Amazon Forest Fires Between 2001 and 2006 and Birth Weight in Porto Velho

TL;DR

This study investigates the impact of Amazon forest fires between 2001 and 2006 on birth weights in Porto Velho, using statistical models to analyze data from over 22,000 live births.

Contribution

It provides new evidence linking forest-fire smoke exposure to variations in birth weight in a specific Amazon region.

Findings

Significant difference in mean birth weights between girls and boys.

Regression models explain over half of the variation in birth weights.

Birth weights are affected by forest-fire smoke exposure.

Abstract

Birth weight data (22,012 live-births) from a public hospital in Porto Velho (Amazon) was used in multiple statistical models to assess the effects of forest-fire smoke on human reproductive outcome. Mean birth weights for girls (3,139 g) and boys (3,393 g) were considered statistically different (p-value < 2.2e-16). Among all models analyzed, the means were considered statistically different only when treated as a function of month and year (p-value = 0.0989, girls and 0.0079, boys) . The R 2 statistics indicate that the regression models considered are able to explain 65 % (girls) and 54 % (boys) of the variation of the mean birth weight.

Tables

Table 1. Table 1: Descriptive statistics of the birth weight by sex in the period of 2001-2005.

Sex	Min	1st Q	Median	Mean	3rd Q	Max	DP	CV
Female	1185	2900	3200	3193	3490	5900	450.8375	0.1412
Male	1160	3000	3305	3310	3600	5730	477.6589	0.1443

Table 2. Table 2: Number of heat spots captured by satellite images for the state of Rondonia during the studied period.

Year	January	February	March	April	May	June	July	August	September	October	November	December	Total
2000	0	6	9	0	0	54	116	2196	2437	643	12	6	5485
2001	0	2	0	0	5	31	74	2133	1927	796	91	1	5060
2002	6	8	1	3	10	141	285	1872	6095	1874	146	20	10461
2003	8	2	1	0	34	66	668	4126	2742	1286	134	63	9130
2004	4	4	3	2	7	126	558	3718	6937	1725	99	17	13200
2005	7	0	0	2	12	82	580	6699	8775	1502	158	1	17818
Total	25	22	14	7	68	500	2281	20744	28913	7826	640	108	61154
Source: INPE - Satellite NOAA-12

Table 3. Table 3: Final model for Girls mean birth weight

Coefficients	Estimate	Std. Error	t value	Pr($>\left|t\right|$)
Intercept	3196.454	5.896	542.13	2e-16 ***
$a$	-340.404	32.294	-10.54	4.25e-15 ***
Residual standard error: 44.9 on 58 degrees of freedom (df)
Multiple R-squared: 0.657; Adjusted R-squared: 0.6511
F-statistic: 111.1 on 1 and 58 df, p-value = 4.254e-15
Note: Significance code: 0 ‘***’

Table 4. Table 4: Final model for Boys mean birth weight

Coefficients	Estimate	Std. Error	t value	Pr($>\left|t\right|$)
Intercept	3.321e+03	6.356e+00	522.566	2e-16 ***
$a$	-1.688e+02	2.178e+01	-7.749	1.79e-10 ***
$b$	-4.485e - 03	2.168e - 03	-2.069	0.0431 *
Residual standard error: 41.92 on 57 degrees of freedom (df)
Multiple R-squared: 0.5429; Adjusted R-squared: 0.5269
F-statistic: 33.85 on 2 and 57 df, p-value = 2.044e-10
Note: Significance codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’

Full text

††footnotetext:

*Corresponding author’s address: Taiane S. Prass

Instituto de Matemática

Universidade Federal do Rio Grande do Sul

91509-900 Porto Alegre, RS - Brazil

E-mail: taianeprass@gmail.com

FAX: 55-51-3308-7301

Running head: Birth weight and forest fires in the Amazon.

The authors declare they have no competing financial interests.

1 Introduction

The aggressive occupation of the Brazilian Amazon dates back to the expansion of the agricultural frontier that started in the $1970$s with the opening of roads. Agriculture, mining, and economic development have led to human occupation and emergence of urban centers in the West Amazon; the state of Rondonia bordering Bolivia, is one of the most impacted by forest fires due to its expanding agriculture and mining activities. Government incentives boasted large-scale agriculture projects, always starting with forest clearances by fire (Aragão and Shimabukuro, 2010). Agricultural land claimed from the felled forest has been made into cropland and pasture, which is managed by fire. Despite official efforts to limit further forest land occupation or preservation, the increasing price of meat (green meat) and soy beans has driven forest destruction at alarming rates in the last $10$ years (Aragão and Shimabukuro, 2010); all done by using deliberate but out of control forest fires. The extent of the felled native forest and, the continuing ecological damage to the native flora and fauna remain unknown. However, there have been attempts to assess health impacts on some urban centers.

