Birth Weight, Physical Morbidity, and Mortality
Birth Weight, Physical Morbidity, and Mortality
Table 1 presents demographic characteristics by birth weight category. Table 2 presents the numbers of offspring with various physical morbidity outcomes and Kaplan-Meier product-limit survival estimates for all outcomes.
Model 1 utilized ordinal birth weight across the entire cohort population. Figure 1 presents baseline risk estimates (dark gray columns) with 95% Wald confidence intervals (I-shaped bars) around the hazard ratio. Point estimates for the reference category, 3,501–4,000 g, were equal to 1. There was a strong inverse association between birth weight and mortality after 1 year (LBW hazard ratio (HR) = 2.15, 95% confidence interval (CI): 1.97, 2.34), as well as cardiac-related death (LBW HR = 2.69, 95% CI: 2.05, 3.53).
(Enlarge Image)
Figure 1.
Associations derived from continuous (lines) and ordinal (columns) representations of birth weight when predicting mortality outcomes among offspring born in Sweden during 1973–2007 (death after first year; part A) and 1973–1995 (cardiac-related death; part B). Baseline, population-wide estimates are shown via the solid lines and dark gray columns. Results from sibling-comparison, fixed-effects models are shown via the dotted lines and light gray columns. I-shaped bars, 95% confidence intervals.
Model 2 used a continuous representation of birth weight. Associations with the mortality outcomes were better explained by a quadratic model of birth weight Web Table 2. Figure 1 also presents continuously represented birth weight risk in the baseline quadratic model, model 2 (solid black line). Within the figures, note that a similar interpretation can be drawn from the ordinal columns as from the continuous parameter estimates of model 2.
The associations remained robust when adjusting for offspring- and parent-specific covariates across mortality outcomes in model 3. Thus, across outcomes, the associations were independent of offspring sex, birth order, year of birth, maternal and paternal age at childbearing, highest level of education, and history of criminal conviction. Model 3 results are not presented graphically for ease of interpretation.
Finally, model 4 was a fixed-effects sibling-comparison model, presented in Figure 1 via the dotted line (continuous) and the light gray columns (ordinal). Consistent with a causal inference, birth weight significantly predicted mortality after 1 year (LBW HR = 3.02, 95% CI: 2.52, 3.62) and cardiac-related death (LBW HR = 4.30, 95% CI: 2.27, 8.14) within differentially exposed siblings while controlling for offspring-specific covariates. Interestingly, across mortality outcomes, the magnitudes of association were larger in model 4 (fixed effects) than in population estimates for both the ordinal and continuous models. We also identified a similar (though larger in magnitude) pattern of increased risk across models when predicting infant mortality (results available upon request). Parameter estimates for baseline, adjusted, and fixed-effects models for continuously represented birth weight are shown in Web Table 3. Web Table 4 presents parameter estimates for ordinal models 1 and 4 as verification of model specification.
Figure 2 presents baseline and fixed-effects results for ordinal and continuously measured birth weight across physical morbidity outcomes in the study population. There was a strong inverse association between birth weight and hypertension (LBW HR = 1.58, 95% CI: 1.37, 1.82) that persisted after adjustment for covariates and whose magnitude was robust in fixed-effects analyses (LBW HR = 1.31, 95% CI: 0.92, 1.86). Similarly, there was an inverse association for ischemic heart disease (LBW HR = 2.52, 95% CI: 1.70, 3.73) that was robust across models and remained present in fixed-effects analyses (LBW HR = 2.18, 95% CI: 0.72, 6.13). Pulmonary circulation problems showed an analogous pattern across models (model 2: LBW HR = 1.43, 95% CI: 1.12, 1.83; model 4: LBW HR = 1.41, 95% CI: 0.79, 2.52).
(Enlarge Image)
Figure 2.
Associations derived from continuous (lines) and ordinal (columns) representations of birth weight when predicting hypertension (A), ischemic heart disease (B), pulmonary circulation problems (C), stroke (D), and type 2 diabetes mellitus (E) among offspring born in Sweden during 1973–1995. Baseline, population-wide estimates are shown via the solid lines and dark gray columns. Sibling-comparison, fixed-effects models are shown via the dotted lines and light gray columns. I-shaped bars, 95% confidence intervals.
