Modelling

This chapter describes selected aspects of the development of a model for quantifying the health impacts of HEE improvements. This development has built on preceding research by the authors and other teams and has, in part, been described elsewhere.32 In this chapter two pieces of epidemiological research (methods studies) are reported that have been important in helping to improve the model and its parameterisation and illustrate its application for a policy analysis. The methods studies relate to the quantification of the loss of life expectancy in cold-related death (given questions about the possibility of short-term mortality displacement) and the impact that housing adaptations (and specifically energy efficiency measures) may have on heat risk.

Methods study 1: quantifying life lost as a result of cold and heat

Article extract reproduced from Rehill et al.49 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/http://bmjopen.bmj.com/content/5/4/e005640.

The objective of this methodological analysis was to clarify the degree to which deaths in England associated with hot and cold days are confined to frail people who would have died anyway in the next few weeks or months.

The context is well known. There is ample evidence that mortality increases on and immediately after hot or cold days,3,50–52 and such evidence is used to support public health action.53,54 However, what is less clear is the degree to which life is shortened in temperature-related mortality. This has some importance, as it is possible that the excess of deaths might occur wholly or partly among very frail people who are approaching the end of their life. If so, the degree of shortening of life would be small,55 a phenomenon commonly referred to as ‘harvesting or mortality displacement’, and the excess of deaths sometimes referred to as ‘deaths brought forward’.

There is some evidence to support the occurrence of harvesting, particularly with respect to heat-related deaths.2,56,57 It may also be relevant that the effect of temperature on mortality is greatest among the elderly50,56,58 and perhaps also among those with existing comorbidities,50 both are groups who would be expected to have a relatively short life expectancy. However, the extent to which harvesting accounts for the excess of temperature-attributable deaths remains unclear.

Some daily time-series studies have used distributed lag and related approaches to estimate very short-term harvesting (roughly ≤ 1 month) by looking for mortality decreases in the weeks after hot spells. Estimates from such studies suggest that harvesting may account for around 70% of all heat-associated deaths in 15 European cities59 to fewer than 10% of deaths during the 2003 heatwave in France.60 The proportion may depend on a variety of factors, including the severity of the heatwave.

There is less evidence of short-term mortality displacement for cold-associated deaths, possibly because the longer lag associated with cold effects makes the distributed lag approach to estimating harvesting problematic.55 There is also the methodological issue that it is difficult with daily studies to identify harvesting beyond a month or so61 because of noise and the need to control for seasonal and other temporal variation that is not a result of temperature.

Some studies have demonstrated that higher than average overall deaths in winter lead to lower daily temperature effects on mortality the following summer62,63 (a phenomenon that might be interpreted as indicating a relatively finite pool of individuals who are vulnerable to temperature-related mortality). A cohort study found that years of high summer temperature variability were associated with high mortality.55 These studies bring indirect evidence that some temperature-related deaths are displaced by at least several months but not direct evidence of the extent of displacement as a result of heat and cold.

Estimates of the seasonal excess of mortality in winter are less affected by short-term mortality displacement. The Excess Winter Deaths Index approach used by the UK’s ONS identifies excess mortality over the 4 months of ‘winter’ (defined as December to March), compared with the preceding 4, and following 4 months. However, these excesses could still be biased by displacement from winter to non-winter in the same year. Moreover, an appreciable fraction of the excess winter deaths may be attributable to seasonal factors other than temperature, such as respiratory infections, influenza and behavioural changes.

Given the limited evidence on the extent of mortality displacement as a result of heat and cold from previous studies, this study set out to avoid some of the limitations of current techniques by applying time-series methods to investigate if annual death counts have been associated with annual summaries of temperature extremes, an analysis that that would find association only if those deaths would not otherwise occur in the same year (as the year of high or low temperatures) and so provide evidence on a lower limit of the degree to which temperature-associated deaths represent short-term displacement.

Methods

The study was based on deaths among London residents from all natural causes from September 1949 to September 2006 using two sources of mortality data to cover the whole study period. For 1949–75, we used digitised weekly counts of death registrations published in print by the Registrar General,64 supplementing 1950–64 data from a previous study.16 For 1976–2006, we used daily counts of death collated for other studies65,66 and originally obtained from ONS. The data were for the London Administrative County for 1949–65, and for the larger Greater London thereafter. For all years, we retrieved deaths as a result of all natural causes, cardiovascular causes and respiratory causes. The daily counts were collapsed into weeks to create a complete series of weekly mortality for the entire 57 years of study, starting from 2 October 1949.

