(Click here to return to Research & Statistics page)

(Click here to return to Homepage)

 

 

 

Predicting Infant Mortality

 

 

 

 

 

 

A paper submitted in partial fulfillment of the

Requirements for HCA 5311 Health Care Research Methods:

Design and Analysis

28 November 2000

 

 

 

 

 

 

 

Introduction

            The United States has experienced a significant improvement in infant mortality rates over the last decade.  A large portion of this improvement is attributed to advances in medical treatment of very low birth weight infants (Hamvas, 2000), which illustrates U.S. health care reliance on costly medical interventions rather than on preventive measures.  Greater improvement is possible if health officials focused more on prevention of risk factors contributing to infant mortality.  However, the multiple factors that influence infant mortality are not adequately understood (McCormick, 1999).  A reliable method to assess multiple risk factors would allow health officials to focus interventions on preventive measures.  This can reduce expensive medical requirements and the overall infant mortality in a specific population.

            Infant mortality rates are a concern in this country as they represent an essential indicator of nation’s health (Stuart-Brown, 2000).  A number of indicators predict infant mortality rates.  Per capita income is often cited as an indicator, yet the U.S. infant mortality rate ranks behind many poorer countries, suggesting that wealth does not accurately reflect a nation’s health (St. Peters & Tarlov, A. R., 1999).  Other commonly used indicators include income; mothers’ education level; distribution of available health care; low birth weight and many others (MacDorman & Atkinson, 1999).  However, the predictive ability of these measures is not consistently demonstrated (Robinson & Wharrad, 2000).  Much of the difficulty is due to the multiple risk factors associated with infant mortality, which can serve to obscure risk factors of greater significance.  A focus on preventing the most significant risks is needed to influence infant mortality before birth (Flaherty, 1999).

The purpose of this study is to evaluate seven independent variables as predictors of infant mortality within a population.  Variables were selected for their ability to include multiple risks associated with infant mortality as described by various studies (Lerner, 1995; Luker, 1997; Short, 1999;).  Establishing confidence in aggregate predictors of infant mortality allows health officials to identify populations with multiple risk factors.

Statistical Methods

This study hypothesized that infant mortality is a function of seven selected independent variables.  If this relationship holds, a model is created to predict a population’s infant mortality risk if provided that population’s data for a risk factor.  The study also explores the statistical correlation between independent variables as justification to use aggregated risk factors to predict infant mortality.

Multivariate regression analysis, using SPSS for Windows v. 10.0.5 (standard version), was applied to examine the relationship between infant mortality and seven risk factors.  Independent variables were selected for their ability to represent aggregates of many risk factors that influence the dependent variable infant mortality.  Data was obtained for each U.S. State and the District of Columbia (hereafter referred to as “51 states”) for each of the variables.  All data represents 1997 vital statistics as reported and defined by the U.S. National Center for Health Statistics (Statistical Abstract of the United States, 1999).  Alpha probability was set at the p < .05 level for the data analysis.  The Pearson correlation was used to determine which of the seven independent variables best predicted infant mortality, and to determine which variables to include in a final predictive model.

The dependent variable consisted of n = 51 state’s infant mortality rates.  The independent variables consisted of n = 51 states’ vital statistics for: low birth weight; births to unmarried women, births to adolescents; people below the poverty level; people without health insurance; children without health insurance; and per capita income.  All variables represent continuous data (listwise, with no missing observations).  Pearson correlations were used to evaluate inter-relatedness among independent variables, which identified variables to define as aggregates.  The seven variables were analyzed for predictive ability, the most predictive model was constructed, and the final regression equation was used to predict the infant mortality rate for a selected population.  Bexar County, Texas was selected as the population to test the predictive ability of the final model.  Bexar Co. data was obtained from the Texas Department of Health.

Results

Descriptive statistics, correlations and significance levels are reported in Appendix A. Table A2 contains multiple regression analysis results.  The seven independent variable model showed that five correlations, adolescent births, people below the poverty level, people without health insurance, children without health insurance, and per capita income, showed a range of

r = .511, .459, .150, .131 and -.076 respectively, with associated significance levels of p < .001, .001, .147, .180, and .298 respectively.  These five correlations were less significant than the remaining two; therefore, these five variables were excluded from the model (see Table A1).

A model consisting of low birth weight and births to unmarried mothers was constructed.  Multiple regression analysis of these two variables demonstrated a very significant ability to predict infant mortality.  Low birth weight showed t (49) = 3.437, and births to unmarried mothers showed t (49) = 3.487.  Both t-tests had significance levels of p < .001. This model showed R = .820 and R square = .672.  Analysis of variance showed F (2, 48) = 49.098 and

p < .001.  Based on these results, low birth weight and births to unmarried mothers were accepted as the strongest predictors of infant mortality.  The other five variables were rejected as less accurate predictors.

Evaluation of inter-relatedness between independent variables demonstrated that low birth weight and unmarried mothers represent aggregate variables of adolescent births, poverty level, and people and children with no health insurance.  Pearson correlations for these four independent variables were significant at a range of p < .001 to p = .033, indicating that low birth weight and unmarried mothers account for a high degree of aggregate risk to predict infant mortality.  Per capita income showed no inter-relatedness to other independent variables and no significant relationship as a predictor of infant mortality.  Specific results of inter-relatedness are displayed in Table A3.  Data for Bexar Co. low birth weight and births to unmarried mothers were applied to the regression equation of the predictive model.  This application predicted the Bexar Co. infant mortality rate (actual rate = 7.4, predicted rate = 7.3) within less than one standard deviation (1.787).  Results are presented in Table A4.

