Non-linear association of low birthweight with risk factors including women’s BMI: evidence from an international comparison

Abstract

Background

It is widely known that, in addition to physical characteristics such as parents’ height and weight, social attributes such as parents’ education, income level, and employment status have a significant impact on birthweight. However, these results were obtained using data from individual countries, and there were no country-level comparisons of the factors identified as influencing birthweight. A multi-country analysis is essential to determine whether these factors are consistently linked to low birthweight across different economic and healthcare systems.

Methods

Using panel data on low-birthweight fertility rates for 143 countries over the period of 2000–2015, we used five factors (women’s body mass index [BMI], real gross domestic product [GDP] per capita, women’s employment status, healthcare level, and adolescent childbearing) in relation to countries’ low-birthweight rates to determine how these factors relate to each country’s low-birthweight rate and estimate a fixed-effects model. Considering the possibility that these five factors are non-linear rather than linear, we estimated quadratic and cubic functions. We conducted a detailed analysis of women’s BMI and real GDP per capita. Furthermore, we considered the 2008 global financial crisis as an exogenous natural experiment for the low-birthweight rate and conducted a difference-in-differences (DID) analysis to confirm the possibility that the correlation between women’s employment rate and low-birthweight rate is a causal relationship.

Results

All five factors were identified as important risk factors. Of the five, all but adolescent childbearing were found to have a non-linear rather than a monotonous linear relationship with low birthweight. The low-birthweight rate improved sharply with improvement in women’s average BMI below 28. Furthermore, the results of the DID analysis suggest that women’s employment plays an important role in the relationship between low-birthweight rates and GDP.

Conclusions

These results provide a useful policy tool for achieving the goal of the 65th World Health Assembly to “reduce the incidence of low birthweight.” In particular, improvements in women’s average BMI, real GDP per capita, and women’s employment rates in low-income countries may be linked to reductions in the incidence of low birthweight.

Background

Birthweight is an indicator of infant health that influences later life. Barker and Osmond (1986) [1] and Barker (1990) [2] found that low birthweight was significantly and negatively associated with adult health. Recent studies in the social sciences have established that birthweight has significant negative effects on social attributes, such as income and occupation, in addition to adult health [3,4,5]. Moreover, high birthweight is significantly negatively associated with adult health [6, 7]. Furthermore, analysis of Japan, the U.S., and India Yamane and Tsutsui (2022) [8] showed that low and high birthweight matter to quality of life. Their study demonstrated that the associations between birthweight and quality of life are widely diverse across countries depending on economic status.

The determinants of low birthweight include parental smoking, alcohol and drug use, employment status, age at birth, maternal pre-pregnancy emaciation tendency, preterm delivery, and psychological stress during pregnancy [9,10,11,12]. Economic studies have reported that parents’ physical characteristics, mothers’ years of education, and fathers’ socioeconomic status (education, income, occupation, etc.) are risk factors for low birthweight [4].

How well do the factors found in intra-country studies explain low-birthweight rates in each country? This is the research question addressed in the present study. In affluent countries, a fetus has a higher probability of being born healthy because the mother is likely to be well nourished and the medical system is well equipped to handle the risks of childbirth. However, the opposite may be true in poor countries. In May 2012, the 65th World Health Assembly set a goal of a 30% reduction in the low birthweight rate between 2012 and 2025; however, as of 2015, the goal is far from being achieved.¹

Exploring the factors associated with low birthweight through international comparisons is important to overcome this situation [13]. This study investigates whether the critical factors of low birthweight that have been identified with intra-nation data can also be confirmed using international data. International comparative studies are important to identify policy issues that should be addressed nationally. Despite its importance, to the best of our knowledge, no international comparative study has been conducted to examine the determinants of low birthweight. This study addressed these gaps.

