The Butajira project in Ethiopia: a nested case-referent study of under-five mortality and its public health determinants.

The Butajira project in Ethiopia: a nested case referent study of under-five mortality and its public health determinants D. Shamebo,1 A. Sandstro5m,2 L. Muhe,3 L. Freij,4 1. Krantz,5 G. L6nnberg,2 & S. Wall2 During one year of follow-up, 306 deaths of children under the age of 5 years were included in a concur- rent case-referent study that was based on a population estimated at 28 780 in 1987. A total of 612 live referents, matched for age, sex and study area, were also selected from the study population through density sampling. Data were collected by lay reporters by verbal autopsy. For the study period the estimated cumulative under-five mortality rate was 293 and the infant (0-11 months old) mortality rate was 136 per 1000. Major probable causes of death were diarrhoeal disease or acute respiratory infections (ARI). The relative importance of parental and environmental characteristics was assessed using conditional multiple logistic regression analysis. Under-five mortality was associated with paternal illiteracy, maternal ethnicity, and not being in the committee of people's organizations. Parental factors affected the infants relatively more than they did the children, especially with regard to ARI mortality. This was also noted with "absence of window", a proxy measure for evaluating the type of housing. In terms of etiological fractions a greater number of under-five deaths could be ascribed to parental than environmental conditions, with relatively more infants being affected than children. Introduction Diarrhoeal disease and acute respiratory tract infec- tions (ARI) are the leading causes of morbidity and mortality in developing countries among under-5- year-olds (1-3). Factors identified as determinants of these disease entities are matemal education and occupation, duration of breast-feeding, place of resi- dence, household income, infant's birth weight, and the birth order and interval (1, 3-6). The beneficial effect of mothers' education on the survival of under-5-year-olds has repeatedly been discussed (7-11). Improvements in access to water and sanita- tion have also contributed to a reduction in child mortality (12-14). 1 Department of Community Health, Faculty of Medicine, Addis Ababa University, Addis Ababa, Ethiopia (deceased). 2 Department of Epidemiology and Public Health, University of Umea, S-901 85 Umea, Sweden. Requests for reprints should be sent to Dr S. Wall at this address. 3 Department of Pediatrics, Faculty of Medicine, Addis Ababa University, Addis Ababa, Ethiopia. 4 Department of Medical Microbiology and Immunology, Faculty of Medicine, University of Gothenburg, Gothenburg, Sweden. 5 Department of International Health Care Research (IHCAR), Karolinska Institute, Stockholm, Sweden. Reprint No. 5398 Epidemiological studies in developing countries, particularly in Africa, are often cross-sectional or longitudinal and based on prospective home visits (2, 6, 10). The case-referent approach, however, which is cost-effective, has not been used as much as one would expect (15-17). Ideally data are needed either from a well organized system of registration of vital events or from a prospective cohort of new- borns to study determinants like the ones mentioned above. The former approach is not often used as there are generally no such reliable systems operat- ing in developing countries; the latter approach is frequently avoided because of the costs involved and problems of follow-up over a long period of time. One has therefore often resorted to estimations of mortality rates by indirect techniques such as reports of recalled events from interviews with mothers (18-20). In 1986 a census was performed in ten study populations sampled from 86 rural communities in the Butajira district, 130 km south of Addis Ababa. The population is organized through Peasants' Asso- ciations (PA) in the rural areas (where farming is the main occupation) and Urban Dwellers' Associations (UDA) in the towns. Most people in the district live in traditional round huts (tukuls) with thatched roofs. Water sources, for all purposes, are either unprotec- ted rivers or wells. The census was followed by Bulletin of the World Health Organization, 71 (3/4): 389-396 (1993) © World Health Organization 1993 389 D. Shamebo