We found an inverse association with stroke in the baseline model (LBW HR = 1.59, 95% CI: 1.28, 1.96) that was robust in magnitude in the adjusted and fixed-effects models (LBW HR = 1.37, 95% CI: 0.83, 2.25). The baseline model predicting type 2 diabetes also showed an inverse association, where lower birth weight was associated with increased odds (LBW HR = 1.79, 95% CI: 1.50, 2.14). This association was also robust in the adjusted model and when using fixed-effects modeling (LBW HR = 1.71, 95% CI: 1.14, 2.56).
Similar to our mortality results, the magnitudes of association were larger following fixed-effects modeling using continuously measured birth weight (model 4) as compared with magnitudes from population estimates (models 2 and 3). Parameter estimates from baseline, adjusted, and fixed-effects models for continuously represented birth weight are presented in Web Table 3, while ordinal parameter estimates (models 1 and 4) are presented in Web Table 4 for model specification verification.
In sensitivity analyses, we first tested whether gestational age influenced the results. In particular, we limited the cohort to full-term (≥37 weeks) births and found that results were not biased by premature birth Web Figure 1, though the reduced number of persons at the lowest birth weights who were born full-term contributed to large confidence intervals around these estimates. In addition, although the sample has been shown to have reliable gestational ages whether measured via last menses or ultrasound, we examined whether the removal of extreme gestational ages (<23 weeks and ≥42 weeks and 6 days) affected the results by performing an analysis that included all persons, regardless of their gestational age. The results gave interpretations commensurate with those of the main analyses and sensitivity analyses limiting the sample to full-term births only (results available upon request). Second, we tested an assumption of the sibling-comparison design by exploring whether results from families with more than 1 offspring would generalize to offspring without siblings. The results suggested that estimates were not biased by differences between families with only 1 offspring and those with more than 1 offspring Web Figure 2.
Results
Table 1 presents demographic characteristics by birth weight category. Table 2 presents the numbers of offspring with various physical morbidity outcomes and Kaplan-Meier product-limit survival estimates for all outcomes.
Mortality
Model 1 utilized ordinal birth weight across the entire cohort population. Figure 1 presents baseline risk estimates (dark gray columns) with 95% Wald confidence intervals (I-shaped bars) around the hazard ratio. Point estimates for the reference category, 3,501–4,000 g, were equal to 1. There was a strong inverse association between birth weight and mortality after 1 year (LBW hazard ratio (HR) = 2.15, 95% confidence interval (CI): 1.97, 2.34), as well as cardiac-related death (LBW HR = 2.69, 95% CI: 2.05, 3.53).
(Enlarge Image)
Figure 1.
Associations derived from continuous (lines) and ordinal (columns) representations of birth weight when predicting mortality outcomes among offspring born in Sweden during 1973–2007 (death after first year; part A) and 1973–1995 (cardiac-related death; part B). Baseline, population-wide estimates are shown via the solid lines and dark gray columns. Results from sibling-comparison, fixed-effects models are shown via the dotted lines and light gray columns. I-shaped bars, 95% confidence intervals.
Model 2 used a continuous representation of birth weight. Associations with the mortality outcomes were better explained by a quadratic model of birth weight Web Table 2. Figure 1 also presents continuously represented birth weight risk in the baseline quadratic model, model 2 (solid black line). Within the figures, note that a similar interpretation can be drawn from the ordinal columns as from the continuous parameter estimates of model 2.
The associations remained robust when adjusting for offspring- and parent-specific covariates across mortality outcomes in model 3. Thus, across outcomes, the associations were independent of offspring sex, birth order, year of birth, maternal and paternal age at childbearing, highest level of education, and history of criminal conviction. Model 3 results are not presented graphically for ease of interpretation.
