Does Low Birth Weight Matter ? Evidence from the U . S . Population of Twin Births

A widely held position is that spending resources to reduce the incidence of low birth weight (LBW) births represents one of the best investments in improving infant health and welfare. The emphasis of public policies (e.g. Medicaid, WIC, maternal smoking cessation campaigns) on LBW rests crucially on the claim that any policy that works to reduce the incidence of LBW births will improve infant health outcomes. This paper empirically evaluates this claim – that birth weight can reliably substitute for direct health outcomes of interest (e.g. mortality) when evaluating and predicting the effects of health policies and interventions. If this claim is true, then the component of birth weight that is exclusively driven by environmental factors (and not by genetic factors) should be highly correlated with other known measures of infant health, including mortality. We examine this type of birth weight variation by focusing on within twin-pair correlations between birth weight and various health outcomes, using data on the population of twin births in the United States, which we construct using a new matching algorithm on available linked birth-infant death micro-data. While there are substantial differences in birth weights between twins, we find that the heavier twin is no more likely to survive past one-year of life than the lighter twin. This finding is consistent across demographic groups, causes of death, and years of birth and is insensitive to various econometric specifications. We find that another continuous proxy of health at birth, the 5-minute APGAR score, is more highly associated with infant mortality both between and within twin pairs, but is not correlated with within twin-pair differences in birth weight. Furthermore, we examine a policy application – maternal smoking cessation – that provides a concrete example of the implications of our findings. While smoking during pregnancy has a substantial association with birth weight and LBW incidence, it has little relation with either infant mortality or the 5-minute APGAR score. The evidence suggests that birth weight may provide a misleading proxy for measuring the effectiveness of health interventions. Introduction It is a widely held position that, whether at the individual or governmental level, spending resources on reducing the incidence of low birth weight (LBW) births (defined to be less than 2,500 grams) represents one of the best investments in improving infant health and welfare. This position is based on the observation that LBW infants have been shown to have higher mortality rates, poorer health as children or adults, reduced cognitive ability, and eventually lower education levels and labor market earnings relative to their normal weight counterparts. In fact, reducing the incidence of LBW births has been a goal of several public policies targeting infant health, including Medicaid, the Women, Infants, and Children (WIC) program, and maternal smoking cessation campaigns. The emphasis of public policy on LBW rests crucially on the claim that any policy that works to reduce the incidence of LBW births will necessarily improve infant health outcomes, and in particular, reduce infant mortality. This paper empirically evaluates this claim – that birth weight is a valid policy “marker” for infant health. By “valid policy marker”, we mean a variable that can reliably substitute for the actual outcome of interest when evaluating and thus predicting the health effects of policies and interventions. For example, within a human capital framework, education could provide a valid policy marker for improved earnings capacities, since the presumed structural link between educational attainment and earnings implies that any intervention that works to increase educational attainment (and other measures of human capital) would generally result in higher individual earnings. On the other hand, a concrete example of a misleading policy marker from recent medical research is the level of cholesterol as a marker for cardiovascular heart disease risk among postmenopausal women. A recent, highly publicized randomized trial found higher heart disease rates among the women treated with estrogen plus progestin, even though previous studies had established that the therapy reduced cholesterol levels, a generally accepted indicator for increased risk of coronary heart disease. 1 1 An intermediate outcome clinical trial established that estrogen/progestin replacement therapy reduces cholesterol levels, a generally accepted indicator for increased risk of coronary heart disease, among postmenopausal women. As a result, it was widely believed that one of the health benefits of taking estrogen after menopause would be a 2 As we show below within a simple econometric framework, if birth weight constitutes a valid policy marker, then it must be true that the component of birth weight that is exclusively driven by environmental factors has a strong correlation with other known measures of infant health, including mortality. A strong correlation would be consistent with the notion that any policy that succeeds in reducing the incidence of LBW births would improve health and reduce infant mortality. On the other hand, a weak correlation between LBW and mortality would call into question the conjecture that birth weight is a sufficient index of overall infant health. The main confounding problem is that genetic factors undoubtedly contribute to the determination of birth weight and overall health in unknown ways, and hence may entirely explain the strong cross-sectional relation between birth weight and mortality. Thus, in order to account for genetic variation across families, and isolate birth weight variation that is primarily or entirely driven by environmental factors, we analyze the population of twin births in the U.S. and examine whether within twin-pair differences in health are correlated with within twin-pair differences in birth weight. Using a new algorithm to match twins with their siblings in National Center for Health Statisticslinked birth-infant death microdata, we analyze the population of twin births in the United States for the years 1985-1986, 1989-1991, and 1995 – a sample of over 350,000 twins. While the average twin weighs only 2,400 grams, there are substantial differences in birth weights between twins of the same mother. For example, on average the heavier twin weighs 300 grams more than the lighter twin at birth. This differential is 50 percent larger than the average birth weight difference between the infants of mothers who did and did not smoke during pregnancy. The large variation in birth weights within twin pairs leads to precise fixed effects estimates of the effect of birth weight on infant mortality. It also allows us to examine the significant nonlinearities in the infant mortality-birth weight relation. There is a strong and robust cross-sectional correlation between infant mortality and birth weight between twin pairs, particularly at birth weights below 1,000 grams. In addition, there are substantial reduced incidence of heart attack and stroke. However, a recent, randomized controlled trial found higher heart 3 differences in several measures of health within twin pairs. Surprisingly, however, we find that within twin-pairs the heavier twin is no more likely to survive past one-year of life than the lighter twin, even among twins born at very low weights. This finding is consistent across demographic groups, causes of death, and years of birth and is insensitive to econometric specification. Two interpretations are possible: 1) the within twin-pair comparison yields a “true” structural estimate of the “effect” of birth weight on infant mortality, and that estimate is small and in many cases zero, or 2) the evidence is simply inconsistent with the notion of a “one-to-one” relation between birth weight and important measures of health: some interventions might simultaneously reduce LBW and mortality, while others might reduce LBW without impacting (or perhaps having a perverse impact on) mortality. Either interpretation calls into question the validity of using birth weight alone as a substitute for actual health outcomes of interest in evaluating and predicting the effects of policy interventions. We find that another continuous proxy of health at birth, the 5-minute APGAR score, is more highly associated with infant mortality both between and within twin pairs, but is not correlated with sibling differences in birth weight. Twin differences in birth weight are also uncorrelated with the use of assisted ventilation after birth, the most common post-birth intervention for at-risk infants. On the other hand, twin differences in APGAR scores are highly associated with assisted ventilation incidence. In our analysis, birth weight stands out as the one infant outcome that appears uncorrelated with other known measures of health, after accounting for common genetic factors within twin-pairs. Finally, we examine the implications of our findings for an important policy application: the validity of using birth weight to evaluate infant health benefits of maternal smoking cessation. Maternal cigarette smoking during pregnancy has been identified as the largest modifiable risk factor for LBW incidence in developed countries (Kramer 1987). Further, a recent editorial in the Journal of the American Medical Association advocates the prevention of all maternal smoking as an “optimal public health outcome”, since “maternal smoking is a significant risk factor for LBW infants, which, in turn, disease rates among the women treated with estrogen plus progestin (Women’s Health Initiative Investigators, 2002). 4 influences infant mortality” (Vogler and Kozlowski, 2002). In our analysis of both singletons and twins, we find that while smoking during pregnancy has a substantial association with birth weight and LBW incidence, it has little relation with eith