et al. continuous household surveillance based on monthly home visits. Vital events have been recorded since 1987, starting with a baseline population of 28 780 living in 6258 households (21). The dynamic popula- tion has during the first four years of surveillance generated close to 125 000 person-years of follow- up, around 21 000 of which were under-5-year-olds. This population-based framework can be used for epidemiological analyses and interventions. Demographic surveillance provided age- and cause- specific mortality rates, from which life expectancies as well as mortality pattems were estimated (21). This paper, which analyses the determinants of under-five mortality using a concurrent case-referent technique, focuses on parental and environmental correlates and discusses their relative importance for the planning of public health interventions. Materials and methods Four possible sets of determinants of infant (0-11 months old) and child (1-4 years old) mortality have been identified and analysed as preconditions for mortality according to a conceptual framework (Fig. 1). The postulated causal web behind under- five mortality includes parental determinants (age, education, religion, marital status, and ethnicity) coupled with environmental factors, such as sanita- tion and hygiene and modified by the parents' health behaviour and practices. On the basis of climate and geographical conditions, the study area was divided into highland and lowland. All the children within the ten study populations could be regarded as a cohort followed over time through demographic surveillance (21). Trained interviewers collected information on births, deaths, probable causes of death, and migrations during monthly visits to individual households. All under- five deaths from 1 October 1988 to 30 September 1989 were potential cases in the study (Fig. 2). For each case, two live referents (matched for age, sex and study area) were selected from the Fig. 1. Factors that determine under-five mortality. Reproductive history Parental --.Environmental Feeding DseaseMoaltfactors factors history hstoryy V4093W0s Health behaviour and practices Fig. 2. Timing of data collection. Demographic F surveillance C E Births , .mmigrations N 1 u 1 Jan 87 1 Oct 88 1 Sept 89 1 Jan 91 Deaths tigrations Incident CASE-REFERENT study onL infant and child J mortality continuously updated register of the study popula- tion. The households with cases and referents, so identified, were then visited within a period of one month by two specially trained field workers (non- medical personnel) and information was collected by the use of a pretested, structured questionnaire on demographic characteristics, sociocultural pattems, type of housing, and environmental factors (Fig. 3). Disease history and behavioural factors as well as reproductive history, feeding practices and health behaviour were also included in the questionnaire. The interviewers were closely supervised by a nurse in charge of the field station of the demographic sur- veillance. Out of the 351 deaths registered during the monthly demographic surveillance, 45 (13%) could not be included in the study because the families had left the area after the death, or refused to participate, or were not at home. Thanks to the study base, within which referents and cases were matched, we were able to take some baseline characteristics for the 45 cases that were lost to follow-up; these, when compared with the 306 cases that were included in the study, were not different with regard to sex. However, a greater proportion of children aged 1-4 years (15%) than infants (7%) was lost to follow-up. Losses to follow-up were also greater among cases in the highlands (17%) than in the lowlands (11%). A verbal autopsy procedure, with symptom- prompted questions in a structured questionnaire, was used for a broad classification of probable immediate causes of death (22-24). The records were independently reviewed by three physicians, who then determined the diagnostic category by consensus. Three categories were used: ARI (in- cluding measles and whooping cough), diarrhoeal diseases, and other causes (unspecified fevers, mal- nutrition, accidents, etc.). The method was validated prior to implementation in this study (L. Muhe et al., Department of Pediatrics, Addis Ababa University, manuscript in preparation). The interview with the