Neonatal Mortality and its Correlates: Practical implications of Data Analytics across densely populated states in India

Abstract:

The Neonatal Mortality Rate in India is among the highest in the world. Non-availability of trained manpower especially in states with large population along with poor healthcare infrastructure is one of the major hurdles in ensuring quality neonatal care. This is a study of causes of high neonatal mortality rates in selected states of India such as Assam, Bihar, Chhattisgarh, Jharkhand, Madhya Pradesh, Odisha, Rajasthan, Uttar Pradesh, Uttarakhand to understand causes and improvements in healthcare.

Background:

The data has been downloaded from Open Government Data Platform India (data.gov.in), under the Indicators of Annual Health Survey (2012-2013). For this analysis, we have extracted the records related to the rural areas in the states of Assam, Bihar, Chhattisgarh, Jharkhand, Madhya Pradesh, Odisha, Rajasthan, Uttar Pradesh, Uttarakhand. In our extracted data, we deal with data from the 293 districts of the selected sates.

Methodology:

Development and Multiple Linear Regression analysis are performed using SAS-JMP pro 13.

Data Exploration:

The data set extracted is subdivided into Urban and Rural. The rural data collection was performed on a total of 30656 sample units covering a rural population size of 34374122. In a total of 284 districts surveyed the extracted data has no missing records for rural. However, there are missing records for urban data.

The data set consists of nearly 294 rows in which 70% is used for development of the model and the rest 30% for validation. Random sampling method is used for differentiating the rows in the data set.

Factors selected for Analysis:

·       Ante natal care 
·       Post Natal Care 
·       Janani Suraksha Yojana (JSY) 
·       Immunization, vitamin a & iron supplement and birth weight  
·       Breastfeeding and supplementation 
·       Mortality

Since we are studying factors associated with Neonatal mortality rate we have eliminated few factors which are irrelevant for our study. As per the AHS survey children surveyed were of the age group of 12-23 months, hence we have not considered Childhood Diseases as a factor of analysis.

Data Transformation and finding the best model:

• After eliminating the outliers multivariate analysis identified collinearity between below pairs:

·  POST NATAL CARE - New born(s) who were checked up within 24 hrs. of birth (%) – Rural, POST NATAL CARE - Mothers who received Post-Natal Check-up within 1 week of delivery (%) – Rural, POST NATAL CARE - Mothers who received Post-natal Check-up within 48 hrs. of delivery (%) – Rural

·  ANTE NATAL CARE - Mothers who consumed IFA for 100 days or more (%) – Rural, ANTE NATAL CARE - Mothers who had Full Antenatal Check-up (%) – Rural

• Bidirectional selection to reduce independent variable

Fit Model was plotted with all variable relevant to the neonatal mortality i.e., schooling status, work status (children), illness, injury, family planning, etc. have not been considered as these are irrelevant for neonatal mortality assessment. A neonatal death is defined as a death during the first 28 days of life (0-27 days), thus all parameters related to children (12-23 months as per the data) and adults have been eliminated.

After bidirectional selection, R Squared adjusted value 0.61, i.e., approx. 62% accuracy and F ratio as <0.0001* is reached which is very significant. The residuals are also closely clustered near the best fit line. This model gave the best results so far (all models have not been discussed here) and further analysis was performed.

*Note: For a good model R^2 adjusted should be closer to 1, however the dataset is small depending on the number of states surveyed, i.e. only 9 states of India in the year (2012-2013).

• Correlation of Estimates:

From the below picture, it has been evident that the variables considered for analysis are independent in nature.

Data Analysis:

• Validating Assumptions:

1.     Residuals must be normally distributed

·       The Residuals are almost normally distributed
·       Normal Q plot is linear

2.   Predicted – Residual plot should not have a pattern

No significant pattern in bivariate fit of predicted NMR and residual NMR. It supports a linear model.

Results:

As per the analysis we find 6 major factors affecting Neonatal Mortality Rate (NMR) in the selected rural states of India. The prediction expression is as follows:

 NMR increases with the following major factors:

· The children who did not receive breast milk.

· The quality of Ante natal care and Post Natal care in the district (Despiteit’s a positive factor, but from the prediction, we can infer that health centers in the rural state didn’t have adequate facilities ultimately contributing to NMR)

· Children who didn’t receive vaccination at the time of birth is also contributes to the rise of NMR.
· Mothers who availed financial assistance for institutional delivery under Janani Suraksha Yojana have a little contribution to the increase in NMR. It has been noted that financial assistance resources are not utilized efficiently.

NMR Decreases with the below factor:

. Children breastfed within an hour makes higher chance of reducing the NMR.

Conclusion:

As per the analysis factors contributing to NMR in the states of India are Immunisation given to the children, Quality of Ante natal care & Post-natal Care, Supplementation & Breastfeeding and Government Schemes for financial assistance while delivery.

Disclaimer: The analysis has been done using SAS-JMP Pro13 as part of my masters study at NUS and is purely for educational purpose. The content is my understanding of the subject, gathered during study and following journals on the subject. The cover image has been downloaded from the internet. This is preliminary analysis and will be modified with more knowledge acquired in this area.