Weekly counts were aggregated into years commencing in autumn (which we defined to begin in the first week of October), rather than using conventional calendar years, so as to align temperature and mortality data for the same winter season (instead of artificially splitting winter at 1 January and noting that cold can be delayed by as long as 3 weeks2). Heat-related mortality effects are predominantly seen within 1–3 days,50 so the October boundary would mean that each year included most deaths resulting from hot weather in the warm season months up to and including September. Weeks were numbered sequentially from the start date and organised into years of 52 or 53 weeks, with each year starting in early October, but then dropping all weeks numbered as the 53rd to leave summed counts of deaths for 57 52-week years, for simplicity in the regression model.

Boundary and International Classification of Diseases classification changes

The change in the administrative definition of London in 1966 led to a sharp artificial jump in death counts, which we modelled using two indicator variables to make ‘steps’ going into 1965–6 and 1966–7. Other, smaller discontinuities considered in sensitivity analyses were similarly allowed for, as were discontinuities in International Classification of Diseases67 classifications (affecting only cause-specific analyses) for which no bridging code was available.

Meteorological data

Daily temperature data were derived from the Heathrow Airport meteorological monitoring station (the only station with data available for the whole period), which were downloaded from the British Atmospheric Data Centre (now the Centre for Environmental Data Analysis; see artefacts.ceda.ac.uk/badc_datadoes/index_test.html). To capture heat or cold exposures in one annual summary statistic, we first computed daily measures of ‘heat’ and ‘cold’ [i.e. degrees Celsius above (below) the heat (cold) threshold]. For simplicity, we assumed a V-shaped temperature–mortality model common in daily time-series studies in which there is a common threshold for heat and cold effects. To inform the choice of the minimum mortality temperature (apex of the V; common cold and heat threshold) we reviewed 11 previous studies examining temperature and mortality in London, but found no clear consensus among studies, probably because of differences in methods (e.g. lags, shape of model). For our primary analysis we chose a value of 18 °C of daily mean temperature as a common threshold both for cold and heat, as it approximated the mean over the reviewed studies. These ‘heat-degrees’ for each day were the number of degrees Celsius above 18 °C of the daily mean temperature, whereas ‘cold-degrees’ was the number of degrees below 18 °C of the daily mean temperature. The annual means of these measures, ‘annual-heat’ and ‘annual-cold’, were used in our analyses. Owing to uncertainty over the common threshold temperature, the choice of threshold was prioritised for exploration using sensitivity analyses.

Statistical analysis

We carried out a quasi-Poisson time-series regression analysis, with yearly all-cause natural deaths as the outcome, and the main exposures of interest being annual-heat degree days and annual-cold degree days. We undertook a primary analysis based on a model informed by a priori judgement but explored sensitivity to assumptions in that model.

We used indicator variables to account for steps in the mortality series in 1965–6 and 1966–7 owing to a boundary change (discussed in Boundary and International Classification of Diseases classification changes). In the cause-specific analyses we also included four further indicator variables to reflect steps expected because of International Classification of Diseases classification changes. To control for long-term trends in the model, we included a natural cubic spline function assuming 1 degree of freedom for every 10 years of data (5 degrees of freedom in total), equating to a roughly 10-year moving average. We chose this degree of flexibility by judgement to allow control for gradual changes in population size, age structure and death rates, while leaving enough variability to use in analyses.

We adjusted for influenza epidemics by including as explanatory variable the proportion of deaths each year that were classified as caused by influenza.

Alternative model assumptions, including different degrees of confounder control, were considered in sensitivity analyses.

For comparison with the annual time-series estimates, we also undertook a simple time-series regression of weekly counts following conventional methods, using the same heat-degrees and cold-degrees daily measures aggregated to weeks (‘week-heat’ and ‘week-cold’). We controlled confounding by seasonal and other time-varying risk factors by stratifying by year and month (344 strata) and using a conditional quasi-Poisson model (equivalent to a time-stratified case crossover). Because of the known lag between cold and mortality excess, the cold variable included in the model was the mean of the daily cold-degrees over that week and the previous one.

All analyses were performed in Stata, version 11.2. The annual data set and core code are available from the Dryad Digital Repository (URL: http://dx.doi.org/10.5061/dryad.02k83).49

Results

There were a total of 3,530,280 deaths from natural causes in the 57 years of our analysis: 1949–50 to 2005–6. Allowing for the sharp increase in the annual count of London deaths attributable to the change in the definition of London’s administrative boundary in January 1966, there was evidence of a gradual decline in counts of death from about 1970.

