Exposure-response relationships for personal exposures to fine particulate matter (PM2.5), carbon monoxide, and black carbon and birthweight: Results from the multi-country Household Air Pollution Intervention Network (HAPIN) trial

Abstract Background Household air pollution (HAP) from solid fuel use is associated with adverse birth outcomes, but data on exposure-response relationships are limited. We examined associations between HAP exposures and birthweight in rural Guatemala, India, Peru, and Rwanda during the Household Air Pollution Intervention Network (HAPIN) trial. Methods We recruited 3200 pregnant women between 9 and <20 weeks of gestation. Women randomized to the intervention arm received a liquified petroleum gas (LPG) stove and fuel during pregnancy, while control arm women continued using biomass. We measured 24-hr personal exposures to particulate matter (PM2.5), carbon monoxide (CO), and black carbon (BC) once pre-intervention (baseline), twice post-intervention, and birthweight within 24 hours of birth. We examined the relationship between the average prenatal exposure and birthweight/weight-for-gestational age z-scores using multivariate-regression models. Findings Results showed an inter-quartile increase in average prenatal exposure to PM2.5 (74.5 g/m3) and BC (7.3 g/m3) was associated with a 14.8 (95% confidence interval [CI]: -28.7g, -0.8g) and 21.9g (95% CI: -37.3g, -6.1g) reduction in birthweight and reduced weight-for-gestational age z-scores of -0.03 (95% CI: -0.06, 0.00) and -0.05 (95%CI: -0.08, -0.01) standard deviations, respectively. We found no associations for birthweight or weight-for-gestational age z-scores with CO exposures. Interpretation Results provide support for continuing efforts to reduce HAP exposure alongside other drivers of low birthweight in low- and middle-income countries.


Introduction
Identifier NCT02944682). The specific study areas in each country (Jalapa Municipality,148 Guatemala; Villupuram and Nagapatinam districts of Tamil Nadu, India; Department of 149 Puno, Peru; and Eastern Province, Rwanda) were selected based on high prevalence of 150 cooking with biomass, low background ambient PM2·5 concentrations, and acceptable field 151 feasibility as assessed during an 18-month period of planning and formative research. 20,21 152 Between March 2018 and February 2020, we recruited a total of 3200 (800 per country) 153 non-smoking, pregnant women who were between 18 and ≤35 years of age, between 9 and 154 ≤ 20 weeks of gestation (determined via ultrasound), and who used biomass as a primary 155 fuel. In accordance with the trial protocol, half of the participants in each country were 156 randomized to an intervention arm that received a liquefied petroleum gas (LPG) stove and 157 a continuous supply of LPG fuel following enrollment and throughout their pregnancy, while 158 the balance served as controls and continued to rely chiefly on solid biomass for cooking. 159 160 Personal exposure monitoring during pregnancy 161 Prenatal personal exposure monitoring protocols and results have been described 162 previously. 19,22 Briefly, at each study site, pregnant women participated in three 24-hr 163 personal exposure assessments, once at baseline (between 9 and <20 weeks of gestation) 164 and twice after randomization into the control or intervention arms (between 24-32 weeks 165 and 32-36 weeks of gestation, respectively). During each session, women wore customized 166 vests or aprons fitted so that instrumentation was situated close to their breathing zone. 167 PM2·5 monitoring was performed using the Enhanced Children's MicroPEM™ (ECM) (RTI 168 International), which collects (a) gravimetric samples on pre-weighed 15mm Teflon filters 169 (MTL,USA) utilizing a 2·5 micron impactor at a flow rate of 0·3 liters per minute and (b) real-170 time nephelometric data. 23 BC was estimated post-sampling on the ECM filters using the 171 SootScan® Model OT-21 Optical Transmissometer (Magee Scientific, USA). CO monitoring 172 was performed using the Lascar EL-CO-USB-300 DataLogger (Lascar Electronics, USA). 173 Participants were instructed to always wear the vest or apron during the 24-hr 174 measurement period, except when sleeping, bathing, or when conducting other activities 175 during which the equipment could not be safely worn. During these times, they were 176 instructed to keep the vest or apron nearby. Additionally, data were collected on 177 sociodemographic and household characteristics and activity patterns that may influence 178 exposure. 179 180 Procedures for assuring data quality, weighing filters, and estimating missing gravimetric 181 data based on nephelometry have been described previously. 22 Briefly, gravimetric data 182 quality assurance involved a combination of threshold values for flow rates, inlet pressure, 183 and sampling duration, as well as visual inspection of damaged filters by weighing room 184 technicians. In cases where nephelometric but not gravimetric data were available, PM2·5 185 exposure was estimated based on nephelometric data, using an instrument-specific 186 regression coefficient for the association between nephelometric and gravimetric data for 187 that specific ECM instrument as described previously. 22  is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint  baseline measurement, while the post-intervention exposure was estimated using one or  195  both personal measurements performed after intervention. This allowed for exposure  196  changes resulting from the introduction of the intervention to be weighted according to the  197  length of time participants had the intervention during gestation. An unweighted average  198  of the baseline and other available (1-2) gestational period measurements was used for  199 controls, as they continued using biomass as the primary cooking fuel throughout gestation.

