We performed a population-based cohort study in nine primary healthcare centers (Cempaka Putih, Johar Baru, Kemayoran, Kramat, Jatinegara, Kampung Melayu, Matraman, Paseban, Rawa Bunga) in Jakarta, Indonesia, from March 2016 until September 2020 (Soesanti et al. 2023). Jakarta, the capital city of Indonesia, stands out as a prime example of significant exposure to air pollution, with around 70–80% of the overall air pollution in Jakarta to be attributed to traffic sources. This is driven by a large number of vehicles, including 16.1 million motorcycles, 4.3 million cars, and public transportation systems that contribute substantially to pollution levels in the broader Jakarta region (Soesanti et al. 2023; Central Bureau of Statistics 2003).

We obtained ethical approval from the Institutional Review Board of the Faculty of Medicine University of Indonesia/Cipto Mangunkusumo General Hospital, Jakarta, Indonesia (reference number: 895/UN2.F1/ETIK/2015). Written informed consent was obtained from all the participants before their enrolment (Soesanti et al. 2023).

This study began by enrolling 413 pregnant women who lived within the primary care center catchment area and were easily reachable by phone (Soesanti et al. 2023). Recruitment was performed by midwives at the early stages of pregnancy (in the first or early second trimester with gestational age < 20 weeks). Over the years, the number of recruited pregnant women varied, with 69 participants in 2016, 132 in 2017, 122 in 2019, and 90 in 2019 (Soesanti et al. 2023). Forty-nine participants dropped out for various reasons (refused to continue or moved out of town), 13 were excluded because of early miscarriages, and one was excluded because of an ectopic pregnancy, resulting in a final sample size of 348 mother-singleton infant pairs. Further exclusions were made due to missing data on air pollution measurements (4 pairs), incomplete data on outcome measurement (28 pairs), and extremely low birth weight (1 pair, BW 700 g, GA 27 weeks). As a result, 315 mother-infant pairs were included in this current study. The pregnant women were followed until delivery, and their infants were followed up until six months of age.

A full description of the demographic, socio-economic, and pregnancy-related information has been described in detail previously published paper (Soesanti et al. 2023). We used structured questionnaires during enrollment to collect data on various maternal characteristics, including maternal age, parity, abortion, working status, household income, level of education, prior drug use, and smoking (active and passive) history prior to pregnancy (Soesanti et al. 2023). Paternal characteristics, including age, level of education, and smoking habits prior to their spouse’s pregnancy were also obtained at enrolment (Soesanti et al. 2023). Level of education was categorized as elementary, high school, and under-post graduate, while family income was categorized as below or above the minimum monthly wedges per capita in Jakarta (≥ 290 USD) (Soesanti et al. 2023).

We used a structured questionnaire administered at each trimester to assess pregnancy-related factors, including pregnancy complications, medications, and smoking habits (Soesanti et al. 2023). Weight gain during pregnancy, denoted as delta BMI was derived by subtracting the pre-pregnancy BMI from the BMI recorded at delivery. Most pregnant women delivered at the primary care center. Referral to a secondary health care center for complicated pregnancy or delivery was following the national health care policy (Soesanti et al. 2023). At delivery, maternal blood pressure, body temperature, gestational age, and complications during delivery were documented (Soesanti et al. 2023).

We used repeated interviewer-led questionnaires to obtain evidence of infection at each follow-up, performed at 1 month (infection between 0 and 1 month of age), 2 months (between 1 and 2 months of age), 4 months (between 2 and 4 months of age) and 6 months (between 4 and 6 months of age) in accordance with the national program on vaccination. We obtained information on infection as categorical “yes” and “no”. We asked questions about history of fever (defined as axilla temperature > 38 °C), symptoms of cough, runny nose, wheezing, shortness of breath, diarrhea, and vomiting during the time interval. The symptoms were categorized into respiratory tract infection (upper and lower), gastrointestinal infection, and other infections.

Lower respiratory tract infection (LRTI) was defined as pneumonia or bronchiolitis diagnosed by a physician. To obtain this information, we asked mothers two specific questions: (1) Has your infant been diagnosed with pneumonia by a physician? (2) Has your infant been diagnosed with bronchiolitis by a physician? Additionally, we cross-checked this data with hospital medical records. We also documented the medication received and the duration of hospitalization associated with pneumonia or bronchiolitis. For upper respiratory tract infection (URTI) symptoms, we asked about fever, cough, runny nose, and stuffed nose by asking mothers the following questions: (1) Has your infant had a fever? If the response was “yes”, we proceeded with the following questions: (1) Has your infant had a cough? (2) Has your infant had a runny nose? (3) Has your infant had a stuffed nose? Gastrointestinal (GI) infection was defined based on the presence of specific symptoms obtained through the following questions: (1) Has your infant had diarrhea? (2) Has your infant had vomiting?

We also recorded any episodes of physician-diagnosed urinary tract infection, sepsis, central nervous system (CNS) infections, tuberculosis, and dengue fever. We categorized the infection episodes as other non-specific infections if the infants only have a fever without any other symptoms that can be classified as URTI, LRTI, or GI infection or without any specific diagnosis from a physician. Any episodes of hospitalization, length of stay, diagnosis, and medication given during hospitalization were also recorded.

