Paper 104 MS

Paediatric and Perinatal Epidemiology 1998, 12, 182198

Traditional nutritional and surgical practices and their effects on the growth of infants in south-west Ethiopia

Department of Community Health, Jimma Institute of Health Sciences, Jimma, Ethiopia, bResearch School of Medicine and cDepartment of Psychology, University of Leeds, Leeds, and dDepartment of Psychology, University of Durham, Durham, UK

Makonnen Asefaa,b, Jenny Hewisonc and Robert Drewettd

a

Summary. A 1-year birth cohort of 1563 infants was seen bi-monthly for the first year of life. They comprised all identified infants born in Jimma town, south-west Ethiopia, in the year starting 1 Meskren 1985 in the Ethiopian calendar (11 September 1992). Growth in infancy is poor in this town, as it is in Ethiopia more generally: mean z-scores for both weight and length were more than 1.5 SD below the median of the NCHS/WHO reference population by 1 year of age, and infant mortality was 115/1000. In this paper we examine the weight gain of singletons in relation to background variables and to traditional nutritional and surgical practices in the families. Confirming work in other areas, sanitation, water supply, the income of the family and the mother's literacy were important determinants of weight gain. Almost all the infants were initially breast fed, and about 80% were still breast fed at 1 year. Many were also given cows' milk from 4 months onwards. Breast feeding had beneficial effects up to 8 months of age, and cows' milk had beneficial effects after 6 months of age. Supplementary feeds of solids and semi-solids were given at appropriate ages, but had no detectable benefit. Water was given inappropriately early, but did no detectable harm. Episodes of diarrhoea, fever or persistent cough each reduced weight gain. Catch-up in weight then took up to 8 months, probably because of the poor nutritional quality of supplementary feeds. The incidences of local traditional operations in the first year were: circumcision 63% in males and 4% in females, uvulectomy 35% and the extraction of milk teeth 38%. Although circumcision had no detectable adverse effect on weight, uvulectomy and milk teeth extraction both reduced weight gain. Address for correspondence: Dr R. F. Drewett, Department of Psychology, University of Durham, South Road, Durham DH1 3LE, UK.

182
Ahed Bhed Ched Dhed Ref marker Fig marker Table marker

# 1998 Blackwell Science Ltd.

Ref start

Ref end

Paper 104 MS
Traditional nutritional and surgical practices Introduction
Ethiopia is one of the three poorest countries in the world, with a GNP per head of about $120.1 Its population is large (estimated at 52.8 million in 19912) and young, with 17.8% under the age of 5 according to the 1984 census.3 Infant mortality in the country as a whole is estimated at 130/1000,1 and the National Rural Nutrition Survey of 1992 showed that 64% of children aged 659 months were stunted (height-for-age more than standard deviations (SD) below the WHO/NCHS reference population median) and 47% were underweight (weight-for-age more than 2 SD below the WHO/NCHS median). Very poor weight gain in infancy has been shown to predict subsequent mortality46 and it is also associated with subsequent cognitive deficits.7,8 Ethiopia is known in the West for the famines in the north of the mid-1980s and the emergency relief they triggered, but it is important to appreciate the great geographical and cultural diversity of the country. Jimma is a regional capital in the south-west in Keffa, about 335 km from Addis Ababa (Fig. 1). Keffa is a coffeegrowing area, and Jimma town is about 1760 m above sea level, with a temperature range from 48C to 308C. It has ample rainfall and is covered with green vegetation.

183

Figure 1.

Location of the project, Jimma Town in Keffa province, south-west Ethiopia.

# 1998 Blackwell Science Ltd. Paediatric and Perinatal Epidemiology, 12, 182198

Paper 104 MS
184 M. Asefa et al.
The Jimma Institute of Health Sciences, in which this project was based, was founded in 1983 and seeks to provide a community-oriented and team-based education in medicine, nursing, midwifery, sanitation, pharmacy and laboratory medical sciences. National health policy emphasises the promotion of self-reliance and community involvement in health activities,2 and in this study we sought to examine the effect of family-level activities on growth in infancy, concentrating on traditional nutritional and surgical practices. We report here on their effects on weight gain in the first year. The aim of the analysis presented was (a) to examine variation in individual growth trajectories over the first year and (b) to examine the extent to which poor growth can be explained in this population by the cumulative impact of adverse circumstances and child-care practices in the first year.