Mean daily temperature exceeded the threshold of 18 °C on 11.1% of days and was below this threshold on 88.7% of days. For each year during the study period, the mean cold-degrees over the year (degrees below 18 °C) was, on average, 7.4 °C and varied between 6.2 and 9.0 °C; the mean heat-degrees (degrees above 18 °C) was much lower (0.2 °C) and varied only from 0 to 0.6 °C.

The results for the estimated increase in mortality for each degree of cold and heat across the year, as determined by the regression model, are shown in Table 6. Cold years were associated with increased deaths from all causes. For each additional degree of cold across the year, all-cause mortality increased by 2.3% (95% CI 0.7% to 3.8%), after adjustment for influenza and secular trend. The effect of cold was greater in those aged ≥ 65 years, and for deaths from cardiovascular disease (2.9% per degree) and respiratory disease (7.6% per degree), but the CIs for each of these subgroup analyses were wide.

The estimated increase in mortality with heat (1.7% per degree) was also very imprecise (95% CI 2.9% to 6.5%), as were the age- and cause-specific heat-related mortality estimates, which makes comparisons unreliable.

A range of sensitivity analyses (reported in Appendix 2) showed that only when the spline used to allow for mortality trends over time was very inflexible (three knots) did the association of mortality with cold change substantially (losing statistical significance). This model, however, fitted the data appreciably less well by Akaike information criterion. The association of heat with mortality was more sensitive to model choices but estimates of mortality increment because of heat remained very imprecise in all models. Residual analysis of the main model did not suggest problems of poor model fit.

The comparative conventional month-stratified time-series analyses on weekly data for the same period gave all-cause estimated cold and heat effects as 2.0% (95% CI 2.0% to 2.1%) and 3.9% (95% CI 3.6% to 4.1%) per degree Celsius below and above the 18 °C threshold, respectively. CIs were much narrower than for the annual data estimates and point estimates for age- and cause-specific deaths showed clear pattern of higher RRs in the elderly and in deaths owing to cardiovascular and, in particular, respiratory disease.

Discussion

These analyses provide evidence on a central question for interpreting the public health importance of temperature-related mortality, namely the degree to which heat- and cold-related deaths can be attributed to a very limited shortening of life of perhaps just days or weeks. If the majority of temperature-related deaths were explained by such harvesting, most observers would consider temperature-related mortality to be much less concerning than if there were evidence for a reduction of life by months or years. This analysis does not provide a full answer to this question but suggests a lower limit. Specifically, we found that, over six decades in London, colder years were associated with increased mortality, with 2.3% (95% CI 0.7% to 3.8%) more deaths occurring per degree of cold during the year. This suggests that cold-related deaths were brought forward by at least around half a year. Had the excess of daily deaths associated with preceding weeks of cold weather (which typically occurs in the UK between December and March) been ‘due to occur’ within the same year as defined in this study – that is, before the following October – no association between cold and mortality would have been detected. Our findings were generally robust to variations in the modelling assumptions and regression techniques used.

To our knowledge, no other studies have looked at long-term mortality in relation to long-term temperature defined on an annual basis. The results of our analysis for heat are broadly supported by the cohort analyses that show that years of high summer temperature variability are associated with high mortality.68

It is instructive also to compare our estimates with those from published daily time-series and related studies. Two London studies57,69 used the same cold threshold as us (18 °C). If all the excess deaths in these daily studies had been displaced from beyond the next October and there were no longer lag effects, they should reveal the same increments of risk as in the annual studies, setting aside the slightly different time periods of the studies (1988–92 and 1993–6). The daily studies both found increments of daily mortality of 1.4% per degree below 18 °C, broadly similar to the estimates based on our subsidiary weekly analysis (2.0% per degree Celsius). The fact that our estimate from analysis of annual data was positive, and also similar to our own weekly and the other published estimates, strongly suggests that (most of) the excesses in the daily and weekly studies are, indeed, displaced from beyond October (i.e. from the next year). The fact that the annually-based estimate is somewhat larger might also be interpreted as giving some suggestion that there may be adverse mortality effects operating at much longer lags than were detectable in the daily studies (lags of 3 days and 2 weeks), although the difference could be a result of chance, as 1.4% and 2% falls within the wide CI for this study (95% CI 0.7% to 3.8%).

