Censuswide conducted a survey of 1,001 currently practicing healthcare professionals involved in pregnancy, childbirth, postnatal or newborn care to understand their perceptions of how extreme heat impacts pregnant women and their babies.

Fieldwork took place via an online survey in Australia, Brazil, India, the UK and Zimbabwe between 17 September and 1 October 2026. There were 200 espondents from each geography, except Brazil where there were 201. The questionnaire was administered in Portuguese in Brazil and in English in Australia, India, the UK and Zimbabwe.

Eligibility to participate was determined on the respondents’ frontline role in healthcare and specialist role and responsibilities in maternal and post-partum care (of pregnant women and their babies). This avoided any potential inconsistencies between professional terminology and the structure of how maternal healthcare differs between markets. Eligible roles included obstetricians, gynecologists, doctors, midwives and other midwifery professionals, nurses, family physicians/GPs focused on maternal care, perinatologists, maternal-foetal medicine specialists and neonatologists.

Questionnaire development

The survey was developed by Censuswide with design input from Wellcome and the Climate Opinion Research Exchange using a careful research process, with feedback from Wellcome and Climate Opinion Research Exchange to inform design. The questions were designed to gain insights from healthcare professionals that help provide a picture of concerns around the impact of extreme heat on pregnancy.

The questions capture healthcare professionals’ reported experiences and professional judgement. The findings demonstrate what respondents have observed and their views on the relationship between extreme heat and maternal, foetal and newborn health, but should not be presented as clinical case-record evidence or as independently establishing causation.

Sampling and data collection methodology

Polling was conducted by Censuswide, a member of the British Polling Council.

Respondents were recruited via Censuswide’s own, verified online panel which has over 300,000+ adults.

Completed responses are reviewed for quality, including completion times and response patterns. Responses identified as failing the quality criteria are excluded, and replacements are subject to quality checks before inclusion in the final dataset.

A minimum quota of 25% midwives or midwifery professionals was applied in each market, equivalent to around 50 respondents per country. The remaining sample falls naturally across the other eligible professions, with controls applied where necessary to prevent a single role from disproportionately dominating a market.

Data quality

All completed responses were subject to a structured data-quality review before analysis - checks were conducted at respondent level and, where relevant, across the cumulative sample. Data-quality criteria were applied independently of the substantive survey findings, and respondents were not removed simply because their answers were unusual or differed from the overall pattern.

• Survey validation and eligibility checks. Programmed survey validation measures were reviewed, including the survey attention/red-herring check, automated-response controls, age consistency checks and screening or qualification information. Responses were also reviewed for invalid or impossible routing or profile combinations.
• Completion-time checks. Timings across the substantive questionnaire were reviewed to identify respondents completing the survey unusually quickly. Speeding was considered alongside other quality indicators and in the context of the respondent's overall answer pattern, rather than relying solely on total interview duration.
• Straightlining and response-pattern checks. Substantive grid questions were reviewed for repeated identical answers and other low-variation response patterns. Single instances were treated cautiously, while repeated patterns across multiple substantive question batteries were considered stronger evidence of poor engagement.
• Cross-question consistency. Answers to related questions were compared to identify material logical inconsistencies. This included checking consistency between respondents' stated experiences or needs and selections made in subsequent questions. Apparently unusual relationships were not treated as invalid unless they represented a genuine contradiction.
• Multiple-response checks. Multi-select questions were reviewed for mutually contradictory selections, such as choosing a substantive response alongside an exclusive response option. Selection limits were also checked where relevant.
• Engagement checks. Patterns of unusually low engagement across eligible multi-select questions and other substantive sections were used to identify respondents requiring closer review. Low engagement alone was not automatically treated as sufficient grounds for removal.
• Duplicate and similarity checks. Respondent identifiers and technical information were reviewed for potential duplicate records. Substantive answer patterns were also compared to identify unusually high similarity between respondents. Shared technical identifiers alone were not treated as proof of duplication; corroborating response-pattern evidence was required.
• Professional-role and profile checks. Professional role and related profile information were reviewed for impossible or materially inconsistent combinations. Unusual but plausible profiles were retained unless supported by other evidence.
• Device and sample-source diagnostics. Device and sample-source information was reviewed for unusual concentrations or differences in quality-flag rates. These variables were used diagnostically rather than as standalone reasons for excluding an individual respondent.
• Randomisation and order-effect checks. Randomised response-option ordering was compared with selection behaviour to identify potential primacy or order effects. Where these appeared systematically across the sample, they were treated as questionnaire-level measurement effects rather than evidence that individual respondents were invalid.
• Statistical and distributional diagnostics. Response distributions, unusual answer combinations and response-similarity measures were used to identify cases for further inspection. Statistical rarity alone was not treated as evidence that a response was invalid.
• Researcher review. Programmed checks were used to identify potential issues, but final data-quality decisions also involved researcher review of the respondent's overall response pattern and the context of individual flags. Indicators were considered together to avoid double-counting related behaviours, and respondents were classified according to the strength and independence of the available evidence.

The data-quality process combines programmed checks with researcher adjudication. While some individual mechanical checks can in part be reproduced programmatically, the final inclusion or exclusion decision is not generated by a single automated rule set and includes a human sense-check of the respondent's overall response pattern and the context of any flags identified. For this reason, it would not be possible to provide code that independently reproduces every final data-quality decision into a replication code.

Data weighting

Results are unweighted. Suitable, comparable population benchmarks were not identified for weighting the specialist eligible audience across all five countries. The findings describe the achieved sample of healthcare professionals surveyed and should not be interpreted as nationally or globally representative workforce estimates.

Analysis and reporting

The published results are descriptive percentages for the combined country sample, and for each country separately, with cross-tabulations comparing midwives/midwifery professionals with other eligible healthcare professionals.

Respondent bases will accompany the findings. As a reporting guideline, we do not recommend standalone headline claims based on subgroups with fewer than 50 respondents. This threshold is a practical reporting safeguard rather than a guarantee of statistical reliability. The existing wording distinguishing the healthcare professionals surveyed from the wider national or global workforce should remain.

Download the full data set here.

About this research

Opinion polling uses structured questionnaires to collect people’s views and reported experiences at a particular point in time. This study focuses on healthcare professionals involved in pregnancy, childbirth, postnatal or newborn care, providing insight into their experiences, concerns and perceptions of preparedness for extreme heat. The findings describe the participating professionals’ responses; they do not measure the incidence of medical conditions or independently establish causal relationships between heat and health outcomes. They should be considered alongside clinical and epidemiological research, with reference to the survey’s sampling approach and the respondent bases accompanying each result.

For further information:

For further information, contact: Wellcome_Pregnancyandheat@freuds.com