Subjects and methods

There is no vital events registration in Jimma, so pregnant women were identified through a network of 40 traditional birth attendants (TBAs), each of whom was given responsibility for about 500 homes. The TBAs reported daily to a project interviewer, and pregnancies were monitored jointly by the TBA and project interviewer until the birth. All infants born from 1 Meskren 1985 to 5 Pagume 1985 in the Ethiopian calendar (11 September 1992 to 10 September 1993 in the Gregorian Calendar) were eligible for study. An initial visit was made to them as soon as possible after birth, and then they were seen at 2, 4, 6, 8, 10 and 12 months. If an infant's birth was missed they were recruited as soon as they were identified. There were 10 interviewers in all. Interview schedules were prepared in Amharic and Oromo. All the interviewers spoke both languages, and all the mothers could speak one or the other. Information was collected by interview (e.g. the mother's marital status), by direct observation (sanitation and water supply), by testing (literacy) and by measurement (weight and length). The type of sanitation and the principal water supply for each home were documented by direct observation at the first visit. Family income was recorded as equivalent cash income in birr (2 birr was about US$1 at the time of the study). Maternal literacy was recorded in the interview and then checked by a direct test with a simple text. At each visit we asked whether the baby was breast fed; fed cows' milk or formula; fed solids or semi-solids; or given water. Three illness symptoms were recorded. We asked whether the baby had had diarrhoea (three or more watery stools per day), a persistent cough (lasting at least 4 days), an episode of fever, or any other illness or accident, since the last visit. The incidence of traditional operations was also recorded: these were uvulectomy (section of the uvula palatina using a horsehair), the extraction of milk teeth, circumcision (in males) and the operation in girls which is also known as circumcision (type 1 clitoridectomy).9 The weight of the infant was recorded to the nearest 100 g using a
# 1998 Blackwell Science Ltd. Paediatric and Perinatal Epidemiology, 12, 182198

Paper 104 MS
Traditional nutritional and surgical practices 185
Salter spring balance, and length to 1 cm using a locally produced wooden length board. Details of other information collected, of the training of interviewers and of the planning and implementation of the study, have been described elsewhere.10

Results
In all, 1563 births were identified, though the data we report here come only from the 1502 singletons. Details of the sample size at different ages, of the actual ages at which infants were weighed and of known deaths are given in Table 1. Through their mothers or principal carers, all the infants identified were successfully recruited into the study, though some were later withdrawn or lost to follow-up: 91% of scheduled visits were accomplished and 86% of the infants were successfully followed for a year or to an earlier death. The analyses we report include data for the periods for which they are available from children who were lost to follow-up.

Birthweight and weight gain

Growth curves for weight in infancy can be described using a fractional polynominal,11 for example a Reed model,12,13 which was the best-fitting of five growth models for weight over the first year in Congolese infants.14 In the analyses reported here, a Reed model was used with four terms: a constant (C), a linear term (M, age in months of 30 days), a (natural) log term (LM) and a reciprocal term (1/M2). For the linear term, the origin (age at birth) is 0; for the log and reciprocal terms it is 1, so that they can be used from birth. The coefficients were estimated using a random coefficient regression model specified as a two-level model in ML3-E.15,16 Different infants grow in different ways, and in a random coefficient model each infant's growth data are fitted by
Table 1. Target age for each visit, known deaths and sample sizes (number weighed) at each visit. The actual age that each child was weighed (median and 10th and 90th centiles) is also shown Target age (days) 0 60 120 180 240 300 360 Known deaths in preceding interval 44 26 15 20 20 16 Actual age when weighed (days) 10th 50th 90th centile centile centile 1 60 120 180 299 299 359 2 60 120 180 300 300 360 3 62 122 182 302 302 363

Visit 1 2 3 4 5 6 7

Number weighed 1456 1406 1317 1267 1222 1184 1151

# 1998 Blackwell Science Ltd. Paediatric and Perinatal Epidemiology, 12, 182198

Paper 104 MS
186 M. Asefa et al.
Table 2. Parameters estimates for Reed model. All the estimates are shown separately for males and females. The response variable is weight in kg. Males Estimate SE 1.79 70.17 3.32 1.44 0.0999 0.0133 0.0967 0.107 Females Estimate SE 1.93 70.0844 2.76 1.21 0.0946 0.0130 0.0942 0.101

Parameter Fixed Effects C M LM 1/M2 Random effects At level 2 (between subjects) Variances C M LM 1/M2 Covariances M/C LM/C (1/M2)/C LM/M (1/M2)/M (1/M2)/LM At level 1 (within subjects) Variances C

3.583 0.0661 3.63 3.58 0.372 73.25 73.55 70.443 70.385 3.41

0.388 0.00658 0.357 0.445 0.0468 0.361 0.412 0.0471 0.0500 0.388

2.96 0.0639 3.35 2.97 0.3456 72.88 72.92 70.425 70.364 3.02

0.331 0.00607 0.323 0.378 0.0416 0.318 0.351 0.0431 0.0445 0.341

0.106

0.00339

0.0860

0.00284

The fixed effect estimates show parameter estimates for the terms of the Reed model and specify the average growth curves shown in Fig. 2a and b. Of the level 2 (between-subject) random effects the variances show how the parameter estimates vary from infant to infant in the population; the covariances show how the individual estimates are related. The level 1 random effect is essentially the error variance showing how the weights of individual infants vary around their own growth curve.