Estimates for the impact that heat had from this study were, unfortunately, very imprecise. In particular, the CI for increase in mortality per degree above 18 °C (95% CI –2.9 to 6.5%) includes all of the estimates made in daily studies, including the one found in the one daily study57 using the same 18 °C threshold (1.3%), and also the estimate presented in the current study from weekly data (3.9%). This limit in precision precludes making any conclusion from this analysis on the presence of absence of short-term harvesting for heat-related deaths. The precision limitation is a result of the limited number and variation of hot days annually in London. Annual studies in cities with more days above their heat thresholds would probably have more power.

As in all epidemiological studies, our study was potentially subject to residual confounding by uncontrolled risk factors that change over time, including, for example, changes in the size and demographic structure of the London population, as well as declining age-specific deaths rates owing to changes in health care and risk factors (e.g. smoking). In common with other time-series regression studies,20 we relied mainly on a smooth (spline) function of time to control for these, under the assumption that changes in such time-varying confounders would probably be smooth. Although we cannot rule out residual confounding, it is reassuring that increasing the flexibility of the spline or adding further ‘step’ functions changed the estimated cold effect very little. The much lower estimate for cold when using a less flexible spline (see Appendix 2) has little credibility given the much poorer fit of this model (Akaike information criterion).

How and whether or not to control for influenza is problematic, as it is possible that it might be on the causal pathway between cold and death, although there is no direct evidence that cold is a direct cause of influenza. Our approach, using the proportion of total deaths resulting from influenza as covariate in the regression is, we believe, conservative.

Air pollution is an established risk factor and also changes over time. We were unable to control for this, as we were unable to assemble consistent data for the complete period. However, data on black smoke for the period 1976–2003 were available,70 and an analysis revealed a weak negative correlation between mean annual cold and black smoke during this time, after accounting for year (r = –0.25). This suggests that confounding from air pollution is unlikely to have been substantial, and control for it would be more likely to increase estimates of cold effects.

Despite the large population of London and the long duration of our series, power and precision was limited for some analyses, especially for heat effects, and for age- and cause-specific analyses, the results of which were too imprecise to allow much interpretation, even if point estimates followed broadly the expected pattern.50,71,72 The more precisely estimated patterns that we found in the weekly analysis match those found by others.

Finally, our approach to summarising cold and heat in each year is not the only one possible. We chose it so as to maximise the link between this study and daily studies. However, in fact, the annual measure of cold adopted (degrees below 18 °C) was very highly negatively correlated, in temperate London, with the annual mean temperature (r = –0.97). The heat measure was also correlated, but less strongly so (r = 0.81). Thus, analyses using mean annual temperature would have resulted in very similar estimates of cold effects and broadly similar heat effects.

The importance of this evidence is that it suggests that, in the UK, most people who die as a result of cold in daily time-series studies lose at least 6 months of life expectancy, and possibly much more. Our methods do not allow us to derive an estimate of the actual degree of life shortening, but rather they help to define a lower limit for it. The fact that it appears to be no less than around 6 months strengthens the case that cold-related deaths are a serious public health problem and hence that policies aimed at reducing vulnerability to cold through such measures as home insulation have important benefits to health and specifically to life expectancy. This has a bearing on the quantification and valuation (monetisation) of the health gains from improved protection against cold-related mortality.

Methods study 2: dwelling characteristics as determinants of heat risks to health

This second methods study examined the relationship between dwelling characteristics and the risk of heat-related death. The context here is that energy efficiency measures that improve the thermal performance of a dwelling in protecting against outdoor cold are also likely to influence indoor temperatures during hot weather. Previous analyses have examined indoor heat risk assuming the same relationship between temperature and mortality as observed with outdoor temperature.73 This suggests that dwelling characteristics are as, if not more, important, for health [e.g. the urban heat island effect, the name given to the increment of temperatures in urban centres compared with surrounding areas resulting the thermal properties (heat absorption, capacity, conductance and albedo) of the surfaces and materials found in urban landscapes, the reduced evapotranspiration from reduced natural vegetation, and the waste heat produced from anthropogenic activities].74

Methods

Analyses were based on mortality records (n = 322,840) of people aged ≥ 65 years dying in the summer months (June, July, August, September) of the years 2001–6. These records were linked to dwelling characteristics by the full unit (seven-digit) postcode of residence (212,110 postcodes) and to outdoor temperature data (daily maximum temperature in degree Celsius at lags of 0–1 days) using data from a national-scale Weather Research and Forecasting (see www.mmm.ucas.edn/weather-research-and-forcasting-model) weather model with spatial resolution of 5 × 5 km. The association between mortality and maximum daily temperature was based on a time-stratified case crossover design. There were two status analyses: conditional logistic regression to estimate region-specific heat slopes and their interaction with a dwelling overheating index, and subsequent meta-analysis of the results with random effects across regions.

