Guidance

Quality and methodology information: Cold mortality monitoring reports

Published 18 February 2026

Applies to England

About this report 

This report explains the quality and methodology information (QMI) for the ‘Cold mortality monitoring reports’ official statistics published by the UK Health Security Agency (UKHSA).

This QMI report helps users understand the strengths and limitations of these statistics, ensuring UKHSA is compliant with the quality standards stated in the Code of Practice for Statistics. The report explains: 

• the strengths and limitations of the data used to produce the statistics 

• the methods used to produce the statistics 

• the quality of the statistical outputs

About the statistics 

Cold weather can cause people to become unwell through hypothermia, and falls and injuries from snow and ice. Cold temperatures also place stress on the body, increasing the risk of heart attack, stroke, lung problems and other diseases, and contribute to the spread of infectious diseases such as influenza. This can lead to increases in deaths, both during a period of cold weather and in the days or weeks afterwards. During the winter, UKHSA and the Met Office work together to issue Cold-Health Alerts (CHAs) if the weather is cold enough that it has the potential to affect people’s health. 

The annual ‘Cold mortality monitoring report’ provides information on deaths during cold episodes each year to inform public health actions. The statistics show the number of cold-associated deaths, both for the total impact of cold and for the direct impact of cold independently of impacts on influenza circulation. The total cold-associated deaths are also broken down by age, sex, region, cause of death and place of death. The report also discusses the timing of cold-associated deaths, and compares the cold-mortality relationship in the latest 5 years with a previous 5-year period.

The ‘Cold mortality monitoring report’ is being published as a new official statistic in development in 2026, demonstrating that it is produced in line with the standards of trustworthiness, quality and value in the Code of Practice for Statistics. UKHSA also produces annual heat mortality monitoring reports, which are currently available for summers from 2016 to 2024.

Geographical coverage: England

Publication frequency: Annual

Changes to this document

18 February 2026: QMI report first published.

Contact 

Lead analyst: Mo Davies 

Contact information: extremeevents@ukhsa.gov.uk

Suitable data sources 

Statistics should be based on the most appropriate data to meet intended uses. 

This section describes the data used to produce the statistics. 

Data sources 

Data on deaths is derived from the Office for National Statistics (ONS) death registrations data. Final annual registrations data is used for deaths registered up to 2024, and provisional weekly registrations data is used for deaths registered from 2025 onwards. For each winter, data is extracted for deaths which occurred from October through to April of the following year. This is to allow analysis of delayed impacts of temperature on mortality throughout the winter, which is defined as November through to March of the following year.

Daily mean temperature data from the Met Office at regional level is used in the statistical model. This is obtained through the UKHSA Environmental Public Health Surveillance System. An overall daily mean temperature for England is then obtained as a population-weighted average of the regional daily mean temperatures. Cold episodes are defined as a period of 2 or more consecutive days where the England overall daily mean temperature was 2°C or lower. This is consistent with the decision-making aid threshold used for issuing Yellow Cold Health Alerts.

Data on influenza activity is produced by the UKHSA Vaccines Analysis team. It is derived by multiplying weekly Royal College of General Practitioners (RCGP) influenza-like illness consultation rates by weekly influenza swab positivity rate using RCGP swabbing data. This provides an estimate at weekly level of the level of flu circulation in the England population overall.

The latest available ONS mid-year population estimates are used to calculate rates per million population.

Data quality 

The data that we use to produce statistics must be fit for purpose. Poor quality data can cause errors and can hinder effective decision making.

We have assessed the quality of the source data against the data quality dimensions in the Government Data Quality Framework.

This assessment covers the quality of the data that was used to produce the statistics, not the quality of the final statistical outputs. The quality summary section below explains the quality of the final statistical outputs.

Strengths and limitations of the deaths data 

The strengths of the data include: 

• a comprehensive record of deaths occurring in England because the registration of deaths is mandatory

• review of deaths data carried out by ONS on a provisional basis throughout the year, as well as on a final basis annually

• a high level of completion of all required information for analysis because death certificates have required fields

• using international standards to code underlying causes of death

The main limitation of the data is that there are death registration delays, particularly for deaths referred to a coroner. This means some deaths which have already occurred but not been registered will not be included in the analysis.

Accuracy 

Accuracy is about the degree to which the data reflects the real world. This can refer to correct names, addresses or represent factual and up-to-date data. 

The ONS Mortality statistics in England and Wales QMI report gives information on accuracy and validation checks performed.

Completeness 

Completeness describes the degree to which records are present. 