their own most appropriate curve. This is achieved by allowing the coefficients to vary from infant to infant. Within-subject (level 1) and between-subject (level 2) variation are estimated separately. Level 2 variances for each of the four coefficients estimated for the Reed model describe how the coefficients (and so the growth curve) vary from child to child. The associated level 2 covariances describe how the coefficients are associated. The level 1 variance describes the variation in the infants' weights around their own individually fitted growth curve, for example caused by temporary weight loss or measurement error. The modelling procedures used are more fully described in the previous paper.10
# 1998 Blackwell Science Ltd. Paediatric and Perinatal Epidemiology, 12, 182198

Paper 104 MS
Traditional nutritional and surgical practices
(a) Males (b) Females

187

Weight (kg)

Age (months)

Weight (kg)

Age (months)

Figure 2. Mean weight of the infants plotted in relation to SD lines of the NCHS/WHO reference population: (a) males and (b) females.

The parameter estimates can be seen in Table 2. Separate estimates are given for males and females. The fixed effects are estimates of the population averages, and the average growth curves specified by these coefficients can be seen in Fig. 2, where they are related to the NCHS/WHO reference population.17 Although birthweights were not available, as most of the population delivered at home, they can be calculated approximately from these parameter estimates as the sum of the constant and reciprocal terms. This is 3.23 kg for males and 3.14 kg for females. These values, which are based on the whole population, not just infants delivered in hospitals or health centres, are close to the average of the NCHS reference population. Growth over the first 34 months is also close to the NCHS median, but there is a substantial relative decline from 4 to 12 months, with an average weight at 12 months about 1.5 SD below the NCHS median. The same trend was seen in length, and the infant mortality recorded in the study was 115/1000.10

Background variables
In considering the effect of traditional surgical and nutritional practices we needed to take into account some of the background variables. On the basis of previous work and exploratory analysis, we considered three types of variable: the sanitation and water supply of the home; the family's income; and the mother's literacy. Table 3 shows the different types of sanitation and water supply, as identified in the interview and checked by direct observation. Water supply was predominantely from protected springs (37%) and public taps (35%), with smaller numbers taking their water from rivers, ponds and unprotected springs (10%) or from private taps (18%). Sanitation was provided mostly by pit latrines (70%), with a smaller proportion using flush toilets (6%). No sanitation was available to 25% of families. The table entries give the numbers and percentages in each category, and Haberman standardised residuals.18 These are standardised deviations from the numbers that would be expected in each cell if there were
# 1998 Blackwell Science Ltd. Paediatric and Perinatal Epidemiology, 12, 182198

Paper 104 MS
188 M. Asefa et al.
Table 3. Sanitation and water supply showing the type of sanitation and principle water supply of the households in the study Sanitation Shared Private Shared pit Private pit flush flush None latrine latrine toilet toilet 30 2% +4.3 38 3% +3.5 162 11% +1.8 130 9% 0.0 14 1% 76.0 2 0% 72.2 376 25% 6 0% 72.6 17 1% 71.8 170 11% +0.1 168 11% +0.9 82 6% +1.0 8 1% 70.6 451 30% 20 1% 70.4 31 2% 70.6 219 15% 70.1 192 13% 70.9 119 8% +2.3 11 1% 70.6 592 40% 0 0% 71.0 0 0% 71.3 6 0% 71.3 9 1% 70.0 10 1% +2.7 2 0% +1.8 27 2% 0 0% 71.4 1 0% 71.2 2 0% 74.0 21 1% +0.6 19 1% +3.5 10 1% +8.2 53 4%

Water Supply River or pond Unprotected spring Protected spring

All 56 4% 87 6% 559 37% 520 35% 244 16% 33 2% 1499 100%

Public tap Private tap in compound Private tap in house

Total

For each type of water supply the first row gives the number of families and the second a rounded percentage (percentages less than 0.5 appear as 0%). The third row gives Haberman standardised residuals. Large positive residuals (e.g. +4.3 for `none' for sanitation, and `a river or pond' for water supply) show a positive association; large negative residuals (e.g. 76.0 for `none' for sanitation and `a private tap in the compound' for water supply) show a negative association.

no association between type of water supply and type of sanitation, and can be interpreted as z-scores. High positive scores show positive associations, low negative scores show negative associations. Overall, the Pearson chi-square value for a test of independence is 208 with 20 d.f. (P 5 0.01). In general, these residuals show that a better water supply was associated with better sanitation, as one would expect. In order to estimate the effects of these background variables on growth, we began with the growth-curve model specified in Table 2, then added the water
# 1998 Blackwell Science Ltd. Paediatric and Perinatal Epidemiology, 12, 182198

