A systematic review and meta-analysis on heat thresholds for maternal health and birth outcomes

Of the 3667 studies identified for screening, we included 81 studies, 6 additional studies through backward citation chasing, and another 15 studies through forward citation chasing performed in March 2026. This resulted in a total of 103 studies for this systematic review (Fig. 1). These were published between 1999 and 2026, across six continents, with most research coming from USA (n = 24) and China (n = 24) (Fig. 2 and Table 1, Supplementary Table 1 and Supplementary Table 2 and Supplementary Table 3). Studies were conducted spanning four of five Köppen-Geiger climate classifications, including A- tropical, B- arid, C- temperate, and D- continental²⁰. Only four studies cover data from low-income countries (LIC) classified according to the World Bank Atlas Method of fiscal year 2024-25²¹ (Supplementary Table 3). More than half of the studies were cohort design studies (n = 62), followed by case-control designs (n = 23), cross-sectional designs (n = 17), and a single human-controlled trial (Supplementary Table 7). Sample sizes ranged from 40 to 5,056,018 (Supplementary Table 6). More comprehensive information for descriptive statistics can also be found in the Supplementary Data 1.

PRISMA 2020 (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) flowchart, illustrating the study selection process using Covidence for literature organisation. Green, italicised categories indicate studies whose full text was reassessed after refining selection criteria.

Bubbles represent study locations where data were extracted, with some studies contributing data from multiple locations, resulting in more than 103 bubbles. The size of each bubble indicates the sample size of that specific location. If a study did not report sample sizes for individual locations but provided a total sample size across multiple locations, the per-location sample size was estimated by dividing the total sample size by the number of locations presented in the study. The background map shows the annual average of monthly maximum temperature extracted from the National Centres for Environmental Information from the National Oceanic and Atmospheric Administration (NOAA), https://www.ncei.noaa.gov/. Source data are provided as a Source Data file.

Heterogeneity of Heat Indicators

Overall, there was considerable heterogeneity in heat indicators used across studies. Air temperature was used by 78 studies, including Tmax, Tmean, and Tmin. In total, 18 studies used apparent temperature (AT), of which 13 examined ATmax, and eight used ATmean. Five studies used the Wet-Bulb-Globe Temperature (WBGT), one reported WBGT hourly data, three reported WBGT mean, and one reported WBGTmax. Universal Thermal Climate Indicator (UTCI) was used by three studies; two reported UTCImean, and one reported both UTCImax and UTCImean (Table 1 and Supplementary Table 5).

Summary of effects

In total, 2298 effect estimates (996 odds ratios (OR), 741 risk ratios (RR), 420 hazard ratios (HR), 120 regression coefficients, and 21 prevalence ratios (PR)) were extracted across 24 health outcomes (see Supplementary Table 4 for study count per health outcome). The direction of effect estimates was largely harmful (2155 effect estimates), with few studies (n = 4) reporting the opposite effect (143 effect estimates). Studies with protective effects assessed gestational hypertension (GHTN) and preeclampsia (PE)²², congenital anomalies²³, PTB^24,25, and small for gestational age (SMA)²⁴, which is in line with previous research¹⁸. Of the 2298 effect estimates, 1183 addressed the effects of heat wave exposure. In total, 2020 effect estimates were used to conduct meta-analyses. While average temperature (Tmean) and percentiles did not yield any meaningful association with adverse maternal health and birth outcomes (OR), maximum temperature (Tmax) explained a statistically significant proportion of the variance in the risk of adverse maternal health and birth outcomes (R² = 28%, p < 0.001). Although the overall number of ATmax studies was sufficient for a meta-regression, multiple studies did not report tangible ATmax heat thresholds or additional information required for the analysis, such as baseline heat exposure data or non-exposed controls, preventing an ATmax-based meta-regression from being conducted. Alternative heat indicators were not appropriate for the meta-analysis due to a lack of statistical power. For the Tmax meta-regression, 197 ORs from 19 studies were used after converting all effect estimates to ORs. This total reflects multiple effect estimates per study with multiple effect estimates at the same temperature level, as many studies reported results stratified by factors such as gestational period, geographic location, or several health outcomes. In the Tmax meta-regression, only lag 0 effect estimates were included due to substantial heterogeneity in lag reporting across studies. Lag effects indicate longer-term, delayed effects, which should be interpreted separately from immediate effects of heat. All lag 0 effect estimates that met the inclusion criteria described in the Methods were included in the meta-analysis. The Tmax meta-regression yielded the equation below, which defines the relationship between Tmax and the OR of experiencing adverse pregnancy and birth outcomes at any given maximum temperature.