For a data set to be complete, all records are included, and the most important data is present in those records. This means that the data set contains all the records that it should and all essential values in a record are populated. 

Completeness is not the same as accuracy as a full data set may still have incorrect values. 

Death certificates contain several mandatory fields and so the data used in this report for age, sex, place of death, cause of death and geographical area of death is highly complete. Age and sex are 100% complete in provisional data. Geographical area and cause of death are over 99% complete, and place of death is over 98% complete.

Uniqueness 

Uniqueness describes the degree to which there is no duplication in records. This means that the data contains only one record for each entity it represents, and each value is stored once. 

Some fields, such as National Insurance number, should be unique. Some data is less likely to be unique, for example geographical data such as town of birth. 

Checks are performed by ONS to identify and remove duplicate death registrations.

Consistency 

Consistency describes the degree to which values in a data set do not contradict other values representing the same entity. For example, a mother’s date of birth should be before her child’s. 

Data is consistent if it doesn’t contradict data in another data set. For example, the date of birth recorded for the same person in 2 different data sets should be the same. 

ONS deaths data reflects information as recorded on death certificates based on information provided by the informant, medical examiner and coroner (if applicable). Data is not required to match with data on other records. The ONS user guide to mortality statistics gives more information on how death certificates are completed.

Timeliness 

Timeliness describes the degree to which the data is an accurate reflection of the period that it represents, and that the data and its values are up to date. 

Some data, such as date of birth, may stay the same whereas some, such as income, may not. 

Data is timely if the time lag between collection and availability is appropriate for the intended use. 

Deaths are subject to registration delays, meaning that deaths by date of occurrence are always somewhat incomplete. ONS has published further information on the impact of registration delays on mortality statistics in England and Wales.

Based on historic patterns in death registration delays, it is expected that when extracting data for this analysis in February 2026, over 99% of deaths that occurred in winter 2024 to 2025 have been registered. 

Deaths referred to coroners are subject to longer registration delays, meaning there is likely to be a higher proportion of missing deaths among deaths where the underlying cause of death is in the category ‘External causes’.

Validity 

Validity describes the degree to which the data is in the range and format expected. For example, date of birth does not exceed the present day and is within a reasonable range. 

Valid data is stored in a data set in the appropriate format for that type of data. For example, a date of birth is stored in a date format rather than in plain text. 

The ONS mortality statistics in England and Wales QMI report gives information on accuracy and validation checks performed.

Strengths and limitations of the temperature data

The strengths of the data include: 

• a comprehensive record of temperatures measured by the Met Office in England

• validation and data quality checks carried out by the Met Office

The main limitation of the temperature data is the use of an England-wide average temperature when estimating the relationship between temperature and mortality. This might obscure patterns at regional level where a region experienced very different temperatures to the rest of England.

Strengths and limitations of the influenza data

The main strength of the influenza data is that it provides a record of influenza circulation across the England population, calculated using reliable methods aligned with other UKHSA publications.

The limitations of the data are: 

• the use of an England-wide average, which might obscure patterns at regional level or in specific subgroups of the population such as older adults

• the data only being available at weekly level

Sound methods 

Statistical outputs should be produced using appropriate methods and recognised standards.

This section describes how the statistics were produced and quality assured. 

Data set production 

Method for statistical modelling of cold-associated mortality

Estimates of modelled mortality were obtained from a statistical model based on the observed temperature-mortality relationship over several recent winters. The winters of 2019 to 2020 and 2020 to 2021 were excluded from modelling, due to the significant impacts of the COVID-19 pandemic on mortality in these periods.

Mortality data for deaths occurring from 2013 to 2025 was obtained from ONS (using final annual registrations data for deaths registered 2013 to 2024, and provisional weekly registrations data for deaths registered since the end of 2024).

Temperature data was obtained from the Met Office, using a latitude-longitude grid at 0.1 degree resolution to derive daily mean temperature data at regional level for the same period. An overall daily mean temperature for England was calculated for each date as a population-weighted average. 

Temperature and mortality data is joined based on date. A quasi-Poisson regression model is fit using the distributed lag non-linear modelling framework, to estimate the relative risk of mortality at each temperature. The relative risk is an index representing the risk of mortality relative to a reference temperature: if a temperature has a relative risk of 1.1, this indicates a 10% higher risk of death compared to the reference temperature. A relative risk of 0.9 indicates a 10% lower risk of death compared to the reference temperature. The reference temperature was set at 14.4°C, the Minimum Mortality Temperature (the temperature at which risk of mortality is lowest) for the England population overall.

