Early Initiation of Breastfeeding and the Risk of Infant Diarrhea in Rural Egypt

Protocol Approvals

The protocol for this study was approved by institutional review boards of the US Naval Medical Research Unit-3, the US National Institute of Child Health and Human Development, and the World Health Organization.

Setting

The cohort was assembled in two villages located in the Abu Homos district of Beheira Governorate, ∼40 km from Alexandria. Abu Homos is a rural area, and most adults there are engaged in agricultural occupations.

Assembly of the Cohort

In January 1995, a house-by-house census of the two study villages was undertaken. At this time, households were characterized according to selected sociodemographic characteristics as well as to various aspects of household hygiene, including water use and facilities for defecation. Neonates ≤28 days were enrolled for the study beginning in February 1995, after ascertaining that they had neither major congenital abnormalities nor current illnesses serious enough to require hospitalization, and after obtaining written informed consent from their parents (for parents who were illiterate, the form was read to the parent in the presence of a witness; if the parent consented to the protocol, this consent was documented by a fingerprint). Because the focus of the present analysis was to assess the impact of early initiation of breastfeeding, we included only those neonates who had ever been breastfed. Moreover, because patterns of breastfeeding and infant morbidity may be substantially different for twin births, we included only singleton births in our analysis.

Enrollment of newborns ended in September 1997. A total of 211 neonates were identified, of whom 204 were singleton births. Only 3 (1%) of these neonates had never breastfed. Among the remaining 201 neonates, all of whom met the eligibility criteria cited above, the 3 for whom the day of onset of breastfeeding was unknown were excluded, leaving a total of 198 initially breastfed infants for analysis. The median age at enrollment for this analyzed cohort was 11 days (interquartile range, 7–16 days).

Surveillance of the Cohort

After enrollment each infant was visited at home twice weekly. At each visit the mother of the infant was interviewed about the passage of loose or liquid stools by the infant on each day since the last scheduled visit. Diet was monitored by queries about whether the infant still received breast milk, as well as about what additional foods and liquids had been introduced into the infant's diet since the last visit. These home visits also enabled us to monitor all deaths and outmigrations occurring in the study cohort. Of the scheduled twice-weekly visits, 95% were completed successfully.

Surveillance continued through February 1998. Before inspection of the data, we decided to limit the primary analysis to the first 6 months of life, because we considered it unlikely that early initiation of breastfeeding per se would exert a preventive effect on diarrhea beyond this age. Of the cohort, 190 (96%) were followed until 6 months of age. For 4 infants, the surveillance phase of the study stopped before they had reached 6 months of age. One infant died, whereas an additional 3 infants were lost to follow-up.

Study Definitions

We established definitions of key variables before analyzing the data. A “diarrheal day” was defined as 1) passage of three or more loose or liquid stools in any 24-hour period (for breastfed infants, this also required a statement by the mother that the stools had become more frequent or less formed than usual for the infant); or 2) passage of at least one loose or liquid stool with visible blood in a 24-hour period. An episode of diarrhea was bounded by three or more consecutive nondiarrheal days.

A child was considered to be breastfed if breast milk constituted any portion of the child's diet. Exclusive breastfeeding denoted diets consisting of breast milk only, without other solids or liquids. All other breastfed infants were considered “partially breastfed.” “Early initiation” of breastfeeding referred to breastfeeding that started on the day of birth or during the next 2 days of life, because breast milk contains an appreciable amount of colostrum only during this interval.^{5 ,6} Late initiation indicated breastfeeding that began after the third day of life.

Analytic Strategies

In simple two-group comparisons, the Student t test (or the Mann-Whitney U test when parametric assumptions were not fulfilled) was used for dimensional variables, and the χ² test (or Fisher's exact test for dichotomous variables when data were sparse) was used for categorical variables. To compare the duration of breastfeeding in different groups, we constructed Kaplan-Meier survival curves and compared these curves statistically with the log-rank test.

The incidence rate of diarrhea for an infant was calculated by dividing the number of episodes of diarrhea by the number of person-days at risk. The latter was computed as the potential number of days of follow-up, minus the number of days with missing information and the number of days included in the diarrheal episodes after the first day of each episode. We excluded days after the first day of each episode from the denominators because the child was not at risk for a new episode during these days.

To compare incidence rates of diarrhea in different groups of infants, we used multiple logistic regression models. In these models, we took each day at risk for diarrhea (eg, the days included in the denominators for calculation of incidence rates) as the unit of analysis. With these units, we fitted models in which the occurrence or nonoccurrence of a first day of a new episode of diarrhea was the dichotomous dependent variable, and timing of breastfeeding initiation, together with relevant covariates were fitted as independent variables. Because we classified each day of follow-up in this manner, it was possible to specify time-varying independent variables, such as current diet, in these models. Moreover, because different days of follow-up for each child were not statistically independent entities, it was necessary to adjust for the intercorrelations of repeated observations on different days for individuals with generalized estimating equations (GEE)⁷ (SAS version 6.12, PROC GENMOD, SAS Institute, Inc. SAS/STAT software, Changes and Enhancements through Release 6.12, SAS Institute, Inc, Cary, NC). Coefficients of independent variables in these models were exponentiated to estimate odds ratios (ORs) values for the associations between independent variables and the occurrence of diarrhea on each day, and SE values of these coefficients were used to calculate the 95% CI for each OR. Because the risk of a new diarrheal episode on any particular day was rare (2% per day of follow-up), these OR values are interpretable as ratios of relative rates (RR), and are referred to as such in the text.⁸ We present data for GEE models fitted with exchangeable variance–covariance matrices for the intercorrelation of daily observations in individuals. However, refitting the models with both first-order autoregressive and unstructured matrices gave virtually identical estimates, indicating that the findings were robust with respect to the assumed pattern of intercorrelation of observations. All statistical tests were interpreted in a two-tailed manner to estimate P values and to compute confidence intervals; P values <.05 were considered significant.