Parinaz Mehdipour1, Iman Navidi2, Mahboubeh Parsaeian2, Younes Mohammadi2, Maziar Moradi Lakeh3, Ehsan Rezaei Darzi2, Keramat Nourijelyani4, Farshad Farzadfar5. 1. Department of Epidemiology and Biostatistics, School of Public Health, Tehran University of Medical Sciences, Tehran, Iran, Non-Communicable Disease Research Center, Endocrinology and Metabolism Research Population Science Institute, Tehran University of Medical Sciences, Tehran, Iran. nourik@tums.ac.ir, f-farzadfar@tums.ac.ir. 2. Department of Epidemiology and Biostatistics, School of Public Health, Tehran University of Medical Sciences, Tehran, Iran, Non-Communicable Disease Research Center, Endocrinology and Metabolism Research Population Science Institute, Tehran University of Medical Sciences, Tehran, Iran. 3. Community Medicine Department, Iran University of Medical Sciences, Tehran, Iran. 4. Department of Epidemiology and Biostatistics, School of Public Health, Tehran University of Medical Sciences, Tehran, Iran. 5. Non-Communicable Disease Research Center, Endocrinology and Metabolism Research Population Science Institute, Tehran University of Medical Sciences, Tehran, Iran, Endocrinology and Metabolism Research Center, Endocrinology and Metabolism Institute, Tehran University of Medical Sciences, Tehran, Iran.
Abstract
BACKGROUND: Searching for the latest methods of estimating mortality rates is a major concern for researchers who are working in burden of diseases. Child mortality is an important indicator for assessing population health care services in a country. The National and Sub-national Burden of Diseases, Injuries, and Risk Factors (NASBOD) is conducted in Iran with comparative methods and definitions of Global Burden of Disease (GBD) 2010 to estimate major population health measures including child mortality rate. The need to have accurate and valid estimation of under-5 mortality rate led to apply more powerful and reliable methods. METHOD: The available datasets consist of under-five mortality rates from different sources including death registration systems and summary birth history (SBH) questions from censuses and Demographic Health Survey. These datasets are gathered at national and sub-national levels. We have five time series of under-five mortality rates from SBH method that each one contains 25-year time period. We also calculated Child mortality rates from death registration for 5 years. The main challenge is how to combine and integrate these different time series and how to produce unified estimates of child mortality rates during the course of study. By synthesizing the result of other models, Gaussian Process Regression (GPR) is used as the final stage for generating yearly child mortality rates in this study. GPR is a Bayesian technique that uses data information and defines several hierarchical prior parameters for model. In corporation of GPR and MCMC methods, predicted rates are updated using data and defined parameters in model. This method, also captures both sampling and non-sampling errors and provides uncertainty intervals. The existence of uncertainty for predicting mortality rate is one of the considerable advantages of GPR that distinguish it from other alternative methods. DISCUSSION: Estimating accurate and reliable child mortality rates at national and sub-national levels is one of the important parts of NASBOD project in Iran. Gaussian Process Regression with its special features improves achievement of this goal. GPR is a serious competitor for other supervised mortality predictive methods. This article aims to explain the application and preferences of GPR method in estimating child mortality rate.
BACKGROUND: Searching for the latest methods of estimating mortality rates is a major concern for researchers who are working in burden of diseases. Child mortality is an important indicator for assessing population health care services in a country. The National and Sub-national Burden of Diseases, Injuries, and Risk Factors (NASBOD) is conducted in Iran with comparative methods and definitions of Global Burden of Disease (GBD) 2010 to estimate major population health measures including child mortality rate. The need to have accurate and valid estimation of under-5 mortality rate led to apply more powerful and reliable methods. METHOD: The available datasets consist of under-five mortality rates from different sources including death registration systems and summary birth history (SBH) questions from censuses and Demographic Health Survey. These datasets are gathered at national and sub-national levels. We have five time series of under-five mortality rates from SBH method that each one contains 25-year time period. We also calculated Child mortality rates from death registration for 5 years. The main challenge is how to combine and integrate these different time series and how to produce unified estimates of child mortality rates during the course of study. By synthesizing the result of other models, Gaussian Process Regression (GPR) is used as the final stage for generating yearly child mortality rates in this study. GPR is a Bayesian technique that uses data information and defines several hierarchical prior parameters for model. In corporation of GPR and MCMC methods, predicted rates are updated using data and defined parameters in model. This method, also captures both sampling and non-sampling errors and provides uncertainty intervals. The existence of uncertainty for predicting mortality rate is one of the considerable advantages of GPR that distinguish it from other alternative methods. DISCUSSION: Estimating accurate and reliable child mortality rates at national and sub-national levels is one of the important parts of NASBOD project in Iran. Gaussian Process Regression with its special features improves achievement of this goal. GPR is a serious competitor for other supervised mortality predictive methods. This article aims to explain the application and preferences of GPR method in estimating child mortality rate.
Authors: Peter M Macharia; Emanuele Giorgi; Pamela N Thuranira; Noel K Joseph; Benn Sartorius; Robert W Snow; Emelda A Okiro Journal: BMC Public Health Date: 2019-02-04 Impact factor: 3.295