201
Birthweight Outcome Measurements 202 Following a standard protocol, birthweight was measured within 24 hours of birth by a 203 trained field worker or nurse using a Seca 334 mobile digital baby scale. 204 Newborns were weighed naked to the nearest 10 g and duplicate measurements were 205 recorded on tablet-based REDCap forms. If the first two measured birthweights differed by 206 more than 10 g, a third measurement was taken. The average of the measurements was 207 used in the data analysis. Infants were typically assessed at health facilities where they were 208 born. Each scale was calibrated weekly in the field offices before deployment using standard 209 5-lb and 10-lb weights; scales not within ±2·5% of the standard weight were re-calibrated. In 210 cases in which we were unable to reach the child during the prescribed 24-hr window-due 211 mainly to COVID restrictions or critically ill infants admitted to ICUs or referral hospitals-we 212 used measurements provided by the facility, if available, but conducted sensitivity analyses 213 to compare results. 214 215 As gestational age is a potential mediator in the causal pathway between HAP exposure and 216 birthweight, we did not adjust for it in the E-R models; had we done so, its inclusion would 217 not allow estimation of the total effect of exposure. 25 However, we additionally estimated z-218 scores for weight adjusted by gestational age defined using INTERGROWTH tables 219 (intergrowth21.tghn.org) as a secondary analysis. These weight-for-gestational age z-scores 220 were derived by subtracting off the standard INTERGROWTH sex-specific weight for a given 221 gestational age and dividing by the INTERGROWTH standard deviation of that weight. 222 Measurements were considered invalid if the gestational age at birth was greater than 300 223 days or if the birth weight-for-gestational age z-score did not fall between -6 and 5. 224 225 Exposure-Response Modeling 226 The statistical analysis plan was agreed upon in advance and published with the trial 227 registration prior to unblinding. Analyses were independently replicated by a second 228 member of the study team. E-R analyses were modeled separately for each pollutant (PM2·5, 229 BC, and CO) and birthweight/birthweight-for-gestational age z-score. 230 231 Covariate selection for models was guided by a directed acyclic graph (DAG) ( Figure S3). A 232 minimal set of potential confounders or strong risk factors (e.g., infant sex) were identified 233 in systematic reviews of birthweight, 3,6 and from previous studies of HAP and birthweight. 10-234 12,24,26 We used 10% change-in-estimate (CIE) methods as outlined in Greenland (1989)  is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted August 8, 2022. ; https://doi.org/10.1101/2022.08.06.22278373 doi: medRxiv preprint strata (one in Rwanda, one in Guatemala, two in India, six in Peru). Eighteen and twenty-six 242 subjects were missing BMI and hemoglobin, respectively, and we created a category of 243 missing for these variables so that they were not excluded from the analysis. 244 245 For both birthweight and z-scores, we first fitted linear models with different exposure 246 metrics (i.e., linear, log linear). We then evaluated nonlinear categorical (quartile modes), as 247 well as quadratic, 2-piece linear and restricted cubic spline model with three knots 28 248 models, and assessed model fit using Akaike's Information Criterion (AIC). The knots for the 249 2-piece spline were chosen based on AIC (using quartile cutpoints initially and then 250 narrowing down), while knots for restricted cubic splines were placed at the 5 th , 50 th , and 251 75 th percentiles of exposure. We also used thin plate smoothing splines via generalized 252 additive models, with penalization determined by generalized cross-validation score, using R 253 package mgcv. We also examined effect modification by country, as well as by infant sex, via 254 interaction terms between our exposure metrics and these variables.
Ethics/registration/funding 257 The study protocol was reviewed and approved by institutional review boards (IRBs) or 258 Ethics Committees at Emory University (00089799) The funding sources were not involved in study design, collection, analysis, and 268 interpretation of data, or decisions to submit the paper for publication.

