Epidemiological studies

Long-term trends in cold mortality

Winter-related mortality, based on the proportionate excess in death rate during December to March compared with other months of the year, has been falling in England and Wales since at least the 1950s (Figure 3).15 There is evidence that cold-related mortality burden has been falling for, most probably, the last century or so.16 This may be a result of a wide range of factors associated with socioeconomic development, including improvements in housing, infrastructure, technology, underlying population health and health care.

FIGURE 3 Number of excess winter deaths and 5-year central moving average, England and Wales: 1950/51–2015/16. The dashed box indicates the years corresponding to those used in the analyses in this report. Based on data from the Office for National Statistics (ONS).15 (Figure not reproduced here; see the original article.)

In a first set of analyses, we examined the pattern of changes in annual burden of cold-related deaths since the mid-1970s. Over that time, evidence from the United Kingdom Housing Energy Fact File 201317 indicates a steady rise in winter indoor temperatures in homes, especially since the mid-1980s (Figure 4). This rise has been seen in parallel in dwellings with and without central heating, but average winter temperatures in homes with central heating have been around 1.5 °C greater than in homes without it. The proportion of homes with central heating (mainly gas) has also risen substantially (see Figure 4b). By 2010, only around 9% of homes were still without central heating.

FIGURE 4 Changes over time in home heating and household costs and expenditure on fuel. (a) Trend in winter indoor temperatures; (b) homes with and without central heating; (c) average (UK) fuel prices (2011 prices); and (d) average weekly expenditure. Based on data tables from the United Kingdom Housing Energy Fact File 2013.17 (Figure not reproduced here; see the original article.)

The cost of fuel (expressed in terms of pence per kilowatt hour at 2011 prices) has fluctuated somewhat, especially for electricity and oil, but overall household expenditure on fuel, lighting and power had been in gradual decline until the early 2000s when there was then a relatively sharp increase (see Figures 4c and d). By 2010/11, household expenditure on fuel was broadly in line with the long-term trend of the last four decades.

Our primary focus was to determine the impact of WFPs as well as fluctuations in fuel price. The WFP is a non-means-tested, tax-free annual cash payment, first introduced in 1997, which is paid to individuals who are in receipt of the State Pension or other qualifying benefits.18 It is paid at £200 per eligible household in which the oldest person is aged < 80 years and £300 for households containing a person aged ≥ 80 years.

Methods

For the period 1975–2012, we based analyses on the following data sets:

1. regional (government office region – a standard administrative region) time series of counts of death by day, stratified by age group and by broad cause-of-death group
2. meteorological data, specifically including daily minimum and maximum temperatures and relative humidity
3. regional weekly data of influenza A and B cases reported to the Public Health Laboratory Service, as well as annual survey-based estimates of influenza vaccine coverage for people aged ≥ 65 years
4. national data on domestic fuel costs, adjusted to 2011 prices, as well as data on trends in home heating, including the proportion of homes with central heating and indoor temperatures, derived from the United Kingdom Housing Energy Fact File 201317
5. year indicators as markers of the periods covered by WFPs (first introduced from the winter of 1999/2000).

The association between daily mortality rates and outdoor temperature was analysed using Poisson regression methods, adjusted for long-term trend and seasonality, day of the week and reported influenza cases. Model parameterisations were informed by recent research.1,3,19 Adjustment for trend and seasonality was based on use of cubic spline curves generated using an adaptation of Stata^®’s (StataCorp LP, College Station, TX, USA) frencurv command, specifying seven knots per year in keeping with previous studies, as a compromise between providing adequate control for unmeasured confounders and leaving sufficient information from which to estimate temperature effects.19 The (non-linear) temperature mortality function was fitted using natural cubic spline functions of temperature, implemented using Stata’s mkspline command with specification of two internal and two boundary knots. This function was applied to the mean of the maximum daily temperature over a lag period of 0–13 days (a single term of the mean of the daily maximum temperatures over this period), a lag structure chosen to capture the typical lags of the effect of cold exposure on mortality.1,19,20 Maximum temperature was chosen as the index providing the best fit in analyses of all-cause mortality in the UK.

To test for change in the association between low temperature and mortality we fitted a simplified model in which the (log) risk of mortality at lags 0–13 days was assumed to be a linear function of temperature below a fixed cold temperature threshold assumed to be at 10 °C, a temperature chosen following inspection of plots of the temperature-related mortality function as a convenient reference point close to the minimum mortality temperature for lags of 0–13 days.