95% confidence intervals are also calculated for the relative risk at each temperature, representing uncertainty due to random variation in daily deaths. Exposure-response curves showing the relationship between temperature and mortality and the confidence intervals are presented in the report.

This modelled relationship is applied to the actual temperatures during the cold episodes of winter 2024 to 2025 to obtain modelled predictions of cold-associated mortality related to each episode. Further detail on the modelling methodology is available. The model was adapted from the model described in the link above to take account of differences between heat and cold, including:

• increasing the lag to 14 days to account for longer delays between cold weather and mortality impacts
• setting knots at appropriate percentiles of the temperature-mortality relationship
• including an adjustment for the effects of COVID-19, using the number of deaths with COVID-19 recorded on the death certificate on that day as a variable, in line with work developed at UKHSA for attribution of deaths in the EuroMOMO model

Specific parameter settings in the model, such as the location of knots, were determined through running the model with different options and comparing model performance statistics.

Method for calculating estimates adjusted for flu

The temperature-mortality relationship was also modelled with an additional adjustment for the effect of flu.

Data on influenza activity at weekly level was obtained from the UKHSA Vaccines Analysis team. The weekly value was applied uniformly to all days across that week, and joined to the temperature and deaths data based on date. A single parameter was included in the model for the effect of influenza activity, with a lag of up to 2 weeks.

The modelling method described above was repeated with other parameter settings and the reference temperature remaining the same, and the temperature-mortality relationship with and without the adjustment for flu is presented in the report.

Method for comparing recent winters with previous winters

The temperature-mortality relationship was modelled over the latest 5-year period (2018 to 2019, 2021 to 2022, 2022 to 2023, 2023 to 2024 and 2024 to 2025) and a previous 5-year period (2013 to 2014, 2014 to 2015, 2015 to 2016, 2016 to 2017 and 2017 to 2018). The modelling method described above was repeated for both subsets of data with the same parameter settings and reference temperature, and the temperature-mortality relationships in each period are presented in the report.

Methods for breakdowns of cold-associated mortality 

The number of cold-associated deaths was broken down by region, age group, sex, place of death and cause of death. This was done by applying the method for modelled cold-associated mortality above to each subset of the data, using the same parameter settings and reference temperature.

Grouping by region was done based on the Lower Super Output Area (LSOA) of the location of death, aggregated to region boundaries. Where location of death was not recorded, the location of residence was used instead. Rates per million population were calculated using ONS mid-year population estimates by local authority, aggregated to region boundaries.

Grouping by age was done based on the age at death as recorded on the death certificate. This is based on Grouping B of the Government Analysis Function harmonised standard for age groups, but with an additional breakdown of the ‘aged 75 years and over’ group into ‘aged 75 to 84 years’ and ‘aged 85 years and over’. The additional detail for older age groups is used because cold-associated mortality is greatest for older age groups. Greater uncertainty with small numbers of deaths does not allow using smaller groupings below the age of 65. Rates per million population were calculated using ONS mid-year population estimates aggregated to the same groups.

Grouping by sex was done based on sex as recorded on the death certificate. This is the sex as reported by the informant to the registrar, and may differ from legal sex, sex as recorded in health records, or gender identity. Deaths with unknown or other sex recorded were excluded from the breakdown by sex due to small numbers. Rates per million population were calculated using ONS mid-year population estimates by sex.

Grouping by place of death was done based on the coding of communal establishments on death certificates. These were aggregated to the 5 statistical categories used by the National End of Life Care Intelligence Network (PDF):

• Own residence
• Hospital
• Care home
• Hospice
• Other places

Deaths with unknown place were grouped with ‘Other places’. Rates per million population are not applicable to grouping by place of death.

Grouping by cause of death was done using the underlying cause of death in the ONS deaths data set. This is determined by internationally standardised rules, the Multi-causal and Uni-causal Selection Engine [MUSE] 5.8, for assigning an underlying cause of death based on all causes of death recorded. More information is available in the ONS user guide to mortality statistics.

Causes were aggregated into the policy-relevant groups in the report using the following International Classification of Diseases, Tenth Revision (ICD-10) codes: 

• C00 to C97 for cancer
• F01, F03 and G30 for Dementia and Alzheimer’s
• I00 to I99 for all circulatory diseases
• J09 to J18 for influenza and pneumonia
• J40 to J47 for chronic lower respiratory diseases
• S00 to Y98 for external causes

Deaths with any other cause or unknown cause were grouped as ‘All other’. Rates per million population are not applicable to grouping by cause.

