The pregnancies, outcomes, and exposures are simulated. They are not a UAE registry, not patient records, and not monitoring data. The odds ratios below are properties of the synthetic data-generating process. They are not estimates of the health effects of heat or air pollution in the UAE, and they do not support a causal claim.
Design
The cohort has 10,000 simulated pregnancies in Dubai, Abu Dhabi, and Sharjah. Variables include maternal age, gestational week, region, trimester, temperature, PM2.5, preeclampsia, gestational diabetes, and birthweight in grams. Mean pregnancy temperature is the primary exposure. A high-heat category uses a simulated threshold of 35°C or above. Early, mid, late, and cumulative exposure windows were generated to practise window analysis. They are not genuine longitudinal measurements.
Preeclampsia is the primary outcome. Gestational diabetes and birthweight are secondary. Binary outcomes use logistic regression. Birthweight uses linear regression. Models adjust for the co-exposure, maternal age, trimester, and region where that is the specified adjustment set.
Adjusted synthetic estimates
| Association | Estimate | 95% CI |
|---|---|---|
| Temperature and preeclampsia, per 1°C | OR 1.090 | 1.061–1.120 |
| PM2.5 and preeclampsia, per unit | OR 1.020 | 1.010–1.031 |
| Temperature and gestational diabetes, per 1°C | OR 1.059 | 1.036–1.083 |
| Temperature and birthweight, per 1°C | β −34.91 g | −37.85 to −31.97 |
Exposure windows
| Window | OR per 1°C | 95% CI |
|---|---|---|
| First trimester | 1.082 | 1.056–1.109 |
| Second trimester | 1.082 | 1.056–1.108 |
| Third trimester | 1.070 | 1.045–1.096 |
The temperature-by-trimester interaction p-values were 0.803 for the second trimester and 0.064 for the third. The associations are positive in each window. The interaction test does not provide strong evidence that they differ by stage, so the project does not claim a critical window.
Effect modification, shape, and sensitivity
There was no strong evidence of effect modification by socioeconomic index (p = 0.186) or green-space access (p = 0.235). Both modifiers are synthetic. A natural spline did not improve fit over a linear temperature term (likelihood-ratio p = 0.829). That result describes this simulation. It is not evidence that real temperature–health relationships are linear.
| Exposure definition | Odds ratio | 95% CI |
|---|---|---|
| Overall mean temperature | 1.090 | 1.061–1.120 |
| Cumulative temperature | 1.096 | 1.068–1.125 |
| High heat versus lower heat | 1.606 | 1.323–1.939 |
Cumulative exposure is an average of the synthetic early, mid, and late temperatures. It is not a distributed-lag non-linear model. A real DLNM would need time-varying exposures linked to pregnancy dates.
Workflow
Thirteen R scripts take the work from cohort generation through descriptive analysis, outcome models, interactions, non-linearity, cumulative exposure, sensitivity analysis, and a final summary. Code, CSV outputs, visualisations, the protocol, and the report are in the repository: UAE-Heat-Air-Pollution-Pregnancy-Epidemiology.