J. biosoc. Sci. (1992) 24, 433-445
ESTIMATING INFANT MORTALITY RATES PROSPECTIVELY IN HONDURAS DAVID HUBACHER*, PATRICIA BAILEY*, BARBARA JANOWITZ*, FIDEL BARAHONAt AND MARCO PINELt * Family Health International, Research Triangle Park, North Carolina, USA, and iHonduran Ministry of Public Health, Tegucigalpa. Honduras Summary. In order to determine the validity of infant mortality estimates based on retrospective reporting, the Honduran Ministry of Health carried out a follow-up survey of women interviewed in a 1987 national survey. Women were interviewed approximately 14 months after the baseline survey and were asked about the outcomes of their pregnancies and the survival status of their young children. The overall infant mortality rate calculated from the follow-up survey was lower than that obtained from the baseline survey, due to the particularly low rate among the group of women who were pregnant at the time of the baseline survey. Possible explanations for this low rate are discussed. Introduction
Infant mortality rates are often used as an indicator of the level of socioeconomic development of a country. Reliable data to estimate infant mortality rates can come from several sources. Most industrialised countries derive their estimates from vital registration systems since births and deaths are well documented. In many countries, however, births and deaths are not registered as carefully as required to derive trustworthy estimates; censuses and sample surveys are often used as complementary. A census, usually defined as a complete canvassing of the population, does not always contain the information necessary to estimate infant mortality rates, is very expensive, and in many countries is undertaken infrequently. Sample surveys are therefore often relied upon to generate national estimates of infant mortality. From survey data, infant mortality rates can be estimated by indirect or direct methods (United Nations, 1983). Indirect estimates use data on the number of children ever born and surviving for women in different age groups (as defined at the time of the survey). Rates are calculated making assumptions about the age pattern and trends in mortality and fertility. Several models exist to estimate levels and trends of infant mortality. Direct methods compare the reported number of infant deaths to the number of live births in a given period; they are preferred over indirect techniques when sample sizes are sufficiently large and data quality is high. Large cross-sectional sample surveys alone do not guarantee better estimates, since validity is based on the 433
434
D. Hubacher et al.
accuracy of retrospective reporting of births and deaths. Common sources of error are the omission of births and deaths as well as misreporting of dates of birth and age at death. The most common omissions occur for children who no longer live with the mother, deaths that occurred several years before the survey, and children dying soon after birth who are incorrectly reported as still births rather than live births (United Nations, 1983; Rutstein, 1984). Omission of events is probably the most serious error because this leads to underestimates of the true levels of infant mortality. Prospective studies are another data collection alternative that can be used to estimate infant mortality rates; they can largely eliminate the possibility of omission as well as decrease the likelihood of misreporting dates of important events (Victora et al., 1987; Bailey et al., 1990). This paper presents the experience of such a study undertaken in Honduras in 1988; the historical context in which the prospective study was conceived is first reviewed. Trends in infant mortality
Infant mortality in Honduras has declined substantially over the past 30 years by perhaps as much as 70%, from 166 per 1000 live births in 1960 to 48 in 1987-88 (Guzman & Santos, 1988; Honduran Ministry of Public Health, 1989). However, the estimates of infant mortality come from several sources, use different techniques, and are not always in agreement (Fig. 1). The vital registration system in Honduras is a poor source of data for estimating infant mortality rates, as it substantially undercounts both births and deaths. The 1974 and 1988 Censuses provided mortality estimates as did a series of national surveys. These include two demographic surveys carried out under the direction of the Latin American Centre for Demography (CELADE), one in 1971-72 (EDENH-1) and the second in 1983 (EDENH-II), and two smaller national sample surveys—the 1981 Contraceptive Prevalence Survey (CPS) and the 1984 Maternal-Child Health/Family Planning Survey (MCH/FP). All of these surveys have used the Trussell models for indirect estimation of infant mortality. The 1981 CPS and 1984 MCH/FP surveys provided the lowest rates of infant mortality, but neither survey was designed specifically to focus on mortality. However, the Ministry of Health decided to make infant mortality a priority in the 1987 Epidemiology and Family Health Survey (EFHS) (Honduran Ministry of Public Health, 1989), and a large sample and a complete birth history were included in order to construct a strong data base for computation with direct and indirect techniques. Also, it was decided to select a subset of women (those who were pregnant and/or with an infant) to interview at a later date in order to measure infant mortality prospectively. Estimates from this project, the follow-up survey, would verify the estimates derived from the retrospective data collected through the birth histories of the 1987 survey. Estimates derived from the 1987 EFHS and the follow-up survey using life table analyses strengthen the evidence of a decline in infant mortality levels. This paper focuses on the methodological issues distinct to a prospective study of this type and attempts to explain why this technique has produced different levels of estimates for infant mortality in Honduras.