mother/father/guardian was conducted in a single session for both the cases and referents. WHO Bulletin OMS. Vol 71 1993390 Study of under-five mortality in rural Ethiopia Fig. 3. Information collected on potential public health determinants of mortality. Environmental factors Sanitation Source of wA Water consi Latrine stan Livestock ir Household No. of mem No. ofchild Relationshil Membershil Father organizati Age Ethnicity Housing Religion Type of hou Literacy Size of houc Marital status Type of roo No. of wives Type of floc Occupation Number of i Social position Cooking firE Type of fuel WHO 93206 Source of Ii Statistical methods The 306 cases and 612 referents were analysed retain- ing the matching, i.e., as 306 triplets. For bivariate analyses, Mantel-Haenszel procedures (25) were used to calculate odds ratios as estimates of relative risks. Assuming a statistical power of 80% and a two-sided significance level of 5%, the possibilities of "detecting" a certain excess relative risk may be calculated as a function of the exposure prevalence. Thus, if 20% of the referents have a certain charac- teristic, which is thought of as a risk factor for under-five mortality, we are bound to "find" this if it is associated with a relative risk of 1.6 or more. If, however, the same exposure specifically affects infants only (128 cases), the relative risk has to be at least 2.0. For the modelling of under-five mortality as determined by postulated determinants and to assess the relative importance of various variables, we have applied conditional multiple logistic regression (CLR) analysis using the EGRET software program (26). Initially, risk categories of each of the charac- teristics selected were identified by means of bivariate matched analyses. These variables were then included in the logistic regression analysis. vater ;umption idard n house biers iren ps ip of people's tion use se fing windows es in the house ight Mortality Infant Children From the associated beta-coefficients, odds ratios and their 95% confidence limits were generated. Significant (P <0.05) factors from the multivariate analyses were included as indicator components in summary indices of the parental and the environmen- tal risk factor "load", respectively. These were trans- formed into attributable risks ("etiologic" fractions) to estimate the potential impact of parental and envi- ronmental determinants on infants and children, as well as for boys and girls, highland and lowland communities, and deaths due to acute respiratory infections and diarrhoeal disease respectively. Results During the one year of follow-up a total of 351 cases of under-five deaths were registered. This corres- ponds to a cumulative under-five mortality rate of 293 and an infant mortality rate of 136 per 1000, with higher mortality among boys than girls. The majority of deaths occurred in the four lowland com- munities implying a threefold greater under-five mortality risk compared with the highland communi- ties. The major probable causes of infant deaths were ARI (33%) and diarrhoeal disease (23%); for chil- WHO Bulletin OMS. Vol 71 1993 Parental factors Mother Age Marital status Ethnicity Religion Literacy No. of children 391 D. Shamebo et al. dren aged 1-4 years, diarrhoea accounted for 32% and ARI for 20% of the deaths. The results given below show the relative importance of parental and environmental characteristics as assessed by the matched case-referent analyses. Parental factors Since the population of mothers was homogeneous with respect to literacy (93% were illiterate) and marital status (96% were married), these variables could not be assessed as potential determinants. Also, because the correlation between maternal and paternal ethnicity as well as religion was almost unity, we opted to use the maternal factors in these cases. Table 1 shows that all the potential parental risk indicators seem to be associated with excess under- five mortality, except for polygamy. Three of these (Silti ethnicity, illiteracy of the father, and non-mem- bership of people's organizations) were still retained in the multivariate model. When these factor levels were related to a parental score (0, 1, 2, 3), the dose- response association (Table 2) showed a 4.6-fold risk of under-five mortality for a child with all ver- sus a child with none of the indicators, while still matched for age, sex and