Discussion

This study demonstrates that infant mortality and births to unmarried mothers are very strong predictors of infant mortality.  Unmarried mothers accounted for as much of the risk associated with infant mortality as low birth weight, which is significant because the literature does not ascribe this magnitude of importance to births to unmarried mothers.  Per capita income and having no health insurance are often cited as having a strong influence but did not significantly correlate to infant mortality, which may be explained by disparities of income that exist within populations.  Poverty and adolescent births showed significant correlations to infant mortality, but these were not as predictive as low birth weight and births to unmarried mothers.  Correlation of four independent variables (excluding per capita income) showed a high correlation to low birth weight and to unmarried mothers, suggesting that these two aggregates contain multiple risk factors associated with infant mortality.  These findings imply that the degree of risk commonly associated with certain risk factors is not as closely related to infant mortality as some studies indicate.  Health planners should focus interventions on preventable risks that are most closely associated with infant mortality.  Properly defined aggregate variables may be useful to this effort.  This study is limited in that it does not present a complete statistical analysis of all risk factors associated with infant mortality.  Meta-analysis applied to a comprehensive literature review of all suspected risk factors would help to clarify the true relationships between multiple risk factors and infant mortality rates in the U.S.

Appendix A

Table 1

 

            Low Birth Weight                                                     .767                                         (.001)                

Unmarried Mothers                                                  .769                                         (.001)                                

Adolescent Births                                                     .511                                         (.001)                                

                Below Poverty Level                                                .459                                         (.001)                                

                People Without                                                         .150                                         (.147)                                

Insurance

 

Children Without                                                      .131                                         (.180)                                

Insurance

 

Note.  Table 1 displays Pearson correlations for multivariate regression analysis of seven independent variables with infant mortality, indicating the probability (p < .05) that a risk factor is related to infant mortality.  Low birth weight and births to unmarried mothers show the strongest relationship.

 

 

 

 

Table 2

 

Multivariate Regression Analysis Results for Infant Mortality Against Two Independent Variables (N = 51)

 

 

Low Birth Weight                .672              49.098                48            (.001)                      .06173                          .541

            and

 

 

 

Table 3

              Adolescent         Poverty             Children                       People                     Per Capita

 

Low Birth               .584 (.001)        .461 (.001)          .265 (.030)                   .259 (.033)                 -.036 (.400)

Weight

 

Unmarried              .580 (.001)        .657 (.001)          .347 (.006)                   .367 (.004)                   .048 (.368)

Note.  Table 3 illustrates that low birth weight and births to unmarried mothers represent aggregate variables to adolescent births, poverty, and not having health insurance.  Per capita income was not significantly related.

 

 

 

 

 

 

 

 

Low Birth Weight (X₁)                               7.3                                                                    Y' = a + b₁X₁ + b₂X₂

 

Births to Unmarried Mothers (X₂)          29.3                                                    Y' = .06173 + (.541)(7.3) + (.112)(29.3)

 

Note.  Table 4 illustrates Bexar Co., Texas actual values of low birth weight and births to unmarried mothers as applied to the regression equation of the predictive model.  Y'represents the predicted value, a = the constant from the predictive model, and b = the coefficient for the appropriate variable.  The model predicted Bexar Co. infant mortality rate within one standard deviation (infant mortality standard deviation = 1.787).

 

 

 

 

 

           

References

            Flaherty, L. (1999).  Achievements in public health, 1990-1999: Healthier mothers and babies.  Journal of Emergency Nursing, 26, 151-152.

Hamvas, A. (2000).  Disparate outcomes for very low birth weight infants: Genetics, environment, or both?  The Journal of Pediatrics, 136, 427-428.

Lerner, R. M. (1995).  America’s youth in crisis: Challenges and options for programs and policies.  Thousand Oaks, CA: Sage.

Luker, K. (1997).  Dubious Conceptions: The politics of teenage pregnancy.  Cambridge, MA: Harvard University Press.

Lynch, J. W., Smith, G. D., Kaplan, G. A., & House, J. S.  (2000).  Income inequality and mentality: Importance to health of individual income, psychosocial environment, or material conditions.  British Medical Journal, 320, 1200-1204.

MacDorman, M. F., & Atkinson, J. O.  (1999).  Infant mortality statistics from the 1997 period linked birth/infant death data set.  National Vital Statistics Report, 47(23), 1-7.

McCormick, J. E. (2000).  Prenatal care: Effectiveness and implementation.  American Journal of Public Health, 90, 805.

Pollack, M. M., Koch, M. A., Bartel, D. A., & Rapoport, I. (2000).  A comparison of neonatal mortality risk prediction models in very low birth weight infants.  American Academy of Pediatrics, 105, 1051-1057.

Robinson, J. & Wharrad, H. (2000).  Invisible nursing: Exploring health outcomes at a global level.  Relationships between infant and under –5 mortality rates and the distribution of health professionals, GNP per capita, and female literacy.  Journal of Advanced Nursing, 32, 28-40.

Short, K. (1999).  Examining experimental poverty measures 1990 to 1997.  U.S. Census Bureau. [On-Line].  Available: http://www.census.gov/hhes/poverty/povmeas/examin.html

St. Peters, R. F. & Tarlov, A. R. (2000).  The Society and Population Health Reader, Volume 2: A State and Community perspective.  American Journal of Public Health, 90, 979-980.

Statistical Abstract of the United States (119^th ed.).  1999.  Austin, TX: Hoover’s Business Press.

Stuart-Brown, S. (2000).  Is the ethos of medical practice in community paediatrics compatible with that in public health?  Archives of Disease in Childhood, 83, 101-103.

(Click here to return to Research & Statistics page)

(Click here to return to Homepage)