Paper 104 MS
Traditional nutritional and surgical practices
Table 4. Effects of sanitation and water supply Parameter Water Supply River or pond Unprotected spring 6M Protected spring 6M Public tap 6M Private tap in compound 6M Private tap in house 6M Sanitation None Shared pit 6M Private pit 6M Shared flush toilet 6M Private flush toilet 6M Estimate SE Ratio P

189

0.0363 70.00201 0.0778 70.0111 0.0515 70.00310 0.158 0.0333 0.158 0.0671

0.0794 0.0195 0.0650 0.0160 0.0654 0.0161 0.0700 0.0171 0.106 0.0264

2.27 1.88 2.55

5 0.05 5 0.10 5 0.05

0.0554 0.0350 0.0609 0.0348 70.00523 0.0225 0.0919 0.0586

0.0327 0.00787 0.0310 0.00742 0.0937 0.0224 0.0709 0.0170

4.45 4.70

5 0.01 5 0.01

3.44

5 0.01

The parameter estimates are for both sets of variables fitted at level 2 and entered together. In each case the main effect shows an estimated difference at birth, and the interaction terms (6M) a difference in weight gain per month, compared with no sanitation, or a river or pond water supply

supply and sanitation variables as factors. The factors can be entered in either order (water supply or sanitation first) and their effect evaluated using a likelihood ratio statistic referred to a chi-square distribution. Both factors have effects that are clearly statistically significant, whichever the order of entry. For water supply, the likelihood ratio statistic is 73.4 when entered first and 47.5 when entered after sanitation (10 d.f. P 5 0.01, in both cases). For sanitation, it is 60.8 when entered first and 34.9 when entered after water supply (8 d.f. P 5 0.01, in both cases). The parameter estimates are shown in Table 4. For water supply, each other type of facility was compared with a reference category (river or pond). In order to evaluate an effect on growth, the coefficient of each level was estimated as a main effect and a linear interaction with time. The main effect (e.g. 0.0363 for an unprotected spring) is then the estimated difference in birthweight compared with infants in the reference category; the interaction with time (70.00201 for an unprotected spring) is the difference in weight gain per month compared with infants in the reference category. This is not a large difference (about 0.024 kg
# 1998 Blackwell Science Ltd. Paediatric and Perinatal Epidemiology, 12, 182198

Paper 104 MS
190
Table 5. Parameter Income (log birr) 6M Literacy 6M

M. Asefa et al.
Effects of income and of literacy Estimate 0.163 0.0806 0.0296 0.0524 SE 0.0341 0.00818 0.0271 0.00668 Ratio 4.77 9.86 7.85 P 5 0.01 5 0.01 5 0.01

The income estimates show how weight increases with each 10-times increase in income. The literacy estimates show initial weight and weight gain for infants born to literate mothers (compared with those who were not literate).

over 12 months). The large difference comes with private taps, in the compound (0.399 kg over 12 months), and particularly in the house (0.805 kg). For sanitation, each type was compared with no sanitation. Except for a `shared flush toilet', each type significantly improved on no sanitation, and the effects were quite large (0.4180.703 kg over 12 months). The lack of benefit associated with a `shared flush toilet' may mean that the benefits of the sanitary provision were outweighed by the cross-infection resulting from sharing; but there were only 27 families in this category, so the standard error of this estimate is high and it is hard to place confidence in it. Income and the literacy of the mother are also important background variables, and were also estimated as main effects and in interaction with time. Income varied from none to 2000 birr per annum. Median income was 120 birr, with a quarter of the families having an income less than 60 birr and three-quarters less than 200 birr. Income was transformed to the log of income (to base 10) and used as a continuous variable. Seventy-four per cent of the mothers in the study were literate (1112/1502). Of the mothers who reported in the interview that they could read, only eight (51%) could not read a simple text when asked to do so. Both variables had statistically significant associations with the infants' weights. The log-likelihood differences were 144.5 and 61.0 with two degrees of freedom for the income and literacy variables respectively, entered separately after the 10 parameter growth curves (P 5 0.01 in each case). The parameter estimates are given in Table 5. For income, the interaction term shows an increase in weight gain over a year of 0.967 kg for an infant born to a family with an income of 10 birr (log 10 = 1), 1.934 kg for an infant born to a family with an income of 100 birr (log 100 = 2) and 2.901 for an infant born to a family with an income of 1000 birr (log 1000 = 3), each compared with an infant born to a family with 1 birr. A number of mothers were unable to provide an estimate of income in cash terms. Mean weight and weight gain for this group were estimated by coding them with a dummy variable, and were 0.2592 (SE 0.0821) and 0.1505 (SE 0.01915) respectively. The children born of literate mothers gained on average an additional 0.629 kg over the year compared with those born to illiterate mothers.
# 1998 Blackwell Science Ltd. Paediatric and Perinatal Epidemiology, 12, 182198

Paper 104 MS
Traditional nutritional and surgical practices 191
These, then, were the important structural, economic and educational background variables, against which the child-care practices that are the specific focus of this paper were evaluated. In all the subsequent analyses, therefore, we began with the growth curves, then entered simultaneously the sanitation, water supply, economic and literacy variables in the same form as they are given above. All estimates for other effects came from variables entered subsequently and were therefore adjusted for these background variables.