Ignotti et al. (2010) showed an association of atmospheric pollution (due to forest burning) with occurrences of respiratory diseases (in children and the elderly) in the Brazilian Amazon region. Respiratory diseases in the elderly due to forest fires accounted for $50$ to $80\%$ of mortality rate in the state of Rondonia between the years of $1998$ and $2005$ (Castro et al., 2009). Also in the West Amazon, Mascarenhas et al. (2008) measured particulate matter $<2.5$microm (resulting from forest fire) and showed its association with increased rates of respiratory diseases in children $<10$ years of age.

This study is a first attempt to assess the impact of atmospheric pollution brought about Amazon forest fires on reproductive outcomes (birth weight) in a large urban center - the city of Porto Velho - located in the most impacted region of deforestation and human occupation in the Amazon.

2 Materials and Methods

**Data acquisition **

This is part of our ongoing project on environmental hazards to children’s health in the West Amazon (Dórea et al., 2007); the research protocol was approved by the Ethics Committee of Studies for Humans of the Universidade Federal de Rondonia. We had access to the birth records of children born at the Hospital de Base, the largest public hospital facility caring mostly for the lower middle class in the city of Porto Velho (capital of the state of Rondonia, West Amazonia). A parent publication dealing with ethylmercury exposure derived from Thimerosal-containing vaccines given to newborns appeared elsewhere (Dórea et al., 2009).

We collected data of live births occurring from $2001$ to $2006$; these data were recorded by hospital staff and comprised live birth-weight, sex, and date of birth. The total number of observations for the period was $n=22012$ with a sex proportion of $48.4\%$ of females and $51.6\%$ males. Forest fire information was based on data collected by the satellite NOAA-12 with records summarized monthly.

**Data summarization **

Figure 1 shows the observed values of the birth weights for girls (left plot) and for boys (right plot), during the studied period. The lack of observations during the period from 12/02/2002 to 02/03/2002 corresponds to interrupted services at the maternity ward due to building repairs on the hospital.

The descriptive statistics for the birth weight data are presented in Table 1. The range of variation is wider for girls ($4715$g) than for boys ($4570$g); moreover, both, minimum and maximum weights are higher for girls than for boys (see Table 1). Nevertheless, the coefficient of variation (CV) is relatively low for both girls and boys.

The data are also illustrated in Figures 2(a) and 2(b) as a histogram and a box-plot, respectively. While the histogram shows an overall symmetry between girls and boys regarding ranges of birth weight, the box-plot indicates the presence of outliers in the data. The frequency of outliers for boys is $1.66\%$ and, for girls is $1.02\%$.

Table 2 shows a summary of heat spots captured by satellite images for the state of Rondonia during the studied period; the total number of heat spots varied on a yearly basis, almost doubling from $2001$ to $2002$ then decreasing from $2002$ to $2003$, rising again from $2003$ to $2004$ and from $2004$ to $2005$. Altogether, the number of heat spots practically doubled from $2003$ to $2005$. An illustration of heat spots is graphically shown in Figure 3; it is worth noting that their occurrence is higher during the second semester, which is the dry season (see Figure 3 and Table 2). In tandem with this observation, if forest fires affected birth weights more light children would be expected during the first semester than in the second semester. Furthermore, for years with a higher number of heat spots one would also expect more light birth weights.

It can be noticed that the greatest number of fires occurred during the months of August and September of any given year, which correspond to the peak of the dry season. Considering that the last semester of pregnancy could be the most determinant of birth weight, it is expected that March to June should show effect. We speculate that the smoke pollution from forest fires can be unfavorable for the first months of a pregnancy because they could affect the fetus initial development, thus modulating weight gain throughout pregnancy. Besides, a quite well developing fetus has more metabolic chances of gaining weight than one with an initial developmental hindrance.