Attributable deaths n_i on each day i were computed using the formula n_i = N_i × (RR_i− 1)/RR_i, where RR_i = exp[β × (t_i − t₀)], N_i is the total number of deaths on day i, RR is the relative risk on day i and t₀ is the ‘no-cold-effect’ baseline temperature. Daily attributable deaths were then summed over days for the year to compute the annual burden of cold-related deaths. For this, years were defined to run from July to June to capture cold-related deaths during whole winters.

Results

The regression results confirmed the well-established relationship between low outdoor temperature (over lags 0–13 days) and mortality risk.

Figure 5 plots the annual estimates of (all-cause) cold-related mortality for the London region for the period of analysis. In both plots, points indicated by black dots represent data for the years before the introduction of the government’s WFPs and the green dots data for the years after the introduction of WFPs.

FIGURE 5 The annual estimates of (all-cause) cold-related mortality for the London region. (a) Changes over time in cold-related deaths and mean of daily maximum temperatures in winter (December, January, February, March); and (b) the relationship between cold-related deaths and mean winter temperature, 1975–2011. (Figure not reproduced here; see the original article.)

It is apparent that there has been a steady but progressive reduction in the annual number of cold-related deaths over time, an observation that is broadly consistent with analyses of the trend in the Excess Winter Mortality Index15 published by the Office for National Statistics (ONS) (see Figure 3). Although it is not possible to make a robust interpretation of the exact pattern of the decline in cold-related deaths, the reduction appears to have been relatively greater over the period leading up to the year 2000 than in subsequent years.

As suggested by Figure 5b, the annual burden of cold-related deaths is directly related to mean winter temperature, with higher burdens in colder years. Thus, the overall decline in cold-related deaths over time is in part, but not wholly, explained by the gradual rise in mean winter (December, January, February, March) temperatures during the period of analysis (see Figure 5a), remembering that the computation of cold-attributable deaths was made by reference to the fixed cold temperature threshold of 10 °C.

Table 1 shows the results of analyses of the variation in the gradient of association between outdoor temperature and mortality in relation to both the introduction of WFPs and the annual domestic fuel price. We found evidence that since the introduction of WFPs in 1997 the gradient of association has been somewhat weaker than in earlier years (p = 0.02 for the test of statistical interaction). The period since 1997 has also been accompanied by higher indoor temperatures (see Figure 4).

There is also evidence that years with higher than average domestic fuel prices have a somewhat stronger relationship between low outdoor temperatures and mortality [per cent increase in mortality per degree Celsius fall in temperature of 2.49%, 95% confidence interval (CI) 2.32% to 2.66%] than years with below-average fuel prices (1.97%, 95% CI 1.77% to 2.18%). This is compatible with high fuel prices having a detrimental effect on cold-related deaths.

Discussion

The analyses presented in this section reflect year-to-year changes in cold mortality. The analyses are also dependent on a number of assumptions about model parameterisation and the temperature baseline used to define cold-attributable mortality in a given year.

These appear to indicate a slight weakening of the relationship between low outdoor temperature and mortality since the introduction of the WFPs in 1997, a result that is compatible with an econometric analysis by Iparraguirre.21 However, the period since 1997 has also been accompanied by changes in many other factors that may have reduced the risk of cold-related death. These include the efficiency of the thermal insulation of houses, improved heating systems (which are more likely to account for the temporal increase in indoor temperatures shown in Figure 4) and the reduced incidence/fatality of important ‘temperature-sensitive’ diseases, such as ischaemic heart disease, whose rates have declined rapidly in recent years. Therefore, it is not possible to conclude with certainty that the contribution of WFPs to the cost of heating the home has been the main contributory factor in this decline, although it is likely to have had some benefit if families chose to spend at least part of the WFP on warmer indoor temperatures.

The result for fuel prices is consistent with what might be predicted if the cost of fuel is a constraint on the temperature to which people choose to heat their home, the presumption being that after years of higher fuel prices people have used less fuel than they would otherwise do. Fuel prices rose appreciably from 2004 onwards, largely as a result of changes in wholesale prices and energy suppliers recouping the cost of renewable investments, but the years of high fuel prices do not occur in one block throughout the period of analysis. As with the WFP analysis, a cautious interpretation is needed because of the possibility that there are other factors than high fuel price that have been contributory factors.