Nutritional variables
Of the 1502 singleton infants born, all but two were initially breast fed (the mothers of these two died in childbirth, and the infants themselves subsequently died too). Figure 3 shows the proportion of initially breast fed infants still breast fed at different ages, together with the proportion who have been fed cows' milk solids or semi-solids, or given water. These proportions were derived from life table analysis using the `survival' program in SPSSx. We expected the effects of these variables to vary with the age of the infant having solids at 1 month and 12 months are quite different things. These variables, therefore, were also considered in interaction with time. Simple interactions (with age in months) were generally sufficient, but for breast feeding a more complex interaction with the log and reciprocal term was used. Table 6 shows the change in the log-likelihood for each of these variables, in interaction with the age of the infant, entered alone after the control variables. The parameter estimates for breast feeding and cows' milk, adjusted for each other, are presented as their calculated effects at different ages. The effects of weaning from the breast are clear. Breast milk had large beneficial effects early in life, effects which were gradually attenuated over time. By 8 months there was no longer any demonstrated advantage in being breast fed. For cows' milk the effect was in the opposite direction. Cows' milk early Breast fed in life had an adverse effect; after 6 months its effect was beneficial. These effects of cows' milk were small, however, and there was no detectable Water effect at all of feeding solids or semiMilk solids. Nor did giving the infant water Solids have any demonstrable adverse effect.
Proportion Age (months)

Figure 3. Proportions of infants breast fed, fed cows' milk, supplemented with solids or semi-solids and given water.

Illness symptoms and traditional surgical practices

In considering effects of the nutri-

# 1998 Blackwell Science Ltd. Paediatric and Perinatal Epidemiology, 12, 182198

Paper 104 MS
192 M. Asefa et al.
Table 6. Nutritional practices parameter estimates. These show the effects of not being breast fed, and being fed cows' milk, solid (or semi-solids) or water (a) Parameter Background variables Not breast fed Cows' milk Solids Water Log-likelihood 13157.0 13105.5 13146.0 13153.5 13154.3 Difference 51.5 11.0 3.5 2.7 d.f. 3 2 2 2 P 50.01 50.01

(b) Estimated effects Not breast fed Cows' milk

Age (months) 1 4 7 10 70.443 70.059 70.249 70.028 70.047 0.004 0.101 0.036

In (a) the log-likelihoods are for each variable entered separately. The estimates are adjusted for sanitation, water supply, family income and maternal literacy and the variables are entered in interaction with the age of the infant, allowing for different effects at different ages. In (b) the overall estimated effects of the two statistically significant variables, no breast milk and cows' milk, are given for four ages 3 months apart. The estimates are of average weight differences in kg compared with infants still breast fed and not receiving cows' milk respectively, and are adjusted for one another.

tional variables it would not in general be appropriate to adjust for illness episodes. Breast feeding, for example, protects an infant against diarrhoea;19 we did not, therefore, estimate its effects adjusted for diarrhoea. The traditional surgical practices are another matter. Although circumcision is a ritual operation, both teeth extraction and uvulectomy are sometimes carried out in response to illness episodes, so we needed to adjust for these. The mother was initially asked in the interview whether the infant had had an episode of diarrhoea, persistent cough or fever since the last visit. Table 7 shows the numbers reporting each symptom at each visit. There were no major seasonal trends in these symptoms.
Table 7. Illness episodes showing the number and proportion of infants who were reported to have had one or more episodes of diarrhoea, persistent cough or fever since the last visit. Data are only included if the mother was seen within 710 to +20 days of the scheduled date of the visit Diarrhoea n % 63 127 158 223 303 300 4 10 13 18 26 26 Cough n % 59 96 130 116 107 85 4 7 10 10 9 7 Fever n % 30 52 94 111 108 98 2 4 7 9 9 9

Age 60 120 180 240 300 360

n 1414 1317 1258 1214 1177 1142

# 1998 Blackwell Science Ltd. Paediatric and Perinatal Epidemiology, 12, 182198

Paper 104 MS
Traditional nutritional and surgical practices 193
Table 8. Effects of gaving had an episode of diarrhoea, persistent cough or fever since the last visit, and estimates of residual (lagged) effects 2,4,6 and 8 months later. These estimates are for the weight deficits, compared with a child who did not have the illness episode, and are adjusted for sanitation, water supply, family income and maternal literacy Parameter Diarrhoea Current visit Lagged 2 months Lagged 4 months Lagged 6 months Lagged 8 months Cough Current visit Lagged 2 months Lagged 4 months Lagged 6 months Lagged 8 months Fever Current visit Lagged 2 months Lagged 4 months Lagged 6 months Lagged 8 months Estimate 70.289 70.150 70.131 70.0729 70.0307 70.116 70.111 70.119 70.0499 0.0118 70.135 70.107 70.0584 70.00227 70.00628 SE 0.0159 0.0212 0.0277 0.0344 0.0444 0.0209 0.0263 0.0321 0.0385 0.0458 0.0228 0.0300 0.0381 0.0481 0.0609 Ratio 18.18 7.08 4.73 2.12 5 5 5 5 P 0.01 0.01 0.01 0.05