In order to identify a possible relation between the numbers of heat spots that occurred during pregnancy and related birth weight we considered a multi-step approach. First we used ANOVA to identify whether the sex or the semester, the month or the year of birth, had any influence on the birth weight. After that we considered a regression model that took into account the information obtained in the first step to explain the mean birth weights as a function of the number of heat spots.

3 Results and Discussion

As a result of the apparent differences shown in Table 1, a preliminary analysis to assess the influence of sex was necessary in order to establish if further analysis would have to consider this variable in the statistical model. We tested this hypothesis and, indeed, there was a statistically significant difference between the means (F = 349.19 with p-value $<$2.2e-16 at a confidence level of $99.9\%$). Due to sex effects on birth weight, we included it in the analysis to assess the influence of the heat spots on birth weight. Due to the variation in number of heat spots over the years, several ANOVA tests were performed in order to identify the time (year) effect on birth weight; each sex was analyzed independently.

When considering year as the only variable we cannot reject the null hypothesis that annual means are equal in both cases (girls: F = 0.4854, p-value = 0.7465; boys: F = 1.0118, p-value = 0.3997). Because heat spots occur during the dry season, the time effect was also analyzed as a function of semesters only. In this case also no statistically significant differences between the means for the first and second semesters (girls: F = 1.814, p-value = 0.1781; boys: F = 2.3279, p-value = 0.1271).

Figure 4(a) shows mean birth weight per semester for each studied year. One observes that only in 2003 the average in the first semester is higher than the average in the second semester for both sexes. Figure 4(b) shows the mean birth weight by year for each fixed semester. It is worth noting that for the boys there was a noticeable decrease in birth weight in the first semester of $2002$ that might be a bias introduced due to lower number of observations.

The analysis done independently for semesters and for years showed no statistically significant difference; therefore, we decided to test the combined effects of year and semester in a single statistical model. In this test we hoped to scrutinize for relationship of birth weights and number of heat spots occurring in a specific period of a specific year. In that way if there was a mean birth weight of a semester that would differ from another, it could possibly be detected.

For girls, when we consider the semester by year there was no statistically significant effect (F = 0.8683, p-value = 0.553); also when we analyze a fixed semester and compare the mean for different years, there was also no statistically significant difference (F = 1.0743, p-value = 0.3674) for the first and second semesters (F = 0.4282, p-value = 0.7884). However, for boys, using a similar model, there was a statistically significant effect (F = 2.7001, p-value = 0.0039). As shown in Figure 4(a), the mean differing from the others is the one from the first semester of $2002$. Because of the confounding of fewer observations in that specific semester (first semester of $2002$) these results are considered an analytical artifact. By comparing the means for different years in a fixed semester, we can reject the null hypothesis that they are equal (F = 2.7583 with p-value = 0.0263 and for the second semester, F = 2.72554 with p-value = 0.0278); Figure 4(b) clearly shows an anomaly in the first semester of $2002$, and in the second semester of $2003$. Although we realize that the low mean weight observed during the first semester is unlikely to be related to forest fires the mean birth weight observed in the second semester of $2003$ could possibly be related to the high number of heat spots ($34$) registered in the first semester of $2003$.

Figures 5 and 6 illustrate mean birth weight per month by year; the effects of months throughout the studied years showed a significant effect for both sexes (girls: F = 1.2440, p-value = 0.0989; boys: F = 1.5104, p-value = 0.0079). Figure 5 shows that for both girls and boys, the salient mean birth weight corresponds to March-2002. Considering months as a fixed effect over the years there was no significant difference between the means, except for March, in the case of girls (F = 5.9700, p-value = 9.938e-05) and of boys in March (F = 2.648, p-value = 0.0323) and November (F = 4.8289, p-value = 0.0007). It should be noted that the semester mean (as a function of year) showed significant differences; indeed, from Figure 6, one observes that for March, this was the case with both boys and girls. From Table 2 it can be seen that March-$2003$ showed the highest number of heat spots, compared to March from other years.

Mean birth weight was statistically different between boys and girls; therefore we used a model for each sex to explain mean birth weight, not individual birth weight. The models used for girls and boys accounted for month, semester, and year of birth as well as number of heat spots. For the variable ‘number of heat spots’ we considered number of cases corresponding to a gestation period and overall number for a specified gestation period. After a careful step-by-step run of several statistical models we could conclude that:

Girls:

• a)

Mean birth weight for girls does not depend on month, semester, or year when these variables are analyzed separately.