435
Estimating infant mortality in Honduras 180 160 140-
1201008060-
D O
A 40 20
1955
Census, 1974 CPS,1981 EDENH-H, 1983 MCH/FP, 1984 EFHS, 1987* Vital Statistics (adjusted) Census, 1988 Follow-up, 1988* 1960
1965
1970
1975
1980
1985 1988
Fig. 1. Infant mortality estimates from various sources, 1959-88. Sources: Guzman & Santos (1988); Honduran Ministry of Public Health (1989, 1990); INCAP (1990). * Direct rate estimates.
Methods In 1987 the Honduran Ministry of Health and the Association for Family Planning in Honduras administered the EFHS with technical assistance from Management Sciences for Health and Family Health International. This household probability survey covered a wide range of maternal and child health topics with an emphasis on estimating levels of fertility and infant mortality. Fieldwork for the EFHS was carried out between June and November, 1987; 10,159 women between the ages of 15 and 44 were interviewed. Of these, 2832 women were eligible for a follow-up interview, either because they were pregnant (995) or had a child under the age of 1 (784), or were pregnant and had a child under the age of 1 (53). The follow-up survey was conducted from September 1988 to February 1989, and the fieldwork followed the same visiting sequence as for the EFHS, thus allowing the
436
D. Hubacher et al.
maximum period possible between the two surveys (mean: 14 months; range: 11-18 months). The overall response rate for the follow-up survey was 74%. Of the 733 women not re-interviewed, 93% were not interviewed because they had moved, were not at home (after several contacts), or had died (five). Only six women refused a follow-up interview. Response rates by follow-up group were: pregnant women: 76-2% (758/995); women with infant: 73-5% (1312/1784); pregnant women with infant: 54-7% (29/53). The response rate in the capital city of Tegucigalpa was only 63%. Shortly before the fieldwork started, large sections of Tegucigalpa, where women eligible for the follow-up survey lived, were devastated by a hurricane. Sample and weighting A reasonable response rate is of utmost concern in any prospective study. The 74% response rate was lower than hoped for and led to close examination of those who were and were not re-interviewed. Table 1 shows the distribution of three groups of women (eligible, completed interviews, and lost to follow-up) according to household and maternal characteristics. Ideally, the characteristics of women with completed interviews and those of women lost to follow-up would mirror those of eligible women. However, this was not the case and there are important differences between the two groups. The lost to follow-up group has a higher proportion residing in urban areas and a higher proportion with more household amenities than the group with completed interviews. Also, the women in the lost to follow-up group are younger, better educated, and have fewer live births than the interviewed women. All chi-square tests of independence are significant at the 0-01 level. Because these characteristics are associated with infant mortality, the interviewed women, who are disproportionately representative of women of lower socioeconomic status, would be likely to exhibit higher infant mortality rates than the complete sample of women. Before calculating infant mortality rates, it was therefore necessary to adjust for non-response, and weighting factors were applied to the available data to compensate for loss to follow-up. The complete weighting scheme accounted for non-response at three different levels (Dominik, 1989). The first two levels were concerned with response rates at the area level (given the area sampling unit, or sector number), and the household level (giver the number of eligible respondents in the household); these numbers pertain to the 1987 EFHS. The third component of the overall weighting was concerned with response rates in the follow-up survey given a set of three identifying variables. These were chosen to define broad groups necessary to classify the women by follow-up group (pregnant, including those who were pregnant and had a live infant, or with infant), sanitary facilities (toilet, outhouse/other, or none), and parity (0 1,2-3,4-5, or 6 +). These three variables produced a set of 24 unique combinations of characteristics with response rates ranging from 42% to 87% (the overall follow-up response rate was 74%). The final weighting factor is the product of the inverse of the response rates for each of the three levels; these weightings were applied to each woman for whom complete information was available. These adjustments correct the data so that they better reflect the aggregate of responses that would have been expected had all 2832 women been successfully interviewed in the follow-up.