At a Tmax of 30.55  °C, we determined a 5% risk increase of experiencing adverse pregnancy and birth outcomes. At 34.11  °C we identified a 10% risk increase, and at 39.07  °C, a 20% risk increase, with 24.53  °C marking no risk increase (Fig. 3a). We found a small residual, between-study heterogeneity in our meta-regression (τ²) of 0.007, indicating small differences in magnitude of effect estimates across studies. However, the relative proportion of remaining variability due to true between-study differences was moderate (I² = 43%), likely reflecting variations in population characteristics and settings, including differences in age, living conditions, and access to cooling, amongst others. We were unable to stratify the meta-analysis by gestational period due to substantial variability in definitions and assessment of gestational timing. Therefore, the results include effect estimates aggregated across all pregnancy periods. Two studies by Ren et al., included in the meta-analysis, were based on the same underlying cohort and therefore reported the same sample size, but used different methodological approaches, with the first paper assessing optimal heat thresholds for early warning²⁶, while the second examined regional, socioeconomic moderators for preterm birth²⁷. Health outcome specific meta-regressions were also conducted, with PTB being the only health outcome with sufficient statistical power for a meaningful result. PTB specific thresholds at 30.05  °C marked a 5% risk increase, 33.34  °C, a 10% risk increase, and 38.64  °C, a 20% risk increase. Only minor differences between PTB specific heat thresholds, compared to thresholds identified for overall adverse maternal health and birth outcome, could be determined (Fig. 3b). Furthermore, with the clear over-representation of studies from Asia, Europe, and North America in this review, with generally cooler climates, we could conduct a temperate climate zone-specific meta-regression. The climate zones were classified according to Köppen-Geigers simplified climate zone classifications²⁰, representing slightly lower thresholds for adverse pregnancy and birth outcomes, with 29.03 °C marking a 5% risk increase 31.86  °C, 10% risk increase, and 35.64  °C, 20% risk increase (Fig. 3c). Heterogeneity statistics for both subgroup analyses showed small heterogeneity statistics (see Figs. 3b, 3c). No other disaggregation was possible. To ensure that the findings were not driven by a single study, we conducted leave-one-study-out sensitivity analyses by refitting the overall meta-regression after excluding each study in turn. The temperature slope remained positive and statistically significant in all runs, with thresholds varying minimally at lower threshold levels (\(\pm\) 2 °C) and somewhat more for higher thresholds, also reflected in the larger confidence intervals depicted in Fig. 3a below (Supplementary Data 4 for sensitivity analysis).

Random effects non-linear (quadratic) meta-regression analysis depicting the relationship between maximum daily temperature (Tmax) and (a) adverse maternal health and birth outcomes included in the 21 studies with available relevant data, (b) PTB included in 11 studies with available relevant data, and (c) adverse maternal health and birth outcomes in a temperature climate zone specifically included in 13 studies with available relevant data. Statistical tests were two-sided, and p-values for Fig. 3a-c were statistically significant (p < 0.001). Y-axis represents the log odds ratio (log OR), while the right Y-axis provides its corresponding odds ratio (OR). Data are presented as a black solid line, which corresponds to the predicted effect estimates from the fitted quadratic meta-regression model (random-effects). The shaded area represents the 95% confidence interval around the regression line. Each circle represents an individual study’s effect size (log OR/OR) plotted against the corresponding Tmax, with bubble size proportional to the statistical weight of each data point. Vertical dashed blue lines indicate temperature thresholds corresponding to 5%, 10%, and 20% increased risk. I2 refers to the between-study heterogeneity, τ2 represents the between-study variance, and Pseudo R2 depicts the proportion of between-study heterogeneity explained by the model. Source data are provided as a Source Data file. Supplementary Data 1 includes estimated risks for any kind of adverse pregnancy and birth outcomes included in the 103 studies from 25 to 40 °C Tmax.

