The study population was comprised of offspring with congenital ear malformations and healthy infants from 1 January 2010 to 31 December 2015. The study population was recruited from the Maternal and Child Health Certificate Registry of Liaoning Province, which is managed by Liaoning Women and Children’s Health Hospital. Details on the registry have been previously published [20, 21]. In brief, the registry is an active hospital-based monitoring system that is maintained in accordance with the national monitoring program. All 31 provinces in China have launched the active monitoring system. Similar to other provinces, Liaoning Province submits data to a national database maintained by the China Birth Defects Monitoring Network [22]. All 14 major cities in Liaoning Province (Shenyang, Dalian, Anshan, Fushun, Benxi, Dandong, Jinzhou, Yingkou, Fuxin, Liaoyang, Panjin, Tieling, Chaoyang, and Huludao) were covered by the Maternal and Child Health Certificate Registry. Data on congenital malformations were provided by Maternal and Child Health Certificate Registries in 14 cities of Liaoning Province and collected from Liaoning Women and Children’s Health Hospital [22]. During the study period from 2010 to 2015, approximately 6,000 cases of congenital malformations were reported each year from all maternity units in the province [23].

Cases with congenital ear malformations (n = 1676) were registered in the Maternal and Child Health Certificate Registry of Liaoning Province and classified as live births, stillbirths, or pregnancy terminations after prenatal diagnosis of congenital anomalies that were born or terminated from 2010 to 2015. According to the International Classification of Diseases, 10th Revision, Clinical Modifications [ICD-10-CM], congenital ear malformations include two major malformations: microtia (including anotia; n = 361; ICD10:Q17.2, Q16.0) and other external ear malformations (except for microtia and anotia; n = 1315; ICD10:Q17).

Controls were selected as previously described [20]. In brief, we divided Liaoning Province into southeastern, western, and central regions according to geographical features. Thereafter, five cities from the three regions (southeastern region: Dalian; western region: Fuxin, Chaoyang, and Huludao; central region: Shenyang) were representatively selected based on the degree of air pollution and socio-economic characteristics. The control group was a random sampling of 1.5% of live births without congenital ear malformations from the five cities in the three regions by random birth-year sampling, which was not case-related (mismatched). Subjects with missing or implausible covariate information were excluded and not included in the final analysis. We conducted this study according to national and local regulations. The Institutional Review Board of Liaoning Women and Children’s Health Hospital reviewed and approved the study protocol, and the study was conducted in compliance with local and national regulations.

The data collection process was carried out as previously described [21]. In brief, each newborn or terminated fetus was inspected by an experienced obstetric or pediatric specialist after birth. Cases were registered and coded based on the International Classification of Diseases, 10th Edition (ICD10). For cases of suspected congenital ear malformations that were diagnosed by prenatal ultrasonography, we examined newborns or fetuses again after postpartum check or pregnancy termination. Cases were diagnosed in all monitored hospitals from the 14 cities in Liaoning Province, and experienced obstetric or pediatric specialists immediately interviewed the newborn’s mother in order to complete the Birth Defects Registration Form that was used to gather data, including information on maternal and child demographics, diagnosis of congenital defects, history of early pregnancy, and family history. Forms were submitted to the local Maternal and Child Healthcare Institution in the 14 cities, which submitted forms to the Liaoning Women and Children’s Health Hospital. Data were retrospectively verified by a team of experienced genetics and pediatrics specialists [24].

The data quality control was carried out as previously described [24]. Based on the Chinese Maternal and Child Health Surveillance Workbook, specialists from monitoring institutions at all levels diagnosed the disease, and collected and checked the data, as well as the medical records to ensure the high quality of data. To distinguish between inadequacies and inaccuracies, the experts also carried out an independent retrospective investigation [24].

The environmental air pollutant monitoring station network in Liaoning Province is comprised of 77 monitoring stations in 14 cities (two of which served as controls), and this network assessed air pollutant exposure (except for the two controls) (Figure 1). The 77 air pollutant monitoring stations in Liaoning Province were mainly located in urban areas, where they covered residential areas and represented air pollutant levels in the region. According to the Chinese government’s ambient air quality standards from the China National Environmental Monitoring Centre (CNEMC, http://www.cnemc.cn), all monitoring stations in the 14 cities detected and reported air pollutant concentrations and submitted these data to the Environment Protection Bureau. The exposure levels were assessed using the mean SO₂ concentration at each monitoring station in the mother’s city of residence, and then these values were used to calculate the monthly mean SO₂ concentration (Supplementary Figure S1). Monthly mean SO₂ concentration exposure levels were assigned to each newborn after averaging concentrations from all air pollutant monitoring stations recorded by the Birth Defects Registration Form during the 3 months before conception and the 3 months after conception. The gestational age of cases and controls was estimated according to the due date provided by the obstetrician. We used the date of the last menstrual period to determine the first month of pregnancy. If the date of the last menstrual period occurred in the first half of the month, then this month was considered as the first month of conception. Otherwise, it was considered as the first month before conception.

According to the change in the estimate criterion (10%) [25], we selectively adjusted the corresponding covariates in the model. In the final model, the covariates were maternal age (two categories: <30, ≥30 years), season of conception (four categories: spring [March–May], summer [June–August], fall [September–November], and winter [December–February]), gravidity (two categories: 1, ≥2), parity (two categories: 1, ≥2), maternal education (four categories: elementary school or less, middle school, high school, and college or above), and maternal NO₂ and PM₁₀ exposure levels.