Estimating infant mortality in Honduras
437
Table 1. Percentage distribution of follow-up women by household and maternal characteristics* Characteristics Household Residence Urban Rural Source of water Faucet inside the home Faucet outside the home Other Electricity in home Yes No Sanitary facilities Toilet Outhouse and other None Maternal Age (years) 15-19 20-24 25-29 30-34 35-39 >40 Education None Primary 1 3 Primary 4-6 > 7 years Parity 0 1 2-3 4 5 >6 Follow-up group Pregnant women Women with infants Women with infants and pregnant Total No. of woment
Eligible follow-up women
Completed follow-up interviews
Lost to follow-up
29 71
25 75
40 60
13 42 45
11 42 47
16 44 40
31 69
28 72
39 61
19 32 49
15 33 52
28 29 43
15 29 24 16 11 5
14 27 24 17 12 6
20 35 24 12 6 3
22 33 31 14
23 34 30 13
20 29 32 19
27 31 19 23
24 30 20 26
36 33 18 13
35 63
36 63
32 65
2 100 (2825)
1 100 (2092)
3 100 (733)
All y2 tests of independence for completed and lost to follow-up groups are significant at 001. * At the time of the 1987 Epidemiology and Family Health Survey, t Seven women excluded from table because of missing data.
438
D. Hubacher et al. Results
Infant mortality rates All follow-up infant mortality rates were estimated using weighted data to adjust for non-response. As expected, since non-respondents were of higher socioeconomic status, the estimates were slightly lower than the unweighted rates. The SPSSx procedure SURVIVAL was used to calculate direct rates using standard life table analysis procedures. The 1987 EFHS contained complete histories which were used to estimate infant mortality rates directly. The 1985-87 life table rate of 48 per 1000 is based on a cohort of 5111 live births and is considered the best direct estimate provided by this survey. Several different infant mortality rates were calculated from the follow-up survey (Table 2). The estimate of 42 per 1000 live births uses all the children in the data set. Estimating rates according to the follow-up group (pregnant women and women with Table 2. Infant mortality estimates from the follow-up survey Time
IMR weighted
IMR unweighted
No. oflive births
1986-88 All children 95% CI
42/1000 (34-49)
43/1000
(2690)
1987-88 Newborns 95% CI
26/1000 (16-36)
28/1000
(771)
1986-88 < 1 year olds 95% CI
48/1000 (38-58)
49/1000
(1919)
infants) revealed the problem on which this paper focuses; the rate among infants (48 per 1000 live births) is much higher than the rate of 26 per 1000 live births for the pregnant women. When these group rates are examined according to different sociodemographic variables, the well documented patterns emerge (Table 3). In both follow-up groups, for instance, rural infants show a higher mortality rate than urban infants. Other predictable patterns are maintained for household characteristics, sex of the child, and maternal education; data from the 1987 EFHS are shown for comparison. It is particularly important to note these patterns in the newborn group; perhaps the estimate is low because of a weakness in the method, but not in the data. The mortality estimate from the pregnant women (newborns) was much lower than that for the infants. Higher estimates for the newborns in comparison to the infants might have been expected; information on early deaths for the newborns is likely to be more complete than for the older group.
