Gestational Triclosan Exposure and Infant Birth Weight: A Systematic Review and Meta-analysis

Abstract

Background:

Exposure to triclosan, an antimicrobial chemical used in some personal care and cleaning products, has been associated with reduced birth weight in some, but not all epidemiological studies.

Objectives:

We conducted a systematic review and meta-analysis to characterize the relation of gestational triclosan exposure with infant birth weight and identify sources of heterogeneity between studies.

Methods:

We identified original studies measuring urinary triclosan concentrations during pregnancy and reporting their association with infant birth weight, gestational age (GA) adjusted birth weight (g), or GA-standardized birth weight z-scores. Using a random effects model, we estimated differences in these outcomes per 10-fold increase in triclosan concentrations and considered triclosan levels and infant sex as sources of heterogeneity. Using Navigation Guide Methods, we evaluated risk of bias within individual studies and across the body of evidence.

Results:

Among thirteen studies, median triclosan concentrations varied by almost 2-orders of magnitude (0.6–29 ng/mL), with higher concentrations in North American and some European studies compared to Asian ones. Associations between triclosan and birth weight (β:−20g; 95% CI:−65, 26; n=6) were stronger than those for GA-adjusted birth weight (β:−12g; 95% CI:−29, 5; n=9). Triclosan was not associated with GA-standardized birth weight z-scores (β:−0.04; 95% CI:−0.16, 0.07; n=5). The association between triclosan and GA-adjusted birth weight was stronger in studies with median triclosan values ≥ 10ng/mL compared to studies with median values <10 ng/mL (β:−27g; 95% CI:−61, 7; n=4 vs. β:6g; 95% CI:−20, 31; n=5). With a limited number of studies, we observed suggestive evidence that inverse associations were more apparent in studies with ≥ 2 prospective triclosan measures compared to those with one measure.

Discussion:

Available evidence, with “low” risk of bias, provides limited evidence that triclosan exposure and reduces infant birth weight. We observed stronger inverse associations between triclosan concentrations and birth weight in populations with higher triclosan exposure.

1. Introduction:

Triclosan is a man-made antimicrobial chemical used in some personal care and cleaning products (Dann and Hontela 2011; Geer et al. 2017), including some toothpastes, mouthwashes, soaps, and lotions. Human exposure occurs primarily through dermal or oral routes (Rodricks et al. 2010). Triclosan is non-persistent in the body with a biological half-life of ~21 hours (Sandborgh-Englund et al. 2006). While triclosan has a short biological half-life, individuals can be chronically exposed through routine use of triclosan containing products. Up to 75% of the United States (US) population has detectable urinary triclosan concentrations, and generally higher concentrations of triclosan are detected in women, and those with higher education and income (Arbuckle et al. 2015; Calafat et al. 2010; Han et al. 2016). There is concern over the potential health effects of triclosan because it has been found to adversely affect the hypothalamic-pituitary-thyroid axis in both experimental studies using rodents and observational studies of humans (Braun et al. 2018; Brucker-Davis 1998; Johnson et al. 2016; Skarha et al. 2019).

Adequate thyroid hormone concentrations are essential during fetal development and failure for the mother or developing fetus to maintain sufficient thyroid hormone concentrations during gestation has been associated with adverse fetal growth trajectories including reduced birth length and head circumference (Bigsby et al. 1999; de Escobar et al. 2004; Lassen et al. 2016; Philippat et al. 2014; Wolff et al. 2008). Maternal hypothyroidism has been previously associated with lower birth weight (Derakhshan et al. 2020; Korevaar et al. 2017; Monen et al. 2015). Evidence for triclosan to alter maternal thyroid hormone concentrations have been found in both animal (Axelstad et al. 2013; Paul et al. 2010; Rodríguez and Sanchez 2010) and human literature (Johnson et al. 2016; Braun et al. 2018 p. 201; Wang et al. 2017).

Indeed, some, but not all, epidemiological studies report that gestational urinary triclosan concentrations during pregnancy are associated with decreased birth weight and neonatal anthropometry(Calkins and Devaskar 2011; Gishti et al. 2014, 2015; Toemen et al. 2016). The heterogeneity in findings across the epidemiologic literature warrants further investigation in order to better understand the underlying reasons which may be contributing to these disparate findings in order to ultimately determine if triclosan adversely impacts fetal growth.

Thus, we evaluated the relation between gestational urinary triclosan and infant birth weight by conducting a systematic review and meta-analysis of human studies examining maternal exposure to triclosan during gestation and infant birth weight. In addition to applying The Navigation Guide criteria to these studies to assess risk of bias, we evaluated several potential sources of heterogeneity, including different birth weight outcomes, child sex, and median triclosan levels.

2. Methods:

2.1. Systematic Review Methodology:

We followed PRISMA guidelines as a reporting standard for our systematic literature review (Moher et al. 2009). Systematic review methodology has been previously used and validated in clinical sciences, such as the Cochrane Collaboration and Grading of Recommendations Assessment Development and Evaluation (Guyatt et al. 2008; Higgins, PT J and Green, S 2011; Woodruff et al. 2011); however, such protocols for conducting systematic reviews may not be applicable in the context of environmental health (Johnson et al. 2014, 2016; Koustas, E et al. 2014; Lam et al. 2014). For this reason, we used the Navigation Guide in conjunction with PRISMA reporting guidelines. The Navigation Guide differs from PRISMA in that it is a methodological framework for conducting environmental health systematic reviews, (Johnson et al. 2014, 2016; Koustas, E et al. 2014; Lam et al. 2014, 2016; Vesterinen et al. 2015; Woodruff et al. 2011) while the PRISMA guidelines are standards for reporting how a systematic reviews and meta-analyses was conducted. We developed our protocol for our systematic review prior to beginning our initial literature search, and registered it in PROSPERO, registration number: CRD42019147431 (Patti et al. 2019).

2.2. Study Question

Our objective was to estimate the effect of gestational exposure to triclosan on infant birth weight in the context of describing the dose-response relation for risk characterization. Thus, we sought to answer the question: “Among infants, what is the effect of a 10-fold increase in maternal urinary concentrations of triclosan during gestation on infant birth weight?” The “Participants”, “Exposure”, “Comparator” and “Outcomes” (PECO) statement is outlined below (Morgan et al. 2018).

• Participants: Mothers and their newborn infants, the population was human.

