Cross-Country Individual Participant Analysis of 4.1 Million Singleton Births in 5 Countries with Very High Human Development Index Confirms Known Associations but Provides No Biologic Explanation for 2/3 of All Preterm Births

Abstract

Background: Preterm birth is the most common single cause of perinatal and infant mortality, affecting 15 million infants worldwide each year with global rates increasing. Understanding of risk factors remains poor, and preventive interventions have only limited benefit. Large differences exist in preterm birth rates across high income countries. We hypothesized that understanding the basis for these wide variations could lead to interventions that reduce preterm birth incidence in countries with high rates. We thus sought to assess the contributions of known risk factors for both spontaneous and provider-initiated preterm birth in selected high income countries, estimating also the potential impact of successful interventions due to advances in research, policy and public health, or clinical practice.

Methods: We analyzed individual patient-level data on 4.1 million singleton pregnancies from four countries with very high human development index (Czech Republic, New Zealand, Slovenia, Sweden) and one comparator U.S. state (California) to determine the specific contribution (adjusting for confounding effects) of 21 factors. Both individual and population-attributable preterm birth risks were determined, as were contributors to cross-country differences. We also assessed the ability to predict preterm birth given various sets of known risk factors.

Findings: Previous preterm birth and preeclampsia were the strongest individual risk factors of preterm birth in all datasets, with odds ratios of 4.6-6.0 and 2.8-5.7, respectively, for individual women having those characteristics. In contrast, on a population basis, nulliparity and male sex were the two risk factors with the highest impact on preterm birth rates, accounting for 25-50% and 11-16% of excess population attributable risk, respectively (p<0.001). The importance of nulliparity and male sex on population attributable risk was driven by high prevalence despite low odds ratios for individual women. More than 65% of the total aggregated risk of preterm birth within each country lacks a plausible biologic explanation, and 63% of difference between countries cannot be explained with known factors; thus, research is necessary to elucidate the underlying mechanisms of preterm birth and, hence, therapeutic intervention. Surprisingly, variation in prevalence of known risk factors accounted for less than 35% of the difference in preterm birth rates between countries. Known risk factors had an area under the curve of less than 0.7 in ROC analysis of preterm birth prediction within countries. These data suggest that other influences, as yet unidentified, are involved in preterm birth. Further research into biological mechanisms is warranted.

Conclusions: We have quantified the causes of variation in preterm birth rates among countries with very high human development index. The paucity of explicit and currently identified factors amenable to intervention illustrates the limited impact of changes possible through current clinical practice and policy interventions. Our research highlights the urgent need for research into underlying biological causes of preterm birth, which alone are likely to lead to innovative and efficacious interventions.

Fig 1. Overview of analytic approach.

The four country datasets (Czech Republic, New Zealand, Slovenia, Sweden) and the U.S. comparator state (California) are indicated at the top of the figure along with the number of births included in the analysis from each. Each of the five core analyses are represented in the boxes below, and the relationship between analyses is represented by arrows.

Fig 2. Systematic review of previously published multivariate analyses of preterm birth [–36].