The cohort’s exposure to air pollution was assessed using land use regression (LUR) models. These models were developed based on targeted measurements of fine particles and nitrogen oxides in Jakarta (Lu et al. 2023b). Detailed description of the measurements and model development have been described previously in Soesanti et al. (2023) Briefly, the study area encompassed the primary care catchment area of the cohort study, located in the center of Jakarta. Measurements were made at 88 sites across the study area (Fig. 1) and LUR models were developed for PM2.5, soot (a measure of black carbon), nitrogen dioxide (NO2) and the sum of nitrogen dioxide (NO2) and nitrogen oxide (NO), denoted as NOx.

Study area of Jakarta air pollution sampling (Soesanti et al. 2023). Red pin indicates traffic sites, yellow pin: urban background, green pin: urban green, white pin indicates reference site (colour figure online)

LUR models were developed using supervised linear regression procedures, using predictor variables derived from direct field observations of traffic counts and street configuration and global GIS databases of road data from Open Street map and impervious surface information (Soesanti et al. 2023). These models can be found in Table S1. All the models included motorcycle counts at the nearest road as a predictor variable. The LUR models explained 61, 59, 26 and 33% of the measured annual average concentration variability for NOx, NO2, PM2.5 and soot, respectively (Soesanti et al. 2023). Exposure to air pollution was assessed individually using LUR model. However, due to the lack of continuous monitoring data for all four evaluated pollutants in Jakarta, we could not use extrapolation methods and were not able to account for temporal differences between women giving birth at different times (Soesanti et al. 2023). As a result, we could not calculate trimester-specific or full pregnancy-specific exposure. We also could not define specifically the exposure in the first six months of postnatal life. Continuous monitoring data was available from the US embassy only for PM2.5 (AirNow Department of State 2023). We calculated annual average concentration and full pregnancy average concentrations related to temporal variation (Soesanti et al. 2023).

We categorized the confounding variables into three groups: (1) family demographic factors, (2) environmental factors, and (3) infant's factors. Within family demographic factors, variables such as level of education, household income, maternal age at pregnancy, parity (Madsen et al. 2017; MacIntyre et al. 2014; Aguilera et al. 2013; Shi et al. 2021; Lu et al. 2023a), delta BMI, mother's working status during pregnancy, and maternal co-morbidity/gestational complications were predetermined as potential confounders (Madsen et al. 2017; MacIntyre et al. 2014; Aguilera et al. 2013; Shi et al. 2021). Level of education and household income were used as proxies for socioeconomic status (Madsen et al. 2017; MacIntyre et al. 2014; Shi et al. 2021). Smoking status during pregnancy (active or passive) (Madsen et al. 2017; MacIntyre et al. 2014; Shi et al. 2021; Liu et al. 2022) and exposure to pesticides during pregnancy, as reported in the questionnaire, were considered potential confounders related to indoor environmental factors. Gestational age and birth weight were identified as potential confounders related to infant factors and were treated as numerical variables (Madsen et al. 2017; MacIntyre et al. 2014; Aguilera et al. 2013; Liu et al. 2022).

Continuous variables for baseline subject characteristics of the mothers and infants were expressed as mean and standard deviation or median and interquartile range if distributions were skewed, while categorical variables were expressed as number of subjects and its percentage. Air pollutant concentrations were presented as annual average concentrations of PM2.5, soot, NOx, and NO2 with interquartile range (IQR) and minimum–maximum values. The distribution of infection during the first six months of life was tabulated for each age (0–1,1–2, 2–4, and 4–6 months) interval, presented as numbers and percentages.

We included infants with a birth weight ≥ 2500 g for the final analysis, as children with low birth weight have a higher risk of infections in infancy. This restriction aimed to minimize the potential for bias. We analyzed the cumulative incidence of upper respiratory tract, lower respiratory tract, and gastrointestinal infection in the first six months of life separately. The cumulative incidence of infection was denoted as binomial “yes” and “no” for each infection, while air pollutant concentration was entered as a continuous variable.

Multivariable logistic regression adjusted to all potential confounders was performed to investigate the association between traffic-related air pollutants with the cumulative incidence of URTI, LRTI, and GI infections during the first six months of life. Additionally, we analyzed the association between traffic-related air pollutants with the incidence of URTI at each age interval (1–2 months, 2–4 months, 4–6 months) by performing multivariable logistic regression adjusted to all potential confounders. Effect estimates (Odds ratios) were calculated based on the IQR increment of each air pollutant (Soesanti et al. 2023). The IQR was 7.14 μg/m3 for PM2.5, 0.75 × 10–5 per m for soot, 4.68 μg/m3 for NOx, and 3.74 μg/m3 for NO2 (Lu et al. 2023b). Statistical significance was determined when 95% confidence intervals did not include unity, indicated two-sided p values less than 0.05. We used R.4.1.2. to develop LUR models and IBM SPSS version 24 for Mac for statistical analyses.