex-adjusted inverse relationship with BMI regardless of creatinine correction (p-value for trend=0.01; 0.03 when creatinine corrected), but the urinary arsenic concentrations observed (mean: 24.54μg/L) suggested higher exposures than would be expected in the United States (Su et al. 2012).

It has been hypothesized that urinary creatinine may be a collider, or common descendent of two variables, in causal pathways (Greenland et al. 1999; O'Brien et al. 2016). Specifically, urinary creatinine concentrations are influenced by both the body’s hydration status and body composition. Crude measures of obesity like BMI poorly discriminate between fat and muscle; as such, controlling for urinary creatinine in regression models could induce spurious associations (Greenland 2003). Given the biological process of creatinine excretion and the equivocal epidemiological evidence of a relationship between low-level arsenic exposure and obesity, a thorough investigation of appropriate urinary dilution metrics and analytic techniques is warranted. Thus, using data from the National Health and Nutrition Examination Survey (NHANES), we evaluated the association between arsenic exposure and obesity using various dilution adjustment approaches, including creatinine, urinary osmolality (the amount of solute particles contained in urine), and urinary flow rate (the quantity of urine produced over a specified period of time) as alternative metrics of urine dilution (Hays et al. 2015). We also employed a novel method (covariate-adjusted standardization) to obtain an estimate of arsenic exposure that is independent of demographic factors and anthropometric measures, and solely attributable to hydration status (O'Brien et al. 2016; O'Brien et al. 2017).

We used data from the 2009–2012 NHANES survey cycles. Conducted by the National Center for Health Statistics (NCHS) and the Centers for Disease Control and Prevention (CDC), NHANES is a complex survey design that surveys a representative sample of the civilian, noninstitutionalized U.S. population (NHANES 2014). We limited our analysis to adults (≥20y), and complete data for the measures are described below (n=3,097). Females who self-reported being pregnant or breastfeeding, or who had a positive urine pregnancy test at the time of exam, were excluded. We also excluded participants with chronic kidney disease, defined as an estimated glomerular filtration rate less than 60mL/min/1.73 m2 using the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation, or self-report of receiving dialysis in the past 12 mo (Levey et al. 2009). NHANES is approved by the National Center for Health Statistics Research Ethics Review Board. All NHANES participants provide informed consent before taking part in the survey.

Complete data were available for 3,097 adult NHANES participants. Geometric means of the urinary arsenic concentrations, dilution metrics, BMI, and waist-to-height ratio across covariate groups are presented in Table 2. Urinary arsenic concentrations were positively associated with urinary creatinine and osmolality, and inversely associated with urinary flow rates. BMI and waist-to-height ratio were also positively associated with urinary creatinine and osmolality, but no associations were observed with urinary flow rate.

The associations between estimated urinary total arsenic 1 and BMI are summarized in Figure 2. Strong inverse associations between urinary total arsenic and BMI were observed when creatinine was included as an independent covariate in the regression model (Model 2), arsenic concentration was standardized by creatinine (Model 3), osmolality was included as an independent covariate in the regression model (Model 5), and arsenic concentration was standardized by osmolality (Model 6). Null associations were observed between urinary total arsenic and BMI when no adjustment for urinary dilution was made (Model 1), arsenic concentration was covariate-adjusted standardized by creatinine (Model 4), and urinary flow rate was included as an independent covariate in the regression model (Model 7). A marginally positive association was observed between urinary total arsenic concentration and BMI when arsenic was standardized to urinary flow rate through multiplication to obtain excretion rates (Model 8). Appreciably similar associations were observed between estimated urinary total arsenic 2 and BMI (Figure 3).

The associations between estimated urinary total arsenic 1 and waist-to-height ratio are summarized in Figure 4. Inverse associations between urinary total arsenic and waist-to-height ratio were observed when creatinine was included as an independent covariate in the regression model (Model 2), arsenic concentration was standardized by creatinine (Model 3), osmolality was included as an independent covariate in the regression model (Model 5), and arsenic concentration was standardized by osmolality (Model 6). Null associations were observed between urinary total arsenic and waist-to-height ratio when no adjustment for urinary dilution was made (Model 1), arsenic concentration was covariate-adjusted standardized by creatinine (Model 4), urinary flow rate was included as an independent covariate in the regression model (Model 7), and arsenic was standardized to urinary flow rate through multiplication (Model 8). Appreciably similar associations were observed between estimated urinary total arsenic 2 and weight-to-height ratio (Figure 5).

The epidemiologic literature on arsenic and obesity is scant, but limited evidence exists suggesting arsenic exposure may be inversely related to BMI (Ettinger et al. 2014; Grashow et al. 2014; Ronco et al. 2010; Su et al. 2012). In contrast, our findings do not support an association between arsenic exposure, as measured in urine, and obesity. It is possible that low-level arsenic exposure, as seen among the general population of the United States, has no effects on adiposity; occupational studies and studies conducted outside of the United States may be capturing exposures above a potential threshold. Our findings further suggest that urinary creatinine and osmolality may be colliders on the causal pathway from arsenic exposure to obesity as common descendants of hydration and body composition. When urinary creatinine and osmolality were accounted for in our analyses, we observed significant inverse associations with BMI and waist-to-height ratios. In other words, controlling for these variables may have opened a previously blocked backdoor path between urinary arsenic concentrations and obesity. Using covariate-adjusted standardization, adjusting for urine dilution using urinary flow rates, or modeling arsenic excretion rates resulted in null associations between urinary arsenic concentrations and measures of obesity.