area. Stratifying for the matching variables showed a greater impact from parental factors on infants than children and in high- land than lowland areas. The impact was equal on boys and girls, but ARI mortality was clearly more affected than diarrhoeal disease mortality. Environmental factors Sanitary conditions were analysed in terms of the type of water source, distance to this source, volume of consumption, latrine standard, and whether live- stock was kept in the house. Since access to piped water was extremely rare (available in only a few urban households), rivers and wells were the only source for comparison. Housing was assessed in terms of shape and size, space for living, and the availability of a win- Table 1: Influence of parental and environmental risk factors on under-five mortality (based on bivariate matched and conditional logistic regression respectively) Risk No. of No. of Bivariate Multivariate factor cases referents odds ratio odds ratio Parental problem: Mother 15-24 or .35 years 129 235 1.29 (0.97-1.71)a Father >35 years 228 425 1.33 (0.95-1.88) Mother Silti 82 130 1.67 (1.13-2.44) 1.74 (1.21-3.16) Father Silti 81 132 1.41 (0.95-2.08) Mother Muslim 218 440 1.16 (0.81-1.66) Father illiterate 208 378 1.47 (1.06-2.04) 1.45 (1.01-2.06) More than one wife 79 163 1.00 (0.72-1.38) Not in a people's organization 280 511 2.09 (1.33-3.27) 1.95 (115-2.63) Child lacking parent(s) 38 42 1.90 (1.20-3.01) Environmental problem: Water from well 202 398 1.14 (0.69-1.90) Distance to water >5 min 188 396 0.80 (0.56-1.14) Water consumption <6 I/person/day 148 286 1.11 (0.83-1.47) Latrine poor or lacking 262 506 1.37 (0.87-2.16) Livestock in house 252 512 0.94 (0.63-1.42) Siblings under-five 197 403 0.93 (0.69-1.26) Tukul hut 279 543 1.74 (0.88-3.46) - Single room 237 472 1.03 (0.70-1.49) No window in house 227 412 1.57 (1.10-2.25) 1.54 (1.08-2.21) Cooking fires in the house 292 578 1.29 (0.63-2.63) Fuel dung 131 259 1.04 (0.71-1.51) No electricity 293 584 1.31 (0.37-4.61) Living area <6 m2/person 156 301 1.09 (0.82-1.46) a Figures in parentheses are 95% confidence limits. WHO Bulletin OMS. Vol 71 1993392 Study of under-five mortality in rural Ethiopia dow, electricity and facilities for cooking. As only Among parental factors, under-five mortality few of the households had electricity, this predictor was significantly and independently associated with was difficult to assess. In the multivariate model, ethnicity, paternal illiteracy, and non-membership of lack of a window was the only indicator retained; as people's organizations. The effect of maternal educa- an environmental marker it was associated with a tion could not be assessed as almost all the mothers 2.7-fold relative risk of infant mortality (Table 2). were illiterate. In previous studies, education of the The lack of a window when used as a dependent mother has been a strong predictor of infant and variable in a logistic regression was related to the child survival (5, 9, 27); the usefulness of this parental, hygienic, crowding and housing indicators, variable diminishes when one has a homogeneous and showed that crowding and poor housing were population. Mortality was associated with younger both significantly related to the absence of a win- (<25 years) and older (.35 years) ages of mothers. dow. Specifically, single rooms and tukul huts were Similar associations were observed in other studies the underlying characteristics of these components. (5, 11, 28). In our multivariate model, matemal age Of 128 deceased infants, 101 lived in houses without was however not retained. Neither did an excess a window; no sex or area differences were seen. under-five mortality in children lacking one or both However, the ARI mortality risks were again more parents remain in the multivariate model. pronounced than the diarrhoeal disease mortality Silti ethnicity was associated with increased risks. For infants, a fivefold ARI mortality risk asso- under-five mortality. Differences in under-five mor- ciated with lack of a window could be demonstrated. tality by religion (5) and ethnicity may be related to culturally determined differences in health behav- Discussion iour. Clearly, anthropological studies are needed tounderstand this. Factors influencing infant and child mortality in a Paternal illiteracy has been significantly associa- developing country have, to our knowledge, not pre- ted