Overheating index

The derivation of the dwelling-type specific overheating index was based on building physics simulation using the validated modelling software EnergyPlus version 8.3 (US Department of Energy, Washington, DC, USA). Using input data from the EHS,47 this was used to calculate a temperature anomaly for representative dwelling archetypes. The anomaly was defined as the difference between the mean indoor temperature [mean (T_indoor)] and the mean outdoor temperature [mean (T_outdoor)] on days when the 2-day average of daily maximum temperatures, T_{max0–1, outdoor} was above the 97.5th centile of the temperatures distribution (the anomaly for ‘extreme heat’) and on days when the 2-day average of daily maximum temperatures T_{max0–1, outdoor} was between the 93 and the 97.5 centile (‘moderate heat’). For each of 16 dwelling types, eight indices of anomaly were computed, reflecting 2 × 2 × 2 combinations of moderate/extreme heat, daytime/night-time and bedroom/living room temperature.

These archetype-specific temperature anomalies were then linked to (geographically-referenced) Address Point Data Ordnance Survey, Southampton, UK; see www.ordnancesurvey.co.uk/business-and-support/productsaddress-point.html accessed 3 August 2018) to classify dwellings by postcode location, to which mortality data were also linked. This enabled postcode locations and, thus, mortality data to be classified with respect to the overheating potential (temperature anomaly indices) for their place of residence (Figure 22). The mortality data so classified were analysed for the association between mortality risk and outdoor temperature and the statistical interaction with the overheating index.

FIGURE 22 Illustration of the classification of postcode locations with respect to dwelling type and overheating indices (temperature anomalies). Orange and red dwellings have positive temperature anomalies. Based on data from OS MasterMap® Building Heights [FileGeoDatabase geospatial data], Scale 1 : 2500, Updated: 29 November 2014, Ordnance Survey (GB), Using: EDINA Digimap Ordnance Survey Service, <http://digimap.edina.ac.uk>, Downloaded: 2016–09–15 12:27:53.213 and Versik Analytics® UK Buildings [FileGeoDatabase geospatial data], Updated: 1 November 2014. (Figure not reproduced here; see the original article.)

Results

The overall (regionally ‘averaged’) slope for the association between high temperature and mortality was a 3.3% (95% CI 2.5% to 4.0%) increase in mortality risk for each 1 °C rise in temperature (lags of 0,1 days).

Table 7 shows the summary results of the modification of this relationship with each of the eight temperature anomaly indices. The pattern of results is mixed. Only one coefficient for effect modification was nominally statistically significant, namely that relating to the temperature anomaly for the daytime bedroom temperature on very hot days. This indicates a 1.34% (95% CI of 0.37% to 2.32%) increase in the temperature–mortality slope for each degree Celsius increase in temperature anomaly. Therefore, this suggests that people living in dwellings that are likely to develop higher bedroom temperatures during the daytime in very hot weather are more vulnerable to heat-related death.

However, it should be noted that across all eight coefficients the pattern of results does not seem consistent. The point estimates for five out of the eight effect modification coefficients are negative (although with CIs including zero).

Discussion

This analysis was an attempt to obtain direct empirical evidence that people living in dwellings that are more likely to overheat during hot weather would have greater risk of heat-related death. Overall, its results provide only weak, although plausible, evidence that such modification of risk does occur, and specifically that those living in houses in which the bedroom temperature increases more than in most other houses during periods of very hot weather are more vulnerable to heat-related death.

It is not possible to draw a very robust conclusion from this result, especially given that only one out of eight different coefficients examined showed evidence of effect that was clearly different from the null. The overall pattern of results suggests that the evidence for modification of heat risk by dwelling characteristics is greater on very hot days and on hot days and for markers relating to bedroom rather than living room temperatures and for daytime rather than night-time anomalies. Arguably this is a mismatch with expected time–activity patterns. Given the rooms that people are most likely to occupy different times of day, a priori prediction would suggest that the most sensitive markers would be those reflecting daytime living room temperatures or night-time bedroom temperatures.

