Regression Models
 |
In the last decade or so, model-based analyses have become important sources of global information on the prevalence of maternal mortality, largely because of the absence of reliable national level empirical data on maternal mortality throughout the world, particularly in developing countries.
United Nations (WHO/UNICEF/UNFPA) models use statistical regression to predict the proportion of maternal deaths among deaths of women of reproductive age. The regression coefficients are based on models built using countries with good data from both developed and developing countries. The independent (predictor) variables chosen need to be almost universally available and have an intuitive link to maternal mortality. The resulting models explain a high proportion of variance.
Predictors used vary but as shown below include general fertility rate (GFR), skilled attendants at delivery, gross national product (GNP), completeness of vital registration and regional dummy variables among others.
Initial attempts to model the maternal mortality ratio were abandoned in favour of modelling the proportion of maternal deaths of women of reproductive age (PMDF). This made better use of good registration data and used the more reliable characteristics of sisterhood data (the proportion of deaths that were pregnancy-related versus level of mortality). Four models have been developed so far: 1990, 1995, 2000, and 2005. There have also been other models built or proposed.
Advantages:
- models do not require any calculation or collection of maternal mortality data
Limitations:
- they are not empirically based
- precision is unknown in some models
- the quality of data used as predictors may be poor or non-existent
- the data on numbers of adult female deaths (used once PMDF is calculated) may be poor - only as good as the model and the predictor variables it comprises
1990 WHO/UNICEF Model*:
*NB 1990 Model not published
1995 WHO/UNICEF/UNFPA Model:
 |
2000 WHO/UNICEF/UNFPA:
- general fertility rate (births per 1000 women aged 15-49) (GFR)
- percentage of births by skilled attendant (SA)
- gross domestic product per capita based on purchasing power parity conversion (GDP/PPP)
- dummy variable for countries with complete death registration (VRComplete)
- dummy variable for Latin America, sub-Saharan Africa and the Middle East-North Africa (LASSAME)
 |
2005 WHO/UNICEF/UNFPA:
- general fertility rate (births per 1000 women aged 15-49) (GFR)
- proportion of births with skilled attendants (SKA)
- gross domestic product per capita based on purchasing power parity conversion (GDP)
- dummy variable equal to 1 if registration of deaths is 90% or more complete (VRcomplete)
- dummy variable identifying observations from Europe (EUR)
- dummy variable identifying observations from North Africa and the Middle East (MENA)
- dummy variable identifying observations from Western Pacific (WP)
 |
The UN models predict the proportion of maternal deaths among deaths of females of reproductive age (PMDF). This is then applied to estimates of adult female deaths and births to calculate the maternal mortality ratio (MMR), the maternal mortality rate (MMRate) and lifetime risk (LTR). For example, in 2005, the estimated PMDF was applied to the 2005 WHO figures for non-HIV/AIDS reproductive-aged female deaths, to obtain the estimated total number of maternal deaths. The number of maternal deaths divided by the 2005 UNDP estimates of the number of live births gave the point estimate for MMR in 2005. The uncertainty limits were derived from model estimates of the standard error of the forecast.
1990 Model:
 |
Documentation on the development of the 1990 models:
|
|
 |
|
 |
|
|
|
|
|
1995 Model:
|
|
|
|
|
|
Hill KH, AbouZhar C, Wardlaw T (2001) Estimates of Maternal Mortality for 1995. Bulletin of the World Health Organization, 79(3): 182-94. |
2000 Model:
|
|
|
|
|
2005 Model:
|
|
|
|
|