Study population. This study was conducted in the 70 villages of the International Centre for Diarrhoeal Disease Research Centre for Health and Population Research (ICDDR,B) Maternal Child Health Family Planning Project (MCH-FP project) intervention area located 45 km southeast of the capital city of Dhaka in the Matlab thana of rural Bangladesh. Rice, jute and fish are the main agricultural products of this river delta region. ICDDR,B has maintained demographic surveillance of the population since the 1960s and over the years has provided preventive and curative health services in the community in conjunction with ongoing health research projects (Fauveau 1994). The existing network of 80 MCH-FP community health workers and logistical support services was utilized for birth detection, capsule delivery, compliance monitoring and motivational activities during the supplementation study. The MCH-FP community health workers routinely visit all households with a married woman of childbearing age twice a month and provide family-planning services, immunizations and treatment or referral to an ICDDR,B clinic for a variety of health problems. Women participating in the supplementation study received these services as usual. The additional study-specific visits at 2 wk, 3, 6 and 9 mo postpartum were conducted by a separate team of trained field workers.
All women registered in the MCH-FP area who delivered a live infant in the service area between June 14–August 29, 1994, were eligible for the trial. After the report of a live birth was received, an enrollment visit was scheduled while infants were in the eligible age range of 7–21 d. Data collected from the MCH-FP records indicated that 368 women delivered live infants during the enrollment period. Women (300) remained eligible for the trial after excluding those who delivered their infants outside of the service area (n 5 25), those with infants .21 d-old (n 5 35), those whose infants had died (n 5 5) and those with severely ill infants (n 5 3). Of the 300 eligible families, 48 (16%) refused to participate, and 32 births (11%) were not detected due to the absence of community health workers, a time lag in birth reporting or for other unknown reasons. We successfully recruited 220 (73%) of the eligible women into the study, 218 with single and 2 with twin births. This study was approved by The Johns Hopkins University School of Hygiene and Public Health Committee on Human Research and the Ethical Review Committee at ICDDR,B. Informed consent was obtained from all study participants in accordance with guidelines at The Johns Hopkins University and ICDDR,B. le-blinded. Because the cultural practices in the region prohibit some women from leaving their homes during the first few weeks postpartum, women were given a choice of completing the enrollment visit at 2 wk (61 week) postpartum either at the central Matlab clinic or in their homes. During the enrollment visit, they were assigned a study identification number that randomly allocated them to one of three treatment groups and to a follow-up schedule that determined the location of their follow-up visits at 3, 6 and 9 mo postpartum. During clinic visits, women provided a full milk sample and a blood sample. During home visits, women provided a casual milk sample, but no blood sample (details described below). Before beginning the study, individual treatment codes and follow-up schedules were assigned to a sequence of identification numbers in blocks of 18 using a random number table (Smith and Morrow 1991). Each block contained all possible combinations of three treatment groups (vitamin A, b-carotene or placebo) and six follow-up schedules. The six follow-up schedules assigned women to complete clinic visits at either 0.5 and 3, 0.5 and 6, 0.5 and 9, 3 and 6, 3 and 9 or 6 and 9 mo postpartum. When they were not assigned to a clinic visit, the women completed their follow-up visits at home. At each visit, half of the women in each of the three treatment groups completed a home visit and the other half completed a clinic visit.
Counts from returned capsule packs and spot checks in the field. Data collection. Serum retinol and the modified relative dose-response test. On days of clinic visits, women came to the central Matlab clinic and received an oral dose of 3,4-didehydroretinyl acetate (8.8 mmol) for the determination of their MRDR ratio (Tanumihardjo et al. 1996). This was immediately followed by a high-fat cookie and tea with lemon. During the waiting period, women were given other snacks low in vitamin A. A blood sample was collected from the antecubital vein 5 h after dosing. At the 6-mo visit, infants were given an oral dose of 3,4-didehydroretinyl acetate (5.3 mmol) for the determination of their MRDR ratio (Tanumihardjo et al. 1996). Mothers were encouraged to breast feed their infants immediately after dosing and on-demand throughout the waiting period. A blood sample was collected from the femoral vein 5 h after dosing. After collection, the blood samples were slowly expressed into a foil-covered 8-mL glass vial, allowed to clot in the dark at room temperature for;1 h and then centrifuged. Serum (500 mL aliquots) was stored at 220°C for up to 3 mo at the field site, transported to Dhaka on ice, stored at 270°C, then transported in liquid nitrogen to Baltimore, MD, and stored at 270°C until analysis.
Statistical analysis. The sample size for the study was calculated to detect a difference in the proportion of individuals with low vitamin A status between a group who received active supplements (either vitamin A or b-carotene) and the placebo group who did not. Therefore, the data were analyzed by comparing separately the values in the vitamin A to the placebo group and the values in the b-carotene to the placebo group. For categorical variables, comparisons were made using x2 analyses. For continuous variables, comparisons were made using two-sided t tests. Because the MRDR ratios were skewed to high values, the data were log transformed to normalize distributions prior to statistical testing, and geometric means are reported. Although P values ,0.05 are considered as statistically significant, values #0.10 are reported to indicate potentially important trends in the data. Statistical analyses were conducted using SPSS 7.5 (SPSS Inc., Chicago, IL).