Nonetheless, it seems very probable on physiological grounds that the risk of heat-related mortality would be greater in dwellings that are likely to exhibit higher indoor temperatures during hot weather. It is also worth remembering that this analysis depends on indirect methods of classification for heat exposure, specifically using building physics models to classify the overheating propensity of different dwelling types and then to assign dwelling types to postcode locations to which mortality data were linked using the postcode of residence. All of the steps entail an imperfect link and hence there is likely to be a degree of misclassification, which is likely to introduce a conservative bias (i.e. against finding a modification effect using such indirect markers). This, and the plausibility of genuine effect modification, adds some weight to the interpretation that the observed association reflects a true causal mechanism, even if it is imprecise.

Therefore, the evidence is far from conclusive but does offer a weak indication of effects that would be entirely compatible with reasonable expectation. The evidence needs to be further strengthened and made more precise if it is to be truly useful in helping to inform assessments of the role of building characteristics as determinants of heat-related health risk. Its importance is perhaps more qualitative. Energy efficiency measures more often act to increase, rather than decrease, the propensity to overheating and, thus, could represent a potentially important unintended adverse consequence of retrofit energy efficiency upgrades that may come to have increasing significance in the context of increasing temperatures under climate change. At present, our best judgement is that level of increased risk is relatively modest, at least while summer temperature distributions are not much more extreme than the present-day average. Helping to improve the empirical evidence on this risk would, however, be an important goal for future research, and an immediate precautionary practical step would be to consider the additional measures that might be included alongside energy efficiency upgrades to help protect the indoor environment against high temperatures in summer. This would be an important adaptation in any case, given the likely increase in summer temperatures as this century progresses.

Model application: comparison of winter fuel payments and home energy efficiency interventions

To illustrate an application of the health model, we present a model-based analysis comparing the current policy on WFPs with subsidised HEE investments in terms of health and other outcomes.

Among the UK’s most formidable long-term challenges, shared by many countries, are the public health burdens of an overconsuming and increasingly ageing population and arguably the ‘biggest global health threat of the 21st century’ climate change.75 Step changes will be needed to achieve vital policy goals, especially in the transition to a low carbon economy,76 but they present major opportunities for improving population health, particularly if synergies between policy areas can be exploited.

Nowhere is this better exemplified than in the housing sector, an important focus of public policy, both because of the probable contribution of inadequate home heating to the large and persisting burdens of winter-related mortality and morbidity8 and because of its central importance to the abatement of GHG emissions.76 Substantial net benefits for health could be achieved if an accelerated programme of HEE were widely implemented,13 assuming appropriate safeguards.77 However, progress to date has not kept pace with stated objectives.

We examined how such benefits could be more rapidly realised. Specifically, we compare two strategies that seek to have a similar impact on health and household incomes but through methods that, on environmental grounds, are diametrically opposed: (1) a policy of WFPs, introduced as part of the government’s initiative to tackle FP among pensioners, and whose inevitable consequence is to increase use of heating fuels through subsidy; and (2) an expanded programme of HEE improvements. Our intention is to test the potential to accelerate the attainment of health and environment goals at acceptable cost by more efficient alignment of policy approaches.

Methods

We compared two specific forms of policy:

1. A policy of fuel subsidy based on the 2014/15 entitlement of the UK government’s Winter Fuel Allowance scheme: £200 per household per year for households containing someone born on or before 5 January 1952 or £300 per household per year for households containing someone over the age of 80 years. The £200 WFP is equivalent of 4000 kWh of heating energy at our assumed 2015 price of £0.05 per unit,78 if all were spent on improving indoor temperatures.
2. An expanded programme of HEE improvement, installing in each qualifying dwelling all energy efficiency interventions (including loft insulation, double glazing, solid and cavity wall insulation, boiler replacement and gas central heating) when such measures are currently (2011 data) absent or suboptimal. We assumed that ‘qualifying households’ were the same as those entitled to receive the Winter Fuel Allowance (policy 1), ensuring the same target households and, thus, direct comparability of the two strategies.

We considered timetables for the HEE improvements based on reinvesting savings from WFPs, entitlement to which was assumed to lapse once a dwelling had undergone a full energy efficiency upgrade. However, even with substantial seed money in year 1 of £0.7B, half of the current WFP annual budget for England, simple calculations showed that the investment programme would not be complete until after 2090. Therefore, we assumed a substantial upfront capital investment of £20B to allow the upgrade to all 4.5 million target dwellings over 5 years. This equates to a rate of upgrades of around one dwelling per minute over 5 years, which is the rate required for the whole stock to 2050 to meet climate change abatement targets. The annualised cost of this capital investment over the lifetime of the upgrades is very similar to that of current WFPs, assuming continuing very low rates of interest.