The analyses suggest that there has been a progressive decline in the annual burden of cold-related deaths since the mid-1970s, which probably reflects the continuation of a trend of reducing winter-/cold-related mortality going back many decades, a trend that has occurred despite the gradual rise in the proportion of people aged > 65 years, who are known to be more vulnerable to the effects of low temperatures. The factors contributing to this decline cannot be determined with confidence but are likely to include a wide range of factors associated with improving socioeconomic conditions, including improving population health and health care and more effective health protection measures,16 for example the introduction in recent years of the Cold Weather Plan22 for England.

Improvements in housing, specifically in the effectiveness of home heating and winter indoor temperatures, are also likely to be contributory. If ambient temperatures also increase as a consequence of climate change, this would be expected to lead to a gradual reduction in cold-related deaths over time, although there is unlikely to be an appreciable reduction over the next decade or so. The next section examines evidence that is relevant to this question.

The impact of home energy efficiency interventions: 2002–10

In this section, evidence relating to the uptake and impact of HEE interventions in England over the period 2002–10 is described.

The core data items used in this analysis are shown in Figure 6. Central to the analysis is HEED, which provides data on the energy performance and installation of energy efficiency retrofits in England since 1993, and is managed by the Energy Saving Trust as a repository of energy efficiency activities in the UK. It includes data from installers, industry accreditation bodies, energy suppliers, government-funded programmes, local authorities and home surveys on a range of dwelling and energy efficiency parameters.23 The data are collected at the dwelling level but were available for linkage in our analyses at the level of the full-unit (seven-digit) postcode. In the UK, the unit postcode relates, on average, to around 14 households or approximately 40 individuals.

FIGURE 6 Simplified diagrammatic representation of the implementation of energy efficiency measures for a small number of English dwellings. Black signifies pre-intervention dwellings and green signifies post-intervention dwellings. The change in intervention status of individual dwellings provides a theoretical basis for comparisons before and after intervention or cross-sectionally across dwellings with and without HEE interventions. The HEED now contains data on approximately 16 million dwellings that have undergone energy efficiency improvements. (Figure not reproduced here; see the original article.)

The database is broadly representative of the English stock in terms of size and dwelling type, although flats are somewhat under-represented, and it has wide geospatial coverage. It has been estimated to contain data on 90% of the energy efficiency interventions that took place between 2002 and 2007.24

As shown diagrammatically in Figure 6, having data from HEED that are reasonably complete for interventions across the English housing stock provided the opportunity to study the impact that HEE interventions have as a natural experiment, because of the potential to examine differences in health outcomes based on multiple before-after comparisons and/or cross-sectional comparisons between dwellings with and without HEE improvements as might be done at time points t₁ and t₂ (dotted lines), for example.

The approach that was proposed had two principal elements (Figure 7):

1. use of the HEED in combination with empirical data and building physics modelling to characterise the timing of energy efficiency interventions and their effect on the indoor environment of dwellings (changes in winter indoor temperatures and indoor air quality)
2. linkage of data on those interventions with weather and mortality/morbidity data to assess the impact that HEE interventions have had on health.

FIGURE 7 Diagram of the principal data linkages for examining the impact of HEE interventions on health. (Figure not reproduced here; see the original article.)

However, it transpired that the estimates of changes to the indoor environment consequent to these energy efficiency interventions were very small, making it very unlikely that impacts on health would be detectable by direct observation of changes in routine health statistics aggregated at postcode level (see Temperature changes, Changes in permeability and air quality. Therefore, the assessment of the impact of these HEE improvements on health was based on an indirect approach rather than by direct linkage of health records to intervention data.

Defining intervention-related changes to the indoor environment

Installing energy efficiency retrofits to improve the energy performance of dwellings (Figure 8) has a number of positive and negative consequences for the indoor and outdoor environment and for households more generally.

FIGURE 8 The uptake of energy efficiency measures in England: 1997–2012. The HEED sample representative of the EHS (n = 168,998 dwellings). a, Data on glazing, boiler or heating system replacement were available only to 2007. For further details see Hamilton et al.25 (Figure not reproduced here; see the original article.)

Our focus here was limited to changes to the indoor environment, specifically winter indoor temperatures and air quality, although the impacts that this has on the energy needs and cost of space heating are briefly considered in the section on health impact modelling (see Energy use).

Because of the range of different energy efficiency measures installed across the English stock [including, but not limited to, insulation of the roof space, cavity walls, solid walls, double (triple) glazing and improvements to heating system/boiler], we classified the impact of each in terms of change in winter indoor temperatures and air quality [specifically the change in concentrations of fine particulate matter (PM_2.5) of both indoor and outdoor origin, radon, STS and mould risk] using a combination of empirical data and building physics modelling.