In this study, we considered that women’s BMI may be an important factor in low birth weight. We hypothesize that the probability of having a low-birthweight infant is lower in countries with higher BMI levels and that poorer countries have a higher percentage of low-birthweight infants because of their poor sanitation and healthcare systems. Maternal obesity is primarily associated with macrosomia (large-for-gestational-age infants) rather than with low birthweight in general. However, if the average BMI of women in a country becomes too high, the percentage of infants with low birthweight may increase due to pregnancy complications, such as preeclampsia and placental dysfunction [14]. Thus, BMI is indirectly associated with low birth weight through these complications. In addition, when the level of medical care becomes very high, it helps ensure that low-weight fetuses are born safely. If this is the case, the low-birthweight rate will have a U-shaped relationship with BMI and income level. Furthermore, this study considers employment status and average mother’s age at childbirth as important factors for low birthweight.

To test these hypotheses, this study used data collected from 143 countries, spanning 2000–2015, which Blencowe et al. [13] collected. We estimated fixed-effects models to control for unobservable variables, such as race and cultural background. To derive policy implications through international comparisons, the results should suggest causes rather than associations. In this case, estimating fixed-effects models using panel data has certain merit compared to cross-sectional analysis because it reveals the rule of within-individual time variation. In addition, we conducted a difference-in-differences (DID) analysis using the 2008 global financial crisis as an exogenous natural experiment. To the best of our knowledge, there are no studies on low birthweight using DID analysis, although the impact of the global financial crisis on fertility, health, and healthcare systems has been studied [15, 16]. Furthermore, whereas most of these studies are country-by-country comparisons of European countries, this study analyzed 143 countries, both developing and developed.²

The remainder of this paper is organized as follows: Sect. 2 proposes the hypotheses, presents the definitions of the variables used in this study, and explains the methods of analysis. Section 3 presents the empirical results. Section 4 discusses the results and concludes the paper.

Methods

Hypotheses

According to [17], having a BMI either lower or higher than the standard, which ranges from 18.5 to 24.99, is undesirable for health. Thus, we hypothesized that women with a standard BMI would have a lower probability of delivering low-birthweight infants. In particular, thin women are more likely to give birth to underweight infants due to undernutrition and other reasons [18]. Thus, we hypothesized the following:

Hypothesis 1

The low-birthweight rate decreases, though diminishingly, as women’s average BMI for the respective country increases and may have a U-shaped curve with respect to the average BMI of women.

As countries with a low real GDP per capita have poor nutritional status and healthcare systems, they are more likely to fall ill when their health deteriorates [19]. Thus, the low-birthweight rate is expected to be higher in poorer countries. This situation improves as countries become richer but to a diminishing degree. Therefore, we propose the following hypothesis:

Hypothesis 2

The low-birthweight rate is higher in countries with lower real GDP per capita. However, the degree of the decline diminishes.

Many other factors may be involved in the probability of low birthweight [10, 11, 13, 20]. This study addresses women’s employment status, level of medical care, and mother’s age at birth.

Regarding employment status, we use women’s employment rate and vulnerable employment rate, refers to women’s work in their family-operated small businesses.³ Vulnerable employment is more prevalent in poorer countries, and as societies mature, women shift to regular businesses. Though we imagine that harder work generally raises the low-birthweight rate, it is difficult to ascertain the effect of the shift from “vulnerable employment” to “employment.” Therefore, this study does not have a priori expectations but examines the data to ascertain the reality.

All else being equal, a society with a well-developed healthcare system keeps mothers in good health and reduces the prevalence of low-birthweight infants. However, as medical technology becomes more highly developed, more lives will be saved, but the probability of low-birthweight infants is likely to increase. Therefore, we propose the following hypothesis:

Hypothesis 3.1

Countries with higher levels of medical care have lower rates of low-birthweight babies because they can maintain maternal health. However, in countries with more advanced medical technology, the decline in the incidence of low birthweight is progressively diminishing because babies that would have otherwise been stillborn are saved.