The impact had on the indoor environment and health

The impact that the outlined policies had on health and other outcomes was estimated by refinement of a model, HIDEEM,31 originally developed for use by the DECC to assess the impact had on health of energy efficiency investments in the housing sector. Core to the model is the 2010/11 EHS,79 which contains data on 16,000 dwellings in England and their 31,000 occupants. The indoor environment (temperatures, indoor pollutants) and energy performance [heating energy and carbon dioxide (CO₂) emissions] of EHS dwellings were modelled by linking their data with that of simulations carried out in a validated building physics and airflow model, CONTAM, developed at the US National Institute of Standards and Technology.33 Simulations were carried out for 16 representative archetypes of dwelling forms (seven types of houses, nine types of flats), under a range of assumptions about air permeability (‘building leakiness’) and ventilation regimens, and the results were matched to the 2010/11 EHS on the basis of dwelling type, floor area and notional permeability. The result was a (modelled) distribution of indoor environmental conditions (indoor temperature and air pollutants) for a representative sample of English dwellings, which was then ‘perturbed’ to estimate changes in personal exposures following policy interventions.

Changes in winter indoor temperature and fuel use were determined by a method based on the empirical relationship between indoor temperature and the E-value of the home, as described in Hamilton et al.26 and previously applied in modelling for the National Institute for Health and Care Excellence.80 The E-value is the heating power input (watts) required to maintain a 1-kelvin temperature difference between the indoor and outdoor environments and is, thus, an indication of whole-house energy efficiency. This empirical relationship, shown in Figure 2, was derived from temperature measurements of dwellings that underwent energy efficiency upgrades as part of the Warm Front scheme.26 A very similar functional form has more recently been obtained by the authors from data in the 2011 EFUS,30 as yet unpublished. The gradient of the relationship between the SIT and E-value indicates the degree to which the benefits of energy efficiency are taken as having higher indoor temperatures and how much as energy/cost saving. With an assumed average cost for heating energy of £0.05 per kWh, the relationship provides a basis for calculating the change in (winter) indoor temperature associated with WFPs, depending on the E-value of the dwelling (but see also Beatty et al.81).

Quantification of health impact

The health impact that changes in indoor air quality and temperature had was modelled using life table methods similar to those of the IOMLIFET model,45 which estimates patterns of survival in the population over time using age- and cause-specific mortality rates. Life tables were set up using 2010 age-specific population and (disease-specific and all-cause) mortality data for England and Wales from ONS, with separate life tables for men and women. We modelled changes in five exposures (indoor temperature, STS, indoor and outdoor sources of PM_2.5, radon and mould) and the selected mortality outcomes listed in Table 2. Morbidity impacts for these same outcomes were estimated from the mortality estimates by applying age- and cause-specific ratios of years of healthy life lost owing to disability (YLD) to the overall YLL derived from the World Health Organization’s Global Burden of Disease data.

Capital costs for HEE improvements and the costs of energy for home heating were based on data published by the Department for Communities and Local Government (now the Ministry of Housing, Communities and Local Government).82 We did not include costs associated with modelled changes in contacts with the NHS.

From the models we determined four parameters of impact that are associated with the WFP and HEE policies: health (the sum of QALYs for all modelled health effects relating to changes in temperature and indoor air quality), CO₂ emissions (based on assumed trends in the carbon intensity of domestic fuel use), energy costs adjusted for projected trends in domestic fuel mix and price and a measure of socioeconomic differentials in impact. These were assessed over time horizons of 5 years and 20 years, with discounting of costs and benefits at 3% per annum.

Results

Across England as whole, we estimated that a total of 4.5 million dwellings were occupied by households qualifying for WFPs. The comparison of the effects of the two policies is summarised in Table 8. For all of the parameters listed, the HEE programme appears more beneficial than using equivalent funding to perpetuate the WFPs.

Our scenario proposes fairly full implementation of all reasonably straightforward retrofit HEE measures, which would probably achieve close to 0.4 °C increase in SIT, which is nearly twice the temperature increase achieved even if WFP recipients spend that money on additional heating.

In terms of health benefits (measured as QALYs), again the HEE programme appears to offer greater gains than those attributable to WFPs. However, one of the clearest and very obvious differences is that the HEE programme reduces the stock energy demand and associated CO₂ emissions, whereas the WFPs act to increase them by encouraging householders to burn more (or effectively subsidise) fuel.