• Exposure: Maternal exposure to triclosan during the gestational period assessed via urinary triclosan concentrations. Exposure to log-10 transformed maternal urinary triclosan concentrations during gestation.

• Comparator: Mother-infant dyads exposed to lower concentrations of triclosan during gestation compared to those with higher concentrations of triclosan during gestation. An incremental increase of log-10 transformed maternal urinary triclosan concentrations during gestation on infant birth weight.

• Outcomes: Our primary outcome of interest was infant birthweight, a marker of fetal growth. To make our review as comprehensive as possible, we included all measures of birth weight reported across, and within included studies. We considered infant birth weight measured in three different ways: birth weight measured in grams, birth weight measured in grams with multivariable adjustment for gestational age, and gestational age standardized z-scores of birth weight.

2.3. Data Sources

We searched PubMed and Medline on May 29^th, 2019, with search terms identified within the supplement (Supplemental Table S1). The Medical Subject Headings (MeSH) database was used to compile our search terms used for triclosan and birth weight. Our search was limited to studies published in English. While we did not limit our search based on publication date, it is worth noting that triclosan did not enter the consumer market until the 1970s (FDA 2016 Decision and History). We updated our search on March 9, 2020, to identify any new studies. We also used several databases to identify synonyms for triclosan, and combined “triclosan” with all identified synonyms using the “OR” statement. To identify synonyms for triclosan, we used PubChem, Sigma-Aldrich, and ChemSpider (https://pubchem.ncbi.nlm.nih.gov/#query=triclosan; https://www.sigmaaldrich.com/catalog/search?term=triclosan&interface=All&N=0&mode=match%20partialmax&lang=en&region=US&focus=product; http://www.chemspider.com/Chemical-Structure.5363.html). Finally, we supplemented these results by hand-searching references of included studies in order to identify any additional publications not previously recognized.

2.4. Study Selection

Original studies were included if they quantified the association between maternal gestational triclosan, measured in urine samples, and reported associations with at least 1 of the 3 birth weight outcomes of interest (grams (g), grams plus adjustment for gestational age, gestational age standardized birth weight z-scores). References compiled from both search engines were screened for duplicates. Two of 3 reviewers (M.P., N.H., P.G.) independently screened the remaining articles at the title and abstract level to determine eligibility for full text screening. At least 2 reviewers independently screened the remaining articles through full text review. Studies were excluded if they met any of the following criteria. 1) they did not report original data; 2) the study was performed on animal subjects; 3) maternal triclosan exposure was not quantitatively measured in any biological sample; 4) infant birth weight was not quantitatively measured as outcome; 5) the study was not available in English.

2.5. Data Extraction

At least 1 of 3 authors (M.P., N.H., P.G.) independently extracted data related to study characteristics, exposure measures, outcome information, effect measures, and secondary analyses from all included articles. A second author reviewed all extracted information. All 3 reviewers assessed and discussed discrepancies to ensure accuracy amongst all 3 reviewers. We extracted relevant estimates of association reporting the relation between maternal triclosan exposure and infant birth weight from the main text or any supplemental material.

For the meta-analysis, we extracted the fully adjusted, linear effect estimate and 95% confidence interval. It should be noted that amongst those studies that were considered meta-analyzable, exposures were transformed and scaled differently across studies. In order to address these discrepancies, we re-scaled all exposure-outcome effect estimates to represent a change in outcome per 10-fold increase in triclosan, when possible. We were unable to include studies that did not report continuous associations between prenatal triclosan concentrations and infant birth weight. Additionally, we contacted 5 of 14 corresponding study authors to request additional results that were not present in their published articles, resulting in usable data from 4 of the 5 studies who’s authors we contacted.

2.6. Risk of Bias in Individual Studies

We evaluated risk of bias across multiple domains described in other applications of the Navigation Guide for each of the included studies (Higgins and Green 2011; Viswanathan et al. 2008). Briefly, we considered the following domains: recruitment strategy, blinding, confounding, exposure assessment, outcome assessment, incomplete outcome data, selective outcome reporting, and conflicts of interest.

Within each domain, authors assigned articles a rating of low, probably low, probably high, or high risk of bias based on specific instructions (Supplemental Table S2). For example, we identified a list of important confounders within our protocol based on subject matter expertise and use of a directed acyclic graph (DAG) (Supplemental Figure S1, Supplemental Table S3). We identified maternal age, pre-pregnancy body mass index (BMI), a measure of socioeconomic status, and maternal smoking as important confounders. We also considered adjustment for race/ethnicity in cohorts within the United States. We included child sex and gestational age as precision variables. Note, this was only considered in cases where z-scores were not used, as z-scores are standardized by gestational age and infant sex. Two of 3 possible review authors (M.P., N.H., P.G.) independently recorded risk-of-bias determinations for each included study. All reviewers collectively determined the final risk of bias rating for all studies.

2.7. Quality of Evidence Across Studies

Considering the full body of evidence across all included studies, we rated the quality of evidence as “low quality”, “moderate quality”, or “high quality” (Supplemental Table S4). Since we only considered human observational studies, the initial rating of quality of evidence was “moderate” as based on previously defined methodologies (Johnson et al. 2014; Koustas, E et al. 2014; Vesterinen et al. 2015; Woodruff and Sutton 2014). Next, based on previously established methods (Balshem et al. 2011), we considered several “downgrade” or “upgrade” factors in adjusting our overall rating (Table 1). Downgrade factors included risk of bias, indirectness, inconsistency, imprecision, and publication bias. Upgrade factors consisted of large magnitude of effect, dose response, and if confounding would minimize the overall effect. For each item downgrade or upgrade factors ratings ranged from −2 to 2 as integers with a rating of 0 indicating no change from the initial quality of rating. Review authors (M.P., N.H., P.G.) independently evaluated the quality of evidence across studies, and as a group determined the final decision.

Table 1.