Heat stress indices (HSI) and heat-health thresholds

In total, 18 studies used apparent temperature (AT), whereof 13 used maximum AT (ATmax), and eight used mean AT (ATmean). ATmax studies assessed PTB (n = 10)^{25,28,29,30,31,32,33,34,35,36}, LBW (n = 1)³⁷, adverse pregnancy outcomes (APO) (n = 1)³⁸, neural tube defects (NTDs) (n = 1)³⁹, and non-accidental deaths (NAD) (n = 1)²⁵; see Table 2 for more information. Although the number of ATmax studies exceeded the minimum required for meta-analysis (n = 13, minimum > 10), the studies did not consistently meet the criteria required for meta-regression. Therefore, ATmax studies are described descriptively below. Eight of ten PTB studies reported an increased risk of PTB as heat increases, and two studies found no association^25,33. The ATmax study assessing LBW reported a weak but significant association with increased heat³⁷, and the NTD study found no statistically significant association³⁹. For NADs, the highest risk increase reported by Kent et al. was found at the 90^th percentile for ≥1 day (2.0%, 95% CI 0.3, 3.8)²⁵. Both protective and harmful birth outcomes were reported by van Zutphen, with the highest effect found for congenital cataracts (OR 1.45, 95% CI 1.10, 1.90)³⁸. The reported PTB risk increases ranged between 15% and 22% (OR 1.15 95% CI 1.06, 1.24⁴⁰), (HR 1.17 95% CI 1.04, 1.32, ≥40 °C vs 20  °C³⁴), (OR 1.22 95% CI 0.94, 1.58 at lag 2, 36  °C vs. 31  °C³⁰). Asta et al. found a large risk increase of PTB (RR 1.94, 95% CI 1.32, 2.85) at lag 0–3 for Venice, comparing the 90th to the 75th percentile of local ATmax³⁵. Three studies reported PTB risk increases per 1  °C increase in ATmax, ranging from 2% to 7.1% (OR 1.02 95% CI 0.98, 3.03²⁹), (OR 1.07 95% CI 1.04–1.11 for Barcelona³¹), (OR 1.07 95% CI 1.05–1.09 for Rome³¹), (OR 1.091 95% CI 0.86, 2,87)³², all at lag 0-2. Furthermore, a 5–7% increased risk of PTB was reported for every 10  °F (5.6  °C) at lag 5 and lag 6, and lag 06³⁶. Finally, a weak association between PTB and ATmax was reported for longer-duration heatwaves (ORs ranging between 1.01–1.03, with most CIs for shorter heat durations crossing 0)²⁸. ATmean was used in eight studies. ATmean studies assessed PTB (n = 6)^{28,36,40,41,42,43}, stillbirth (n = 1)⁴⁴, APO (n = 1)³⁸, and LBW (n = 1)⁴³. All of these found a significantly increased risk of adverse outcomes as heat increases, except when assessing LBW. Again, similarly to when using ATmax, van Zutphen found inconsistent results across adverse birth outcomes, with the highest effect reported for congenital cataracts (OR 1.47, 95% CI 1.11, 1.94). For PTB, a 4% risk increase (OR 1.04, 95% CI 1.01, 1.06) was reported overall⁴⁰, and a 11.63% risk increase was found for every 5.6  °C ATmean increase (OR 1.12 95% CI 1.05, 1.20⁴²) at lag 06. Mohammadi et al. reported the greatest risk increase of PTB at lag 0 (RR 1.61, 95% CI 1.41, 1.83), at the 99^th percentile of the AT distribution⁴¹. Similar to their ATmax findings, Huang et al. reported weak associations between ATmean and PTB during longer-duration heatwaves (OR 1.01-1.03), with most CIs for shorter heat durations crossing 0)²⁸. A strong PTB risk increase was reported by Zou et al. during late pregnancy (OR 3.08, 95% CI 1.45, 6.53) at 29.63  °C. Finally, Avalos et al. reported a significant association between ATmean and PTB at lag 06 (11.6%, 95% CI 4.1, 19.7) per 10  °F (5.6  °C), higher than those reported for ATmax³⁶. For stillbirth, a 10.4% increase in risk of stillbirth was found for every 10  °F (5.6  °C) increase in ATmean (95% CI 4.4%, 16.8%) at lag 2–6⁴⁴. All AT studies were conducted in temperate climate zones following the updated Köppen-Geiger climate classifications²⁰, except for one study carried out in a continental climate zone³³, and two^30,41 in an arid climate zone.

