Contents - Previous - Next
This is the old United Nations University website. Visit the new site at http://unu.edu
Olivia Yambi, Michael C. Latham, Jean-Pierre Habicht and Jere D. Haas
Anthropometric measurements were taken on 2,452 children between 6 and 36 months of age at baseline and at two-month intervals thereafter in rural Tanzania. The children were followed for one year. All deaths occurring in this group were recorded using a village registration system introduced as part of a village nutrition status monitoring system. The relationship between the anthropometric indicators (weight for age, height for age, and weight for height, as well as weight increment) and subsequent mortality was assessed. The results indicate that over the one-year period, nutrition status was a significant predictor of mortality, with the probability of survival lower in children of low nutrition status. Low weight for age (<60% of the standard) was associated with a nine-fold increase in risk compared to weight for age above 80%; low weight for height (<80%) earned an almost fourfold increase in risk compared to weight for height>90%; and low height for age had a twofold increase in risk compared to the normal categories. An overall linear relationship was found between nutrition status and mortality, suggesting a gradual increase in mortality as nutrition status deteriorates. Incremental weight appears to be a good short-term predicator of mortality.
The relationship between anthropometric indicators of nutrition status and subsequent mortality has been documented in hospitalized children  and in free living population groups [2, 3].
Studies in the Indian subcontinent have reported both a linear increase in risk and a threshold phenomenon, indicating an increase in risk only when nutrition status had deteriorated to very low levels [3, 4]. This has raised questions as to the type of risk associated with mild to moderate malnutrition. In Africa, one group found no relationship between anthropometric indicators and mortality , raising the question of geographical and cultural specificity in the predictive ability of anthropometric indicators. Subsequent studies in Africa, in Guinea-Bissau and Malawi, reported an increased risk of mortality with poorer nutrition status and demonstrated no threshold effect [6, 7]. The predictive ability of weight gain in terms of risk of death has been inconclusive, mainly because so little work has been done investigating this, but it has been suggested that this is a good short-term indicator .
Knowledge of the relationship between nutrition status and functional outcomes has both programmatic and policy implications. Our study was undertaken to contribute to this knowledge by assessing the discriminatory power of anthropometric indicators as predictive of mortality in the specific population studied. The results were intended to guide decisions and operational guidelines for nutrition programmes in Tanzania.
Subjects and methods
A longitudinal study with a one-year follow-up period was undertaken between November 1985 and January 1987. It was conducted in 16 villages in the Iringa region of Tanzania that met specified eligibility criteria and were already participating in a large nutrition programme. All children between the ages of 6 and 36 months at the time of first measurement were enrolled in the study on consent of village leaders and parents, resulting in a total of 2,452 children. The start of the investigation in January coincided with the middle of the rainy season (the long rains in the region started in November), when weeding of the fields was in progress.
Preliminary information on the villages was collected and preparations for the study were finalized during November-December 1985. Baseline measurements and data collection started in January 1986. Height. weight, and arm circumference were measured. subsequently. weights were recorded every other month and heights were measured every four months' yielding a total of seven weight and four height measurements for each child surviving to the end of the follow-up. Each time the measurements were made, a two-week morbidity recall was undertaken by interviewing one parent, usually the mother. Other information collected included household characteristics, maternal histories, child feeding practices, and the survival status of children at each measurement.
Causes of death were determined by consulting the records of village health workers and reviewing information on the children's growth charts. At each contact the health of the children was discussed with their parents, nutrition and health advice was given, and, where warranted, appropriate actions were taken. The village health workers made household visits to check on the status of children as part of the regular programme implementation.
All anthropometric measurements were taken by the principal investigator and one other trained nutritionist; interviewing for morbidity was done by a trained medical assistant; and other information was collected by two other members of the team throughout the duration of the study.
All statistical analyses were done using the SAS and BMDP packages on the Cornell University IBM mainframe computer. As the actual dates of birth of all children were obtained, the exact ages of the children at the time of each measurement were calculated using the appropriate SAS programme. Distribution characteristics of the main dependent variables were checked for normality. The WHO/NCHS reference standards [8, 9] were used to convert anthropometric measurements into indicators for weight for age. height for age, and weight for height. Both percentage of medians and Z-score values were used in the analysis.
