Annex: The Keneba pregnancy supplementation study

Annex The Keneba pregnancy supplementation study T.J. Cole,1 F.A. Foord, M. Watkinson, W.H. Lamb, & R.G. Whitehead Introduction The village of Keneba in the Gambia (West Africa) has been closely studied by the UK Medical Research Council's Dunn Nutrition Unit since 1974. Before that time the village had been regularly monitored, starting in 1949, and provides a complete census and accurate dates of birth for all the inhabitants except the oldest. Keneba is a rural subsistence farming commu- nity, predominantly of Mandinka people, with a pop- ulation of about 1000 in 1980. The prevailing climate is dry and gets progressively hotter from November to June, with a rainy season from July to October. The onset of the rainy season leads to a substantial energy deficit in all the adults, as a result of depletion of food supplies and the high energy cost of preparing the land for planting. Consequently the incidence of low birth weight among babies and weight loss in adults is appreciably increased during the rainy season, particularly in the months of August and September. Objective. To test the effect of dietary supplementa- tion during pregnancy on birth weight and gestation- al age. Population. All Keneba mothers of known parity giving birth to a singleton live-bom baby during the period April 1976 to May 1984, with the exception of the months January to April 1978 when the princi- pal investigators were on leave. Design This was an unmatched case-control study, with his- torical controls. The dietary supplement was introdu- ced in May 1980, so that babies bom prior to June 1980 (i.e. with less than 1 month's supplement) were controls, while those bom from June 1980 onwards were cases. The majority of mothers had more than one baby in the study period (with a median of 2, and a maximum of 4). ' MRC Dunn Nutrition Unit, Cambridge, England. Infant variables 1. Birth weight (kg) was obtained within 24 hours of birth by a trained midwife, and recorded to 10 g using a Salter spring balance and tarred sling. 2. Gestational age (weeks) was assessed, from 1978 to 1984, by the method of Dubowitz within 5 days of delivery. Over the period of the study six different paediatricians did the assessment. 3. Head circumference (mm) was measured, from 1978, within 24 hours by a trained midwife, using a fibreglass tape. Maternal variables 1. Age (years) was accurate to the nearest year or better. 2. Parity (count) indicated the number of live births (including the current one) plus last trimester stillbirths. 3. Height (mm) was obtained before pregnancy in a regular clinic for women at risk of pregnancy. 4. Weight (kg) was measured to the nearest 50 g using a Salter balance. 5. MUAC (mm) was measured on the left arm with a fibreglass tape. 6. Triceps skinfold (mm) was measured on the left arm with a Holtain skinfold calliper. Timing of variables. Matemal anthropometric meas- urements were obtained at 6-weekly intervals during pregnancy. For convenience three periods were considered: pre-pregnant, mid-pregnant and late- pregnant. These were definite in terms of the days prior to delivery, with period 1 (pre) being from -365 to -280 days, period 2 (mid) -160 to -120 days, and period 3 (late) -40 to -1 days. Sample size. A total of 379 babies were seen during the study period, 182 controls and 197 cases from 187 mothers. A subset of 288 babies, 103 controls and 185 cases, had gestational age measured. Training and equipment. All the balances and stadi- ometers were regularly calibrated, and the measure- ments were carried out by trained field workers supervised by Dr Andrew Prentice. Pre-test study. No formal pre-test study was carried out, but the data collection had operated for two 72 WHO Bulletin OMS: Supplement Vol. 73 1995 Annex years prior to the introduction of dietary supple- ments. Completeness of response. All pregnant women were enrolled into the study, and all but 5 babies in hospi- tal (3 controls, 2 cases) had their birth weights recorded. Adherence to protocol. The supplement, consisting of biscuits and tea, with mean energy intake 430 kcal/d, was provided in a purpose-built building and closely supervised by field workers. Data coding, entry and clean-up. The data were coded by Dr Andrew Prentice, punched by a trained and experienced data processor in Cambridge, and analysed by Dr Tim Cole using the statistical pro- gram Genstat 4. Data checks were carried out by studying histograms and scatter plots of the data, and as a result some aberrant data points, including two miscoded birth weights, were identified and corrected. Overall quality. The quality of the data is believed to have been high throughout the study, with close supervision at all stages of the data collection. The main deficiency of the study was the use of historical controls, which was necessary both for ethical rea- sons and to avoid compliance problems in the con- trol group. However, there was no trend in birth weight during the four control years, so the results are likely to be valid. A second minor concem is the definition of the period (July to January) over which the supplement affected birth weight. This was obtained from the data to maximize the significance of the supplement season interaction, rather than using an external defi- nition for the timing of the rainy season. On the other hand, there was no reason a priori why the rainy season and the period when the supplement was effective should coincide, so that using a data- derived definition was the only realistic solution. Re-analysis Methods. The new analysis is based on the 379 Gambian singletons of mothers with known parity described in Prentice et al. (2). The aim of the analy- sis is to relate the two outcome variables, birth weight and head circumference, to maternal anthro- pometry during