Sample Size

Sample size was calculated based on estimates of the proportion of babies breast-fed (partially or exclusively) at 3 months, the effect of the intervention on increasing that proportion, and the effect of breastfeeding duration on reducing infectious morbidity. From a prior Belarussian Ministry of Health survey, we estimated that 50% of women who initially breastfed at the control (non-intervention hospitals) would still be breastfeeding (to any degree) at 3 months. Based on the available evidence concerning the effectiveness of the individual components of the BFHI, we anticipated that the intervention would reduce breastfeeding discontinuation by 3 months from 50% to 35%, i.e., that 65% of mothers exposed to the intervention vs 65% of mothers not exposed to the intervention would still be breastfeeding their infants (to some degree) at 3 months. Three months of any breastfeeding was chosen as the primary basis for calculating breastfeeding prevalence based on the data of Howie et al5; in initially breast-fed infants who were weaned at 13 weeks vs those breast-fed any degree and for at least 13 weeks, the relative risk of gastrointestinal infection (the primary outcome) associated with early weaning was approximately 2.

Based on evidence available at the time we planned the trial, we estimated that approximately 60% of Belarussian infants would experience at least one gastrointestinal infection in the first year oflife. Given a relative risk of2 and the (control) 50:50 split ofbreastfeeders to nonbreastfeeders at 3 months, this overall 60% figure translated into proportions of infants experienceing one or more episodes of gastrointestinal infection of approximately40%and 80%,respectively [(.50)(.40)+(.50)(.80)=.60]. If the experimental intervention was successful in changing the split of breastfeeders to nonbreastfeeders from 50:50 to 65:35, then the corresponding overall proportion of infants in the experimental group with one or more episodes of gastrointestinal infection was expected to be (.65)(.40)+(.35)(.80)=.54.

Because the design is based on randomizing clusters rather than individuals, we anticipated the unit of statistical analysis would be the hospital/polyclinic study site. We therefore used sample size techniques for paired cluster randomization,46following an approach similar to that employed in a large community-based study of smoking cessation.47 In addition to the elements involved in usual sample size estimates, factors to be considered in this context are the variation in outcome between unpaired clusters (study sites), the effect of matching in reducing this variation, and the size ofthe clusters (number of subjects per site), which was expected to vary considerably by site.

Under the null hypothesis, the average proportion ofinfants with one or more episodes of gastrointestinal infection was expected to be .60. We assumed three different degrees ofvariability between (unpaired) study sites for this proportion: 95% would be expected to have values between .50 and .70 (high variability), .52 and .68 (moderate variability), or .54 and .66 (low variability). These 95% ranges correspond to approximate standard deviations (o) of .05, .04, and .03,respectively, assuming an approximately normal underlying distribution. In addition, we considered three different degrees of "success" in pairing, as reflected by the within-pair correlation coefficient, p: ineffective pairing (p=0), moderately effective pairing (p=.4), and highly effective pairing (p=.7). Even if matching was ineffective, the relative efficiency of a paired analysis to an unpaired analysis would still be .90.48

Assuming 500 infants enrolled at each maternity hospital, a design using 15 pairs of study sites would provide a power of greater than 80% to detect a significant difference between the two groups at a 2-sided a -level of 0.05, even assuming a "worst-case scenario" of high variability (o=.05) between (unpaired) sites and totally ineffective pairing (p=0), provided that the risk of one or more episodes of gastrointestinal infection was in fact reduced to halfby prolonged (>3 months) breastfeeding. The design also provided power of over 80% to detect a halving of the prevalence of 2 or more respiratory infections from an assumed level of 60%, for o <04, and only marginally effective matching (p> 0.25). The design ensured well over 90% power to detect an increase in prevalence of any breastfeeding from 50 to 65% at 3 months and from 25 to 30% at 6 months, and an increase in predominant breastfeeding from 35 to 50% at 3 months. Owing to the infrequency of hospital admission and (especially) death for infectious illnesses, we anticipated that even this large sample size (500 infants per site x 30 sites = 15,000) would be insufficient to detect differences in hospitalization rates or death from infection.

In order to ensure that 15 hospital pairs would be included in the final study sample, we randomized 17 pairs to provide a margin of security against withdrawals or unforeseen logistical problems at a few hospitals. As it turned out, two of the maternity hospitals (one experimental and one control) refused to carry out their allocated intervention following randomization. Moledechno Maternity Hospital (Minsk region) was allocated to the control intervention and refused to stop its efforts to become more baby-friendly. Luninets Maternity Hospital (Brest region) claimed to be unable to physically arrange its maternity wards to accommodate rooming-in and was therefore unwilling to accept its assignment to the experimentalintervention. We then re-paired the remaining hospitals from these two original pairs, Soligorsk (experimental, Minsk region) and Stolin (control, Brest region).

New Mothers Guide to Breast Feeding

New Mothers Guide to Breast Feeding

For many years, scientists have been playing out the ingredients that make breast milk the perfect food for babies. They've discovered to day over 200 close compounds to fight infection, help the immune system mature, aid in digestion, and support brain growth - nature made properties that science simply cannot copy. The important long term benefits of breast feeding include reduced risk of asthma, allergies, obesity, and some forms of childhood cancer. The more that scientists continue to learn, the better breast milk looks.

Get My Free Ebook


Post a comment