It should be noted that physics-based energy simulation models provide evidence on the ‘technical potential’ for energy savings and temperature change of energy efficiency interventions, which may differ from actual changes that reflect the effect of complex sociotechnical interactions of the dwelling occupier, the building and responses to the external and indoor environment. Poor installation, technical limits, environmental factors, changes in user preferences or practices and allowing for unmet needs to be satisfied, such as ‘temperature take-back’,26 may all give rise to appreciable variations in the actual changes in indoor conditions from dwelling to dwelling and from those predicted by the ‘idealised’ assumptions of building physics models. Therefore, our preference has been to base estimates of change on empirical data whenever possible.

Temperature changes

It has been shown that the energy performance levels of English dwellings has a significant influence on temperatures6,27,28 and that changing the energy performance characteristics of the dwelling through retrofits affects temperatures.29 It has also been shown that our indoor temperatures during wintertime are associated with risk of mortality, especially from cardiovascular disease.6

To classify dwellings with regard to winter indoor temperatures, we used a modification of a method described by Oreszczyn et al. 2006.27 It entailed two principal steps:

1. defining the empirical relationship between the energy efficiency characteristics of a dwelling and winter indoor temperatures measured under standardised conditions
2. determining the pre- and post-intervention energy efficiency characteristics of intervention dwellings using recorded data on dwelling type, age and other parameters in combination with building physics simulations.

Applying the results from step 2 to the relationships defined in step 1 provides the estimate of change in indoor temperature resulting from specific energy efficiency upgrades.

Central to these analyses is the concept of the standardised indoor temperature (SIT). The SIT is a marker of winter indoor temperature under standardised measurement conditions that has been used in the analysis of two data sets of indoor temperature monitoring in dwellings of known energy efficiency characteristics in the English stock. These data sets were from the previously published evaluation of the Warm Front energy efficiency scheme27 and the more recent Energy Follow-Up Survey (EFUS)30 (which is linked to the EHS).

Both data sets contain multiple serial measurements of indoor (living room and bedroom) temperatures during the heating season. For each dwelling in these surveys, regressing the indoor temperature on outdoor temperature (as a polynomial function) and adjusting for time of day allowed for the estimation of the mid-afternoon indoor temperature when the outdoor temperature was 5 °C, as illustrated in Figure 9. This is the SIT. At the same time, each dwelling was classified with respect to its energy efficiency using a measure of thermal performance known as the E-value. The E-value represents the heating power (in watts) required for the whole house to increase the temperature difference between indoor and outdoor environments by 1 kelvin (1 °C).

FIGURE 9 The derivation of the SIT: dwelling-specific regression analyses of the relationship between indoor and outdoor temperature was used to define a living room temperature on the day with a maximum outdoor temperature of 5 °C (the SIT). Data from across multiple dwellings of different energy efficiency levels were then used to define the general relationship between SIT and energy efficiency as reflected by the dwelling E-value (see Figure 10). (Figure not reproduced here; see the original article.)

Using results of SIT and E-value across multiple dwellings provided the empirical relationship shown in Figure 10. This is the core function used to estimate changes in indoor temperature of intervention dwellings in the HEED as energy efficiency interventions alter dwelling energy efficiency characteristics defined in terms of its E-value.

FIGURE 10 The relationship between SIT and whole house E-value [the number of watts required to maintain a temperature difference of 1 Kelvin (1 °C) between indoor and outdoor environments]. Improving the energy efficiency of the dwelling from E to E – δE results in an increase in SIT from T to T + δT. (Figure not reproduced here; see the original article.)

Improving the energy efficiency of the dwelling from E to E – δE results in an increase in SIT from T to T + δT.

The approach was used to estimate the potential change in indoor temperatures following retrofits that would affect the notional thermal performance of the dwelling31 and subsequent impact that it has on cold-related mortality/morbidity.32

To estimate the likely indoor temperatures experienced during winter conditions, a detailed building physics model was used to estimate the SITs of the English dwelling stock for different energy performance levels.33 Dwelling characteristics in the EHS were used to characterise the energy performance of the dwelling (i.e. its E-value), which was then used to predict the SIT at 5 °C in the living room and bedroom. Two model runs were conducted to estimate indoor temperatures: one run modelled the energy performance of the dwellings as described in the EHS (i.e. ‘baseline’), whereas the second run included the introduction of all available energy efficiency retrofits and the corresponding changes in E-value and SITs (i.e. ‘modified’). The retrofits included loft insulation to 250 mm, cavity wall insulation, solid wall insulation, double-glazing installation, condensing boiler installation and draught proofing.