Nearly all the creatinine excreted in urine is produced in skeletal muscle (Barr et al. 2005). Skeletal muscle is positively associated with stature, as taller individuals have longer bones and muscles (Janssen et al. 2000). Similarly, skeletal muscle mass is correlated with body weight; not only is muscle denser than adipose tissue, but heavier individuals require more muscle mass for movement (Janssen et al. 2000). In comparison with normal-weight individuals, obese individuals have more skeletal muscle, albeit with lesser amounts of muscle relative to body fat (Janssen et al. 2000). As such, obese individuals excrete more creatinine than their normal-weight counterparts (Gerchman et al. 2009). It is unclear whether more refined methods of measuring adiposity (e.g., bioelectrical impedance, dual-energy X-ray absorptiometry) would be useful in disentangling fat from muscle mass rendering creatinine a noncollider. Urine osmolality and creatinine are highly correlated (Yeh et al. 2015). Although osmolality is generally considered more robust than creatinine because it is less influenced by individual characteristics (e.g., demographics or medical conditions), we observed that both factors similarly introduced bias (Middleton et al. 2016; Yeh et al. 2015). Multiple studies have observed that osmolality tends to be higher in urine samples from obese individuals, indicating more concentrated urine, than in normal-weight individuals (Chang et al. 2016; Hays et al. 2015; Yeh et al. 2015). However, osmolality has been considered the optimal metric of urine concentration, because it provides a measure of the number of all solutes present regardless of their molecular mass or structure (Chadha et al. 2001; Dossin et al. 2003). The solutes that contribute to total osmolality include urea, sodium, potassium, chloride, and creatinine, among others (Chadha et al. 2001). Studies have observed that urinary concentrations of sodium, chloride, and creatinine solutes have been positively associated with obesity (Al-Hayek et al. 2013; Fotheringham et al. 2014; Fram et al. 2015; Ge et al. 2016; Taylor and Curhan 2006). With the exception of creatinine, however, these urinary components are not derived from muscle; rather, urinary solute concentrations likely reflect dietary intakes (Bingham 2003; Taylor and Curhan 2006).

Urinary flow rate has been proposed as a direct metric of hydration status (Hays et al. 2015). In the present analyses, urinary flow rate was observed to be independent of BMI and waist-to-height ratio, suggesting that it is not a collider of the causal path between urinary arsenic and obesity. Flow-rate calculations assume that the bladder is entirely emptied and requires both the total void volume and times of the prior and current void. To calculate excretion rates, analyte concentrations are multiplied by urinary flow rates. However, the time between voids is rarely collected in large-scale epidemiologic studies and may be difficult to accurately obtain in certain populations, such as children.

In the present analysis, we evaluated a general population sample; however, analytical approaches in high-risk populations should also be considered. For example, kidney disease is qualitatively associated with metrics of urine dilution. Urinary creatinine is elevated in individuals with chronic kidney disease, whereas osmolality and urinary flow rates are significantly lower among diseased individuals. This contrast suggests that urinary flow rate, although not related to measures of obesity among healthy individuals, may be a collider among persons with chronic kidney disease. Thus, covariate-adjusted standardization may be the most suitable approach for high-risk populations, such as individuals with chronic kidney disease. Children were not included in our analyses because the obesity metrics used are systematically lower in this population. We have no reason to believe, however, that methods to account for urine dilution would perform differentially among children. In fact, a recently published study utilized the covariate-adjusted standardization approach to account for urine dilution in a study of urinary phenols and childhood fat mass in a birth cohort (Buckley et al. 2016).

We acknowledge several limitations of our analyses. We did not evaluate specific gravity, a measure of the number of solute particles in urine as well as their size, because it was not measured as part of the NHANES 2009–2012 surveys. Although specific gravity is considered a more robust measure of urine dilution than creatinine (Pearson et al. 2009), it is associated with muscle mass (Hamouti et al. 2010) and would have likely yielded results similar to those observed with creatinine and osmolality in these analyses. Additionally, conditions like diabetes can alter solute concentrations (e.g., glucose and protein) in urine, further disqualifying specific gravity adjustments (Voinescu et al. 2002). Urinary flow rates were used independently and to calculate arsenic excretion rates. The accuracy of the urinary flow rates is questionable, as participants were asked to self-report the time of their last void. However, we have no reason to believe that any inaccuracies would be related to urinary arsenic concentrations or anthropometric measures. We adjusted for age, gender, race/ethnicity, and survey cycle in all regression models, but as in all observational studies, residual confounding could partially explain our findings. Specifically, we did not assess muscle wasting and diuretic usage, which could directly affect urinary creatinine and osmolality, as well as body mass or size, although we expect their effect to be minimal, given the low prevalence in the general population. Further, we did not evaluate total water or protein intake, which could be common causes of urine diluteness and anthropometric measures (Krieger et al. 2006; Rosinger et al. 2016). Lastly, we simplified the causal diagram by not showing the involvement of the one carbon metabolism pathway. Evidence has been found that suggests that arsenic exposure alters creatinine excretion via folate-mediated single-carbon metabolism and that obesity may alter the body’s ability to metabolize arsenic (Gamble and Hall 2012; Hudgens et al. 2016). Regardless of these relationships, urinary creatinine and osmolality would remain colliders in our hypothesized causal scenario.

We recommend investigators draw a directed acyclic graph of a biomarker of exposure–outcome relationships prior to making adjustments for urine dilution. For studies of arsenic specifically, such adjustments may be unnecessary as 24-h concentrations do not significantly differ from spot urine samples (Hinwood et al. 2002). However, if measurement error is a concern, performing covariate-adjusted standardization, adjusting for urinary flow rates (if collected), or modeling analyte excretion rates may be good options, depending on the specific causal scenario. Given that metrics of urine dilution are often affected by body composition, studies of urinary biomarkers of environmental exposures and obesity or obesity-related conditions should carefully consider how to best correct for variations in urine dilution at the time of measurement.