Summary of rating quality and strength of the body of evidence for association between maternal urinary triclosan concentration and infant birth weight outcomes

Category	Summary of Criteria for Downgrades
Risk of Bias	Evidence streams were rated down if most of the relevant evidence came from studies that had high risk of bias. Rate down only if you judged that there was a substantial risk of bias in the body of available evidence.
Indirectness	Evidence streams were rated down if substantial differences existed between the study population, exposure, comparator, or outcome measured relevant to our study question. Potential sources of indirectness included a study population or intervention/exposure that was so different from that of interest that there was a compelling reason to think that the magnitude of effect would differ substantially, or studies that reported on surrogate end points instead of the outcome of interest.
Inconsistency	Evidence streams were rated down if studies conducted in similar human populations had widely different estimates of effect (unexplained heterogeneity or variability in results). To indicate potential inconsistency: a) point estimates varied widely across studies b) confidence intervals (CIs) showed minimal or no overlap for similar studies of comparable size c) the statistical test for heterogeneity had a low p-value (p < 0.05) d) the I2 was large (> 50%, based on the Cochrane’s guide to interpretation of I2). Studies that were inconsistent with respect to the magnitude of an effect (but not in terms of direction of effect estimates) would not be rated down.
Imprecision	Evidence streams were rated down if most studies had small sample sizes and few events, thus leading to wide confidence intervals.
Publication Bias	Evidence steams were rated down if we thought that studies were missing from the body of evidence that might result in an overestimate or underestimate of true exposure effects. We used considerations from GRADE guidance for evaluating publication bias, with modifications to reflect the Navigation Guide’s primary concern with underestimating the true effects of existing chemical exposure. Considerations for evaluating publication bias included the following: a) the body of evidence was dominated by early studies with negative results, particularly studies that were small in size b) studies were uniformly small (particularly when sponsored or funded by industry) c) empirical examination of patterns of results (e.g., funnel plots) suggested publication bias d) we were able to obtain results of unpublished studies that demonstrated results different from those of published studies e) a comprehensive search of the literature was not performed.
Category	Summary of Criteria for Upgrades
Large Magnitude of Effect	GRADE (Guyatt et al. 2011d) recommends rating the evidence stream up by one category (e.g., from “low” to “moderate”) if there were associations with a relative risk (RR) > 2, and up by two categories (for instance, from “low” to “high”) for those with RR > 5. However, there are limitations to using RR to determine magnitude of effect because RR relies on dichotomous exposure scales and outcomes. Although there is no established cutoff for the continuous scales, we evaluated the evidence judiciously to assess whether the magnitude of effect from the human evidence was compelling enough to justify upgrading the evidence.
Dose Response	The evidence stream was rated up if there were consistent dose–response gradients within one or more studies and/or evidence of dose response across the studies in the overall body of evidence.
Confounding Minimizes Effect	The evidence stream was rated up if consideration of plausible residual confounders or biases would only reduce the magnitude of the observed effect, or would suggest a spurious effect when results show no effect.

^aInstructions and definitions on how to apply these criteria were derived from (Lam et al. 2014).

2.8. Strength of Evidence Across Studies

To determine the rating of the overall strength of evidence, we considered 4 items. First, the previously determined rating for the overall quality of evidence. Second, we considered the direction of effect across studies. Third, our confidence in the effect based on the likelihood that a new study would provide new insights or change the conclusions. Fourth, we considered any other factors within the data that could influence certainty(NTP (National Toxicology Program) 2015). Based on previously established categories and rationale, the overall strength of evidence across the body of evidence could be rated as having “Sufficient”, “Limited”, or “Inadequate” evidence of toxicity or “Evidence of lack of toxicity” (Supplemental Table S4).(IARC (International Agency for Research on Cancer) 2006; Sawaya GF et al. 2007; U.S. EPA. 1996; U.S. EPA (U.S. Environmental Protection Agency) 1991). The authors (M.P., N.H., P.G., J.B.) independently evaluated the strength of the evidence across studies and compared judgments before determining the final rating as a group.

2.9. Statistical Analysis

The primary outcome was infant birth weight. Upon examining the evidence included in this systematic review, we observed variation in the ways that studies accounted for gestational age in their assessment of birth weight (i.e., infant birth weight (g), gestational age adjusted birth weight (g), and gestational age standardized birth weight z-scores). Because infant birth weight is strongly related to gestational age, we chose to conduct separate meta-analyses, one for each of the three different measures of infant birth weight identified among included studies. (Storms and Van Howe 2004).

There were some studies that only reported categorical outcomes, and thus were not included in meta-analyses. These studies were considered in final ratings for risk of bias, quality, and strength of evidence ratings. If an included article reported results for more than one of the pre-specified birth weight outcomes, we considered all reported results. We first considered results obtained from the full study sample in our primary meta-analyses. If results were only reported by child sex, values from each sex were included.

Next, we conducted two meta-regressions. First, we conducted the meta-analyses stratified by child sex. This was only done for the outcome gestational age adjusted birth weight (g), as this outcome had a larger pool of results to consider, and gestational age standardized z-scores already account for child sex. Second, we conducted a third meta-analysis where we separated studies based on the median triclosan concentrations reported in the sample. Studies with median triclosan concentrations <10 ng/mL were identified as “low” triclosan exposure, and those with median concentrations ≥ 10 ng/mL were considered to have “high” triclosan exposure.

We performed random effects meta-analyses using the inverse-variance method to estimate differences in each birth weight outcome per 10-fold increase in triclosan concentrations. We also performed meta-regression analyses using random effects meta-analysis, adjusting for median triclosan concentration (low: <10 ng/mL, high: ≥ 10 ng/mL) and child sex. To evaluate statistical heterogeneity across study estimates, we calculated Cochran’s Q statistic (p ≤ 0.05 for statistical significance) and I² values (Higgins, PT J and Green, S 2011; Johnson et al. 2014; Koustas, E et al. 2014; Lam et al. 2014). To evaluate publication bias, we considered funnel plots for each birth weight outcome. However, these results should be interpreted with caution given that <10 studies were included, and it is possible that test power is too low to characterize bias (Sterne et al. 2011). We also explored differences in reported associations based on the number and timing of triclosan exposures assessed. We did not perform meta-regression for this analysis due to the small sample size. We completed all statistical analyses in R Studio (version 4.0.3) and used R Studio packages metaphor and forestplot (R Core Team 2015).

