Under-five mortality among mothers employed in agriculture: findings from a nationally representative sample

Data

Data for the study was downloaded from the Demographic and Health Survey (DHS) website. The nationally representative household survey, conducted by the International Institute for Population Sciences and Macro International from November 2005 to August 2006, is based upon stratified, multistage cluster sampling with two stages at rural and three stages at urban level. All women between the ages of 15–49 interviewed in 29 states in 5 regions were asked to self-report on the birth and death history. Data is extracted for all child births that had taken place in the 59 months preceding the survey, and time in months is noted for each reported death during this period. These procedures generate a sample of 124,385 female participants at 95% response rate with data on 56,438 births. Those alive at the date of survey are censored for the analysis, and the duration of live birth in months is calculated from the date of birth to the date of survey of that subject for each birth order separately. Data is further extracted to get a sub-sample for our study that consists of mothers employed in the agricultural sector. Information on first to fifth parity is available for a total of 12,291 births that took place in the 59 months preceding the survey. Parity six and above are too few and hence excluded from the study.

Measures

Family characteristics are assessed at each index child using single items regarding region, wealth index, standard of living, religion, caste, size of household and father’s education and occupation. National regions of residence are defined as North, Central, East, North East, West and South. These categorizations are based on those created by the International Institute for Population Sciences (IIPS). The IIPS took information on 30 household items and categorised the household into five quintiles with Poorest to the Wealthiest on a scale of 1 to 5. Religion is denoted as Hindu, Muslim, Christian and others. Caste is categorised as scheduled castes, scheduled tribes, other backward castes and others. Size of household is categorised as ≤5 members, 6–10 members and ≥11 members in our study. Parental education is classed into no formal education, primary education and secondary education and above, including graduation and post-graduation. Father’s occupation is categorised as no employment, employment in agriculture and others.

Characteristics of the child are assessed by gender and size at birth. Child size at birth is categorized as very large, larger than average, average, smaller than average and very small. These categories are combined to above average, average and below average. Characteristics of the mothers under study are her age, education, BMI, and anemia. Since the mother’s age at index child shows a curvilinear relation with child mortality, the mother’s age is coded as ≤21 years, 22–23 years, 24–26 years and >26 years to satisfy linearity and proportionality assumptions for Cox PH regression analysis. Body Mass Index (BMI) is based on height and weight measurements obtained by the interviewer at the time of interview and classed as underweight, normal and obese, according to WHO guidelines. Anemia is categorised as severe, moderate and no anemia.

Utilization of health care services are characterised as before and after delivery. All answers related to access to health facilities are coded according to the NFHS guidelines. Place of antenatal care is coded as at home, Government hospital, or private hospital/village/Non-governmental Organisation. Received Prenatal care, Pregnancy Health Education Index, and Breastfeeding Health Nutrition Education are coded as yes/no. Place of delivery is classed as home or hospital. Birth order is taken from Birth Order 1 to 5. Breastfeeding, Immunization and whether they were told of complications in pregnancy are also coded as yes/no.

Data analysis and statistical methods

Demographic characteristics and outcome measures are described using weighted percentages. Univariate Cox Proportional Hazard (PH) analysis is used to determine factors associated with child mortality. Kaplan–Meier curves are used to satisfy PH assumption. Significant variables at univariate analysis with p-value ≤ 0.10 and modified appropriate categories are considered for multivariate Cox PH regression after checking co-linearity among the explanatory variables. Forward stepwise Cox PH method is applied to assess significant explanatory variables of child mortality among mothers employed in agriculture and adjusted Hazard Ratios (HR) with 95% confidence intervals (CI) are presented. Cross-validation of confidence intervals calculated for HRs derived by multivariate Cox Proportional Hazard analysis are internally validated via bootstrapping re-sampling methods using 100 re-samples (Cox, 1972; Harrell, 2001). The predictive accuracy of the model is tested through calculating the area under the receiver operating characteristic (AUROC) curve (Tripepi et al., 2009). The ROC curve is used to quantify the overall accuracy of a model in differentiating outcome groups and is a good measure of its predictive ability. The area under the curve ranges from 0.5 to a theoretical maximum of 1. Acceptable discrimination is represented by an area under the curve of 0.70–0.79, and good discrimination by an area under the ROC curve >0.80. Perfect discrimination corresponds to a c-statistic of 1 and is achieved if the scores for all the cases are higher than those for all the non-cases, with no overlap. The nonparametric method does not make any assumptions about the distributions of test results between the two groups. It is used to calculate Standard Error (SE) and 95% CI of the ROC. P-value ≤ 0.05 (two tailed) is considered for statistical significance. SPSS 20.0 statistical package is used for the analysis (IBM Corp., 2012).

No identifying information is available in the data precluding the need for any ethical approvals. We wish to thank the DHS for making this data available for our study.

Population characteristics

Family characteristics of the population are described in Table 1. Data is available for a total of 9,532 births in un-weighted and 12,291 births in a weighted sample of mothers employed in the agricultural sector. Of all births, 31.2% occur in the Central region, 23.6% in the East region, 14.9% in the West region, 14.4% in the South region, 14.2% in the North region, and 1.7% in the North East region. 70% belong to poor and poorest wealth index while only 9.6% are in richer and richest wealth index. Nearly 43.6% households belong to low and 40% to medium standard of living. 90% are Hindu, 5.9% are Muslim, 1.8% are Christian and 2.2% are classed as others. 23.1% belong to scheduled castes, 19.1% belong to scheduled tribes and 46.3% belong to other backward castes. 54.3% , 37% and 8% deaths occur in households with ≤5 members, 6–10 members and ≥11 members, respectively. 55% of fathers received education either less than or up to primary school. 57% of fathers are employed in agriculture while the rest are in service elsewhere.