There is an appropriate age for giving birth; adolescent childbearing, which is more common in poor countries, is considered too early, whereas childbearing for those over 40 years old, which is more common in rich countries, is considered too late [21,22,23].⁴ Because only adolescent fertility rates are available, we posit the following hypothesis:

Hypothesis 3.2

Countries with a higher percentage of adolescent (teenage) births have a higher incidence of low birthweight.

Method of estimation

Definitions of all variables used in this study are given in Table 1.

Table 1 Variable definitions and sources

To test these hypotheses, we estimated the quadratic and cubic equations with respect to BMI and GDP.

$$\begin{gathered}\:\:{R_{LBW}}_{it}\, = \,{\alpha _1}BM{I_{it}}\, + \,{\alpha _2}BM{I^2}_{it}\, + \,{\alpha _3}BM{I^3}_{it}\, + \,{\beta _1}GD{P_{it}} \hfill \\\,\,\,\,\,\,\,\,\,\,\,\,\,\, + \,{\beta _2}GD{P^2}_{it}\, + \,{\beta _3}GD{P^3}_{it}\, + \,{\gamma _1}EMPLOYE{E_{it}}\, \hfill \\\,\,\,\,\,\,\,\,\,\,\,\,\, + \,{\gamma _2}WEAK\_LABO{R_{it}}\, + \,{\gamma _3}Medicin{e_{it}}\, + \,{\gamma _4}YOUNG\_MOTHE{R_{it}} \hfill \\\,\,\,\,\,\,\,\,\,\,\,\,\, + \sum\nolimits_{k = 2001}^{2015} {{\delta _k}DYEAR_t^k + {u_i} + {\varepsilon _{it}},\:\:t = 200} 0,\: \ldots \:,\:2015\: \hfill \\ \end{gathered} $$

(1)

Where$\:{u}_{i}$ is the fixed effect of the $\:i$th country, $\:{DYEAR}_{t}^{k}$ is the year k’s dummy variable, and $\:{\epsilon}_{it}$ is the disturbance term. In the estimation of the quadratic function, where $\:{\alpha}_{3}\:\text{a}\text{n}\text{d}\:{\beta}_{3}\:$are assumed to be 0, if the estimate of$\:\:{\alpha}_{1}$ is significantly negative and the estimate of $\:{\alpha}_{2}$ is significantly positive, then R_LBW is a U-shaped function of BMI, consistent with Hypothesis 1. Similarly, if $\:{\beta\:}_{1}$ is significantly negative and $\:{\:\beta\:}_{2}$ is significantly positive, then R_LBW is a U-shaped function of GDP, consistent with Hypothesis 2. To determine the curve estimated by the cubic function, we draw a curve of the fitted value of R_LBW using the quadratic and cubic functions and examine whether its shape is consistent with Hypotheses 1 and 2. If $\:{\gamma}_{1}\:$ad $\:{\gamma}_{3}$ are significantly negative and $\:{\gamma}_{2}$ and $\:{\gamma}_{4}$ are significantly positive, then Hypotheses 3 1 and 3 2 are supported, respectively.

All analyses were conducted using Stata. For the fixed-effects model, we accounted for within-country correlation over time by clustering standard errors at the country level.

This study reports the results using a fixed-effects model with the advantage of controlling for unobservable and fixed factors during the observation period, including racial, ethnic, and cultural backgrounds unique to each country. Another important merit is that it reveals within-country rules, and its estimation results may reflect a causal relationship.⁵

In addition, this study applies a DID analysis to further confirm the causal relationships for policy implications [24]. An important exogenous event in the low-birthweight rate worldwide was the 2008 financial crisis. Therefore, this study uses 2008–2009 as the event period and tests whether changes in the low-birthweight rate from before to after the event significantly differed among countries that experienced a large decline in real GDP during this period (treatment group) and those that did not (control group).⁶ Therefore, we estimated Eq. (2).