The mean weight and length/height for each age in months were plotted on NCHS charts. separate for males and females. to identify the percentile ranks within which the children tracked compared to the reference standards.
Several statistical tests were chosen to assess the relationship between nutrition status and mortality. The t test for group means and the chi-square test with specific cut-off points as appropriate were used to test for differences between dead and surviving children. Relative-risk estimates were calculated for different nutrition-status categories. The Kaplan-Meir method was used to calculate the probability of survival at three-month intervals. The log-rank test was employed to test for differences in survival distribution across nutrition-status categories. Logistic regression was used for multivariate analysis where mortality was the outcome variable. The Cox regression procedure was selected for covariate analysis.
The 2,452 children in the study ranged in age from 6 to 36 months (mean 18.1 ± 6.9 months). Fifty-two percent were female and 48% male. The means of the anthropometric indicators for all the children at baseline expressed as percentage of standard and Z scores are shown in table 1. The greatest deficit in terms of standard deviations below NCHS standards was in height for age. The mean length/height was close to the NCHS fifth percentile up to 12 months of age. after which it fell below this. Moderate and severe stunting (<90% height for age) was found in 26.8% of the children and mild stunting (90%-95% height for age) in 49.1%. The mean weight was between the NCHS fifth and twenty-fifth percentiles up to 12 months of age. Above that age it fell below the fifth percentile. About 1.3% of the children were severely underweight (<60% weight for age) and another 35.3% were between 60% and 80% weight for age. Severe wasting (<80% weight for height) was seen in 3.7% of the children; 21.7% were between 80% and 90% weight for height. Thus, overall, stunting was more common than wasting.
TABLE 1. Mean anthropometric indicators at baseline
TABLE 2. Mean weights for age (percentages of stem card) at baseline and last assessment before death-by cause of death
Weight for age
|Diarrhoea||46||73.6 ± 10.5||69.7 ± 11.5|
|Fever||30||79.3 ± 12.4||75.3 ± 19.6|
|Respiratory infection||7||72.8 ± 12.4||70.9 ± 9.0|
|Measles||3||85.4 ± 26.7||78.2 ± 29.6|
TABLE 3. Means of anthropometric indicators of dead and surviving
(N = 2,364)
|Weight/age (%)||76.3 ± 12.5||82.8 ± 10.4||-4.8||.0001|
|Height/age (%)||90.7 ± 4.9||92.3 ± 4.1||-3.0||.0034|
|Weight/ height (%)||91.2 ± 11/3||95.7 ± 9.0||-3.8||.0003|
Of the 2,452 children, 88 had died by the end of the follow-up period. This gives a mortality rate of 35.9 per 1,000 over the one year. Having obtained the dates of death' we were able to calculate the probability of survival at specified time intervals. Almost 80% of the deaths were recorded during the first six months after the baseline examination. Approximately 49% were in females and 51% in males, with gender-specific mortality rates of 33.8 and 38.1 per 1,000 respectively. The mean ages of children who died were 17.9 ± 6.9 months at baseline and 21.5 ± 7.5 months at the time of death. The largest proportion of deaths was recorded in the 12-18 month age group.
As anthropometric measurements were taken at two-month intervals, they were available to within a maximum of two months preceding death. The mean weight for age at baseline for the children who died was 76.3% ± 12.5%, and for the last assessment prior to death for this group it was 72.4% ± 12.6%. Almost all those who died showed a decline in nutrition status assessed by weight for age between measurements.
The majority of deaths were attributed to diarrhoea, followed by fever due to malaria. respiratory infections, measles, injury, and unspecified causes (table 2). Baseline weight for age was lowest in the group reported to have died from respiratory infections, followed by fever and measles.
Nutrition status and mortality
The means of the anthropometric indicators of nutrition status for children who died were significantly lower than those of survivors (p < .05) (table 3). Survival analysis showed significant differences by nutrition-status categories. Children who had greater weight deficit for age had a higher mortality rate than those with higher weight-for-age measurements (table 4). Comparable results were obtained using height for-age and weight-for-height measurements.