with increased under-five mortality. This was viously been studied by means of the case-referent reported in a Nigerian study (6), where the risk was method with a concurrent and nested approach with threefold greater with non-educated fathers. An a high population participation. The present study interesting finding is the association of lower morta- illustrates the potential usefulness of this approach. lity with close involvement in the work of people's The verbal autopsy technique of classifying probable organisations; there was no obvious confounding causes of death is a useful instrument in the field from literacy, but the relative risk associated with situation and was validated for sensitivity and speci- non-involvement was higher among illiterate than ficity prior to implementation in this study by com- among literate fathers. Because those who become paring the results of the field workers with that of a involved are better educated, have easier access to paediatrician (agreement of 80% and 95% for sensi- health care, and can afford to pay for it, this indica- tivity and specificity, respectively) (L. Muhe et al., tor is a good proxy measure for higher income and unpublished data). power. Table 2: Dose-response associations among subgroups of under-five children, by parental and environmental scores composed of indicator variables from multiple logistic regression Odds ratio for subgroups Factor score All under-fives Infants Children Boys Girls Highland Lowland ARI Diarrhoea Parental: 0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1 1.6 1.4 1.4 1.7 1.5 2.3 1.2 4.6 0.8 2 2.3 2.9 1.7 2.5 2.1 3.4 1.8 4.2 1.3 3 4.6 10.7 2.5 4.5 4.7 15.5 2.8 14.8 2.4 Environmental: 0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 la 1.7 2.7 1.2 1.7 1.7 1.7 1.7 1.8 1.3 a Lack of a window was the sole indicator. WHO Bulletin OMS. Vol 71 1993 393 D. Shamebo et al. Unfavourable housing and environmental condi- tions, like living in traditional tukuls or homes without windows, poor latrines, crowding, cooking fires in the house, and small amounts of water consu- med were represented by the "no window" indicator, which affected infants more than children, specially for ARI. It was shown that this indicator was actual- ly a proxy measure for housing and crowding characteristics rather than for parental and hygienic factors; these associations have been reported by others (5, 9, 12, 13, 28). In this study, however, the water source did not come out as equally important as in other studies, mainly because the rural popu- lation was homogeneous with respect to this vari- able, in terms of both water quality and distance from the source. The identification of certain parental and envi- ronmental factors which represent major determi- nants of under-five mortality does not mean that the factors are the causal ones. Rather, they are to be interpreted in varying degrees as representatives of parental and environmental causal factors. Regarding the entry-points for intervention, the relative impor- tance of parental and environmental factors may be considered in various subgroups. For these purposes the multivariate assessments were made separately with parental and environmental score factors. Speci- fic patterns emerged where the parental factors seemed to affect the infants relatively more than they did the children. The greater vulnerability of infants was also related to poor housing conditions. The fact that the potential impact of parental scores seemed less in the lowlands might be due to the epidemic situation since individuals were affected more by a local outbreak of malaria and meningitis (21) than by "endemic" poverty. When comparing ARI and diar- rhoea mortality, it is reasonable to assume that par- ental influence through health-care-seeking behav- iour predisposes to ARI, especially among infants. This will be further analysed in a forthcoming study. Alternatively, the relative importance of the various factors can be assessed in terms of attribu- table risks ("etiologic" fractions), i.e., the propor- tions of cases that are estimated to be "due to" un- favourable parental or environmental characteristics. Thus, 62% of infant deaths could be attributed to parental factors as compared to 40% of child deaths (Fig. 4); the potentially higher vulnerability of infants to poor environmental conditions is also demonstrated through the