$$\begin{gathered}\:R\_LB{W_{it}} = {\delta _1}GROU{P_i} + {\delta _2}AFTE{R_t} + {\delta _3}(GROU{P_i} \times \:AFTE{R_t}) \hfill \\\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\, + {\lambda _{1\:}}EMPLOYE{E_{it}} + {\lambda _2}WEAK\_LABO{R_{it}}\: \hfill \\\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\, + {\lambda _{3\:}}BIRTH\_STAF{F_{it}} + {\lambda _{4\:}}YOUNG\_MOTHE{R_{it}} + \:\:{u_i} + {\varepsilon _{it}}\:\:\:\:\: \hfill \\ \end{gathered} $$

(2)

where AFTER is a dummy variable coded as 1 for 2009 and later, and 0 for 2007 and earlier. GROUP is a dummy variable coded as 1 when the sum of real GDP growth in 2008 and 2009 is in the bottom 50% (treatment) and 0 when it is in the top 50% (control). The coefficient $\:{\delta}_{3}$ of GROUP×AFTER$\:\equiv\:$CROSS indicates the impact of the financial crisis.

For the DID analysis, we implemented an interaction-based approach, ensuring that standard errors were also clustered at the country level for robustness.

$$\:\:\:\:\:\frac{{{\partial ^2}R\_LB{W_{it}}}}{{\partial \:GROU{P_i}\partial \:AFTE{R_t}}} = {\delta _3}$$

(3)

If $\:{\delta\:}_{3}$ is significantly positive, it supports the interpretation that the negative association between GDP and LBW suggests a causal relationship.

We further estimate Eq. (4) to test whether the global financial crisis affected the low- birthweight rate through the employment rate.

$$\begin{gathered}\:R\_LB{W_{it}}\, = \,{\delta _1}GROU{P_i} + {\delta _2}AFTE{R_t} \hfill \\\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\, + \,{\delta _4}EMPLOYE{E_{it}}\, \times \:GROU{P_i}\: \hfill \\\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\, + {\delta _5}AFTE{R_t} \times \:EMPLOYE{E_{it}} + {\delta _7}EMPLOYEE_{it}^2 \times \:GROU{P_i} \hfill \\\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\, + {\delta _8}AFTE{R_t} \times \:EMPLOYE{E_{it}} \hfill \\\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\, + \left( {{\delta _3} + {\delta _6}EMPLOYE{E_{it}} + {\delta _9}EMPLOYEE_{it}^2} \right)\, \times \:\left( {GROU{P_i} \times \:AFTE{R_t}} \right) \hfill \\\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\, + + {\lambda _{1\:}}EMPLOYE{E_{it}} + {\lambda _2}WEAK\_LABO{R_{it}}\, + {\lambda _{3\:}}BIRTH\_STAF{F_{it}} \hfill \\\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\, + {\lambda _{4\:}}YOUNG\_MOTHE{R_{it}} + {u_i}\: + {\varepsilon _{it}} \hfill \\ \end{gathered} $$

(4)

In this case,

$$\:\frac{{{\partial ^2}R\_LB{W_{it}}}}{{\partial \:GROU{P_i}\partial \:AFTE{R_t}}} = {\delta _{3\:}} + {\delta _6}EMPLOYE{E_{it}} + {\delta _9}EMPLOYEE_{it}^2$$

(5)

where $\:{\delta}_{3}$ represents the impact of the financial crisis on R_LBW through channels other than $\:EMPLOYEE$, and $\:{\delta}_{6}{EMPLOYEE}_{i,t}+{\delta}_{9}{EMPLOYEE}_{it}^{2}$ represents the impact through changes in $\:EMPLOYEE$.

Results

Our dataset consists of a maximum of 2,145 country-year observations from 143 countries for the period 2000–2015 (see Table S1 in the Supplementary Material for the list of countries). However, the number of observations varies by variable owing to data availability constraints, with the minimum number of observations being 1,304.