Estimating infant mortality in Honduras
439
Table 3. Infant mortality rates from EFHS and follow-up survey, by selected characteristics Follow-up surveyt Characteristics
1987 EFHS*
Newborns
< 1 year olds
Residence Urban Rural
38 52
24 27
45 49
Electricity Yes No
35 54
16 32
34 54
Floor Dirt Other
56 36
34 17
56 37
Sex Male Female
54 42
31 21
54 43
Maternal education < 3 years > 4 years
55 39
49 4
62 30
No. of birthsj
(5113)
(771)
(1919)
* 1985-1987 births, t Weighted data. * Missing data on some variables reduced number of births in the analysis.
Low infant mortality among pregnant women One reason why such low rates were found for the pregnant women could have been that they were disproportionately representative of women of higher socioeconomic status. The sociodemographic characteristics of the pregnant women were therefore compared with those of the women with infants. Table 4 shows the proportion of women in the two groups that possess particular household and maternal characteristics favourable to better family health. Because the percentages of women are virtually the same for the two groups on all characteristics, it was concluded that the differences in infant mortality rates could not be explained by differences in these household or maternal factors. There was no reason to expect that pregnant women would have higher socioeconomic status than women with infants, but this possibility had to be ruled out. What then explains the very low infant mortality based on pregnant women? The problem can be examined from two different approaches, the under-reporting of deaths and of pregnancies.
440
D. Hubacher et al.
Table 4. Distribution (%) of pregnant women and women with infants by household characteristics, and maternal characteristics (means), based on EFHS, 1987 Characteristics
Pregnant women Women with infants
Household characteristics (%) Electricity Surfaced floor Non-wood fuel Toilet/outhouse Refrigerator Radio TV Adequate garbage disposal Piped water source < 1 hr to health centre Urban residence
32 43 19 51 14 60 26 44 58 56 28
32 41 19 52 14 61 23 44 58 58 30
Maternal characteristics (mean) Age (years) Education (years) Parity (live births)
261 4-4 3-9
26-7 41 3-7
Unreported infant deaths An undercount of infant deaths is the most obvious source of possible error to explain the low mortality rate. There are four distinct possibilities. Table 5 compares the age of death for three groups: all children born between 1982 and 1987 in the EFHS, the newborns of the cohort of pregnant women, and infants born since 1986. If, for example, the proportion of deaths occurring within the first 2 days of life was very low, it could be hypothesised that deaths were undercounted and perhaps misreported as stillbirths. However, the proportion of deaths occurring in thefirst2 days of life for the newborns is over 50%; this compares to 49% and 46% for the infant group and EFHS children, respectively. Thus, there is no evidence to suggest that infant deaths in the newborn group were misreported as stillbirths. This comparison of age at death would provide stronger evidence if the number of neonatal deaths to the newborn group were larger. A second possibility of undercounting deaths is linked to the denial of pregnancy. At the time of the second interview, fifteen women who had reported they were pregnant in 1987 denied the pregnancy and were subsequently removed from the data base. If pregnancy denial was in lieu of reporting an infant death, the overall infant mortality rate would be grossly underestimated. However, some of the fifteen women may have had irregular menstrual cycles and may not have been pregnant. According to data from the 1987 interviews, two women had not had a menstrual period since their last live birth, while most had not had a period in the last 40-70 days. The date of last menses for four women exceeded 5 months. On the other hand, some may have
Estimating infant mortality in Honduras
441
Table 5. Distribution of age at death for infants of the EFHS and follow-up survey (unweighted data) 1987 EFHS ilHrpn ch 1.11 11U1 til born 1982-87 Age at death
Follow-up survey, 1988 Newborns
< 1 year olds*
No.
%
No .
%
No.