3. Results:

3.1. Data Sources

Our search retrieved 37 unique records, 15 of which met inclusion criteria (Figure 1). Of 15 included studies, 8 were based in North America (Aker et al. 2019; Etzel et al. 2017; Ferguson et al. 2018; Geer et al. 2017; Goodrich et al. 2019; Lester et al. 2018; Messerlian et al. 2018; Wolff et al. 2008), 3 were from Europe (Lassen et al. 2016; Philippat et al. 2012, 2014), and the remaining 4 were from Asia (Ding et al. 2017; Huo et al. 2018; Ouyang et al. 2018; Wu et al. 2018). Included studies were published from 2008–2019 and involved 34 to 1,822 study participants (Table 2). All studies measured triclosan exposure in maternal urine, while 1 study also assessed exposure from maternal plasma biomarkers (Geer et al. 2017). We only considered results from triclosan measured via maternal urine biospecimens collected before delivery.

Figure 1.

Flowchart describing literature search and screening process for studies relevant to triclosan exposure and infant birth weight outcomes

Table 2.

Summary of study characteristics on studies examining associations between maternal urinary triclosan concentrations and infant birth weight

Source	Location	Study period	Study Design	Births (n)	Birth Weight Measures	Covariates Adjusted for	Exposure measures (n)	Measurement timing	Concentration (range)
Ding et al., 2017	China	2010 – 2013	Cross-sectional	496	Grams	Maternal age, pre-pregnancy BMI, pregnancy weight gain, parity, passive smoking, household monthly income, infant sex	1	Birth	Median (25th, 75th percentile) 0.5 ug/g creatinine (<0.1a, 2.4)
Huo et al., 2018	China	2012 – 2014	Cross-sectional	1006	Grams adjusted for gestational age	Maternal age, pre-pregnancy BMI, parity, maternal education, passive smoking, infant sex, gestational age, delivery mode,	1	Birth	Median (25th, 75th percentile) 0.6 ng/mLb (0.2, 2.6)
Ouyang et al., 2018	Shanghai, China	2012 – 2013	Cross-sectional	620	Grams adjusted for gestational age	Maternal age, pre-pregnancy BMI, parity, maternal height, maternal education, passive smoking, gestational age, gestational diabetes mellitus, creatinine	1	Birth	Median (min, max) ng/mL Low: 0.8 (<0.1a, 1.4) Medium: 2.7 (1.5, 5.0) High:13.3 (5.0, 95.2 )
Wu et al., 2018	Wuhan, China	2014 – 2015	Cohort	850	Z-score	Maternal age, pre-pregnancy BMI, pregnancy weight gain, parity, maternal height, maternal education, infant sex, paternal height	3	13, 23.6, and 35.9 weeks	Median (25th, 75th percentile) 0.7 ng/mLb (< 0.1a, 2.5)
Wolff et al., 2008	New York City, NY (USA)	1998 – 2002	Cohort	367	Grams adjusted for gestational age	Pre-pregnancy BMI, maternal education, maternal smoking, race, marital status, infant sex, gestational age, creatinine	1	25 – 40 weeks	Median (25th, 75th percentile) 11.0 ug/L (2.9, 42.0)
Etzel et al., 2017	Cincinnati, OH (USA)	2003 – 2006	Cohort	387	Grams, Grams adjusted for gestational age, Z-score	Maternal age, pre-pregnancy BMI, maternal education, maternal serum cotinine concentrations, income, maternal race, marital status, prenatal vitamin use, Beck Depression Inventory score	2	16, 26 weeks	Median (min, max) 16.0 ng/mL (< 2.3a, 1501.0)
Geer et al., 2017	Brooklyn, NY (USA)	2007 – 2009	Cohort	185	Grams	Infant sex	1	3rd trimester	Median (25th, 75th percentile) 9.1 ug/L (2.7, 42.9)
Goodrich et al., 2019	Michigan (USA)	2012 – 2015	Cohort	56	Grams adjusted for gestational age, Z-score	Infant sex, gestational age, specific gravity	1	8 –14 weeks	Median (25th, 75th percentile) 15.4 ug/L (5.6, 73.3)
Messerlian et al., 2018	Boston, MA (USA)	2012 – 2016	Cohort	213	Grams,	Maternal age, pre-pregnancy BMI, maternal education, maternal smoking, in-vitro fertilization (IVF) based vs. Non-IVF based treatment, season	3	6, 21, and 35 weeks	Median (25th, 75th percentile) 9.7 ng/mLb (3.6, 33.1)
Ferguson et al., 2018	Boston, MA (USA)	2006 – 2008	Cohort	476	Z-score	Maternal age, pre-pregnancy BMI, race/ethnicity, health insurance provider	3	18, 26 and 35 weeks	Median (25th, 75th percentile) 13.5 ug/Lb (4.4, 50.3)
Lester et al., 2018	Canada	2008 – 2011	Cohort	1822 (MIREC), 68 (P4)	Grams,	Maternal age, pre-pregnancy BMI, parity, maternal education, maternal smoking, income, place of birth, concentrations were standardized for time of day, time since last void, specific gravity	1(MIREC), 1 (P4)	6 – 13 weeks (MIREC), Before 20 weeks (P4)	Median (25th, 75th percentile) 14.4 ug/L (2.9, 121.2)
Aker et al., 2019	Puerto Rico	2011 – 2017	Cohort	867	Z-score	Maternal age, passive smoking, alcohol use, insurance type, specific gravity	3	16–20, 20–24, and 24–28 weeks	Median (25th, 75th percentile) 15.8 ug/Lb (3.6, 139.3)
Philippat et al., 2012	France	2002 – 2006	Cohort	191	Grams, Grams adjusted for gestational age	Pre-pregnancy weight and height, parity, maternal smoking, maternal education, gestational age, recruitment center, creatinine, concentrations were standardized for hour of sampling, time elapsed between sample collection and freezing, season and day of sampling, and gestational age at collection	1	24 – 30 weeks	Median: (5th, 95th percentile) 24.1 ug/L (1.6, 634.0)
Philippat et al., 2014	France	2003 – 2006	Cohort	520	Grams	Pre-pregnancy weight, parity, maternal height, maternal smoking, passive smoking, maternal education, recruitment center, paternal height, creatinine, concentrations were standardized for hour of sampling, time elapsed between sample collection and freezing, gestational age at collection,	1	22 and 29 weeks	Median (5th, 95th percentile) 29.0 ug/L (< 2.3a, 732.0) 30.0 ug/L (<2.3a, 755.0)
Lassen et al., 2016	Denmark	2010 – 2012	Cohort	514	Grams adjusted for gestational age	Pre-pregnancy BMI, parity, maternal smoking, gestational age	1	28 weeks	Median (5th, 95th percentile) Girls: 1.0 ng/mL (< 0.1a, 536.0) Boys: 1.0 (<0.1a, 335.0)

BMI: Body Mass Index, IVF: in-vitro fertilization, USA: United States of America, MIREC: Maternal-Infant Research on Environmental Chemicals Study, P4: Plastics and Personal-Care Products use in Pregnancy (P4)

^aLimit of Detection

^bAdjusted for specific-gravity

^cStudies that reported birth weight outcomes are z-scores referenced: Wu et al., 2018 and Aker et al., 2019 (Villar et al. 2014), Etzel et al., 2017 (Oken et al. 2003), Goodrich et al., 2019 (Fenton et al. 2013), Ferguson et al., 2018 (Cantonwine et al. 2016)

^dNote that the study design is reflective of the analysis conducted for each included article, which may differ from the larger cohort from which the data were derived.

