425 women were randomly assigned—212 to the breastfeeding group and 213 to the formula feeding group. The characteristics of the study population have been previously described in detail.1
Briefly, median age was 23 years, and 325 (76%) were married. The women had few previous births, with a median parity of one (IQR zero to two). The median age at first coitus was 17 years, and the women had a lifetime median of three sexual partners. The median haemoglobin was 109 g/L and 207 (51%) were anaemic (haemoglobin <110 g/L). The median CD4 count was 407 cells/mL (IQR 268–528). CD4 cell counts were less than 200 cells/mL in 47 (12%) women, 200–499 in 218 (57%), and greater than 500 in 116 (31%). The median HIV-1 viral load was 42 700 virions/mL (IQR 10 245–161 884). HIV-1 subtyping was done for 320 women, of whom 225 (70%) had subtype A, 65 (20%) had D, 22 (7%) had C, 1 (0·3%) had G, and 7 (2%) had recombinant subtypes. The median vitamin A concentration was 0·86 µmol/L and 216 (70%) women had concentrations less than 1·05 µmol/L. Women who were randomly assigned to breastfeed or formula feed had closely similar baseline and delivery characteristics, as previously described ().1
Characteristics of women in breastfeed and formula feed groups at baseline
Of the 212 women in the breastfeeding group, one died during pregnancy, eight were lost to follow-up before delivery, and six had no vital status information after delivery (). Of the 213 women in the formula feed group, five were lost to follow-up before delivery and eight had no vital status information after delivery. Of the remaining 397 women, 24 died; 15 during the first year and nine during the second year of follow-up. Our analysis dataset includes only those women who were alive at the time of delivery and for whom information about vital status after delivery was available. Thus, 197 women in the breastfeeding group and 200 in the formula group were included in the analysis of the relation between lactation and maternal death.
Of these 397 women, 39 (20%) of 197 women in the breastfeeding group and 33 (17%) of 200 in the formula feeding group were lost to follow-up. The most usual reason for loss to follow-up was infant death (21 [54%] women in the breastfeeding and 19 [58%] in the formula feeding group, p=0·8), because those who lost an infant were no longer requested to come to the research clinic, although they were given the option of doing so. Most other women who were lost to follow-up changed residence or left Nairobi for their rural homes. Median time of lost to follow-up was 10·2 months in the breastfeeding group and 8·6 months in the formula group (p=0·3). Women classified as lost to follow-up were not substantially different from those who completed the study with respect to age, parity, or any baseline physical examination or laboratory test characteristics. However, those lost to follow-up were of lower socioeconomic status because they were less likely to live in a home with more than one room (22% [16/92] vs 42% [136/324], p=0·002), less likely to have flush sanitation (65% [47/72] vs 77% [251/325], p=0·03), more likely to have a shared toilet (88% [63/72] vs 73% [236/325], p=0·008), and less likely to have a refrigerator (zero vs 5% [17/325], p=0·05) than those who completed the study. A stratified analysis of correlates of loss to follow-up in each group had similar results, except that not all differences in socioeconomic status indicators were significant.
Compliance with breastfeeding was defined by any use of breastmilk, and that with formula feeding by total absence of breastfeeding. Compliance with the randomly assigned feeding practice was 189/197 (96%) in the breastfeeding group and 141/200 (71%) in the formula group (p=0·001). Of compliant women in the breastfeeding group, 95% were breastfeeding at 3 months, 90% at 6 months, and 80% at 12 months. On the basis of maternal report, the proportion of infants in the breastfeeding group who received 50% or more of their feeds as breastmilk was 79% at 6 months, 32% at 12 months, and 2% at 18 months. Median duration of breastfeeding was 17 months (range <1 week to >24 months), and median age of the infants at introduction of weaning foods was 3·8 months.
18 of the 197 women in the breastfeeding group and six of the 200 women in the formula group died during follow-up. The cumulative probability of death was higher in the breastfeeding group than in the formula group (, log rank test, p=0·009). At 24 months, the cumulative mortality was higher in women in the breastfeeding group than in those in the formula group, with Kaplan Meier survival analysis (11% vs 4%, p=0·02). The difference in the cumulative mortality rates in the two groups was significant by as early as 6 months after delivery (p=0·02). 22% of the overall excess mortality had taken place by 6 weeks, 55% by 6 months, 72% by 1 year, and 100% by 18 months. Overall, there was a three-fold increased risk of dying for women in the breastfeeding group (RR 3·2, 95% CI 1·3–8·1, p=0·01), with Cox proportional hazard regression models. 69% of maternal deaths in the breastfeeding group were attributable to breastfeeding. In Cox regression models, the association between random assignment to the breastfeeding group and maternal death persisted after log plasma viral load and log CD4 count at enrolment were controlled for (4·7, 1·8–12·0, p=0·001).
Mortality of mothers in breastfeeding and formula feeding groups
Of the women who died, ten (48%) had CD4 counts of less than 200 cells/mL at enrolment and nine (43%) had counts of 200–499 cells/mL. Viral loads were high, with 12 (71%) of 17 having enrolment viral loads greater than 100 000 virions per mL, including four with loads more than 1 000 000. There was a significant association between maternal death, and CD4 counts and viral load at enrolment. Compared with women who had CD4 counts of greater than 500 cells/mL the relative risk of death for those with counts of 200–499 cells/mL was 2·4 (0·5–11·2, p=0·3) and 14·7 (3·2–67·4, p=0·001) for those with counts less than 200 cells/mL. Women with a viral load at enrolment of more than the median of 42 700 virions per mL had an 8-fold increased risk of dying (8·0, 1·8–34·8, p=0·006).
Precise information about causes of maternal death was missing because of restricted availability of diagnostic tests and reliance on verbal autopsies for deaths occurring outside Nairobi. However, 15 (63%) of 24 women who died were classified as having AIDS on the basis of clinical diagnoses of Kaposi’s sarcoma, wasting syndrome, tuberculosis, and cryptococcal meningitis. An additional five women (21%) had illness possibly related to HIV-1 infection (eg, pneumonia, chronic diarrhoea, or chronic cough) at the time of death. For the remaining four women, no information about cause of death was available.
The difference between weight shortly after delivery and weight at 6 months postpartum has been used as a measure of the metabolic demands of breastfeeding.12
We compared every women’s weight at her earliest postnatal visit, between 0·5 and 3 months, with weight at the latest visit, between 5 and 9 months after delivery. Women in the breastfeeding group lost more weight during this postpartum period than did those in the formula feeding group (mean 0·17 vs
0·00 kg per month, 95% CI for difference 0·02–0·33, p=0·03). There was a significant relation between weight loss during follow-up and mortality, with a RR of 3·4 (2·0–5·8) for each kg lost per month. The median weight loss per month was 0·7 kg for women who died and 0·05 kg for women who remained alive (p=0·03). After adjustment for weight change during follow-up, there was still an association between breastfeeding and maternal death (2·9, 1·1–7·6).
Of the 24 mothers who died during follow-up, one had an infant whose mortality status was unknown, 12 had infants still alive at the latest follow-up, and 11 had infants who died during follow-up. Of these 11 infants, six died after the mother’s death, one died at the same time as the mother, and four died before the mother. Maternal death was associated with infant death overall (3·1 for infant death, 1·6–5·9, p<0·001). There was an even stronger association between maternal death and subsequent infant death (5·6, 2·4–13·1, p<0·001), even after controlling for infant HIV-1 infection status (7·9, 3·3–18·6, p<0·001).