Descriptive statistics

Table 2 presents the descriptive statistics for the variables used in the estimation. First, we defined “low-income countries” as the bottom 30%, “middle-income countries” as the mid- 40%, and “high-income countries” as the top 40% based on the average GDP for 2000–2015. The median rate of low birthweight decreased as GDP increased. The median BMI was lowest in low-income countries but did not differ between middle- and high-income countries.

Table 2 Descriptive statistics of the variables

Preliminary analysis

Table 3 presents the correlation coefficients between R_LBW and key variables. In the low-income countries, R_LBW and BMI were negatively correlated at the 1% level, whereas in the high-income countries, they were significantly positively correlated. Similarly, R_LBW and GDP are significantly negative at the 1% level for the low- and middle-income countries but significantly positive at the 1% level for the high-income countries. Regardless of country affluence, R_LBW correlates negatively with the female employment rate (EMPLOYEE) and BIRTH_STAFF but positively with vulnerable employment (WEAK_LABOR) and YOUNG_MOTHER (adolescent fertility rate). Simple correlations were consistent with our hypotheses.

Figure 1 shows the means of R_LBW and BMI for the entire world as well as for low-income, middle-income, and high-income countries. In the low-income countries, R_LBW declined substantially, by 2% in the 16 years examined, whereas it declined only slightly in the middle-income countries, and it even rose, though not significantly, in the high-income countries. In aggregate, it has declined slightly worldwide. Moreover, BMI increased significantly not only in low-income countries but also in middle- and high-income countries. An increase in BMI in low-income countries is regarded as an improvement in --health, whereas in middle- and high-income countries, it may indicate the opposite.

Table 3 Correlation coefficients between Low-birthweight rate and the key variables

Fig. 1

Worldwide changes in low birthweight and BMI from 2000 to 2015. Error bars indicate the standard error of the mean (SEM)

Main result

Table 4 presents the estimation results for Eq. (1). Column (1) shows the estimates of the quadratic function of BMI and GDP. The estimate of BMI was significantly negative, and that of the squared term was significantly positive, indicating a U-shaped curve. However, the bottom of the U-curve was 28.83, indicating that most observations were in the decreasing domain. In contrast, the estimates on GDP and GDP² were insignificant. The estimates of EMPLOYEE and EMPLOYEE² were significantly positive and negative, respectively, indicating an inverse U-shape with a peak at 41%, which is close to the median of EMPLOYEE for high-income countries (Table 2). WEAK_LABOR and its squared values were not significant. BIRTH_STAFF and its square were significantly negative and positive, respectively, indicating a U-shape with a bottom at 74.7%. YOUNG_MOTHER was significantly positive. These results support our hypotheses, with the exception of Hypothesis 2.

To examine whether Hypothesis 2 is valid, columns (2) and (3) present the results in the case that the BMI and GDP variables are included separately. In column (3), the estimates of GDP and GDP² become significantly negative and positive, respectively, which is consistent with Hypothesis 2. Column (4) presents the estimates of the cubic function of BMI and GDP. The estimates are qualitatively the same as those in column (1), except that BIRTH_STAFF and BIRTH_STAFF² become insignificant. The estimate of BMI³ was significantly negative. Figure 2 shows the predicted values of R_LBW calculated from the estimates in columns (1) and (4) of Table 4, which are overlaid on the scatter plots of R_LBW and BMI. The curve of the quadratic function is U-shaped, whereas the curve of the cubic function reveals that R_LBW decreases strongly until the inflection point at 26.7 and then decreases gradually thereafter. However, the two curves are fairly similar; the only qualitative difference is that the quadratic function predicts an increase in R_LBW after the bottom, that is, BMI = 28.8, and the cubic function predicts a decrease.⁷

Table 4 Estimation results of Eq. (1)

To further evaluate the model’s predictive performance, Supplementary Figure S1 compares the actual and predicted low-birthweight rates. The correlation between them is 0.63, significant at the 1% level.