3.2. Study Characteristics

A third (n=5) of included studies only reported outcomes for infant birth weight (g),(Ding et al. 2017; Geer et al. 2017; Lester et al. 2018; Messerlian et al. 2018; Philippat et al. 2014) and 20% (n=3) only reported gestational age standardized birth weight z-scores (Aker et al. 2019; Ferguson et al. 2018; Wu et al. 2018). The remaining studies reported 2 of the birth weight outcomes of interest (Huo et al. 2018; Lassen et al. 2016; Ouyang et al. 2018; Wolff et al. 2008)(Goodrich et al. 2019; Philippat et al. 2012), while only 1 reported results for all 3 (Etzel et al. 2017).

Half of the studies collected multiple urine samples throughout the gestational period (Aker et al. 2019; Etzel et al. 2017; Ferguson et al. 2018; Messerlian et al. 2018; Philippat et al. 2014; Wu et al. 2018), while the remaining 6 obtained 1 urine sample during gestation (Geer et al. 2017; Goodrich et al. 2019; Lassen et al. 2016; Lester et al. 2018; Philippat et al. 2012; Wolff et al. 2008). Among these, collections occurred during the first 20 weeks gestation in 2 studies (Goodrich et al. 2019; Lester et al. 2018), in the latter 20 weeks of gestation in 5 studies (Geer et al. 2017; Lassen et al. 2016; Philippat et al. 2012, 2014; Wolff et al. 2008), and throughout the gestational period in the other 5 studies (Aker et al. 2019; Etzel et al. 2017; Ferguson et al. 2018; Messerlian et al. 2018; Wu et al. 2018). There were 3 studies that collected maternal urine samples at the time of delivery (Ding et al. 2017; Huo et al. 2018; Ouyang et al. 2018).

3.3. Range of Triclosan Exposure

Median triclosan values varied by approximately 2 orders of magnitude (0.6 to 20 ng/mL), most markedly by geographic location (Figure 2). The lowest medians were in Asian cohorts, with higher values in North American cohorts.

Figure 2.

Median maternal urinary triclosan concentrations by geographic location in studies examining association between triclosan exposure and infant birth weight outcomes

3.4. Risk of Bias in Individual Studies

Included studies were consistently rated as “low” or “probably low” risk of bias across all domains (Figure 3, Supplemental Table S5a-S5n). Judgments of “probably high” risk of bias were given to 7 studies due to concerns for recruitment strategy, confounding, exposure assessment, and selective outcome reporting. Only 1 study received a judgment of “high” risk of bias due to concerns over residual confounding (Geer et al. 2017). Risk of bias ratings across studies did not differ based on year of publication, geographic location (Asia, North America, Europe), number of triclosan exposure measures ascertained (at birth, only 1, more than 1), types of outcomes reported (all birth weight outcomes, infant birth weight (g) and gestational age adjusted birth weight (g), or range of exposure (low: median level <10ng/mL, high ≥ 10ng/mL).

Figure 3.

Summary of risk of bias domains for individual studies examining associations between triclosan exposure and infant birth weight outcomes

3.5. Triclosan and Birth Weight Meta-Analyses

Of the 15 studies included, 14 presented results that could be included in the meta-analyses. One study only reported results based on categorical triclosan concentrations (Ding et al. 2017) which are summarized along with other categorical results reported from 5 included studies (Figure 4) (Etzel et al. 2017; Huo et al. 2018; Lassen et al. 2016; Ouyang et al. 2018; Philippat et al. 2012). Two studies reported outcomes for gestational age and child sex standardized birth weight z-scores (Goodrich et al. 2019; Philippat et al. 2012). Note that referenced data for birthweight z-scores were different amongst the studies, but specific to the geographic location where the study took place (Cantonwine et al. 2016; Fenton and Kim 2013; Oken et al. 2003; Villar et al. 2014).

Figure 4.

Associations Between Categories of Maternal Urinary Triclosan Concentrations and Infant Birth Weight

Ratio of change in infant birth weight relative to the lowest category of maternal triclosan concentrations (reference group). Lines represent the dose response for each independent study. Single points represent the effect for studies and are plotted at the midpoint of each category of triclosan concentration.

^a Birth weight measured in grams

^b Birth weight measured in grams, adjusted for gestational age

^c Gestational age and child sex standardized birth weight z-scores, transformed to grams

3.5.1. Gestational Age Standardized Birth Weight Z-Scores

Among 5 studies reporting data on gestational age standardized birth weight z-scores, each 10-fold increase in maternal urinary triclosan concentration was associated with a 0.04 (95% CI: −0.16, 0.07; 𝑝 = 0.23) decrease in gestational age standardized birth weight z-score (Figure 5a) (Etzel et al. 2017; Ferguson et al. 2018; Wu et al. 2018; Goodrich et al. 2019; Aker et al. 2019). Similar to studies of birth weight adjusted for gestational age, there was minimal evidence to suggest heterogeneity (I²=32.03% and Cochrane’s Q=5.61; p=0.23).

Figure 5.

Meta-analyses for associations of urinary triclosan concentrations during gestation and infant birth weight outcomes

*Boys only, **Girls only

All values represent change in birth weight per 10-fold increase in urinary triclosan concentrations. Error bars indicate 95% confidence intervals.