Two metamodels were constructed from the detailed baseline and modified Health Impact of Domestic Energy Efficiency Measures (HIDEEM) model runs using the regression model of the average (living room and bedroom) SIT as a function of dwelling age, dwelling type, number of bedrooms and loft insulation level. These variables corresponded to those available within HEED. Model 1 predicts the baseline energy performance of the dwellings as described in the EHS, whereas model 2 includes the installation of all eligible retrofits for the UK housing stock. Because HEED provided only limited information on the dwelling, the meta-models included only a limited number of independent variables to predict the SIT.

The result of this classification is shown in Figure 11. A notional pre-intervention E-value is assigned on the basis of dwelling characteristics and, similarly, a post-intervention E-value is assigned reflecting the effect of the specific form of energy efficiency change. From these E-values, the empirical relationships are used to determine the corresponding SITs and hence the overall change in SIT. It is this change in temperature that is used as the basis for the calculation of the temperature-related impact on mortality and morbidity.

FIGURE 11 The classification of dwellings with regard to energy efficiency (E-value) and SIT. ID, identifier; LR, living room. (Figure not reproduced here; see the original article.)

Changes in permeability and air quality

Intervention-related changes in dwelling permeability, ventilation characteristics and consequent pollutant levels were characterised using CONTAM version 2.4c (National Institute of Standards and Technology, US Department of Commerce, Gaithersburg, MA, USA), a validated multizone airflow and pollutant transport simulation tool.33 Five contaminants were modelled: STS, PM_2.5 from internal sources, PM_2.5 from external sources, radon and moisture (as a precursor of mould). A series of pollutant sources and sinks were placed within appropriate building zones. External concentrations of pollutants were specified when relevant. The CONTAM models then predict the concentrations of the pollutants within each building zone every 15 minutes for a year. In modelling the indoor environment, we assumed that occupants opened windows during summer daytime hours (i.e. 09.00 to 17.00), trickle vents were open at all times and extract fans were used whenever present in the kitchen or bathroom.

From the CONTAM model outputs, annual average indoor pollutant concentrations (weighted to reflect exposure levels experienced around the home with 45% of the occupant’s time spent in the living room, 45% spent in the bedroom and 10% spent in the kitchen) within the archetype dwellings for a range of permeabilities (in the range 0.5 to 30 m³m^–2hour^–1). Interpolation using a fourth-order polynomial was used to estimate concentrations for permeabilities that were not directly modelled in CONTAM. The permeability for each EHS dwelling was calculated using a physics-based method that uses details that are relevant to air infiltration, that is openings and cracks in the fabric (walls, windows, flues, vents, etc.), and normalises this using a fabric-to-volume ratio. Each EHS dwelling was matched into one of the 16 × 4 archetype and ventilation categories and the exposure concentrations were calculated using the pollutant models. The heat loss owing to overall permeability for each dwelling is calculated by using the infiltration level, building volume and heat capacity of air.34

Changes in household fuel costs

The impact that HEE measures had on energy use was studied in a sample of dwellings with a gas connection (approximately 90% of all dwellings). To do this, the HEED was linked to the Energy Supplier Meter Point data, a database of annualised gas and electricity energy supplier data for the years 2004–7. This linkage was carried out at address level by the DECC and provided for analysis in an anonymised form with only lower-layer super output area identifiers.

The annualised gas demand data were derived from two meter readings at least 6 months apart and corrected for seasonal normal demand (i.e. annual weather correction) and end-user climate sensitivity. The annualising method provides a means of comparing total gas demand between years that removes the effect of cold or warm weather (described fully in Ofgem 201335). The electricity demand data are also derived from meter readings and annualised using a process that allocates meters to domestic annual demand profiles. Electricity data are not corrected for interannual weather.

A nationally representative subsample of approximately 168,998 dwellings with gas and electricity meters was drawn for analysis using the 2011 EHS,36 which is a cross-sectional survey representative of English dwellings and households. The EHS sample frame was constructed to be representative of gas-heated English dwellings and aligned with HEED data using dwelling age, dwelling type, number of bedrooms, government region and household tenure, from which a sample was selected using a simple random sampling.