Fig. 2

Scatter plots of BMI and R_LBW with fitted values of quadratic and cubic functions. Scatter plots of BMI and R_LBW in each country 2000–2015 and fitted values of the quadratic and cubic functions presented in Eq. (1) are shown. The red dots represent the means at each BMI of predicted values calculated using the estimates by the quadratic function shown in column (1) of Table 4, and the white dots represent those estimated by the cubic function shown in column (4). The blue dots represent the observed values of R_LBW

Examination of causality: DID analysis

We first checked whether the parallel trend assumption holds true. Figure 3a presents the average R_LBW values for the treatment and control groups, which show that there was no extreme difference in the movements of the two groups before the onset of the 2008 financial crisis. Figure 3b and c illustrate the groups separately to enlarge the figure and show the movement more clearly. However, the R_LBW of the control group was not affected by the crisis (Fig. 3b). The R_LBW of the treatment group increased during the crisis for a period of time. These figures reveal that the financial crisis was an exogenous event to R_LBW, affecting only the treatment countries, and that the assumption of a parallel trend was satisfied.

Fig. 3

Observed means of R_LBW for parallel trend. Panel a: Trend for both groups. Panel b: Trend for the control group. Panel c: Trend for the treatment group. Group1 (treatment) consists of the countries with real GDP growth rates in 2008 and 2009 in the bottom 50%, and Group 0 (control) represents all others

To further confirm the plausibility of the DID approach, we examined employment rate trends for treatment and control groups (Fig. 4). The observed patterns indicate that employment rates followed similar trends before 2008, supporting the assumption that any post-crisis divergence is attributable to the financial shock rather than pre-existing differences.

Fig. 4

Parallel trends in employment rates. Panel a: Employment rate trends for both groups. Panel b: Employment rate trend for the control group. Panel c: Employment rate trend for the treatment group. Observed employment rates for treatment and control groups before and after the 2008 financial crisis. Group1 (treatment) consists of the countries with real GDP growth rates in 2008 and 2009 in the bottom 50%, and Group 0 (control) represents all others

Columns (1) and (2) of Table 5 present the estimation results of Eq. (2). When R_LBW is regressed on the three variables AFTER, GROUP, and CROSS (= GROUP×AFTER), CROSS is significantly positive at the 1% level, indicating that countries hit harder by the financial crisis have a lower low-birthweight rate than those that were not (column 1). In column (2), when EMPLOYEE and EMPLOYEE² were added, CROSS was significantly positive at the 5% level.

Columns (3) and (4) present the estimation results of Eq. (4). Column (3) presents the results when only EMPLOYEE and EMPLOYEE² were included in CONTROL. Column (4) presents the results of the full specification of the control variable as in column (1) of Table 4. The coefficients of CROSS×EMPLOYEE and CROSS×EMPLOYEE² are significantly positive and negative, respectively, indicating that the impact of the financial crisis via EMPLOYEE significantly increased the low-birthweight rate but diminishingly.

Table 5 The estimates of Eqs. (2) and (4): DID analysis

Discussion

We posited four hypotheses on the relationship between the rate of low birthweight and the supposed risk factors. Data analysis supported the relationship, indicating the significance of our risk factors. However, two problems remained concerning Hypotheses 1 and 2. First, the shape of the function in regions of large BMI was not sufficiently clear. Although the estimation of the quadratic function revealed a U-shaped curve, the bottom of the U-shaped curve was located near the right edge of the observed value, suggesting that the association was decreasing but diminishing in most of the domains. It is not yet clear how the increased BMI will affect the incidence of low birthweight if BMI is further increased in the future. Second, GDP, which was found to be an important risk factor, was insignificant when BMI was added, suggesting that GDP is associated with R_LBW through many paths, including BMI. Examining this complex relationship remains a task for future studies.