Finally, model 4 was a fixed-effects sibling-comparison model, presented in Figure 1 via the dotted line (continuous) and the light gray columns (ordinal). Consistent with a causal inference, birth weight significantly predicted mortality after 1 year (LBW HR = 3.02, 95% CI: 2.52, 3.62) and cardiac-related death (LBW HR = 4.30, 95% CI: 2.27, 8.14) within differentially exposed siblings while controlling for offspring-specific covariates. Interestingly, across mortality outcomes, the magnitudes of association were larger in model 4 (fixed effects) than in population estimates for both the ordinal and continuous models. We also identified a similar (though larger in magnitude) pattern of increased risk across models when predicting infant mortality (results available upon request). Parameter estimates for baseline, adjusted, and fixed-effects models for continuously represented birth weight are shown in Web Table 3. Web Table 4 presents parameter estimates for ordinal models 1 and 4 as verification of model specification.
Physical Morbidity
Figure 2 presents baseline and fixed-effects results for ordinal and continuously measured birth weight across physical morbidity outcomes in the study population. There was a strong inverse association between birth weight and hypertension (LBW HR = 1.58, 95% CI: 1.37, 1.82) that persisted after adjustment for covariates and whose magnitude was robust in fixed-effects analyses (LBW HR = 1.31, 95% CI: 0.92, 1.86). Similarly, there was an inverse association for ischemic heart disease (LBW HR = 2.52, 95% CI: 1.70, 3.73) that was robust across models and remained present in fixed-effects analyses (LBW HR = 2.18, 95% CI: 0.72, 6.13). Pulmonary circulation problems showed an analogous pattern across models (model 2: LBW HR = 1.43, 95% CI: 1.12, 1.83; model 4: LBW HR = 1.41, 95% CI: 0.79, 2.52).
(Enlarge Image)
Figure 2.
Associations derived from continuous (lines) and ordinal (columns) representations of birth weight when predicting hypertension (A), ischemic heart disease (B), pulmonary circulation problems (C), stroke (D), and type 2 diabetes mellitus (E) among offspring born in Sweden during 1973–1995. Baseline, population-wide estimates are shown via the solid lines and dark gray columns. Sibling-comparison, fixed-effects models are shown via the dotted lines and light gray columns. I-shaped bars, 95% confidence intervals.
We found an inverse association with stroke in the baseline model (LBW HR = 1.59, 95% CI: 1.28, 1.96) that was robust in magnitude in the adjusted and fixed-effects models (LBW HR = 1.37, 95% CI: 0.83, 2.25). The baseline model predicting type 2 diabetes also showed an inverse association, where lower birth weight was associated with increased odds (LBW HR = 1.79, 95% CI: 1.50, 2.14). This association was also robust in the adjusted model and when using fixed-effects modeling (LBW HR = 1.71, 95% CI: 1.14, 2.56).
Similar to our mortality results, the magnitudes of association were larger following fixed-effects modeling using continuously measured birth weight (model 4) as compared with magnitudes from population estimates (models 2 and 3). Parameter estimates from baseline, adjusted, and fixed-effects models for continuously represented birth weight are presented in Web Table 3, while ordinal parameter estimates (models 1 and 4) are presented in Web Table 4 for model specification verification.
Sensitivity Analyses
In sensitivity analyses, we first tested whether gestational age influenced the results. In particular, we limited the cohort to full-term (≥37 weeks) births and found that results were not biased by premature birth Web Figure 1, though the reduced number of persons at the lowest birth weights who were born full-term contributed to large confidence intervals around these estimates. In addition, although the sample has been shown to have reliable gestational ages whether measured via last menses or ultrasound, we examined whether the removal of extreme gestational ages (<23 weeks and ≥42 weeks and 6 days) affected the results by performing an analysis that included all persons, regardless of their gestational age. The results gave interpretations commensurate with those of the main analyses and sensitivity analyses limiting the sample to full-term births only (results available upon request). Second, we tested an assumption of the sibling-comparison design by exploring whether results from families with more than 1 offspring would generalize to offspring without siblings. The results suggested that estimates were not biased by differences between families with only 1 offspring and those with more than 1 offspring Web Figure 2.
Source...