271
Participant Characteristics 272 While 3200 women were enrolled in the study, 5 enrollees were determined to be ineligible 273 after randomization and exited the study. After accounting for miscarriages, stillbirth, and 274 withdrawals, the 3195 pregnancies yielded 3060 live births. Of these, 3018 had valid 275 birthweights (others had birthweights measured outside the 24-hr window or the study 276 team was unable to obtain any birthweight measurement, see CONSORT diagram, Figure  277 S1). Sixteen additional births were excluded on account of a gestational age >300 days; 278 weight-for-gestational age z-scores are unavailable in the INTERGROWTH database beyond 279 300 days of gestation. 3002 subjects were thus eligible for inclusion in E-R analyses. These 280 were further restricted by the availability of exposure data for each of the three pollutants 281 of interest (Table 1). 282 283 Participant characteristics are summarized in is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint were 92·2 (83·9) µg/m 3 for PM2·5, 10·0 (7·4) µg/m 3 for BC, and 2.0 (2·9) ppm for CO (Table  294 S6). PM2·5 and BC exposures were highly correlated (Spearman's = 0·79), but correlations 295 between exposure to PM2·5 and CO (Spearman's = 0·34) as well as BC and CO (Spearman's 296 = 0·39) were relatively weak. The intervention resulted in marked reduction in exposure. 297 Post-intervention mean personal PM2·5 was 24.0 μg/m 3 in the intervention arm and 298 70.7μg/m 3 in the control arm. Similar reductions of exposure were seen for BC and CO. 299 300 Exposure distributions are depicted in Figure 1 and described in Supplemental The mean (SD) birthweight of live born infants was 2909 (471) g with mean gestational age 307 at delivery of 39·3 (1·5) weeks; 5·3% of births were classified as preterm (163/3002) and 308 17·7% as LBW (Figure 1). Mean (SD) birthweight was 2921 g (474·3 g) in the intervention 309 arm and 2898 g (467·9 g) in the control arm, a difference of 19·6 g (95% CI: -10·1 g, 49·2 g). 310 311 Exposure-response analyses 312 Results for linear and log-linear models for birthweight and for weight-for-gestational age z-313 score are shown in Table 3 and Table 4, respectively, for each of the three measured 314 pollutants. Quartile models are presented in Tables S1 and S2. In linear models, an inter-315 quartile increase in gestational exposure for PM2·5 (74·5 µg/m 3 ) and BC (7·3 µg/m 3 ) was 316 associated with a change in birthweight of -14.8 g (95% CI: -28·7 g, -0·8 g] and -21·9 g (95% 317 CI: -37·7 g, -6·1 g], respectively (Table 3). For weight-for-gestational age z-scores, the same 318 exposure increases were associated with a decrease of 0·03 (95% CI: -0·06, 0·00) and 0·05 319 (95%CI: -0·08, -0·01) standard deviations, respectively (Table 4). No associations were 320 apparent between CO exposures and birthweight in the linear models or between any of the 321 measured pollutants and LBW prevalence. Quartile analyses (Tables S1 and S2) showed that 322 the decrease in birthweight and z-scores were not monotonic for PM2.5, while decreases 323 were monotonic for z-scores but not birthweight for BC. 324 325 Evaluation of different models indicated that the linear fit presented above was appropriate 326 to model the relationships between the birthweight outcomes and BC. For PM2.5, however, 327 a quadratic (non-linear) fit was better suited to the birthweight outcome (Table S3 and 328 Figure S2), with a positive linear coefficient (0·2325) and a negative quadratic coefficient (-329 0·009), indicating an initial increase in birthweight with higher PM2·5 followed by a 330 subsequent decrease at the higher exposures. Both categorical and cubic spline models 331 supported this relationship ( Figure S2). Linear models fit best for BC for both birthweight 332 and z-scores, as well as PM2.5 and z-scores. Smoothed E-R curves for PM2·5/BC and 333 birthweight and weight-for-gestational age z-scores can be seen in Figure 2 and is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted August 8, 2022. ; https://doi.org/10.1101/2022.08.06.22278373 doi: medRxiv preprint