Each bar represents the difference between the reported odds ratios and 1; positive/negative bars are associated with increased or decreased risks of preterm birth respectively. Where categorical variables were reported, we reported the category having the largest significant odds ratio. We included all risk factors considered by at least 3 studies. Reference categories are the following: non-Hispanic white (ethnicity) (all studies, except Dekker '12 (non-Caucasian)); prenatal care beginning before 13 (Berkowitz '09) or 14 (Rodrigues '07) completed weeks of gestation, prenatal care received during first trimester (Lang '96), "adequate" (Hillemeier '07), received (Kristka '07), or not received (Zhang '12); high education (all, except Lang '96, Rodrigues '07), high-school graduate (Lang '96), 4–8 years of school education (Rodrigues '07); age > 20 (Kistka '07), age 20–29 (Berkowitz '09, Chiavarini '12), age 25–29 (Hillemeier '07), age 25–34 (Lang '96, Xu '14), age 18–30 (Meis '98), age 20–34 (Kramer '92), age 20–35 (Heaman '12, Olsen '95), age < 35 (Di Renzo '11); low perceived stress; healthy BMI; female baby; non-smokers; least deprived population; married. For categorical variables only the category with the largest significant odds ratio is shown; ethnicity: Black (Berkowitz '98, Hillemeier '07, Kristka '07, Meis '98, Lang '96), Caucasian (Dekker '12); education: lowest education (all studies); age: < 15 (Lang '96), < 20 (Berkowtiz '98, Kristka '07, Olsen '95), > 30 (Meis '98), > 35 (Di Renzo '11, Heaman '12, Hillemeier '07, Kramer '92, Rodrigues '07), > 40 (Chiavarini '12); BMI/Obesity: < 20 (Berkowtiz '98, Dekker '12, Kristka '07, Olsen '95), > 25 (Di Renzo '11), > 30 (Zhang '12), > 45 (Xu '14); poverty: high level (Erickson '01, Hillemeier '07, Kristka '07, Xu '14). A missing bar indicates that the risk factor was not considered in the study. Abbreviations: PTB, preterm birth; HPTN, hypertension; DBTS, diabetes; BMI, body mass index; gest., gestational; ART, assisted reproductive technology; OR, odds ratio; H, hospital; R, registry; S, survey; PS, prospective study.

Fig 3. Risk factors for preterm birth across four countries and one comparator U.S. state.

The odds ratios for each risk factor were calculated using five independent logistic regression models. Statistical significance was defined as p < 0.05. The width of the shaded lines are proportional to the reported odds ratios. A missing value indicates that data on risk factor were not available. For categorical variables, the reference categories were age 20–34 (age), non-Hispanic white (ethnicity), healthy BMI (BMI 18.5–24.9), highest education (college graduate or more), least deprived (poverty quintile Q1). Abbreviations: BMI, body mass index; MELAA, Middle Eastern, Latin American or African; PTB, preterm birth; ART, assisted reproductive technology; CS, cesarean section; 20wk, 20 weeks; Pacific P, Pacific people; * p<0.05; ** p<0.01

Fig 4. Top 10 subpopulations with highest probability of preterm birth.

4 datasets (New Zealand, Czech Republic, Slovenia, Sweden) representing a total of ~3 million singleton pregnancies were combined. Subpopulations were defined by their unique combinations of risk factors (top) and ranked by probability of preterm birth (bottom). This analysis was restricted to risk factors common to all four datasets. Subpopulations with prevalence below 1 in 10,000 were excluded. Additional subpopulations are shown in Fig E of S1 Appendix. Abbreviations: PTB, preterm birth; gest, gestational; CS, cesarean section.

Fig 5. Estimated contribution of risk factors on population preterm birth rate and opportunities for various stratified interventions for (A) Czech Republic, (B) New Zealand, (C) Slovenia, (D) Sweden.

Results from the logistic regression analysis were combined with prevalence of risk factor to estimate the impact of each risk factor (see Methods). Risk factors were grouped into three intervention areas, "Research", "Policy and Public Health", "Clinical Practice". Some risk factors (e.g. diabetes, hypertension) could and were classified into multiple categories. Percentage ranges indicated for "Policy and Public Health" and "Clinical Practice" thus reflect scenarios with or without inclusion of these overlapping risk factors. An asterisk represents the observed preterm birth rate in each dataset. Error bars: 95% confidence intervals. Abbreviations: PTB, preterm birth; HPTN, hypertension; DBTS, diabetes; CS, cesarean section; ART, assisted reproductive technology; 20wk, 20 weeks.

Fig 6. Estimated contributions of risk factors and clinical practices to differences in preterm birth rates between countries with VHHDI (left) and Sweden (right).

The left and right bars represent the preterm birth rates for the indicated countries [5]. The size of each step in the “waterfall” was calculated by taking the difference in the estimated impact of risk factor (or clinical practice) between the indicated country and Sweden. The last step, labeled "unknown", represents the percentage not captured by the risk factors and clinical practices shown here. "N/A" indicates that information was not available to estimate the impact of the risk factor. Results for additional countries are shown in Fig F of S1 Appendix. Abbreviations: PTB, preterm birth; HPTN, hypertension; DBTS, diabetes; ART, assisted reproductive technology; 20 wk, 20 weeks.