Identification of distinct risk subsets for under five mortality in India using CART model: an evidence from NFHS-4
# Background The objective of this study was to find the distinct risk subsets or clusters identified by the combination of factors and important factors to classify under five mortality (U5M) in high focused Indian states. # Methods Using population-based cross-sectional data from the National F...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Inishmore Laser Scientific Publishing Ltd
2020-06-01
|
| Series: | Journal of Global Health Reports |
| Online Access: | https://doi.org/10.29392/001c.13169 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1850100872092057600 |
|---|---|
| author | Vineet K Kamal Sharad Srivastav Dolly Kumari Mukesh Ranjan |
| author_facet | Vineet K Kamal Sharad Srivastav Dolly Kumari Mukesh Ranjan |
| author_sort | Vineet K Kamal |
| collection | DOAJ |
| description | # Background
The objective of this study was to find the distinct risk subsets or clusters identified by the combination of factors and important factors to classify under five mortality (U5M) in high focused Indian states.
# Methods
Using population-based cross-sectional data from the National Family Health Survey (NFHS, 2015-2016) on 1, 40, 427 live births of five years preceding the survey occurred to 99,205 women of high focused Indian states with U5M rate above the national level, a recursive partitioning approach based two classification tree models, one without considering missing values and other with missing together approach, were fitted using binary outcome of U5M and independent factors comprising of socioeconomic, demographic, maternal and biological, nutritional and environmental factors.
# Results
There were nine and sixteen sub-groups in model-1 and model-2, respectively. In model-1, breastfeeding = no & birth in past 5 years = (2, 3+ births) and in model-2, breastfeeding = no & birth weight = (<2.5kg, not known) & birth in past 5 years = (2, 3 or more births) were found to be maximum mortality risk sub-groups. In terms of variable importance to predict U5M, model-1 identified birth in past 5 years, breastfeeding, birth order, wealth index, mother‘s age at birth. Model-2 additionally identified delivery complications, birth weight, state, sanitation facility, birth interval, caste, education. Overall correct classification rate was higher for model-1 (66%) than model-2 (64%).
# Conclusions
The main observed risk cluster was combination of two factors like breastfeeding and number of births in past 5 years, which for most people are easily modifiable with appropriate strategies and policies. Finally, to combat U5M in high focused states, identifying risk subsets or clusters is important for targeting and intervening purposes, as the intensity and type of policies and programs may differ according to clusters. This method is suitable to identify complex natural interactions between predictors, important variables and hypothesis generation to inform policy maker on intervention strategies, which may be difficult or impossible to uncover using traditional multivariable techniques. |
| format | Article |
| id | doaj-art-00a5c927f1044700a7c491838e83034b |
| institution | DOAJ |
| issn | 2399-1623 |
| language | English |
| publishDate | 2020-06-01 |
| publisher | Inishmore Laser Scientific Publishing Ltd |
| record_format | Article |
| series | Journal of Global Health Reports |
| spelling | doaj-art-00a5c927f1044700a7c491838e83034b2025-08-20T02:40:11ZengInishmore Laser Scientific Publishing LtdJournal of Global Health Reports2399-16232020-06-01410.29392/001c.13169Identification of distinct risk subsets for under five mortality in India using CART model: an evidence from NFHS-4Vineet K KamalSharad SrivastavDolly KumariMukesh Ranjan# Background The objective of this study was to find the distinct risk subsets or clusters identified by the combination of factors and important factors to classify under five mortality (U5M) in high focused Indian states. # Methods Using population-based cross-sectional data from the National Family Health Survey (NFHS, 2015-2016) on 1, 40, 427 live births of five years preceding the survey occurred to 99,205 women of high focused Indian states with U5M rate above the national level, a recursive partitioning approach based two classification tree models, one without considering missing values and other with missing together approach, were fitted using binary outcome of U5M and independent factors comprising of socioeconomic, demographic, maternal and biological, nutritional and environmental factors. # Results There were nine and sixteen sub-groups in model-1 and model-2, respectively. In model-1, breastfeeding = no & birth in past 5 years = (2, 3+ births) and in model-2, breastfeeding = no & birth weight = (<2.5kg, not known) & birth in past 5 years = (2, 3 or more births) were found to be maximum mortality risk sub-groups. In terms of variable importance to predict U5M, model-1 identified birth in past 5 years, breastfeeding, birth order, wealth index, mother‘s age at birth. Model-2 additionally identified delivery complications, birth weight, state, sanitation facility, birth interval, caste, education. Overall correct classification rate was higher for model-1 (66%) than model-2 (64%). # Conclusions The main observed risk cluster was combination of two factors like breastfeeding and number of births in past 5 years, which for most people are easily modifiable with appropriate strategies and policies. Finally, to combat U5M in high focused states, identifying risk subsets or clusters is important for targeting and intervening purposes, as the intensity and type of policies and programs may differ according to clusters. This method is suitable to identify complex natural interactions between predictors, important variables and hypothesis generation to inform policy maker on intervention strategies, which may be difficult or impossible to uncover using traditional multivariable techniques.https://doi.org/10.29392/001c.13169 |
| spellingShingle | Vineet K Kamal Sharad Srivastav Dolly Kumari Mukesh Ranjan Identification of distinct risk subsets for under five mortality in India using CART model: an evidence from NFHS-4 Journal of Global Health Reports |
| title | Identification of distinct risk subsets for under five mortality in India using CART model: an evidence from NFHS-4 |
| title_full | Identification of distinct risk subsets for under five mortality in India using CART model: an evidence from NFHS-4 |
| title_fullStr | Identification of distinct risk subsets for under five mortality in India using CART model: an evidence from NFHS-4 |
| title_full_unstemmed | Identification of distinct risk subsets for under five mortality in India using CART model: an evidence from NFHS-4 |
| title_short | Identification of distinct risk subsets for under five mortality in India using CART model: an evidence from NFHS-4 |
| title_sort | identification of distinct risk subsets for under five mortality in india using cart model an evidence from nfhs 4 |
| url | https://doi.org/10.29392/001c.13169 |
| work_keys_str_mv | AT vineetkkamal identificationofdistinctrisksubsetsforunderfivemortalityinindiausingcartmodelanevidencefromnfhs4 AT sharadsrivastav identificationofdistinctrisksubsetsforunderfivemortalityinindiausingcartmodelanevidencefromnfhs4 AT dollykumari identificationofdistinctrisksubsetsforunderfivemortalityinindiausingcartmodelanevidencefromnfhs4 AT mukeshranjan identificationofdistinctrisksubsetsforunderfivemortalityinindiausingcartmodelanevidencefromnfhs4 |