In total, seven studies used the Heat Index (HI), assessing PTB (n = 4)^25,45,46,47, stillbirth (n = 2)^47,48, SGA (n = 2)^47,49, PROM (n = 1)⁵⁰, LBW (n = 1)⁴⁷, and non-accidental deaths (NAD) (n = 1)²⁵. Four of seven studies reported an increased risk of adverse outcomes with higher HI. One study found only a weak and non-significant overall association with stillbirth (OR 1.2, 95% CI 0.8, 1.6) at 33.1 °C⁴⁸. Another study reported substantial variation across HI definitions, with null and negative associations for PTB and NAD²⁵. A final study found that heat variability, rather than high heat itself, was associated with increased risk of LBW and SGA⁴⁹. The following studies reported adverse heat-related effects. Another PTB study found a 1% to 3% increased risk associated with HI values between 32.2  °C and 36.7  °C (OR 1.01, 95% CI 1.00, 1.01), (OR 1.03, 95% CI 1.03, 1.03. No stronger associations for cumulative lags were found⁴⁵. A risk increase in PTB during first-trimester heat exposure (HR 1.06, 95% CI 1.01, 1.11), unclear associations for the second trimester, and a strong association during third-trimester heat exposure (HR 1.51, 95% CI 1.34, 1.71) was also reported⁴⁷. The same study assessed stillbirth and found great geographical variation, with non-significant associations in Sweden and Italy, but significantly increased risks in Belgium (HR 1.69, 95% CI 1.48, 1.92) and Greece (HR 1.26, 95% CI 1.17, 1.35)⁴⁷. Another study reported a reduction in days of pregnancy associated with higher HI, with a mean reduction of 1.6 days (95% CI −3.2, 0.6) and a 5.3-day (95% CI −10.1, −0.5) reduction at 32  °C one day prior to delivery⁴⁶. All studies using HI were conducted in temperate climate zones²⁰, except for one conducted in a continental climate zone⁴⁹.

Studies using WBGT (n = 5) assessed APO (n = 2)^51,52, PTB (n = 2)^43,47, umbilical artery resistance (i = 2)^51,53, and foetal heart rate (n = 1)⁵¹, stillbirth (n = 1), SGA (N = 1)⁴⁷, LBW (n = 1)⁴³, and placental abruptions (n = 1)⁵⁴. All five WBGT studies reported adverse heat-related effects. WBGT was associated with increased PTB risk during the third trimester (1.51 CI 95% 1.34, 1.71); notice the identical effect estimates to the HI analysis from this study⁴⁷. The same study also reported increased risk for stillbirth (HR 1.54, 95% CI 1.44, 1.65) and SGA (HR 1.16, 96% CI 1.11, 1.20)⁴⁷. Another study found strong associations between WBGT and PTB throughout pregnancy (OR 3.35, 95% CI 1.39, 8.06)⁴³, as well as LBW (OR 2.89, 95% CI 1.17, 7.14). Placental abruption risk was highest at lag 1 (RR 1.23, 95% CI 1.11, 1.39). WBGT was also associated with a 13.4 beats/min increase in foetal heart rate per 10  °C WBGT increase, with WBGT values ranging between 19.3  °C and 27.3  °C (95% CI 9.5, 17.2). Rekha et al. further reported a 3.1-fold risk increase of APOs at 27.1 °C WBGT (95% CI 1.3, 7.3)⁵². Of these five studies, three were conducted in temperate climate zones^43,47,54, one in arid⁵¹ and one in tropical⁵² climate zones. UTCI studies (n = 4) assessed PTB (n = 2)^47,55, stillbirth (n = 2)^47,55, umbilical artery resistance (n = 2)^51,53, and LBW (n = 1), miscarriage (n = 1), preeclampsia (n = 1), gestational hypertension (n = 1)⁵⁵, and SGA (n = 1)⁴⁷. Three of four UTCI studies reported adverse heat-related effects, one found mixed results, depending on outcome⁵⁵. One study found UTCI to be associated with increased risk of stillbirth at 46.4  °C, lag 0-13 (RR 2.049, 95% CI 1.012, 4.151)⁵⁵, but not with PTB, LBW, spontaneous abortions, preeclampsia, or gestational hypertension. Another study reported a 10.7 beats/min increase in foetal heart rate for each 10  °C UTCI increase at temperatures ranging between 22.1  °C and 34.2  °C (95% CI 7.5, 13.8)⁵¹. Supporting this finding, another study, reported impaired placental-foetal blood flow, measured as changes in umbilical artery resistance (0.963 95% CI 0.981, 1.007)⁵³. Of the four UTCI studies, two were conducted in a temperate climate zone^47,53, and two were conducted in an arid climate zone^51,55, according to the Köppen-Geiger climate classifications²⁰.