TABLE 4. Mortality by nutritional-status category
|Categorya||N||Deaths (no.)||Failure rate||SE|
|Weight for ageb|
|Height for agec|
|Weight for heightd|
a. Percentages of standards
b. Log - rank chi - square = 52.1 (p < .0001).
c. Log - rank chi - square = 25.96 (p < .0001).
d. Log - rank chi - square = 95.75 (p < .0001).
Estimates of relative risk
Estimates of relative risk showed that children who were severely malnourished using the indicator of weight for age, with the cut-off points used in the analysis of Chen et al. , were eight times more likely to die than those categorized as "normal." When 80% weight for age is used (as is usual practice in growth-monitoring programmes), the risk for severely underweight compared to non-underweight children was approximately nine-fold. Moderately and severely wasted children (weight for height <80%)) carried a risk about 2.5 times higher than those with 90% or greater weight for height. In the small group of children with severe stunting (<85% height for age: N= 88) the relative risk of mortality is estimated as about 2. In the moderately stunted group (85%-89% height for age) the relative risk of death was 2.4 times that in the group with height for age 95% or greater. The results of relative risk estimates are summarized in table 5.
TABLE 5. Estimates of relative risk of mortality, by nutritional-status category
a Percentages of standard.
TABLE 6. Ratio of mortality (per 1,000 over 12 months) between the lowest and highest 10% of children, by nutrition-status indicator
|Mortality rate||Ratio (lowest/highest)|
|Lowest 10%||Highest 10%|
To take into account the different numbers of children in each of the nutrition-status categories in the estimation of risk, we compared the lowest and highest 10% in terms of mortality. The results, summarized in table 6, show ratios between the two groups ranging from 3.3 to 6.4.
Relationship between nutrition status and mortality
To identify the type of curve relating nutrition status and mortality, and to compare the results with those of earlier studies, we calculated the mortality rates for 10-percentage-point intervals for weight for age and weight for height and 5-percentage-point intervals for height for age.
FIG. 1. Mortality rate by percentage of standard weight for age
FIG. 2. Mortality rate by percentage of standard weight for height
Figure 1. shows that mortality increased as weight for age decreased, a relationship that was largely linear. Below 90% weight for age, mortality almost doubled for the next 10 percentage points, and it increased about 1.5 times between 70%-79% and 60%-69% weight for age. Below 60% median weight for age, the mortality rate was three times that of the category just above.
Weight for height showed an overall linear relationship to mortality without any threshold effect (fig. 2).
The results for height for age (fig. 3) show a U-shaped curve, with mortality rising above and below the category of 90%-94% height for age. These results may be explained by the very few deaths in the upper values of height for age and a possible effect of age. As more than 50% of the deaths were in children under 18 months of age, we examined the data in two age ranges, with 18 months as a cut-off point. It was in the younger group that the risk of death increased with increase in height for age. In analysis with age at baseline as a covariate when indicators of nutrition status were in the model, age did not show a significant relationship with mortality.
FIG. 3. Mortality rate by percentage of standard height for age
Incremental weight and mortality
Analysis of the relationship between weight increment and subsequent mortality was undertaken by calculating the weight increment between two successive weighings and comparing the values between dead and living children for each of the periods. For the first and second increment periods. children who died had significantly lower weight gain than those who survived (t statistic = -3.22 for the first period and -3.73 for the second period; p < .05). The results for subsequent periods were not significant.
The weight increment between any two adjacent measurements carries error associated with each measurement. Regressing the weight gain on attained weight at the beginning of the interval and controlling for the child's age showed the weight of the child at the beginning of the interval to be a significant predictor of subsequent weight gain. Regression analysis was used to predict mortality from the first period's weight gain with the weight of the child at baseline (weight t1) and age as covariates. Controlling for the initial weight and age of the child, the weight increment was a significant predictor of mortality (beta = -1.743 ± 0.29; p < .05). Weight increment thus was a good short-term predictor of mortality. We could not carry out the regression analysis for subsequent periods due to the small number of deaths beyond the first few months of follow-up.