%
Days 0 1 2 3 4-27
65 56 14 22 105
25 21 5 9 40
5
55
4
45
16 6 3 3 17
36 13 7 7 37
Total
262
100
9
100
45
100
Months
ESTIMATING INFANT MORTALITY RATES PROSPECTIVELY IN HONDURAS DAVID HUBACHER*, PATRICIA BAILEY*, BARBARA JANOWITZ*, FIDEL BARAHONAt AND MARCO PINELt * Family Health International, Research Triangle Park, North Carolina, USA, and iHonduran Ministry of Public Health, Tegucigalpa. Honduras Summary. In order to determine the validity of infant mortality estimates based on retrospective reporting, the Honduran Ministry of Health carried out a follow-up survey of women interviewed in a 1987 national survey. Women were interviewed approximately 14 months after the baseline survey and were asked about the outcomes of their pregnancies and the survival status of their young children. The overall infant mortality rate calculated from the follow-up survey was lower than that obtained from the baseline survey, due to the particularly low rate among the group of women who were pregnant at the time of the baseline survey. Possible explanations for this low rate are discussed. Introduction
Infant mortality rates are often used as an indicator of the level of socioeconomic development of a country. Reliable data to estimate infant mortality rates can come from several sources. Most industrialised countries derive their estimates from vital registration systems since births and deaths are well documented. In many countries, however, births and deaths are not registered as carefully as required to derive trustworthy estimates; censuses and sample surveys are often used as complementary. A census, usually defined as a complete canvassing of the population, does not always contain the information necessary to estimate infant mortality rates, is very expensive, and in many countries is undertaken infrequently. Sample surveys are therefore often relied upon to generate national estimates of infant mortality. From survey data, infant mortality rates can be estimated by indirect or direct methods (United Nations, 1983). Indirect estimates use data on the number of children ever born and surviving for women in different age groups (as defined at the time of the survey). Rates are calculated making assumptions about the age pattern and trends in mortality and fertility. Several models exist to estimate levels and trends of infant mortality. Direct methods compare the reported number of infant deaths to the number of live births in a given period; they are preferred over indirect techniques when sample sizes are sufficiently large and data quality is high. Large cross-sectional sample surveys alone do not guarantee better estimates, since validity is based on the 433
434
D. Hubacher et al.
accuracy of retrospective reporting of births and deaths. Common sources of error are the omission of births and deaths as well as misreporting of dates of birth and age at death. The most common omissions occur for children who no longer live with the mother, deaths that occurred several years before the survey, and children dying soon after birth who are incorrectly reported as still births rather than live births (United Nations, 1983; Rutstein, 1984). Omission of events is probably the most serious error because this leads to underestimates of the true levels of infant mortality. Prospective studies are another data collection alternative that can be used to estimate infant mortality rates; they can largely eliminate the possibility of omission as well as decrease the likelihood of misreporting dates of important events (Victora et al., 1987; Bailey et al., 1990). This paper presents the experience of such a study undertaken in Honduras in 1988; the historical context in which the prospective study was conceived is first reviewed. Trends in infant mortality
Infant mortality in Honduras has declined substantially over the past 30 years by perhaps as much as 70%, from 166 per 1000 live births in 1960 to 48 in 1987-88 (Guzman & Santos, 1988; Honduran Ministry of Public Health, 1989). However, the estimates of infant mortality come from several sources, use different techniques, and are not always in agreement (Fig. 1). The vital registration system in Honduras is a poor source of data for estimating infant mortality rates, as it substantially undercounts both births and deaths. The 1974 and 1988 Censuses provided mortality estimates as did a series of national surveys. These include two demographic surveys carried out under the direction of the Latin American Centre for Demography (CELADE), one in 1971-72 (EDENH-1) and the second in 1983 (EDENH-II), and two smaller national sample surveys—the 1981 Contraceptive Prevalence Survey (CPS) and the 1984 Maternal-Child Health/Family Planning Survey (MCH/FP). All of these surveys have used the Trussell models for indirect estimation of infant mortality. The 1981 CPS and 1984 MCH/FP surveys provided the lowest rates of infant mortality, but neither survey was designed specifically to focus on mortality. However, the Ministry of Health decided to make infant mortality a priority in the 1987 Epidemiology and Family Health Survey (EFHS) (Honduran Ministry of Public Health, 1989), and a large sample and a complete birth history were included in order to construct a strong data base for computation with direct and indirect techniques. Also, it was decided to select a subset of women (those who were pregnant and/or with an infant) to interview at a later date in order to measure infant mortality prospectively. Estimates from this project, the follow-up survey, would verify the estimates derived from the retrospective data collected through the birth histories of the 1987 survey. Estimates derived from the 1987 EFHS and the follow-up survey using life table analyses strengthen the evidence of a decline in infant mortality levels. This paper focuses on the methodological issues distinct to a prospective study of this type and attempts to explain why this technique has produced different levels of estimates for infant mortality in Honduras.