Quality assurance 

The report is produced using R. The production of the figures and the supplementary data tables has been automated, reducing the risk of human error. The code is version-controlled using Git and follows Reproducible Analytical Pipelines principles. The code is independently reviewed by an analyst outside the production team, in line with Aqua Book recommendations. The code is run by a second analyst to check that outputs are reproducible. Further quality assurance is done after running the code. Interim and final outputs are sense-checked, and figures and tables are compared with those in related reports by at least 2 members of the production team.

Confidentiality and disclosure control 

Personal and confidential data is collected, processed, and used in accordance with the UKHSA Privacy Notice. All UKHSA staff with access to personal or confidential information must complete mandatory information governance training, which must be refreshed every year. Information is stored on computer systems that are kept up to date and regularly tested to make sure they are secure and protected from viruses and hacking. UKHSA staff do not store data on their own laptops or computers. Instead, data is stored centrally on UKHSA servers. 

Data presented in the ‘Cold mortality monitoring report’ does not relate to individuals. The figures reported are the modelled number of deaths that can be attributed to cold weather, based on the relationship between cold temperatures and mortality over several winters. There is no risk of including data which identifies an individual.

Geography 

All information in the report is presented for England overall. Some information is also provided at regional level.

Quality summary 

Quality means that statistics fit their intended uses, are based on appropriate data and methods, and are not materially misleading. 

Quality requires skilled professional judgement about collecting, preparing, analysing, and publishing statistics and data in ways that meet the needs of people who want to use the statistics. 

This section assesses the statistics against the European Statistical System dimensions of quality.

Relevance 

Relevance is the degree to which the statistics meet user needs in both coverage and content. 

The statistics are relevant for public health professionals and across government. Results are used for national risk assessment such as the National Risk Register for low temperatures and snow, and to develop understanding of the most vulnerable populations in and following cold episodes. This informs guidance and public health action to protect vulnerable groups.

The report contributes to UKHSA fulfilling obligations by reporting on cold-associated deaths to directly support delivery of the Weather-Health Alerts, a core responsibility of UKHSA under the Civil Contingencies Act 2004 as a category 1 responding organisation, and the monitoring of Goal 2 of Adverse Weather and Health Plan.

Accuracy and reliability 

Accuracy is the proximity between an estimate and the unknown true value. Reliability is the closeness of early estimates to subsequent estimated values. 

Cold-associated deaths can potentially be underestimated due to registration delays. This is unlikely to have a significant impact as almost all deaths occurring in the period modelled will have been registered prior to the analysis.

There is some uncertainty in the figures as cold-associated deaths cannot be measured directly but only estimated through modelling. We therefore report 95% confidence intervals around estimates. There is some additional uncertainty in the figures adjusted for flu, due to the flu data only being available at weekly and national level.

Timeliness and punctuality 

Timeliness refers to the time gap between publication and the reference period. Punctuality refers to the gap between planned and actual publication dates. 

The first cold mortality report on winter 2024 to 2025 is being published during the following winter, approximately 11 months after the end of the reference period, November to March of the previous year. Future annual reports will be published in the autumn, with the next report on winter 2025 to 2026 anticipated to be published in autumn 2026. This is timely as it is before the beginning of the new CHA season and allows time for the results of the report to inform preparedness and planning for the next winter. 

The annual reports are official statistics and are pre-announced at least 28 days in advance. Provisional publication dates for the year ahead are pre-announced online in December and can be found on the UKHSA release calendar.

Accessibility and clarity 

Accessibility is the ease with which users can access the data, also reflecting the format in which the data is available and the availability of supporting information. Clarity refers to the quality and sufficiency of the metadata, illustrations and accompanying advice. 

Tables and visualisations in the report meet accessibility standards.

Coherence and comparability 

Coherence is the degree to which data that is derived from different sources or methods, but refer to the same topic, is similar. Comparability is the degree to which data can be compared over time and domain.

UKHSA has also produced estimates of cold-associated mortality in each winter through the annual influenza surveillance reports, using a different modelling method and definition of cold weather.

The influenza reports use weekly data, and only consider the impact of extremely cold weeks on mortality, defined as an entire week with average temperature below 3°C. This ‘Cold mortality monitoring report’ instead provides an estimate of the impact of cold weather episodes on mortality, including some more moderate or shorter periods of cold. The definition in the ‘Cold mortality monitoring report’ (2 consecutive days below 2°C) is chosen for its relevance to CHA thresholds. Despite these differences, during methodological development of this report, estimates from the 2 methods were found to be coherent.