pregnancy, using linear regression analysis. Birth weight. All the regression analyses include binary variables adjusting for sex, dietary supple- ment, season, season-supplement interaction, and maternal parity, coded for parity 1 and parity 10 or more. The effects of maternal anthropometry are assessed by adding them to the regression. None of the pre-pregnancy (period 1) measurements are of value, not even the mid-pregnancy (period 2) MUAC or triceps skinfold measurements. However, height, mid-pregnancy weight (weight 2) and late-pregnancy weight (weight 3) are all sig- nificant when included in the model simultaneously, with weight 2 having a negative coefficient. If the coefficients for weight 2 and weight 3 were exactly equal and opposite, they could be replaced in the regression by the difference between them, i.e., weight gain. As it is, weight 3 has a slightly larger coefficient than weight 2, suggesting that there is extra size information in weight 3. This shows that both third trimester weight gain, derived as weight 3 less weight 2 and weight 3, and adjustment for gesta- tional age does not materially affect the conclusions. Head circumference. Head circumference is signifi- cantly affected by sex, parity 1 and maternal height. Weight 2 is better than height, and weight 3 is better than weight 2. None of pre-pregnant (period 1) anthropometry variables is significant, not even MUAC 2 and triceps skinfold 2. Thus the important maternal anthropometry vari- ables are, in decreasing importance, late-pregnancy weight, mid-pregnancy weight, and height. The main factor influencing the change in weight from mid- pregnancy to late-pregnancy is of course birth weight, so adjusting for birth weight might remove the need for maternal weight. Introducing birth weight to the regression makes late-pregnancy weight insignificant, but both the sex and parity effects remain significant, showing that they are not mediated by birth weight. Adjusting for gestation does not affect things materially, and adding birth weight causes weight 3 and gestation to become significant. This suggests that the impact of maternal anthropometry on head circumference is mediated through the birth weight, or else that the factors that influence maternal anthropometry also affect birth weight. Conclusions The results show that birth weight is directly related to height and weight in late pregnancy, and also to weight gain in the third trimester. This holds whether or not gestational age is taken into account. The only effect of adding gestation age to the regression is to weaken slightly the maternal anthropometry associa- tions. Part of the maternal weight gain seen in late pregnancy is due to the weight of the fetus, so it is not surprising that the two should emerge as signifi- cantly related. However, there are also separate and significant effects of height and weight, representing maternal size. The fact that their regression coeffi- cients are both positive, so that they operate in the WHO Bulletin OMS: Supplement Vol. 73 1995 73 Annex same direction, means that matemal size rather than shape is important. This implies that measures of weight-for-height such as body mass index (BMI) are likely to be less effective than weight alone for predicting birth weight. Matters relating to head circumference are more difficult. Although height, mid-pregnancy weight and late-pregnancy weight are all significantly re- lated to head circumference when considered separ- ately, the last of them overrides the other two. In addition, mid- and late-pregnancy weight act in con- cert, both with positive coefficients, so there is no evidence for a weight gain effect. Head circumference adjusted for birth weight removes the effect of late-pregnancy weight, show- ing that head circumference and birth weight are influenced to a similar degree by matemal anthro- pometry. The data set used in this analysis is not large, and hence is of low power to investigate the influ- ence of matemal anthropometry on fetal growth. In addition, although the relationships are significant, they are too weak to be of value for prediction pur- poses. This is confirmed by the results of the WHO meta-analysis applied to the Gambian data, where the positive predictive values of matemal anthro- pometry on pregnancy outcome are all small. In detail, the WHO meta-analysis results dis- agree with those given here, in that they show BMI to be a better predictor than weight. This is most likely due to the differing outcome measures used in the two analyses. Birth weight and head circumfer- ence are continuous measurements, and are likely to be related to maternal size across the whole spectrum of infant size. Conversely, outcome measures such as intra-uterine growth retardation, low birth weight and prematurity all focus on the low end of the birth weight distribution. The other important difference between the anal- yses is the effect of season. The regression analysis includes a seasonal adjustment, whereas the WHO analysis does not. Thus pre-pregnant BMI, which has the highest positive predictive value, is a proxy for season. Women are substantially thinner in the rainy season, when low birth weight is common, than in the dry season, and indeed it was this observation that motivated the original study. The results of the study were published, provi- sionally in 1983 (1), and finally in 1987 (2), demon- strating a positive effect of supplementation, of about 220 g during the rainy season but not during the dry season. References 1. Prentice AM et al. Prenatal dietary supplementation of African women and birthweight. Lancet, 1983, 1: 489-492. 2. Prentice AM et al. Increased birthweight and pre- natal dietary supplementation of rural African women. American journal of clinical nutrition, 1987, 46: 912-925. 74 WHO Bulletin OMS: Supplement Vol. 73 1995