This study confirmed earlier results indicating that malnutrition is associated with mortality. It contradicts findings of one study in Africa  that no such relationship exists, but those results may have been due to methodological problems, thus indicating that the relationship is not unique to certain population subgroups. The relationship observed was largely linear, and we could not demonstrate a threshold level above which nutrition status did not become a determinant of mortality in the anthropometric categories used. This finding differs from that of Chen et al. .
The fact that mortality was higher in all categories with deterioration of nutrition status has important implications for efforts to address nutrition problems. For example, only 1.3% of the children were severely underweight, whereas about 40% were moderately underweight. If indeed nutrition status is related to mortality among moderately malnourished children, this group also requires serious attention even if immediate efforts focus on those at greatest risk. The resources required to deal with the larger numbers of children will be higher. After controlling for biological characteristics and other factors not discussed here, poor nutrition status remains significantly associated with mortality. Therefore, programmes aimed at reducing mortality must include explicit efforts to reduce malnutrition and not just some by-product of other general activities.
Our finding of a U-shaped relationship between height for age and mortality raised questions about the adequacy of the sample size and possibly age confounding at different ends of the distribution. An analysis with age showed that this variable was not significantly related to mortality. The study in Guinea-Bissau showed similar aberrant results with an inverted U curve . Since other analyses discussed here show significant association between height for age and mortality, the seemingly aberrant results have to be interpreted in conjunction with them.
Evaluation of indicators showed the benefits of weight for age and weight for height in predicting mortality. Even with the limitations of sample size, weight increment appeared as a significant indicator of risk of death. Whereas the choice of an indicator is eventually determined by the specific operational and programmatic conditions, our results suggest that weight for age is a useful indicator for purposes of risk assessment at the community level. This, combined with weight increment, should assist health workers and parents in identifying children requiring focused attention.
We observed that nutrition status can deteriorate rapidly. Given the attendant high risk of mortality, the frequency of contact in growth-monitoring and growth-promotion programmes may be crucial. Our data show that poor growth increments for the last two weighing periods preceding death were associated with a high risk of mortality. Given the ever increasing burden placed on workers in health institutions, trained village workers could provide continuous follow-up in the communities they serve.
This research was supported with funds from the Iringa Nutrition Programme. The principal investigator thanks all those who made it possible to conduct the research, and especially the Tanzania Food Nutrition Centre, Iringa Region, and the staff of the UNICEF-Tanzania country office at the time of the study.
1. Gomez F, Ramos GR, Frank S, Cravioto M, Chavez JM, Vazquez J. Mortality in second and third degree malnutrition. J Trop Pediatr 1956;2:77-83.
2. Kielman AA, McCord C. Weight-for-age as an index of risk of death for children. Lancet 1978;1:1247-50.
3. Chen LC, Chowdhury AKM, Huffman SL. Anthropometric assessment of protein-energy malnutrition and subsequent risk of mortality among preschool age children. Am J Clin Nutr 1980;33:1836-45.
4. Bairagi R, Chowdhury MK, Kim YJ, Curlin GT. Alternative anthropometric indicators of mortality. Am J Clin Nutr 1985;42:296-306.
5. Kasongo Project Team. Anthropometric assessment of young children's nutritional status as an indicator of subsequent dying. J Trop Pediatr 1983;29:69-75.
6. Smedman L, Sterky G, Mellander L, Wall S. Anthropometry and subsequent mortality in groups of children aged 6-59 months in Guinea-Bissau. Am J Clin Nutr 1987;46:369-73.
7. Lindskog U, Lindskog P, Carstensen J, Larson Y, Gebre-Medhin M. Childhood mortality in relation to nutritional status and water supply: a prospective study from Malawi. Acta Paediatr Scand 1988;77:260-88.
8. Hamill PW, Dritz TA, Johnson CL, Reed RB, Roche AF. NCHS growth curves for children birth-18 years. Vital and Health Statistics Series 11, no. 15. Washington, DC: Department of Health, Education, and Welfare, 1977.
9. WHO. A growth chart for international use in maternal and child health care. Geneva: WHO, 1978.
Contents - Previous - Next