3.5.2. Birth Weight (g)

Of 6 studies that reported infant birth weight in grams (Etzel et al. 2017; Geer et al. 2017; Lester et al. 2018; Messerlian et al. 2018; Philippat et al. 2012, 2014), each 10-fold increase in maternal urinary triclosan concentration was associated with decreased birth weight (β=−20; 95% CI: −65, 26; 𝑝 = 0.40). The value of I² (76.09%) and Cochrane’s Q (Q=15.29; p=0.009) indicate 𝑝 heterogeneity amongst these studies (Figure 5b).

3.5.3. Gestational Age-Adjusted Birth Weight (g)

Among 7 studies examining associations between maternal urinary triclosan concentrations with gestational age adjusted birth weight measured in grams, each 10-fold increase in maternal urinary triclosan concentration was associated with more modest decreases in birth weight compared to studies not adjusting for gestational age (β=−12; 95% CI: −29, 4; 𝑝 = 0.15) (Etzel et al. 2017; Goodrich et al. 2019; Huo et al. 2018; Philippat et al. 2012; Wolff et al. 2008; Lassen et al. 2016; Ouyang et al. 2018). The I² value was small (0%) and Cochrane’s test for heterogeneity was not significant (Q=3.21; p=0.92) (Figure 5c).

3.5.4. Low vs. High Levels of Triclosan Exposure

To maximize the number of included studies, we conducted secondary analyses amongst those studies that reported results for the outcome of gestational age adjusted birth weight (g). Amongst studies with low levels of triclosan exposure (<10 ng/mL, n=5) (Huo et al. 2018; Lassen et al. 2016; Ouyang et al. 2018), there was no association between triclosan and infant birth weight (β=6; 95% CI: −20, 31; 𝑝 = 0.66), (Figure 6). When considering studies with high levels of triclosan exposure ( ≥ 10 ng/mL, n=4) (Etzel et al. 2017; Goodrich et al. 2019; Philippat et al. 2012; Wolff et al. 2008), we observed a modest, inverse association between gestational urinary triclosan concentrations and infant birth weight (β=−26; 95% CI: −59, 7; 𝑝 =0.13). Results from the meta-regression approached significance when comparing associations in low vs. high level exposure studies, (difference in associations β=14; 95% CI: −62, 91; 𝑝 = 0.08).

Figure 6.

Meta-analyses for associations of urinary triclosan concentrations during gestation and infant birth weight in grams adjusted for gestational age by level of exposure

*Boys only, **Girls only

All values represent change in birth weight per 10-fold increase in urinary triclosan concentrations. Error bars indicate 95% confidence intervals.

This suggests that a potential inverse association between gestational urinary triclosan concentrations and infant birth weight in more highly exposed populations, as effect size varies depending on the source population level of triclosan.

3.5.5. Modification by Child Sex

Next, we conducted meta-regressions to evaluate if the association between gestational triclosan exposure was modified by child sex (n=4 studies). Again, we only considered results reported for the outcome of gestational age adjusted birth weight (g). There was not strong evidence that the association between triclosan levels and gestational age adjusted birth weight (g) differed in males (n=4) (β=−7; 95% CI: −31, 17; = 0.60) compared to female infants (n=3) ( β=1; 95% CI: −11, 13; 𝑝 = 0.51), (difference in associations β=−7; 95% CI: −102, 88; 𝑝 = 0.25) (Supplemental Figure S2). However, given the small sample size of eligible included studies for this secondary analysis, these results should be interpreted with caution.

3.5.6. Timing and Number of Triclosan Exposure Assessments

The number and timing of maternal urine samples used to assess triclosan exposure varied across included studies. Studies with only one urine sample collected this sample either at the time of delivery (n=2), or during the gestational period (n=7). The remaining studies (n=5) collected ≥2 urine samples throughout pregnancy and before delivery, averaging repeated urine triclosan concentrations. While we were underpowered to conduct a meta-regression, we observed that studies with two or more prospectively collected triclosan measures reported inverse associations, while those studies with one sample collected prospectively or at the time of delivery reported null, or positive associations. It should be noted that among studies with one sample at the time of delivery, median triclosan exposure levels were among the lowest of all included studies.

3.6. Quality and Strength of Evidence Across Studies

We rated the quality of evidence to be “moderate”. This was based on 3 downgrades (−1) for indirectness, inconsistency, and publication bias, and 2 upgrades (+1) for dose response, and confounding minimizes the effect. We downgraded the evidence stream (−1) for indirectness because across studies, birth weight as the main outcome of interest was not reported consistently. We also downgraded (−1) the evidence stream for inconsistency because of geographic variation in effect estimates and triclosan concentrations. Further, results from Cochrane’s Q statistic of heterogeneity indicated significant heterogeneity (p<0.05) and the I² value was large (>50%) for the birth weight (g) outcome. However, it should be noted that we were limited in our ability to perform formal statistical tests to evaluate publication bias given the reduced sample size of studies based on the 3 birth weight outcomes considered when examining potential publication bias (Supplemental Figure S3).

We upgraded the overall body of evidence based on evidence for dose response for 2 reasons. First, among studies with low triclosan exposure (median value <10ng/mL), associations were generally null, compared to studies with higher triclosan exposure (median value ≥ 10ng/mL), which reported inverse associations. Second, within several studies, we found evidence for a dose-response relationship by observing monotonic associations across categories of triclosan concentrations (Figure 4). Etzel et al. (2017) reported a decrease in gestational age adjusted birth weight z-scores across quartiles of triclosan exposure. Similarly, Ding et al (2017), Huo et al (2018), and Philippat et al. (2012) also reported an inverse trend, such that birth weight (g) decreased with increasing categories of urinary triclosan concentration. However, this was not observed in all studies (Lassen et al. 2016; Ouyang et al. 2018). We judged these collective findings to be consistent of enough evidence to upgrade the overall quality of evidence for dose response (+1), although not strong enough to upgrade the overall quality of evidence by +2.

We also upgraded the evidence because there was no evidence that residual confounding would minimizes the effect of gestational triclosan exposure on birth weight. For example, in Etzel et al., the association between triclosan and birth weight was stronger after adjusting for potential confounders (i.e., negative confounding). The evidence stream was rated up (+1) when considering the attenuation of the magnitude of effect across birth weight outcomes. For example, we observed a larger effect estimate for decreased infant birth weight measured in grams compared to gestational age adjusted birth weight. It is noteworthy that the results from the meta-analysis for gestational age adjusted birth weight were attenuated, relative to infant birth weight measured in grams, suggesting that the association between gestational triclosan concentrations and infant birth weight are confounded by gestational age. Further, results from gestational age and child sex standardized birth weight z-score analysis also show a reduction in the magnitude of effect. These findings suggest that the association between gestational triclosan exposure and infant birth weight are potentially mediated by gestational age. Indeed, some studies observed inverse associations between triclosan levels and duration of gestation (Aker et al. 2019; Etzel et al. 2017).