435
Estimating infant mortality in Honduras 180 160 140-
1201008060-
D O
A 40 20
1955
Census, 1974 CPS,1981 EDENH-H, 1983 MCH/FP, 1984 EFHS, 1987* Vital Statistics (adjusted) Census, 1988 Follow-up, 1988* 1960
1965
1970
1975
1980
1985 1988
Fig. 1. Infant mortality estimates from various sources, 1959-88. Sources: Guzman & Santos (1988); Honduran Ministry of Public Health (1989, 1990); INCAP (1990). * Direct rate estimates.
Methods In 1987 the Honduran Ministry of Health and the Association for Family Planning in Honduras administered the EFHS with technical assistance from Management Sciences for Health and Family Health International. This household probability survey covered a wide range of maternal and child health topics with an emphasis on estimating levels of fertility and infant mortality. Fieldwork for the EFHS was carried out between June and November, 1987; 10,159 women between the ages of 15 and 44 were interviewed. Of these, 2832 women were eligible for a follow-up interview, either because they were pregnant (995) or had a child under the age of 1 (784), or were pregnant and had a child under the age of 1 (53). The follow-up survey was conducted from September 1988 to February 1989, and the fieldwork followed the same visiting sequence as for the EFHS, thus allowing the
436
D. Hubacher et al.
maximum period possible between the two surveys (mean: 14 months; range: 11-18 months). The overall response rate for the follow-up survey was 74%. Of the 733 women not re-interviewed, 93% were not interviewed because they had moved, were not at home (after several contacts), or had died (five). Only six women refused a follow-up interview. Response rates by follow-up group were: pregnant women: 76-2% (758/995); women with infant: 73-5% (1312/1784); pregnant women with infant: 54-7% (29/53). The response rate in the capital city of Tegucigalpa was only 63%. Shortly before the fieldwork started, large sections of Tegucigalpa, where women eligible for the follow-up survey lived, were devastated by a hurricane. Sample and weighting A reasonable response rate is of utmost concern in any prospective study. The 74% response rate was lower than hoped for and led to close examination of those who were and were not re-interviewed. Table 1 shows the distribution of three groups of women (eligible, completed interviews, and lost to follow-up) according to household and maternal characteristics. Ideally, the characteristics of women with completed interviews and those of women lost to follow-up would mirror those of eligible women. However, this was not the case and there are important differences between the two groups. The lost to follow-up group has a higher proportion residing in urban areas and a higher proportion with more household amenities than the group with completed interviews. Also, the women in the lost to follow-up group are younger, better educated, and have fewer live births than the interviewed women. All chi-square tests of independence are significant at the 0-01 level. Because these characteristics are associated with infant mortality, the interviewed women, who are disproportionately representative of women of lower socioeconomic status, would be likely to exhibit higher infant mortality rates than the complete sample of women. Before calculating infant mortality rates, it was therefore necessary to adjust for non-response, and weighting factors were applied to the available data to compensate for loss to follow-up. The complete weighting scheme accounted for non-response at three different levels (Dominik, 1989). The first two levels were concerned with response rates at the area level (given the area sampling unit, or sector number), and the household level (giver the number of eligible respondents in the household); these numbers pertain to the 1987 EFHS. The third component of the overall weighting was concerned with response rates in the follow-up survey given a set of three identifying variables. These were chosen to define broad groups necessary to classify the women by follow-up group (pregnant, including those who were pregnant and had a live infant, or with infant), sanitary facilities (toilet, outhouse/other, or none), and parity (0 1,2-3,4-5, or 6 +). These three variables produced a set of 24 unique combinations of characteristics with response rates ranging from 42% to 87% (the overall follow-up response rate was 74%). The final weighting factor is the product of the inverse of the response rates for each of the three levels; these weightings were applied to each woman for whom complete information was available. These adjustments correct the data so that they better reflect the aggregate of responses that would have been expected had all 2832 women been successfully interviewed in the follow-up.