This study found that the association of the rate of women’s employment and that of low birthweight had an inverse U-shaped curve, with its peak being around 41%. This suggests that an increase in women’s employment rate during the period of 2000–2015 was associated with a decline in the rate of low birthweight in low- and middle-income countries and improved the rate of low birthweight in high-income countries on average. WEAK_LABOR was not significant. Relation between BIRTH_STAFF and low-birthweight rate is a U-shaped curve, with its bottom being approximately 75%. Adolescent birth rate is significantly and positively associated with low-birthweight rate; specifically, a decrease in the adolescent birth rate decreases the low-birthweight rate.

The results can address how to achieve the goal of the 65th World Health Assembly to “reduce the incidence rate of low birth weight.” Table 6 presents the contributions of each individual variable to the reduction in the rate of low-birthweight infants during the period of 2000–2015, which was calculated from the change in the magnitude and estimates of the coefficients shown in Column 1 of Table 4. Regarding the 1.3% decrease in fitted value of R_LBW in 16 years, although the increase in BMI contributed 54% of this reduction, the decrease in YOUNG_MOTHER contributed 30%, the increase in GDP contributed 13%, the increase in BIRTH_STAFF contributed 5%, and the increase in employment rate deteriorated rate of low birthweight slightly.⁸

Which is the most promising political instrument for the future? To answer this question, we must investigate the change in each variable by income-group countries. Over the 16 years examined, R_LBW improved from 16.1 to 10.2% in low-income countries, whereas it deteriorated slightly from 7.2 to 7.8% in high-income countries. These values suggest that low-income countries have room to reduce low birthweight rates. The variables in Table 6 may continue to be policy instruments, but their effectiveness will be weakened because the situations have already been improved even in the low-income countries over the 16 years examined: BMI 22.7 to 25.6, BIRTH_STAFF 50–86%, YOUNG_MOTHER from 9.5 to 7.6 births per 100 females. However, the GDP and employment rate of low-income countries were still substantially lower compared with high-income countries, that is, GDP US$ 4020 vs. US$ 41,728 and EMPLOYEE 17% vs. 46%. While improvements in standard of living and employment are expected to reduce low-birthweight rates, the most effective policy interventions will vary by country and require tailored approaches.

Although this study identifies key determinants of low birthweight, implementing effective policies requires adaptation to each country’s unique economic, healthcare, and social conditions. For example, in low-income countries, improving maternal nutrition and access to healthcare may be the most effective strategy, whereas in high-income countries, addressing obesity-related pregnancy complications may be a greater priority.

Table 6 Contributions of individual variables to the reduction in the rate of low-birthweight babies during the period of 2000–2015

Conclusions

This study identified the risk factors for low-birthweight births using data from 143 countries for 2000–2015. Although many risk factors have been identified using individual data in a country [4, 9, 12], international comparisons of low birthweight problems have not been adequately performed. A multi-county analysis is essential to determine whether these factors are systematically associated with low birthweight across diverse economic and healthcare systems. By comparing multiple countries, this study provides insights into both universal and country-specific determinants of low birthweight, helping to fill this gap in the literature.

In this study, we focused on five risk factors: women’s BMI, per capita GDP, women’s employment status, level of healthcare, and adolescent childbearing. We employed the quadratic and cubic functions in the estimation because the association between the rate of low-birthweight births and these variables is supposed to be non-linear. For policy applicability, this study used a fixed-effects model in which the estimates represent correlations between time-varying variations within a country. Furthermore, using the 2008 financial crisis as a natural experiment, the DID analysis provides evidence suggesting a relationship between GDP and low birthweight rate, where female employment rates play and important role. Thus, the results of this study represent a step toward policy use.

This study has several limitations. Among these, data availability may be the most significant. Variables such as the rate of high-birthweight infants, distribution of BMI, and rate of births by mothers aged over 40 years were not available. Obtaining more detailed data and using a more flexible functional form will be tasks for future studies. DID analysis should to be performed more carefully, considering the influence of unmeasured confounders.