etiologic fractions of 50% and 9%, respectively. Fig. 4 also illustrates the greater impact of parental factors in the highlands and the relative cause-specific impact. Ultimately, "etiologic" fractions show the hypo- thesized potential impacts from interventions- changing the distribution of risk factors or "eradica- Fig. 4. Potential impact of public health determinants, in terms of attributable risks, on under-five mortality. Parental 400 300 200 Envlrn 100 0 100 Number of under-five deaths 200 300 400 ting" certain unfavourable characteristics. Obviously, such intervention would have to be specifically tar- geted, e.g., infants and children, in highland and low- land areas and be differently composed to prevent more diarrhoeal disease than ARI mortality. Thus, it could be shown that paternal illiteracy meant a six- fold greater risk of diarrrhoeal disease mortality and that the potential benefit from close involvement in the work of people's organizations seems greater among children with illiterate fathers as well as in the highland, especially urban, areas. Among the en- vironmental factors, keeping livestock in the house meant a significant and more than twofold under-five mortality risk. A postulated association between ARI mortality and indoor air pollution, as indicated by the presence of cooking fires inside the house, did not reach statistical significance owing to the homo- geneity of this characteristic. This issue will be fur- ther explored in a prospective morbidity study. In a further analysis of parental and environmental fac- tors, D. Shamebo et al. (unpublished data) identified prolonged breast feeding as having a positive impact on infant and child survival, specifically preventing deaths due to diarrhoea; none of the sanitary factors seemed to predispose to ARI mortality, while a poor latrine standard and low water consumption both significantly increased diarrhoeal disease mortality. In conclusion, if the unfavourable parental or environmental factors should be removed, we could hypothesize (based on our multivariate analyses of 306 under-five deaths) obtaining a public health impact of 55% or 31% fewer deaths, respectively. These findings not only challenge the common view that rural populations in the Third World live under homogeneously poor conditions but also support fur- ther epidemiological and action-oriented public health research. WHO Bulletin OMS. Vol 71 1993 Infants Children c Boys Etiologic Girls . cases Highland Lowland ARI Diarrhoea :-: Total I-X...... t:--::::--:: - 394 Study of under-five mortality in rural Ethiopia Resume Projet Butajira en Ethiopie: une etude cas/t6moins de la mortalite chez les moins de cinq ans et ses d6terminants en rapport avec la sant6 publique Au cours d'une periode de suivi d'un an, 306 d6ces d'enfants de moins de cinq ans ont ete inclus dans une etude cas/t6moins basee sur une population evaluee a 28780 habitants en 1987. On a selectionne comme temoins 612 enfants, apparies pour l'age, le sexe et la zone d'etude, par 6chantillonnage dans la population. Les don- n6es ont ete recueillies par des agents de sant6 sans formation m6dicale, par autopsie verbale. Pour la p6riode etudiee, le taux cumulatif esti- m6 de mortalite chez les moins de cinq ans etait de 293 pour 1000 et le taux de mortalit6 chez les nourrissons (0-11 mois), de 136 pour 1000. Les principales causes probables du d6ces etaient une maladie diarrheique ou une infection respira- toire aigue (IRA). L'importance relative des para- metres parentaux et environnementaux a et6 6va- lu6e par analyse de regression logistique multiple. La mortalite chez les moins de cinq ans 6tait associ6e a l'analphabetisme paternel, l'origine ethnique de la mere et avec le fait que les parents ne faisaient pas partie des organisations des comit6s du peuple. Les facteurs parentaux influaient encore davantage sur la mortalit6 des nourrissons, surtout par IRA. Ce type de relation a 6galement 6te observ6 avec "I'absence de fenetre", une mesure approximative d'6valuation du type de logement. En ce qui concerne les rela- tions 6tiologiques, un plus grand nombre de d6ces avant l'age de cinq ans pourrait etre imput6 a des parametres parentaux plut6t qu'environne- mentaux, les nourrissons 6tant relativement plus touch6s que les enfants. References 1. Leowski, J. Mortality from acute respiratory infec- tions in children under 5 years of age: global esti- mates. World health statistics quarterly, 39: 138-1 44 (1986). 