Estimating infant mortality in Honduras
437
Table 1. Percentage distribution of follow-up women by household and maternal characteristics* Characteristics Household Residence Urban Rural Source of water Faucet inside the home Faucet outside the home Other Electricity in home Yes No Sanitary facilities Toilet Outhouse and other None Maternal Age (years) 15-19 20-24 25-29 30-34 35-39 >40 Education None Primary 1 3 Primary 4-6 > 7 years Parity 0 1 2-3 4 5 >6 Follow-up group Pregnant women Women with infants Women with infants and pregnant Total No. of woment
Eligible follow-up women
Completed follow-up interviews
Lost to follow-up
29 71
25 75
40 60
13 42 45
11 42 47
16 44 40
31 69
28 72
39 61
19 32 49
15 33 52
28 29 43
15 29 24 16 11 5
14 27 24 17 12 6
20 35 24 12 6 3
22 33 31 14
23 34 30 13
20 29 32 19
27 31 19 23
24 30 20 26
36 33 18 13
35 63
36 63
32 65
2 100 (2825)
1 100 (2092)
3 100 (733)
All y2 tests of independence for completed and lost to follow-up groups are significant at 001. * At the time of the 1987 Epidemiology and Family Health Survey, t Seven women excluded from table because of missing data.
438
D. Hubacher et al. Results
Infant mortality rates All follow-up infant mortality rates were estimated using weighted data to adjust for non-response. As expected, since non-respondents were of higher socioeconomic status, the estimates were slightly lower than the unweighted rates. The SPSSx procedure SURVIVAL was used to calculate direct rates using standard life table analysis procedures. The 1987 EFHS contained complete histories which were used to estimate infant mortality rates directly. The 1985-87 life table rate of 48 per 1000 is based on a cohort of 5111 live births and is considered the best direct estimate provided by this survey. Several different infant mortality rates were calculated from the follow-up survey (Table 2). The estimate of 42 per 1000 live births uses all the children in the data set. Estimating rates according to the follow-up group (pregnant women and women with Table 2. Infant mortality estimates from the follow-up survey Time
IMR weighted
IMR unweighted
No. oflive births
1986-88 All children 95% CI
42/1000 (34-49)
43/1000
(2690)
1987-88 Newborns 95% CI
26/1000 (16-36)
28/1000
(771)
1986-88 < 1 year olds 95% CI
48/1000 (38-58)
49/1000
(1919)
infants) revealed the problem on which this paper focuses; the rate among infants (48 per 1000 live births) is much higher than the rate of 26 per 1000 live births for the pregnant women. When these group rates are examined according to different sociodemographic variables, the well documented patterns emerge (Table 3). In both follow-up groups, for instance, rural infants show a higher mortality rate than urban infants. Other predictable patterns are maintained for household characteristics, sex of the child, and maternal education; data from the 1987 EFHS are shown for comparison. It is particularly important to note these patterns in the newborn group; perhaps the estimate is low because of a weakness in the method, but not in the data. The mortality estimate from the pregnant women (newborns) was much lower than that for the infants. Higher estimates for the newborns in comparison to the infants might have been expected; information on early deaths for the newborns is likely to be more complete than for the older group.
