Statistical issues in analyzing 24-hour dietary recall and 24-hour urine collection data for sodium and potassium intakes

Dietary recalls and urine assays provide different metrics for assessing sodium and potassium intakes. Means, variances, and correlations of data obtained from these two modes of measurement differ. Pooling of these data is not straightforward, and results from studies employing the different modes may not be comparable. To explore differences between these metrics, the authors used data from the Trial of Nonpharmacologic Intervention in the Elderly (TONE), which included repeated standardized 24-hour dietary recalls and 24-hour urine collections administered over 3 years of follow-up, to estimate sodium and potassium intakes. The authors examined data from 341 control participants assigned to usual care that were collected between August 1992 and December 1995. Dietary recalls yielded estimates of sodium intake that averaged 22% less than those from urine assays and estimates of potassium intake that averaged 16% greater than those from urine assays. Sodium intake estimates were less repeatable (r = 0.22 for diet; r = 0.30 for urine) than potassium intake estimates (r= 0.49 for diet; r= 0.50 for urine), making relations with outcomes more difficult to characterize. Overall, the performance of the two measurement modes was fairly similar across demographic subgroups. Errors in separate estimations of long term sodium and potassium intakes using short term data were strongly correlated, more strongly than the underlying long term intakes of these electrolytes. Because of the correlated measurement error, estimated regression coefficients for linear models including both electrolytes as predictors may be confounded such that the separate relations between these nutrients and outcomes such as blood pressure cannot be reliably estimated by common analytical strategies.