Heat thresholds for adverse maternal health outcomes

Preeclampsia (PE) and gestational hypertension (GHTN) were assessed using Tmax (n = 4)^22,56,57,58, Tmean (n = 3)^22,56,59, and UTCI (n = 1)⁵⁵. Findings were mixed, with several studies reporting decreased or null associations, while others found increased risks during specific gestational periods. Using Tmean, Zhao et al. reported a significantly decreased PE/GHTN risk at a threshold of 28.7  °C (RR 0.59, 95% CI 0.39, 0.89). The same study found a similarly decreased risk using Tmax at 33.2 °C (RR 0.60, 95% CI 0.41, 0.88) at lag days 0-15²². Shankar et al. also reported a reduced PE/GHTN risk during first-trimester exposure to Tmax 30-40 °C (RR 0.95, p = 0.02), but an increased risk during the last trimester (RR 1.07, p = 0.005)⁵⁷. Using UTCI and separating PE and GHTN, Khodadadi et al. found no association with GHTN, but reported an elevated PE risk at 46.4 °C UTCI (RR 1.248, 95% CI 0.872, 1.788) at lag 0-13⁵⁵. Similarly, Tran et al. and Bogan et al. found no significant associations between Tmax and PE neither in early pregnancy (OR 1.02, 95% CI 1.00, 1.03)⁵⁸ nor when comparing PE outpatients (OR 1.006, 95% CI 0.988, 1.023) to inpatients (OR 0.991, 95% CI 0.974, 1.008)⁵⁶. In contrast, Part et al. reported increased PE risk associated with higher Tmean during early gestation (23 °C vs 18 °C), including week 3 (HR 1.76 95% CI 1.12, 2.78) and week 4 (HR 1.79 95% CI 1.19, 2.71)⁵⁹. Most of these studies were conducted in a temperate climate zone, with one study extending to tropical⁵⁷ and arid⁵⁵ climate zones. Premature Rupture of Membranes (PROM) was assessed by three studies, using HI (n = 1)⁵⁰, Tmean (n = 1)⁶⁰, and Tmax (n = 1)⁶¹. All three studies reported adverse heat-related effects. Using HI, Jiao et al. increased PROM risk at 32.8  °C (HR 1.097, 95% CI 1.058, 1.138), 35.6  °C (HR 1.180, 95% CI 1.121, 1.242), 37.8  °C (HR 1.270, 95% CI 1.169, 1.380), and 40.0  °C (HR 1.456, 95% CI 1.150, 1.843)^22,50. Using Tmean, Song et al. reported the strongest association at threshold 32  °C and lag 0–2 (RR 2.161, 95% CI 1.240, 3.764), while also lag 0–1 and 0-3 were significant, however weaker, and lag 0–4 to 0–7 were not significant⁶⁰. Finally, PROM was associated with daily temperature variation, measured using Tmax, with higher PROM rates in hotter months, ranging between 34.1  °C and 36.2  °C⁶¹. Thresholds described for PROM were established for temperate⁵⁰, continental⁶⁰, and arid climate zones²⁰.

OHAT tool for risk of bias assessment

Although the OHAT Risk of Bias Assessment is composed of eleven questions, only seven of these were applicable to almost all studies, since four items of the tool were limited to animal studies or human-controlled trials (HCT). This review included no animal studies and only a single HCT study. The HCT study therefore required the assessment of three additional items on the OHAT tool. Two studies scored negatively on one or more items related to detection bias, confounding bias, and other biases. All other outcomes scored either + low risk of bias or ++ definitely low risk of bias, or NRs when there was a lack of sufficient information to make a judgement (Supplementary Data 3).