ONS published a one-off report on climate-related mortality in England and Wales: 1988 to 2022 as experimental statistics. Within this report an estimate of cold-attributable mortality for each year within the series was provided.

The estimates produced in this analysis are not directly comparable with those published by ONS, as the 2 approaches are designed to assess different aspects of cold-related mortality. This report focuses on deaths occurring during specific cold weather episodes in recent winters linked to CHA thresholds. It uses methods that capture short-term delayed effects and allow for disaggregation according to factors such as influenza, age, sex, region, place of death, and underlying cause. As a result, the estimates reflect current patterns of risk and vulnerability under today’s climate, population, and health system conditions. In contrast, the ONS estimates are based on a much longer time period and describe the average effect of cold weather over 35 years, using a statistical definition of cold days applied across the whole period.

The ONS results therefore provide a long-term, smoother picture of cold-related mortality, while the findings presented in the ‘Cold mortality monitoring report’ describe the estimated impacts of specific recent cold weather episodes. The figures should be interpreted in the context of their different aims and cannot be compared directly.

Uses and users 

Users of statistics and data should be at the centre of statistical production, and statistics should meet user needs.

This section explains how the statistics are used, and how we understand user needs. 

Appropriate use of the statistics

These statistics can be used to: 

• monitor national annual trends in cold episode days and cold-associated deaths

• compare cold-associated deaths between different areas, population groups, settings and causes

• monitor progress against the goals of the UKHSA Adverse Weather and Health Plan to reduce mortality related to adverse weather.

Known users 

Public health and emergency planning professionals, at national, regional and local levels. 

UK Government bodies involved in monitoring of climate change adaptation and impacts on health as a result of adverse weather events.

User engagement 

We are carrying out a user survey to gather your views and shape the future of our publication.

In addition, engagement will focus on proactive communication and targeted user engagement to support clear understanding and appropriate use of the statistics upon release. This will include coordinated press activity, publication of blogs to explain the purpose, methods and limitations of the output, and early engagement with regional and local stakeholders. Two-way feedback will be supported through regional networks and direct engagement with stakeholders, alongside wider dissemination through established forums and conferences. Ongoing methodological scrutiny and peer review will support continuous improvement and alignment with the Code of Practice for Statistics.

We also informally collect and review user feedback. Our user engagement activities include:

• stakeholder webinars such as annual summer preparedness and winter preparedness

• reviews of queries received by the Extreme Events and Health Protection team from stakeholders

• stakeholder workshops on climate-health metrics development run by the UKHSA Centre for Climate and Health Security

Related statistics 

Annual surveillance of influenza and other seasonal respiratory viruses in the UK reports on mortality associated with influenza, COVID-19 and cold weather. This report relies on weekly data and focuses only on extremely cold weeks, defined as a full week where the average temperature is below 3°C. In contrast, this ‘Cold mortality monitoring report’ uses daily data and looks at defined cold weather episodes, including shorter or less extreme periods of cold. This approach aligns with Cold Weather Alert thresholds and allows us to capture impacts that might be missed if we only looked at very severe, prolonged cold. Despite these differences, the results from the 2 approaches are broadly consistent.

In 2023 ONS published experimental statistics on climate-related mortality in England and Wales, 1988 to 2022. This analysis provides an estimate of the average impact of cold weather over 35 years. The ‘Cold mortality monitoring report’ instead focuses on recent winters and estimates the impact of specific cold weather episodes under today’s conditions, including today’s population structure, health system pressures, and patterns of vulnerability. The ONS work provides a long-term, broad view of average cold-related mortality over decades, while this report provides a more detailed and up-to-date picture of the impact of recent cold spells. Because the methods and purposes differ, the figures should not be directly compared, but together they provide complementary insights into how cold affects health in England.

The annual heat mortality monitoring reports provide information on deaths observed during heat episodes each year to inform public health actions. This differs to the ‘Cold mortality monitoring report’, which is based on statistical modelling of how mortality has responded to low temperatures over 5 recent winters.

The weekly all-cause mortality surveillance reports compare the actual number of deaths in England compare with the expected numbers of deaths for each week.

Other assessments of mortality include the number of weekly deaths registered in England and Wales, which is published by ONS.

The Office for Health Improvement and Disparities also produces the Excess mortality within England report, which provides estimates of expected deaths by month of registration for population subgroups and by cause of death.

The different methods used in the UK for mortality assessment, and their varied purposes, are discussed in more detail in Measuring excess mortality: a guide to the main reports.