5.55 4.22 3.71

5 0.01 5 0.01 5 0.01

5.92 3.57

5 0.01 5 0.01

An episode of illness may, however, affect weight not only at the next visit, but also at subsequent visits, if catch-up in weight has not been completed by the subsequent visits. Therefore, each illness variable was also considered in a lagged form, to estimate its effects not only at the visit following the episode, but also at visits 2, 4, 6 and 8 months later. As shown in Table 8 episodes of diarrhoea, persistent cough and fever each had adverse effects on weight. These adverse effects were detectable not Cough only at the immediately following Fever visit, but also at ensuing visits up to 6 months later. Figure 4 shows these Diarrhoea effects as curves for catch-up growth after episodes of each type of illness. Uvulectomy and circumcision Time (months) can only be carried out once and, Figure 4. Catch-up growth after three types in practice, the extraction of milk of illness episode, showing the mean deficit teeth was also only carried out once. (relative weight loss) in kg at the time of the Data on these were analysed using next visit after the illness and 2, 4, 6 and 8 months later. life tables for the three procedures.
# 1998 Blackwell Science Ltd. Paediatric and Perinatal Epidemiology, 12, 182198
Weight (kg)

Paper 104 MS
194
(a)

M. Asefa et al.
(b) Circumcision Uvulectomy Male Teeth extraction

Probability

Probability

Female

Age (months)

Age (months)

Figure 5. Incidence of surgical practices (a) showing the hazard functions for uvulectomy and the extraction of milk teeth and (b) the hazard functions for circumcision in males and for type 1 clitoridectomy in females (also known locally as circumcision). The hazard function gives the probability per day of an unoperated infant having the operation.

The probability of uvulectomy in the first year was 0.354 (SE 0.013) and of having milk teeth extracted was 0.384 (SE 0.013). Figure 5 shows the hazard functions, in order to emphasise the timing of these operations. Uvulectomy was most frequent at 24 months of age; milk teeth extraction was most frequent from 4 to 6 months of age, but was seen throughout the first year. The probability of circumcision was 0.628 (SE 0.019) in males and 0.040 (SE 0.0080) in females. It seemed desirable to check up on the very low level of circumcision reported in females, so a consecutive series of 200 females was given a complete physical examination by a qualified nurse. No girl recorded as uncircumcised was found on examination to have been circumcised; the operation may, of course, take place when the girl is older. Effects of traditional operations on weight gain, adjusted for illness episodes over the same period, are shown in Table 9. Uvulectomy and milk teeth extraction were associated with significantly lower weights at the following visit; the adverse effect of milk teeth extraction was still detectable 4 months later. There was, however, no detectable effect of circumcision on weight.

Discussion
In this population the estimated average birthweight for singletons was reasonably close to the NCHS reference. For males it was 3.23 kg and for females 3.14 kg. The mean for infants actually weighed by 7 days of age was 3.22 kg. Growth over the first 3 months was also close to the NCHS reference. Thereafter there was a steady relative decline in weight (and also in length.)10 As one would expect, the infants' weight gain over the first year was related to environmental and familial background variables. As regards water supplies, the
# 1998 Blackwell Science Ltd. Paediatric and Perinatal Epidemiology, 12, 182198

Paper 104 MS
Traditional nutritional and surgical practices 195
Table 9. Effects of traditional surgical practices. The effects are for weight deficits, attributable to uvulectomy, milk teeth extraction and circumcision and are adjusted for disease symptom episodes as well as for sanitation, water supply, family income and maternal literacy Parameter Uvulectomy Current visit Lagged 2 months Lagged 4 months Milk teeth extraction Current visit Lagged 2 months Lagged 4 months Circumcision Current visit Lagged 2 months Lagged 4 months Estimate 70.0541 70.0363 70.0117 70.0797 70.0697 70.0470 0.0290 70.0180 0.0302 SE 0.0216 0.0229 0.0226 0.0234 0.0224 0.0224 0.0253 0.0247 0.0236 Ratio 2.50 P 5 0.05

3.41 3.11 2.10

5 0.01 5 0.01 5 0.05

Estimates of residual (lagged) effects 2 and 4 months later are also provided. The circumcision effects are averaged over males and females. There was no significant difference between the effects for males and females, but the very small number of operations on females needs to be borne in mind.