• b)

When mean birth weight is analyzed as a function of a specific month and year, a significant difference is detected for March-2002. However, further detailed analysis showed that introducing a constant for March and April of 2002 explains variations in birth weight for girls. As a consequence there was no statistically significant evidence that birth weight for girls can be affected by number of forest fires.

Boys:

• a)

Overall, when the variables month, semester, and year are considered per se, there is no statistically significant difference. However, further detailed analysis showed a significant factor influencing mean birth weight for the period from March to June of 2002.

• b)

In order to introduce the ‘number of heat spots’ variable in the model, we included it as a new variable, denoted by $N(\cdot)$, and tested the following possibilities:

• –

$N(\cdot)$ = Number of heat spots in $i$-th month of gestation, where $i\in\{0,1,\cdots,9\}$,

• –

$N(\cdot)$ = Sum of the number of heat spots from month $i$ to $j$, with $i\leq j$ and $i,j\in\{0,1,\cdots,9\}$,

where, by definition, ‘0’ is the month at which pregnancy has started.

Among all tested models, the best to explain mean birth weight variation (considering the value of the statistic $R^{2}$ in the regression model) were:

• •

For girls: $W(m)=\mu+a\,\mathds{I}_{A}(m)=3196.454-340.404\,\mathds{I}_{A}(m)$, where $m$ is the birth period, $A=\{\mbox{March}-2002,\mbox{April}-2002\}$ and $\mathds{I}_{A}(\cdot)$ is the indicator function of set $A$, which is defined by $\mathds{I}_{A}(m)=1$, if $m\in A$, and $\mathds{I}_{A}(m)=0$, if $m\notin A$. The coefficients of the model, its p-value, and the $R^{2}$ value are shown in Table 3. The value of $R^{2}$ indicates that $65\%$ of the variation of the mean birth weight for girls can be explained by this model. Figure 7(a) illustrates the girls’ monthly mean birth weight (in red) and respective values estimated by the model (in black) for the period of $2002$ to $2005$.

• •

For boys: $W(m)=\mu+a\,\mathds{I}_{A}(m)+bN(m)=3321-168.8\,\mathds{I}_{A}(m)-0.004485N(m)$, where $m$ is the birth period, $A=\{\mbox{March}-2002,\mbox{April}-2002,\mbox{May}-2002,\mbox{June}-2002\}$, $\mathds{I}_{A}(\cdot)$ is the indicator function of set $A$, and the $N$ variable is defined as $N(m)=\sum_{j=0}^{1}n_{j}$, where $n_{j}$ is the number of heat spots after the $j$-th month of gestation and $n_{0}$ is the gestation initial month. The coefficients of the model, its p-value, and the $R^{2}$ value are shown in Table 4. The value of $R^{2}$ indicates that $54\%$ of the variation of the mean birth weight for boys can be explained by this model. Figure 7(b) illustrates the boys’ monthly mean birth weight (in red) and respective values estimated by the model (in black) for the period of $2002$ to $2005$.

The models captured the most accentuated decreases of mean birth weight (Figures 7(a) and 7(b)) however it failed to capture the birth weight fall of November-$2003$.

The smoke and particulate matter generated by forest fires currently going on in the Amazon have been positively associated with morbidity and mortality of vulnerable groups (Mascarenhas et al., 2008; Castro et al., 2009; Ignotti et al., 2010). These effects are caused by impairment of lung function due to physicochemical properties of the inhaled particles (Schwarze et al., 2006). Population based studies have shown an increased risk of low birth weight associated with households that use biomass fuel (Boy et al., 2002; Tielsch et al., 2009; Thompson et al., 2011).

We noticed that, all analysis indicating that the means are not statistically equal, also indicated that differences occurred in the means corresponding to the period with missing observations. Dealing with missing data is always challenging and, based on these findings, we believe that a more complete data set could lead to different models and attendant outcomes. For instance, the indicator function, included in both models to account for the information that is not available, would not be necessary.

We could not control for variables known to interact with reproductive outcomes. On a population based birth weight outcome there was no direct association of a particular year or season. However, our study showed air pollution (derived from forest fires) to have an unfavorable association with birth outcomes.

Acknowledgments

Taiane S. Prass was partially supported by CNPq-Brazil. S.R.C. Lopes was partially supported by CNPq-Brazil, by CAPES-Brazil, by INCT em Matemática and by Pronex Probabilidade e Processos Estocásticos - E-26/170.008/2008 -APQ1.