What an ICC says about a single sample
The study asks whether one urine measurement can stand in for exposure across gestation. The metric used to answer this is the intraclass correlation coefficient, or ICC: the share of total variation in log2-transformed, specific gravity-standardized concentrations that lies between participants rather than within the same participant over time. An ICC near 1 means participants keep their relative rank across samples, so one sample carries information about the rest of pregnancy; an ICC near 0 means a participant's own concentrations move enough that a single value mostly describes that moment.
This distinction is why the paper's answer is not a simple yes or no. The authors measured 154 analytes from 12 chemical classes in three urine samples per participant, collected from 6 to 37 weeks of gestation. They then estimated ICCs separately for each cohort site for 58 analytes detected in more than 60% of samples at that site, and used kappa statistics for 56 analytes detected in 20-60% of samples to check whether detection itself agreed across the three timepoints.
Reliability differed more by site than by chemical
The headline result is heterogeneity. ICCs spanned -0.12 to 0.91, and the pattern was not uniform across the three sites: parabens were consistently high (>0.5) in Illinois but lower in Puerto Rico and New York; benzophenones were higher in New York; antimicrobials were higher in Puerto Rico; most phthalate and polycyclic aromatic hydrocarbon metabolites were moderate to high in Illinois but lower in Puerto Rico; and most aromatic amines were moderate to high in New York but lower in Illinois.
Only three analytes behaved consistently everywhere: benzophenone-1 and propyl paraben had ICCs above 0.5 at all sites, and bisphenol F had ICCs below 0.1 at all sites. The authors read the site differences as a signal that demographic and geographic characteristics, and possibly seasonal exposure patterns, shape how stable a biomarker looks. That interpretation is offered as an explanation, not tested as a cause in this design.
Higher concentrations usually meant steadier biomarkers
The paper groups analytes by the site-specific median concentration relative to the limit of detection, or LOD, the lowest level the assay can reliably distinguish. Analytes with a median-to-LOD ratio above 50 were generally more stable across pregnancy, including benzophenones, propyl paraben, antimicrobials, and naphthalenes. Analytes with a ratio below 10 generally had lower ICCs.
Two exceptions sharpen the interpretation. Benzophenone-8 had a median-to-LOD ratio below 10 yet ICCs above 0.3 at all three sites, which the authors suggest may reflect low-level but near-continuous exposure. Conversely, several phthalate metabolites had high concentrations relative to their LODs but low ICCs, which the authors attribute to sporadic high-intensity exposures rather than steady ones. Concentration alone therefore does not determine reliability.
Detection frequency was also inconsistent over time
For the 56 analytes detected in 20-60% of samples, kappa statistics measuring agreement of detection across the three timepoints ranged from -0.22 to 0.53. As with the ICCs, agreement was generally lower when detection was lower, with exceptions among some phthalate metabolites. Benzyl paraben, several phthalate metabolites, several insecticides and organophosphate esters, and 3,4-dichloroaniline showed consistently low kappa values (below 0.2) at all three sites.
This secondary analysis matters because it addresses a different question from the ICC. An ICC describes how well concentrations track each other; kappa describes whether a sample is above or below the detection threshold at all. A chemical can be unreliable on one measure and not the other, so the two should not be substituted for each other when judging whether one sample suffices.
What the three-site design can and cannot settle
The sample is 90 participants across three Environmental influences on Child Health Outcomes Cohort sites, with three samples each. The sites differed in sampling windows: New York samples were mostly early-to-mid pregnancy, Puerto Rico mostly mid-pregnancy, and Illinois spanned all three trimesters, with the widest collection window. Because ICCs were estimated within each site rather than pooled, the site-specific estimates rest on small groups and should be read as site-level descriptions rather than national averages.
The authors also note that all samples were measured in randomized batches with the same assay method, and that sensitivity analyses using specific-gravity-unstandardized values and LOD/√2 replacement produced similar results. The study characterizes variability but does not link it to health outcomes, so it cannot say whether exposure misclassification from single samples would bias a particular association in a particular direction.
For low-ICC analytes, narrow the exposure window
The practical conclusion follows from the ICC values rather than from any single chemical. For analytes with low ICCs or low kappa statistics at all three sites, the authors recommend interpreting a single measurement as exposure near the time of collection, not as an average across pregnancy. For analytes with consistently high ICCs, such as benzophenone-1 and propyl paraben, a single sample carries more information about the broader pregnancy period.
The paper also quantifies why simply adding samples is not a universal fix: it cites a simulation finding that a chemical with an ICC below 0.2 would need more than 35 samples per person to reduce bias toward the null in health-outcome associations to below 10%. Because that is impractical, the authors list alternatives: a validation sub-study with repeated measures and measurement-error correction, timing collection to critical windows of susceptibility, within-subject pooling of multiple samples, or biospecimens reflecting longer-term exposure such as hair.
Treat benzophenones and bisphenol F as opposite cases
The most decision-relevant split in this paper is between analytes that held their ranking across pregnancy and those that did not. Benzophenone-1 and propyl paraben were the only analytes with ICCs above 0.5 at every site, and benzophenone-8 stayed above 0.3 despite concentrations near the detection limit; these are the cases where a single pregnancy sample is most defensible as a broader exposure marker. Bisphenol F, with ICCs below 0.1 at every site, is the clearest case where one sample should not be read as a pregnancy-wide measure.
Between those poles, the site-specific results argue against importing a reliability estimate from one cohort to another. A reader planning a study should check the ICC for the specific analyte and population at hand, and where it is low, either narrow the claimed exposure window or build repeated measures into the design rather than assuming one sample covers gestation.
Sources used in this answer
Intra-individual variability of urinary non-persistent chemical concentrations during pregnancy in the ECHO Cohort.
Across 90 pregnant participants in three ECHO Cohort sites, ICCs for 58 frequently detected urinary analytes ranged from -0.12 to 0.91 and varied by site; only benzophenone-1 and propyl paraben were consistently above 0.5 and bisphenol F consistently below 0.1, implying that for most non-persistent chemicals a single pregnancy sample better represents exposure near collection time than the whole pregnancy.