Congenital ear malformations were treated as a dichotomous category and analyzed as a dependent variable. From the third month before conception to the third month after conception, the SO₂ exposure concentration was computed as a continuous independent variable and a categorical independent variable based on quartiles of distribution in controls. Characteristics of the categorical variables in cases and controls were presented as frequencies and percentages, and they were compared using the Chi-square test. Means, standard deviations (SD), medians, and percentiles were used to characterize the monthly mean ambient SO₂ concentration in the 14 cities in Liaoning Province. Spearman’s rank correlation was applied to analyze the correlation between three pollutants (SO₂, NO₂, and PM₁₀). Odds ratios (ORs) and 95% confidence intervals (CIs) were calculated by multivariable logistic regression analysis. We also examined the associations between maternal SO₂ exposure, congenital ear malformations, and subgroups in offspring. Specifically, SO₂ exposure concentrations were divided into quartiles (determined from controls), and the other quartiles (second, third, and fourth) were compared to a reference (the first quartile) to calculate ORs. We conducted linear tests using the median values of SO₂ exposure in each category as a continuous variable and congenital ear malformations as the response variable. Additionally, ORs in both per 1-standard deviation (SD) and per 10 μm/m³ increment were also reported. In our analysis, we respectively fitted single-pollutant (SO₂ and the aforementioned covariates) and multi-pollutant models (SO₂+NO₂; SO₂+PM₁₀; SO₂+NO₂+PM₁₀) to assess the effects of different pollutants. Furthermore, subgroup analyses were used to investigate the associations between SO₂ exposure, congenital ear malformations, and maternal age (<30, ≥30 years) using stratification. We added a multiplicative interaction in the regression model between exposure and age to measure the potential interactions.

We performed a sensitivity analysis using propensity score matching (PSM) analysis, which can reduce baseline group differences. Cases with congenital ear malformations (n = 1676) and controls (n = 7950) were matched using 1:1 nearest-neighbor matching with a caliper width equal to 0.2 of SD of the logit of propensity scores. And propensity scores were estimated by multivariable logistic regression model, with adjusting for maternal age, season of conception, gravidity, parity, and maternal education. If a case subject could not be matched to any control subject, then the case subject was excluded. Absolute standardized differences (ASD) were used to confirm the balance of covariates between cases and matched control individuals [26]. An ASD of less than 10% was considered a negligible difference between cases and controls.

For associations in the multivariable logistic regression analysis, we further used a restricted cubic splines (RCS) model to assess their shapes of dose-response association [27]. Nonlinear associations were modeled by penalized cubic splines with 3 equally spaced knots. For the first and the second months after conception, the minimum value was defined as the referent value by default in the SAS macro “%RCS_Reg” [27]. The 5th percentile value was defined as the reference in the other six exposure windows.

All statistical analyses were conducted using SAS software version 9.4 (SAS Institute Inc., Cary, North Carolina, United States). Statistical significance was taken at p-values < 0.05 based on two-sided tests.

One of the strengths of this study is that it provides original findings on the effects of exposure to ambient SO₂ during the 3 months before conception and the 3 months after conception and that the SO₂ concentration is associated with congenital ear malformations risk. The large sample was selected from the Maternal and Child Health Certificate Registry of Liaoning Province from 2010 to 2015, and it included 1676 cases of congenital ear malformations and 7950 controls, which allowed us to investigate the topic. Additionally, maternal SO₂ exposure (data from the Environment Protection Bureau of 14 cities in Liaoning Province) was assessed in detail. The exposure window included the 3 months before conception and the 3 months after conception to ensure that the strength of the association was fully assessed. Finally, we performed sensitivity and subgroup analyses stratified by congenital ear malformations subtypes and maternal age (<30, ≥30 years), and the results showed stability of associations. The RCS combines quantitative data with the strength of association for the occurrence of outcome, enabling a continuous presentation of the association strength dose-response relationship.

However, there are some limitations, and we suggest interpreting our results with caution. Firstly, the assessment of individual SO₂ exposure concentrations may have led to exposure misclassification. Because it is difficult to accurately collect individual exposure levels, we used fixed-site monitoring to represent the exposure to pollutants, which inevitably led to bias in exposure assessments. Therefore, to reduce the misclassification of exposure in future studies, it may be necessary to use more sophisticated models to evaluate SO₂ exposure such as land-use regression models or spatial interpolation models [38]. Secondly, exposure assessment was based on fixed residential addresses, but the residential address of mothers may have changed during conception. For example, the mobility rate was 9% in the northern UK study [39], but it was 3.1% in the Lanzhou study in China [9]. Nevertheless, Lupo et al. [40] have reported that changes in residential addresses during pregnancy had little effect on these assessments. Thirdly, outdoor air pollutant exposure was only captured, but indoor microenvironment air pollutant exposure, such as homes and workplaces, was not, which may have also led to the misclassification of exposure, and the overestimation or underestimation of effects. Fourthly, although the case information of congenital malformations was obtained through the monitoring network and had various quality controls, it was difficult to guarantee absolute consistency in the process of determining and collecting cases due to the differences in diagnosis among different hospitals, and the longest diagnosis time was 7 days after birth, which may have led to the loss of cases. Finally, other unmeasured covariates, such as maternal disease, maternal or paternal smoking, family history of genetic disease, and toxicity exposure during pregnancy, which have been reported in previous studies as possible risk factors for congenital ear malformations [5, 41], may have influenced our results.

The present study suggests that maternal SO₂ exposure is associated with increased risks of congenital ear malformations in offspring. Future studies are still needed to confirm or refute these associations and provide scientific evidence for possible public health intervention during the crucial period of ear development.