Model Calibration \(\rightarrow\) Calibration Targets
We fitted the model to empirical primary data (i.e. not modeled estimates) for various maternal health indicators. We calibrated to data from 1990-2015, reserving estimates from 2016-2020 as a test set to assess the predictive accuracy of our model. We had 22,495 targets in the training set, and 1,525 estimates in the testing set.
Indicator | Training Set (1990-2015): # Countries (# Estimates) | Testing Set (2016-2020): # Countries (# Estimates) | Data Source |
---|---|---|---|
Fertility Indicators | |||
Total fertility rate (TFR) (age 15-49) | 186 (6,633) | 120 (249) | UN World Fertility data[1] |
Contraceptive prevalence (age 15-49 | 74 (235) | 20 (21) | DHS data |
Twinning rate | 107 (107) | 0 (0) | Smits 2011,[2] Pison 2015[3] |
Process Indicators | |||
Anemia prevalence (by severity) | 48 (291) | 16 (48) | DHS data |
Stillbirth rate | 93 (656) | 0 (0) | Blencowe 2016[4] |
Facility delivery % | 121 (286) | 84 (85) | DHS data, WHO GHO database[5] |
C-section rate | 161 (479) | 60 (69) | UNICEF Maternal and Newborn Health Coverage Database[6] |
Spontaneous/operative vaginal delivery % | 24 (48) | 0 (0) | WHO Global Survey on Maternal and Perinatal Health[7] |
Mortality Indicators | |||
Maternal deaths (total) (CRVS data) | 123 (2,311) | 70 (117) | WHO Mortality Database[8] |
Maternal deaths by cause (CRVS data) | 114 (11,216) | 69 (928) | WHO Mortality Database[8] |
Pregnancy-related mortality ratio | 56 (137) | 0 (0) | DHS data |
Miscellaneous (maternal deaths, maternal mortality ratio, pregnancy-related mortality ratio) | 22 (96) | 3 (8) | Country-specific sources |
Total Fertility Rate (TFR): We obtained estimates of total fertility rates from the UN World Fertility Data 2019.[1] The database includes data from civil registration systems, population censuses, and sample surveys available as of August 2019 and covers the time period from 1950 to the present.
Contraceptive Prevalence Rate: We estimated contraceptive prevalence rates (i.e. proportion of women ages 15-49 using any method of contraception) based on DHS data for 4,052,949 women from 256 surveys in 75 countries.
Twinning Rate: We obtained estimates of twinning rates (per 1,000 deliveries) for 107 countries from the medical literature.[2,3]
Anemia Prevalence: We estimated the prevalence of anemia by severity (mild/moderate/severe) based on DHS data for 98,558 pregnant women from 113 surveys in 52 countries.
Stillbirth Rate: We obtained estimates of stillbirth rates from Blencowe 2016,[4] which collated stillbirth data from national routine or registration systems, nationally representative surveys, and other data sources identified through a systematic review, web-based searches, and expert consultation. We restricted the extracted estimates to data which reported the number of stillbirths and were based on the 28 weeks definition of stillbirth. We only included estimates from national routine informations systems or nationally-representative household surveys (i.e. data types 1-3), excluding sub-national data. This yielded 656 estimates from 93 countries.
Facility Delivery %: We estimated the proportion of deliveries which occurred in a facility based on DHS data for 1,717,203 women from 261 surveys in 75 countries. We supplemented these estimates with country-level data from the Global Health Observatory on births attended by skilled health personnel[5] for 108 countries not available in the DHS data. We use the data on skilled birth personnel as a proxy for facility delivery.
C-Section Rate: We obtained data on national c-section rates by country from the UNICEF Maternal and Newborn Health Coverage Database.[6] 548 estimates were available for 166 countries included in the model between 1990 to 2020.
Spontaneous and Operative Vaginal Delivery Rate: We obtained estimates of spontaneous and operative vaginal delivery rates from the World Health Organization Global Survey on Maternal and Perinatal Health (WHOGS): a multicountry, facility-based survey that collected data for all women delivering in randomly selected facilities from randomly selected countries.[7] A total of 24 countries took part in the study. Data collection took place during 2004 and 2005 in Africa and the Americas and during 2007 and 2008 in Asia. As this is a facility-based survey, we compare these estimates to the mode of delivery among facility-based births in the model.
Maternal Deaths: We obtained estimates of total (reported) maternal deaths from the WHO Mortality Database which maintains data from Civil Registration Vital Statistics (CRVS) systems.[8] We extracted all maternal deaths as defined by ICD-8, ICD-9, and ICD-10 codes (see Appendix 2.3), depending on the country/year.
Maternal Deaths by Cause: We obtained estimates of maternal deaths by 8 causes (grouped by ICD-10 code, see Appendix 2.3) from the WHO Mortality Database.[8] We restricted these estimates to observations where the total reported maternal deaths equaled the sum of the specific causes, ensuring that there were no unspecified (or generally coded) maternal deaths included in these calibration targets.
Pregnancy-Related Mortality Ratio (PRMR): We obtained estimates of the pregnancy-related mortality ratio from DHS surveys, which were generally estimated using the sisterhood method to inquire about pregnancy-related deaths which occurred in the seven years preceding the survey. We therefore offset the year of the estimates to the midpoint of the seven-year period preceding the survey when calibrating the model.
Miscellaneous Maternal Mortality Estimates: We obtained various estimates of maternal mortality (maternal deaths, maternal mortality ratio [MMR], pregnancy-related mortality ratio [PRMR]) from available country-specific sources for countries for which we did not already have mortality estimates from the sources noted above. When calibrating the model we took into account the reporting mode (i.e. CRVS or survey) of each reported estimate. We summarize these other estimates of maternal mortality here.
We also compared our model results to estimates from other sources. These estimates were not used to calibrate the model, but allow us to assess the convergent validity of our model results with estimates from other sources.
Indicator | Source |
---|---|
Births (medium/low/high variants) | UN World Population Prospects 2019[9] |
Total fertility rate (TFR) | UN World Population Prospects 2019[9] |
Anemia prevalence (overall) among pregnant women | WHO GHO database[10] |
Contraceptive prevalence | UN World Contraceptive Use 2019[11] |
Maternal deaths | United Nations Maternal Mortality Estimation Interagency Group 2019,[12] GBD 2015 Maternal Mortality Collaborators[13] |
Maternal mortality ratio (MMR) | United Nations Maternal Mortality Estimation Interagency Group 2019,[12] GBD 2015 Maternal Mortality Collaborators[13] |
GMatH (Global Maternal Health) Model - Last updated: 28 November 2022
© Copyright 2020-2022 Zachary J. Ward
zward@hsph.harvard.edu