clear benefits came from piped water, especially private supplies in the house or compound. Shared or private pit toilets and private flush toilets were also associated with better weight gain. These results are consistent with those from many previous studies.20 The beneficial effect of maternal literacy was also as expected.21 These variables and the income of the family were used to provide a reasonable level of control over relevant background variables in assessing effects of the nutritional variables and traditional practices. Breast feeding was started almost universally, and 80% of mothers continued to breast feed for the first year. Its effects were predictably beneficial up to about 8 months of age. In the second half of the first year cows' milk was also beneficial. Its use as a weaning food is widespread in Ethiopia, which has a very large cattle population.22 More surprising was the lack of any detectable beneficial effect of supplementary (solid and semi-solid) weaning foods. This absence of benefit cannot be attributed to a lack of variability in the timing of the presentation of supplementary foods; as Figure 3 shows, about 20% of infants were given solids by 6 months, but about 20% were still given none by 1 year. The foods given were `atimet' and `genfo', which are porridges, and `fetefit', which is a mixture of milk and injera, a sour bread made with teff (Eragrostis teff). Whether these weaning foods were given in inadequate amounts or too infrequently, or lacked the necessary energy, or were so heavily contaminated that they gave no net benefit in the first year, is an urgent topic for research in Ethiopia.23,24
# 1998 Blackwell Science Ltd. Paediatric and Perinatal Epidemiology, 12, 182198

Paper 104 MS
196 M. Asefa et al.
The large and prolonged effects of diarrhoea and other disease symptoms on weight gain may also be partly explained by the poor quality of the available weaning foods. There is, of course, much previous evidence for negative effects of diarrhoea particularly, on growth e.g. in Guatemala,25 Gambia26,27 and Uganda,26 but recent research suggests that `infections may be an important determinant of nutritional status only when the child's usual diet is suboptimal'.23 In Jimma, the diet clearly was suboptimal, and growth was badly affected. We anticipated, but did not find, an adverse effect of water. Water is unnecessary for exclusively breast-fed infants, even in tropical countries.28 It can reduce breast-milk intake and is often contaminated.29 However, there was no evidence in the present study that the widespread giving of water early in life had any adverse effect. On the other hand, two traditional surgical practices did have adverse effects. Both uvulectomy and the extraction of milk teeth led to detectable relative weight loss. From mothers' accounts, collected at the next visit, of the circumstances in which a child died we also have grounds for attributing two deaths to circumcision and three to uvulectomy and/or teeth extraction. In one case, for example, an infant bled heavily after the circumcision and `died at the gates of the hospital'; in another, the milk teeth were extracted and the uvula cut and, according to the mother `there was pus', and the baby died because `he couldn't stand two procedures at the same time'. While these attributions may not constitute proof, they are consistent with other evidence of the hazards associated with these traditional operative practices.30,31 In conclusion, although growth was poor and mortality high in infancy in this population, there are clear pointers to relevant public health interventions. Improved sanitation, especially for the 25% of the population who do not have even pit latrines, is one possibility. A reduction in unnecessary surgical procedures (uvulectomy and the extraction of milk teeth) would also be desirable. Breast feeding is almost universal, and supplements are generally well timed, so the most urgent task is to find out why the supplements, though well timed, provided no detectable benefit to the infant, leading to such poor growth and poor recovery from illness in the weaning period.

Acknowledgements
This study was supported by the Rockefeller Foundation, by UNICEF-Ethiopia, and by the Jimma Institute of Health Sciences. It was approved by the Ethics Committee of the Jimma Institute of Health Sciences and managed by a group that included Dr Fekadu, Head of the Jimma Zonal Health Department, and Ato Fasil, Dr Sileshi and Ato Sileshi of the Jimma Institute of Health Sciences. The Institute's Director, Dr Teklasion, also contributed substantially to its implementation. We are grateful to our interviewers and their supervisors, to the Jimma TBAs, and to the families who took part for their year-long cooperation.
# 1998 Blackwell Science Ltd. Paediatric and Perinatal Epidemiology, 12, 182198