335
Trends for full term births (95% of births) were similar to trends for all births (Table S4). No 336 statistically significant interactions (at the 0·05 level) were observed with infant sex, but 337 female births showed a larger effect than male births for birthweight, and for z-scores 338 (Table S5). Trends were reasonably consistent across countries for the association between 339 PM2.5 and BC with both birthweight and z-scores (Tables S7 and S8). We also ran separate 340 models for our three exposure measurements during gestation, i.e. for baseline, mid-point, 341 and end of gestation measurements (these corresponding roughly to early 2 nd trimester, 342 end of 2 nd trimester, and end of 3 rd trimester). These models, for both birthweight and z-343 score, showed no pattern whereby early or later exposures had stronger effects on the 344 outcome ( is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted August 8, 2022. ; https://doi.org/10.1101/2022.08.06.22278373 doi: medRxiv preprint

Data Sharing 429
Will individual participant data be available (including data dictionaries)?
Yes What data in particular will be shared?
Individual participant data that underlie the results reported in this article, after de-identification (all results summarized in text, tables, figures, and appendices) What other documents will be available?
Study protocol, statistical analysis plan, informed consent form, analytic code When will data be available (start and end dates)?
Beginning 6 months following article publication With whom?
Anyone who wishes to access the data For what types of analyses?
Any purpose By what mechanism will data be made available? is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted August 8, 2022. ; https://doi.org/10.1101/2022.08.06.22278373 doi: medRxiv preprint Log linear 0·02 (-0·02, 0·06) 0·24 7214 1 All models adjusted for mother's education, baseline BMI, nulliparity, diet diversity, food insecurity 19 score, second-hand smoke, baseline hemoglobin, age, infant sex and 10 randomization strata. IQRs 20 for PM2·5, BC, and CO were 74·51, 7·30, and 1·68 respectively. On the log scale, IQRs for PM2.5, BC, 21 and CO were 1·04, 0·85, and 1·40 respectively.

22
. CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted August 8, 2022. ;  Table  6. Results are presented separately for each study site and in combination for the entire trial. Dots are individual datapoints. X-axes are log transformed. Thick solid lines inside the box are medians. The lower and upper hinges (i.e., the ends of the box) correspond to the 25 th and 75 th percentiles. The whiskers (i.e., the lines beyond the box) extend from the hinge to 1·5 * IQR. The panel-wide dotted vertical lines are study-wide medians. In panel (A), the shaded area indicates low birthweight (< 2500 g). In Panel (B), the dashed line is the WHO Interim Target 1 annual guideline value of 35 µg/m3.