Table 2 presents the characteristics of the child and mother. Of the 12,291 total births 457 male and 455 female children died during the chosen period of study. 22% of children are born below average size. Nearly 68% of mothers have no formal education while 15% only attain primary education. 45.5% of mothers are underweight while only 3.2% are obese. Nearly 36% of mothers are not anemic, 41.6% are mildly anemic while 20.4% are moderately anemic and 1.9% are severely anemic.

Utilization of the health care services are presented in Table 3. For antenatal care, the majority (56%) go to government hospitals, 31% go to private hospitals while 13% receive care at home. 48.7% receive prenatal care; 48.6% receive pregnancy health nutrition education; and 10.5% receive breastfeeding health nutrition education. Over 77% are home deliveries while 23% are delivered in hospitals. 99% receive assistance at delivery; pregnancy complications occur in 8.7%. Children receive breastfeeding in 70% of cases. Of the total births, 24.6% are of Birth Order 1, 27.4% are Birth Order 2, 21.3% are Birth Order 3, 15.4% are Birth Order 4 and 11.2% are Birth Order 5. Immunization services are used by 7.1%.

Predominantly, mothers working in agriculture are Hindu, belong to other backward castes, receive less than or up to primary education, and are in poor or poorest households with 6–10 members. The majority of the children born are average or above average size. Most mothers receive no prenatal care and are not told of pregnancy complications. Most of the children are born at home and never immunized. A large number of children who die belong to households with less than or equal to 5 members and were not breastfed.

Univariate analysis

Univariate Analysis of family characteristics associated with child mortality are represented in Table 4. Households at the poorer and poorest wealth indices are risk factors compared with middle (HR 0.72, 0.60–0.88, 0.001) and richer and richest (HR 0.49, 0.37–0.66, p = 0.001) wealth indices. Medium Standard of Living (HR 0.75, 0.63–0.89, p = 0.001) and high standard of living (HR 0.54, 0.42–0.70, p = 0.001) act as protective factors with rise in standard of living corresponding to reduction in child mortality. Other backward castes (HR 0.80, 0.64–0.98, p = 0.03) are at less risk compared to scheduled castes. There is no statistically significant difference between scheduled castes and scheduled tribes. Size of household significantly reduces child mortality for households with 6–10 members (HR 0.57, 0.49–0.67, p = 0.001) and for households with ≥11 members (HR 0.57, 0.44–0.77, p = 0.001) in comparison to ≤5 members. Region, religion, the father’s education and occupation are not significant factors.

Table 5 presents the Univariate analysis of characteristics of child and mother associated with child mortality. Male and female children are equally at risk. Below average child size is a risk factor (HR 1.65, 1.32–20.06, p = 0.001) for child mortality. Risk of child mortality decreases with increase in the mother’s age from less than 21 years to above 26 years. The mother’s education is a protective factor with secondary level and higher acting as a significant protective factor (HR 0.49, 0.38–0.61, p = 0.001). With improvements in the mother’s anemic level, deaths are fewer in number. Not having anemia is a protective factor (HR 0.60, 0.35–1.0, p = 0.05).

Univariate analysis of utilization of health care services are described in Table 6. Prenatal care decreases mortality by 53% (HR 0.47, 0.39–0.55, p = 0.001). Breastfeeding health nutrition education reduces mortality by 59% (HR 0.41, 0.28–0.59, p = 0.001). Being told of pregnancy complications decreases child mortality (HR 0.59, 0.42–0.81, p = 0.001). Breastfeeding is a protective factor, reducing the risk of child mortality by 64% (HR 0.36, 0.30–0.41, p = 0.001). Birth order is a statistically significant factor with survival at Birth Order 1 being at significant risk compared to higher birth orders. Mortality significantly decreases with increasing birth order. Place of antenatal care services, BMI, pregnancy health nutrition education, place of delivery, ever immunization and assistance at delivery did not present themselves as significant factors for under-five mortality among mothers employed in agriculture.

Multivariate analysis

The results of multivariate analysis of factors associated with child mortality are shown in Table 7. A high standard of living significantly reduces child mortality (HR 0.68, 0.52–0.89, p = 0.001) in comparison to a low sandard of living. Size of household reduces mortality in households with 6–10 members by 40% (HR 0.60, 0.50–0.72) and in households ≥11 by 34% (HR 0.66, 0.48–0.91), respectively, in comparison to households with ≤5 members. Increase in the mother’s age corresponds with reduction in child mortality. Receiving prenatal care (HR 0.40, 0.34–0.48, p = 0.001) is a highly significant protective factor in child mortality. Breastfeeding health nutrition education is a significant protective factor (HR 0.45, 0.31–0.66, p = 0.001). Breastfeeding reduces child mortality significantly by 70% (HR 0.30, 0.25–0.35, p = 0.001). Birth order is not a statistically significant factor in multivariate analysis though Birth Order 5 is at a slightly higher risk than Birth Order 1.

The area under the receiver operating characteristic (ROC) curve, measuring the ability of the model to discriminate mortality outcomes (death/no death) and to evaluate the model’s practical ability for correctly discriminating a subject’s outcome is shown in Fig. 1. Bootstrapping analysis shows only a little bias in estimated parameters. The Area Under ROC is 0.75 (95% CI [0.73–0.77]), indicating that the developed model is fair enough for appropriate prediction of child mortality containing true area between 0.73 and 0.77 with 95% surety.