2. Shamebo, D. et al. The Butajira rural health project in Ethiopia: mortality pattern of the under-fives. Journal of tropical paediatrics, 37: 254-261 (1991). 3. Mirza, N.M. et al. Mortality patterns in a rural Kenyan community. East African medical journal, 67: 823-829 (1990). 4. Heard, N.M. & Ncobeni, L. Community-based infant and child mortality rates for peri-urban Pietermaritz- burg. South African medical journal, 70: 283-284 (1986). 5. Oni, G.A. Child mortality in a Nigerian city: its levels and socio-economic differential. Social science and medicine, 27: 607-614 (1988). 6. Adedoyin, M.A. & Watts, S.J. Child health and child care in Okelele: an indigenous area of the city of llorin, Nigeria. Social science and medicine, 29: 1333-1341 (1989). 7. Majumder, A.K. Maternal factors and infant and child mortality in Bangladesh. Journal of biosocial science, 20: 89-98 (1988). 8. Chen, L.C. Primary health care in developing coun- tries: overcoming operational, technical and social barriers. Lancet, 2: 1260-1265 (1986). 9. Aksit, B. & Aksit, B. Socio-cultural determinants of infant and child mortality in Turkey. Social science and medicine, 28: 571-576 (1989). 10. Cleland, J.G. & van Ginneken, J.K. Maternal edu- cation and child survival in developing countries: the search for pathways of influence. Social science and medicine, 27: 1357-1368 (1988). 11. Bailey, M. Factors affecting infant and child mortali- ty in rural Sierra Leone. Journal of tropical paedia- trics, 34: 165-168 (1988). 12. Merrick, T.W. The effect of piped water on early childhood mortality in urban Brazt, 1970 to 1976. Demography, 22: 1-24 (1985). 13. Lindskog, U. et al. Childhood mortality in relation to nutritional status and water supply - a prospective study from rural Malawi. Acta paediatr. Scand., 77: 260-268 (1988). 14. Victora, C.G. et al. Water supply, sanitation and housing in relation to the risk of infant mortality from diarrhoea. International journal of epidemiology, 17: 651-654 (1988). 15. Bartholomew, S. & MacArthur, B.A. Comparison of infants dying from the sudden infant death syn- drome with matched controls. Social science and medicine, 27: 393-397 (1988). 16. Griffin, P.M. et al. Risk factors for fatal diarrhoea: a case-control study of African children. American journal of epidemiology, 128: 1322-1329 (1988). 17. Victora, C.G. et al. Infant feeding and deaths due to diarrhoea: a case-control study. American journal of epidemiology, 129: 1032-1041 (1989). 18. Timaeus, I. et al. Health surveys in developing countries: the objectives and design of an inter- national programme. Social science and medicine, 27: 359-368 (1988). 19. Hill, A.G. & David, P.H. Monitoring changes in child mortality: new methods for use in developing coun- tries. Health policy and planning, 3: 214-226 (1988). 20. Murray, C.J.L. A critical review of international mor- tality data. Social science and medicine, 25: 773-781 (1987). 21. Shamebo, D. et al. The Butajira rural health project in Ethiopia: epidemiological surveillance for health research and intervention. Scandinavian journal of primary health care, 10: 198-205 (1992). 22. Mirza, N.M. et al. Verbal autopsy: a tool for deter- mining causes of death in a community. East Afri- can medical journal, 67: 693-698 (1990). 23. Kalter, H.D. et al. Validation of post-mortem inter- WHO Bulletin OMS. Vol 71 1993 395 D. Shamebo et al. views to ascertain selected causes of deaths in chil- dren. International journal of epidemiology, 19: 380-386 (1990). 24. Pacque-Margolis, S. et al. Application of the verbal autopsy during a clinical trial. Social science and medicine, 31: 585-591 (1990). 25. Mantel, N. & Haenszel, W. Statistical aspects of the analysis of data from retrospective studies of disease. Journal of the National Cancer Institute, 22: 719-748 (1959). 26. EGRET - Epidemiological Graphics, Estimation, and Testing package. Data file definition. Version 0.19.2 © copyright 1985 - 1990, SERC. 27. Tsui, A.M. et al. Maternal and socio-demographic correlates of child morbidity in Bas Zaire: the effects of maternal reporting. Social science and medicine, 26: 701-713 (1988). 28. Hull, T.M. & Gubhaju, B.B. Multivariate analysis of infant and child mortality in Java and Bali. Journal of biosocial science, 18: 109-118 (1986). 396 WHO Bulletin OMS. Vol 71 1993