We rated the overall strength of the evidence as “limited”. This was collectively based on our judgment of “moderate” quality of the body of evidence, as our confidence in the relation between gestational triclosan exposure and birth weight is constrained due to inconsistencies of findings across individual studies. However, the majority of studies were found to be of “low” or “probably low” risk of bias, increasing our confidence in the findings overall. The direction of the association suggests an inverse association between gestational urinary triclosan concentrations with infant birth weight in the meta-analyses conducted, although some studies report null associations. Moreover, when comparing the results of studies with low triclosan exposure, to those with higher exposure, we observed stronger inverse associations with birth weight among those studies with higher median levels of triclosan.

4. Discussion:

Overall, we identified limited evidence of toxicity with regard to triclosan exposure and infant birth weight after systematically reviewing and meta-analyzing fifteen studies examining the association of gestational triclosan exposure and infant birth weight. Across these studies, we observed “low” risk of bias across domains that may affect study results. However, we observed modest heterogeneity in the association between triclosan concentrations and birth weight outcomes, which are suggestive of stronger inverse associations among studies with higher median triclosan concentrations.

Two previous systematic reviews and meta-analyses came to contradictory conclusions regarding the association between gestational triclosan exposure and birth weight. (Zhong et al. 2020; Khoshhali et al. 2020) One concluded no association, while the other concluded that gestational triclosan exposure was associated with greater birth weight. However, neither study applied the Navigation Guide criteria to their evaluation.(Zhong et al. 2020; (Zhong et al. 2020; Khoshhali et al. 2020), which is the recommended methodology for evaluating evidence streams within environmental health (Johnson et al. 2014, 2016; Koustas, E et al. 2014; Lam et al. 2014, 2016; Vesterinen et al. 2015; Woodruff et al. 2011). Applying the Navigation Guide, we systematically evaluated the risk of bias in individual studies, as well as the quality of evidence across studies using previously established methods. This additional analysis of included studies was unique to our meta-analysis and could contribute to some of the differences in our conclusions relative to the prior meta-analyses. Additionally, consistent with the methodology outlined in the Navigation Guide, we also registered our systematic review protocol in PROSPERO (Patti et al. 2019). At the time when our proposal was registered with PROSPERO, no other systematic reviews or meta-analyses investigating the relation between triclosan and birth weight were registered.

There are a number of reasons that could contribute to the difference in our conclusions. These include differences in outcome definitions, attention to residual confounding, and recognition of differences in the range of triclosan exposure across studies. This is particularly relevant given that we report that the triclosan and birth weight association differed in studies with low (<10 ng/mL) vs. high (>10 ng/mL) urinary triclosan concentrations. Moreover, median urinary triclosan concentrations varied across studies by nearly 2-orders of magnitude (see Figure 2). Our findings suggest of a potential inverse association between gestational urinary triclosan concentrations and infant birth weight in more highly exposed populations, though the effect size varied depending on the source population level of triclosan. Future studies should be conducted in samples with higher levels of triclosan exposure, as the current state of the literature may not have enough of an exposure contrast to identify stronger associations, if they truly exist. In addition, the discrepancies in how birth weight was assessed within and across studies (i.e., grams, grams adjusting for gestational age, and z-scores) limited our ability to conduct meta-analyses with all included studies (n=15). Future studies should consider reporting birth weight using multiple definitions.

While previous meta-analysis using a similar groups of studies (Zhong et al. 2020) also found that there was substantial heterogeneity in terms of demographic characteristics of study subjects, timing of exposure measurement (i.e., trimester when urine samples were ascertained), and exposure detection methods in their included studies, our review is unique in that we explored heterogeneity in triclosan exposure range based on geographic location.

Another reason our conclusions may differ from those of prior meta-analyses is how we defined birth weight as an outcome. We considered birth weight measured in grams, gestational age adjusted birth weight, and gestational age standardized birth weight z-scores (also standardized by child sex). Both prior meta-analyses did not differentiate birth weight (g) from gestational age adjusted birth weight, nor did they consider gestational age standardized birth weight z-scores (Zhong et al. 2020; Khoshhali et al. 2020). This is an important distinction that may increase the validity of our outcome definition, as infant birth weight is highly dependent on gestational age (Storms and Van Howe 2004). Future studies should use gestational age standardized birth weight z-scores as not to erroneously adjust for a mediator and enhance comparisons across studies (Schisterman et al. 2009) given that triclosan exposure during pregnancy has been previously associated with reduced gestational age (Aker et al. 2019; Etzel et al. 2017).

Infants born as low birth weight are at an increased risk for a wide range of health outcomes throughout the life course including learning disabilities, as well as attention-deficit disorder or attention-deficit/hyperactivity disorder (Stein et al. 2006). Indeed, some of these same outcomes have been previously associated with prenatal urinary triclosan concentrations (Jackson-Browne et al. 2018, 2019a, 2019b). While the prevalence of low birth weight in the United States is stable (~8% from 2014–2019), low birth weight affects a sizable portion of infants and incurs substantial financial costs to the healthcare system. (Calkins and Devaskar 2011; Martin et al. 2019). Given that that low birth weight is preventable (Bailey and Byrom 2007), it is imperative that we identify modifiable risk factors, such as reductions in triclosan use, in order to have the largest impact at the population level. In the US (U.S. Food & Drug Administration 2017) and Europe (Andriukaitis, 2016), triclosan has been phased out of many consumer products due to the recognition of triclosan causing unacceptable risks to the environment as a known harmful endocrine disrupting chemical (Andriukaitis, 2016). However, triclosan is still used and sold in China (National Health Commission 2019) and Canada, although with restrictions (Health Canada 2016). Given that triclosan has been replaced with benzalkonium chloride, benzethonium chloride, or chloroxylenol in some products (Food and Drug Administration 2016a, 2016b, 2016c), it is critical to ensure that these substitutes do not confer any additional adverse health effects.

