Development of a Parsimonious Design for Optimal Classification of Exclusive Breastfeeding

Abstract

A deuterium oxide dose-to-mother (DTM) technique is used to determine if an infant is exclusive breastfeeding (EBF). However, the DTM method is intensive, requiring seven paired mother-infant samples during a 14-day study period. The purpose of this study was to develop a field-friendly protocol. Data from 790 mother-infant pairs from nine countries were analyzed using a Markov chain Monte Carlo method with Stan. The data were split into (i) model building (565 pairs) and (ii) design evaluation (225 pairs). EBF classification was based on a previously published cut-off for nonmilk water intake. Classification based on the full design was the reference (gold standard classification). The receiver operating characteristics of parsimonious designs were used to determine an optimal parsimonious classification method. The best two postdose windows (days 7-9 and 13-14) yielded optimal categorization with similar performance in the design evaluation data. This postdose two-sample design provided 95% sensitivity and specificity when compared with the full design.

Conflict of interest statement

The authors declared no competing interests for this work. As an Associate Editor for CPT: Pharmacometrics & Systems Pharmacology, Stephen Duffull was not involved in the review or decision process for this article.

© 2019 The Authors CPT: Pharmacometrics & Systems Pharmacology published by Wiley Periodicals, Inc. on behalf of the American Society for Clinical Pharmacology and Therapeutics.

Figures

Figure 1

Graphic illustration of exclusive breastfeeding (EBF) or non‐EBF judgment criteria. The probability scale (Pr) is the probability of being non‐EBF (i.e., Pr non‐EBF = Y). Study pair 1 is classified as EBF (i.e., the probability of being non‐EBF is essentially 0), study pair 2 is also classified as being EBF (because the probability of being non‐EBF is  0.9); and study pair 4 is non‐EBF (with Pr = 1). Rs, water intake rate from sources other than breastmilk.

Figure 2

Individual visual predictive check for the best final model evaluation. Open circles are the observations. The solid lines (median, 2.5% and 97.5% quantiles) are the model‐predicted response. D2O, deuterium oxide; EBF, exclusive breastfeeding.

Figure 3

Proportion of the imputed observations vs. the actual measured observations in the imputed full data set. Light gray bars represent the actual (measured) observations; dark gray bars represent the imputed (simulated) observations.