Estimating infant mortality in Honduras
439
Table 3. Infant mortality rates from EFHS and follow-up survey, by selected characteristics Follow-up surveyt Characteristics
1987 EFHS*
Newborns
< 1 year olds
Residence Urban Rural
38 52
24 27
45 49
Electricity Yes No
35 54
16 32
34 54
Floor Dirt Other
56 36
34 17
56 37
Sex Male Female
54 42
31 21
54 43
Maternal education < 3 years > 4 years
55 39
49 4
62 30
No. of birthsj
(5113)
(771)
(1919)
* 1985-1987 births, t Weighted data. * Missing data on some variables reduced number of births in the analysis.
Low infant mortality among pregnant women One reason why such low rates were found for the pregnant women could have been that they were disproportionately representative of women of higher socioeconomic status. The sociodemographic characteristics of the pregnant women were therefore compared with those of the women with infants. Table 4 shows the proportion of women in the two groups that possess particular household and maternal characteristics favourable to better family health. Because the percentages of women are virtually the same for the two groups on all characteristics, it was concluded that the differences in infant mortality rates could not be explained by differences in these household or maternal factors. There was no reason to expect that pregnant women would have higher socioeconomic status than women with infants, but this possibility had to be ruled out. What then explains the very low infant mortality based on pregnant women? The problem can be examined from two different approaches, the under-reporting of deaths and of pregnancies.
440
D. Hubacher et al.
Table 4. Distribution (%) of pregnant women and women with infants by household characteristics, and maternal characteristics (means), based on EFHS, 1987 Characteristics
Pregnant women Women with infants
Household characteristics (%) Electricity Surfaced floor Non-wood fuel Toilet/outhouse Refrigerator Radio TV Adequate garbage disposal Piped water source < 1 hr to health centre Urban residence
32 43 19 51 14 60 26 44 58 56 28
32 41 19 52 14 61 23 44 58 58 30
Maternal characteristics (mean) Age (years) Education (years) Parity (live births)
261 4-4 3-9
26-7 41 3-7
Unreported infant deaths An undercount of infant deaths is the most obvious source of possible error to explain the low mortality rate. There are four distinct possibilities. Table 5 compares the age of death for three groups: all children born between 1982 and 1987 in the EFHS, the newborns of the cohort of pregnant women, and infants born since 1986. If, for example, the proportion of deaths occurring within the first 2 days of life was very low, it could be hypothesised that deaths were undercounted and perhaps misreported as stillbirths. However, the proportion of deaths occurring in thefirst2 days of life for the newborns is over 50%; this compares to 49% and 46% for the infant group and EFHS children, respectively. Thus, there is no evidence to suggest that infant deaths in the newborn group were misreported as stillbirths. This comparison of age at death would provide stronger evidence if the number of neonatal deaths to the newborn group were larger. A second possibility of undercounting deaths is linked to the denial of pregnancy. At the time of the second interview, fifteen women who had reported they were pregnant in 1987 denied the pregnancy and were subsequently removed from the data base. If pregnancy denial was in lieu of reporting an infant death, the overall infant mortality rate would be grossly underestimated. However, some of the fifteen women may have had irregular menstrual cycles and may not have been pregnant. According to data from the 1987 interviews, two women had not had a menstrual period since their last live birth, while most had not had a period in the last 40-70 days. The date of last menses for four women exceeded 5 months. On the other hand, some may have
Estimating infant mortality in Honduras
441
Table 5. Distribution of age at death for infants of the EFHS and follow-up survey (unweighted data) 1987 EFHS ilHrpn ch 1.11 11U1 til born 1982-87 Age at death
Follow-up survey, 1988 Newborns
< 1 year olds*
No.
%
No .
%
No.
%
Days 0 1 2 3 4-27
65 56 14 22 105
25 21 5 9 40
5
55
4
45
16 6 3 3 17
36 13 7 7 37
Total
262
100
9
100
45
100
Months