The cohorts included in our review represent geographically and socioeconomically diverse populations with a wide range of exposure to triclosan during pregnancy. It is possible, that as more studies becomes available, particularly those evaluating associations at higher levels of triclosan exposure, our confidence in the inverse association of birth weight with triclosan will change. While the magnitude of the associations we observed were modest, they have important population-health implications. Given that birth weight is a strong indicator for a number of health outcomes throughout the life course (Mathews and Driscoll 2017), including neurodevelopment, growth, and overall health (Stein et al. 2006), even minor decreases in individual birth weight can result in major consequences at the societal level, emphasizing that documented decreases in fetal growth are of major public health concern.

Our systematic review and meta-analysis has several strengths and limitations that should be considered. First, our systematic review and meta-analysis followed the Navigation Guide, which has been optimized for conducting systematic reviews in environmental health (Johnson et al. 2014; Koustas, E et al. 2014; Lam et al. 2014). This addition to our analysis was unique relative to other systematic reviews that also investigated triclosan and birth outcomes (Khoshhali et al. 2020; Zhong et al. 2020). A key strength of our review is that we included results from all outcomes reported from included studies. Thus, we were able to independently assess each outcome, demonstrating which measures may capture our outcome of interest best and whether gestational age acts as a mediator. This approach also differed from that of previous meta-analyses, and identified potential reasons for discrepancies across these prior reviews.

One limitation is that we were unable to account for all potential sources of heterogeneity. We speculate that the heterogeneity in results across studies could be attributed to residual confounding, exposure measurement error, or methods to adjust for urine dilution. While most studies were rated low and probably low risk of bias for confounding, there was a lack of consistency in regard to which covariates were included in analyses. Differences in exposure measurement, including the number and timing of urinary triclosan concentrations during pregnancy varied across studies. This is particularly important given that triclosan has a relatively short half-life,(Sandborgh-Englund et al. 2006) which results in moderate withinperson variability that would be expected to attenuate associations towards the null if exposure is non-differentially misclassified (Stacy et al. 2017). Thus, studies with only 1 measure of triclosan would be subject to greater exposure misclassification than studies with more than 1 measure of triclosan (Perrier et al. 2016; Whitehead et al. 2012). Moreover, samples collected at birth may not be representative of triclosan exposure during the gestational period, which was of primary interest for the purpose of this review. While the number of available studies limited our ability to conduct secondary analyses based on timing of triclosan exposure on infant birth weight, we qualitatively evaluated this (Figure 7) and observed some evidence to suggest that inverse associations were more apparent in studies with at least two prospective triclosan exposure measures. Furthermore, different methods of adjustment for urine dilution could contribute to exposure misclassification discrepancies across studies since some accounted for urine dilution using creatinine, specific gravity, or urine osmolality.

Figure 7.

Associations between number and timing of maternal urinary triclosan concentrations and infant birth weight

*Boys only, **Girls only

All values represent change in birth weight per 10-fold increase in urinary triclosan concentrations. Error bars indicate 95% confidence intervals.

5. Conclusions

The available epidemiological evidence of moderate quality and low risk of bias provides limited evidence that gestational triclosan exposure is associated with reductions in infant birth weight. Our results indicate that observed differences in this association across studies are related to the level of triclosan exposure in source populations. Future work should consider evaluating associations between triclosan and birth weight at higher levels of triclosan exposure and evaluate potential sources of heterogeneity such as gestational age, child sex, and geographic location.

Table 3.

Ratings and rationale of the quality and strength of the body of evidence for association between maternal urinary triclosan concentration and infant birth weight outcomes

Criteria for Downgrades
Category	Final Rating	Rationale
Risk of Bias	0	When considering the risk of bias across the studies, we concluded that there was not substantial bias across the body of evidence.
Indirectness	−1	Our outcome of interest is birth weight and the studies included assessed this in a number of different ways. Thus, we had to downgrade for indirectness given that we had to assess three different outcomes (grams, grams + gestational age, z-scores). Additionally, there were substantial differences regarding exposure levels across geographic levels.
Inconsistency	−1	When assessing the I2 values, we are able to determine that the studies that assessed grams as their outcome of interest had considerable heterogeneity, especially compared to the other outcomes (grams + gestational age and z-scores).
Imprecision	0	The majority of studies had moderate to large sample sizes and thus the reported results were fairly precise.
Publication Bias	−1	The body of evidence was not dominated by early studies with null results, as all studies were completed around the same time, with relatively comparable sample sizes. The studies were not uniformly small, nor were they sponsored by industry. A comprehensive search was completed of the literature. The funnel plots do suggest that there may be some cases of bias, such that at the base of the plot there is less symmetry than toward the tip of the plot, and for grams outcome some studies fall outside the funnel itself. Note that results from funnel plots should be evaluated with caution, as we did not have enough studies (>10) to include in analyses. We were also not able to obtain results from unpublished studies that demonstrated different results than the published studies that were included.
Criteria for Upgrades
Category	Final Rating	Rationale
Large Magnitude of Effect	0	The magnitudes of effect estimates from the included studies are modest.
Dose Response	+1	While there is not clear evidence for a dose response relation within studies, there is certainly a dose response effect across all studies (consider our analysis comparing the <10 triclosan to ≥ 10 triclosan exposure levels). Within studies that included categorical analyses, we observed evidence for a dose-response effect, such that birth weight decreased with increasing concentrations of urinary triclosan exposure.
Confounding Minimizes Effect	+1	In order to assess the impact of adjusting for confounding, we would need to compare the crude and adjusted results which only 6 of the 15 studies provided. Of these 6, we generally see that the effect of adjustment on the magnitude of effect is downward in 3 studies, and the magnitude of effect is attenuated in 1 study, but is strengthened in another. When we consider studies that adjusted for grams + GA (compared to just grams), we do see that the averaged summed magnitude of effect is attenuated, suggesting that a more optimal approach to addressing birth weight, but adjusting got GA, an important predictor, does in fact attenuate the magnitude of the results, this is further attenuated when considering the magnitude of effect from z-scores compared to grams+ gestational age adjustment.

Highlights:

• Systematically reviewed literature evaluating gestational triclosan and birth weight

• Evidence was low risk of bias and provided limited evidence for triclosan toxicity

• Gestational exposure to triclosan may reduce infant birth weight

• Triclosan exposure was higher in North America and Europe compared to Asia

• Inverse associations between triclosan and birth weight in higher exposed samples