Not everyone would win by comparison with regard to status quo, as illustrated by the plot of the expected distribution of temperature changes comparing HEE and WFP programmes (Figure 23). The premise here is that the universal WFPs would be withdrawn as dwellings receive full energy efficiency upgrades.

FIGURE 23 Distribution of change in SIT with the HEE relative to WFPs on the assumption that currently universal WFPs would be withdrawn following the installation of HEE measures. (Figure not reproduced here; see the original article.)

Discussion

Comparing the effect on health and GHG emissions of WFPs and HEE investments is seemingly very specific, but it illustrates an important point of principle. Burdens of winter- and cold-related mortality and morbidity remain a major problem for the UK. At the same time, by almost any objective measure, the collective global response to climate change, and that of the UK specifically, has been inadequate. No government has yet succeeded in implementing policies of the scale and ambition needed to alter this, which is doubtless in part because of the perceived economic cost to the Treasury, or to society as a whole, of doing so.

However, the analyses that we present here suggest that, for one major section of the UK population, namely elderly people, the UK Treasury already commits expenditure to combat FP at a very similar annualised cost to that needed to transform the energy efficiency of all homes currently occupied by the same population. Moreover, our models indicate that energy efficiency upgrades would have a greater impact on health than WFPs and would decrease rather than increase CO₂ emissions. In other words, redirecting the WFP budget to HEE interventions would achieve all of the avowed aims for which the WFP is provided and also meet the UK’s HEE investment targets for GHG reduction in this part of the housing sector. Admittedly, such a policy would entail an upfront capital investment, but the debt repayments would be almost the same as the current WFP budget if (possibly appreciable) additional administrative costs are ignored. Moreover, as already noted, the accelerated speed of retrofit installations that this would achieve is needed for the whole period up to 2050 to meet climate change mitigation targets. Therefore, in effect, it represents no more than a prioritisation of the elderly in a programme of investment that should already be occurring and continued for 35 years.

Our comparison of the relative benefits of HEE investment is based on the results of a model, and all models are uncertain. In our judgement, however, it is unlikely that those uncertainties materially affect the general conclusion. The need/opportunity for HEE investment is based on analysis of data from a recent EHS. The current cost of WFPs is known, and although estimates of the capital costs of HEE investment vary somewhat, we used reasonably authoritative figures published by DECC. The estimates of the change in winter indoor temperature for both policies were at least in part based on the same empirical relationship and appear consistent with published before-and-after measurements.27 There is also increasing evidence about the health benefits of HEE interventions,9,83,84 even if our estimated changes to the indoor environment are imprecise. It is unlikely that we have so overestimated the relative advantage of HEE interventions in terms of health or energy costs as to reverse the sign, and it seems quite consistent with expectation that HEE investments decrease CO₂ emissions whereas WFPs do the opposite.

Redeploying WFPs would raise various practical and ethical issues. Even if there are more winners than losers, safeguarding some categories of loser (e.g. the poorest households) would seem important. It is likely that there would be additional administrative costs with HEE investments, which we have not attempted to estimate; neither have we taken account of the fact that a proportion of elderly people will move home or suggested what approach to use for subsequent cohorts of people who reach pension age after the 5-year programme. In addition, of course, there are questions of equity and fairness, including the fact that larger homes, generally occupied by wealthier families, would need, and thus receive, more investment (unless capped) and the fact that private landlords would be potential beneficiaries unless the government prescribed this as a mandatory investment cost that private landlords should bear themselves.

The views of potential target householders show that the policy choice would have different importance for and impact on different households. Climate change objectives are not a priority, and poorer families in particular value, and even depend on, the WFP as a universal entitlement. However, there appears to be a degree of acceptance that HEE improvements could offer relative advantages for health, well-being, security and even pride in one’s home.

Whether the intention is to protect health, to ease household budgets or to limit CO₂ emissions, our estimates suggest that more can be achieved through HEE. Even though the argument is more complex with regard to equity, there is a case to favour HEE over WFP if consideration is given to the responsibility of current to future generations.

In conclusion, for most households, HEE improvements of similar annualised cost to current WFPs achieve greater improvements in indoor temperatures and health, and reduce rather than increase CO₂ emissions. Replacing policies that incentivise additional fuel consumption for home heating with a rapid full-scale programme of energy efficiency could help transform the UK stock to achieve important benefits for health and crucial climate change abatement targets without imposing undue additional financial burdens.