Paper 104 MS
Traditional nutritional and surgical practices References
1 World Bank. World Development Report 1993 Investing in Health; World Development Indicators. Oxford: Oxford University Press, 1993; pp. 1133. 2 Ministry of Health. 1985 Primary Health Care Review. Addis Ababa: The Provisional Military Government of Socialist Ethiopia, 1985; pp.1340. 3 Assefa Hailemarian, Kloos H. Population. In: The Ecology of Health and Disease in Ethiopia. Editors: Kloos H, Zein ZA. Boulder: Westview Press, 1993; pp.1540. 4 Boerma JT, Sommerfelt AE, Bicego GT. Child anthropometry in cross-sectional surveys in developing countries: an assessment of the survivor bias. American Journal of Epidemiology 1992; 135:438449. 5 Pelletier DL, Frongillo EA, Habicht JP. Epidemiological evidence for a potentiating effect of malnutrition on child mortality. American Journal of Public Health 1993; 83:1130 1133. 6 Schroeder DG, Brown KH. Nutritional status as a predictor of child survival: summarizing the association and quantifying its global impact. Bulletin of the World Health Organisation 1994; 72:569579. 7 Smart JL. Malnutrition, learning and behaviour: 25 years on from the MIT symposium. Proceedings of the Nutrition Society, London, 1993; 52:189199. 8 Skuse D, Pickles A, Wolke D, Reilly S. Postnatal growth and mental development: evidence for a `sensitive period'. Journal of Child Psychology and Psychiatry 1994; 35:521545. 9 Toubia N. Female circumcision as a public health issue. New England Journal of Medicine 1994; 331:712716. 10 Asefa M, Drewett RF, Hewison J. An Ethiopian birth cohort study. Paediatric and Perinatal Epidemiology 1996; 10:443462. 11 Royston P, Altman DG. Regression using fractional polynomials of continuous covariates: parsinomious parametric modelling. Applied Statistics 1994; 43:429467. 12 Cole TJ. The use and construction of anthropometric growth reference standards. Nutrition Research Reviews 1993; 6:1950. 13 Berkey CS, Reed RB. A model for describing normal and abnormal growth in early childhood. Human Biology 1987; 59:973987. 14 Simondon KB, Simondon F, Delpench F, Cornu A. Comparative study of five growth models applied to weight data from Congolese infants between birth and 13 months of age. American Journal of Human Biology 1992; 4:327335. 15 Goldstein H. Efficient statistical modelling of longitudinal data. Annals of Human Biology 1986; 13:129141. 16 Prosser R, Rasbash J, Goldstein H. ML3: Software for Three-level Analysis User's Guide for V2. London: Institute of Education, University of London, 1991; pp. 1142. 17 World Health Organisation. Measuring Change in Nutritional Status. Geneva: World Health Organisation, 1983; pp. 5101. 18 Upton JGG. The Analysis of Cross-Tabulated Data. Chichester: John Wiley & Sons, 1978; pp. 1148. 19 Feachem RG, Koblinsky MA. Interventions for the control of diarrhoeal diseases among young children: promotion of breast-feeding. Bulletin of the World Health Organization 1984; 62:271291. 20 Esrey SA, Habicht J. Epidemiological evidence for health benefits from improved water and sanitation in developing countries. Epidemiological Reviews 1986; 8:117128. 21 Cleland JG, van Ginneken JK. Maternal education and child survival in developing countries: the search for pathways of influence. Social Science and Medicine 1988; 27:1357 1368.
# 1998 Blackwell Science Ltd. Paediatric and Perinatal Epidemiology, 12, 182198

197

Paper 104 MS
198 M. Asefa et al.
22 Dejene Aredo. The economy, with particular reference to the agricultural sector. In: The Ecology of Health and Disease in Ethiopia. Editors: Kloos H, Zein ZA. Boulder: Westview Press, 1993; pp. 1540. 23 Brown K. The importance of dietary quality versus quantity for weanlings in less developed countries: a framework for discussion. Food and Nutrition Bulletin 1991; 13:8694. 24 Motarjemi Y, Kaferstein F, Moy G, Quevedo F. Contaminated weaning food: a major risk factor for diarrhoea and associated malnutrition. Bulletin of the World Health Organization 1993; 71:7992. 25 Martorell R, Habicht JP, Yarlbrough C, Leichtig A, Klein RE, Western KA. Acute morbidity and physical growth in rural Guatemalan children. American Journal of Diseases of Children 1975; 129:12961301. 26 Cole TJ, Parkin JM. Infection and its effect on the growth of young children: a comparison of the Gambia and Uganda. Transactions of the Royal Society of Tropical Medicine and Hygiene 1977; 71:196198. 27 Rowland MGM, Cole TJ, Whitehead RG. A quantitative study into the role of infection in determining nutritional status in Gambian village children. British Journal of Nutrition 1977; 37:441450. 28 Ashraf RN, Jalil F, Aperia A, Lindblad BS. Additional water is not needed for healthy breast-fed babies in a hot climate. Acta Paediatrica 1993; 82:10071011. 29 Martines JC, Rea M, de Zoysa ID. Breast feeding in the first six months: no need for extra fluid. British Medical Journal 1992; 304:10681069. 30 Dagnew MB, Damena M. Traditional child health practices in communities in northwest Ethiopia. Tropical Doctor 1990; 20:4041. 31 Prual A, Gamatie Y, Djakounda M, Huguet D. Traditional uvulectomy in Niger: a public health problem? Social Science and Medicine 1994; 39:10771082.

# 1998 Blackwell Science Ltd. Paediatric and Perinatal Epidemiology, 12, 182198