Health impacts

For calculation of health impacts, we based calculations on changes in indoor air and SIT in relation to outcomes for which there is reasonably clear epidemiological evidence of a link6 (Table 2). Cause-specific mortality was modelled using life table methods based on the IOMLIFET model45 but applied to individuals in the EHS36 based on their age, sex and specific exposure changes. Life tables were set up using a 2010 age-specific population and (disease-specific and all-cause) mortality data for England and Wales from ONS, with separate life tables set up for men and women. For each intervention, we calculated changes in years of life lost (YLL) by comparing life tables with and without the relative risks (RRs) for the corresponding exposure change and using lag functions shown in Appendix 4. Impacts that these same outcomes had on morbidity were approximated from the mortality estimates by applying age-specific and cause-specific ratios of years of healthy life lost because of disability (years of life list because of disability; YLD) to the overall YLL derived from the World Health Organization’s Global Burden of Disease data.31 Further details are given in Hamilton et al.32

Results

The uptake of specific forms of energy efficiency interventions by dwelling type, urban–rural status and decile of socioeconomic deprivation, as well as changes in SIT, is summarised graphically in Figure 12.

FIGURE 12 The uptake of specific forms of HEE interventions. (a) HEE retrofits by dwelling type; (b) HEE retrofits by urban–rural area; (c) HEE retrofits (number of interventions) by neighbourhood deprivation decile group; and (d) mean baseline SIT (°C) for different dwelling archetypes. (Figure not reproduced here; see the original article.)

Table 3 tabulates the change in SIT for the main categories of intervention by dwelling age, type and housing tenure. As this shows, for most forms of energy efficiency upgrade the change in temperature is modest but is dependent on property age and type in particular. The increases in temperature are generally greatest for dwellings that are older and, thus, less energy efficient. Loft insulation is associated with very modest increases in temperature, whereas much greater changes are seen for cavity and solid wall insulation.

Across all energy efficiency interventions (Figure 13), the change in the distribution of SITs is modest and averages at just 0.09 °C.

FIGURE 13 Distribution of (a) the pre- and post-intervention SITs; and (b) the difference in post- and pre-intervention SITs. (Figure not reproduced here; see the original article.)

The associated change in health impacts (deaths, life-years and YLDs) is illustrated in Figure 14. The plots show the evolution of the impact that the changes have had over time, noting that the number of overall deaths saved (around 280 deaths in year 1) eventually returns to baseline level but with a reduction in deaths at younger age and an eventual increase in the number of deaths at older age. The annual gain in life-years saved takes several decades to be fully realised and plateaus at over 4000 life-years per year. The morbidity impacts (as YLDs) are approximately 10% of the mortality impacts.

FIGURE 14 Modelled change in deaths saved and life-years gained. Deaths saved (a) overall; and (b) by age group and life-years gained (c) overall; (d) by age group; and (e) a combination of mortality and morbidity impacts summarised as YLLs and YLDs. (Figure not reproduced here; see the original article.)

A summary of the intervention-related change in quality-adjusted life-years (QALYs) per 100,000 population (averaged over a 50-year time horizon) is shown in Table 4 separately for temperature-related health impacts and those owing to change in indoor air quality. In each case, a range is shown to reflect results for different assumptions about ventilation characteristics (whether or not installations have associated purpose-provided vents in order to maintain ‘adequate’ ventilation). Note that the health impacts associated with the ventilation may be negative or positive (depending on assumptions about ventilation changes) and may be as much as an order of magnitude greater than those related to temperature change.

These air quality-related impacts reflect a balance of changes affecting pollutants with outdoor and indoor sources. Unless there is purpose-provided compensatory ventilation (such as trickle vents), energy efficiency measures tend to reduce dwelling permeability and, thus, ventilation. This is beneficial, especially in high outdoor pollution areas, as it reduces the ingress of particle pollution PM_2.5 from the outdoor environment [as shown in Figure 15, which shows a negative shift (reduction) in PM_2.5 derived from outdoor air after intervention], as well as contributing to improved energy efficiency (reduced loss of heated air). However, reduced permeability, and hence reduced air exchange, acts to increase the concentration of pollutants derived from indoor sources, including indoor PM_2.5, radon, second-hand cigarette smoke, humidity/mould, and such pollutants as volatile organic compounds. Again this is illustrated in Figures 16 and 17, which show increases in concentrations of PM_2.5 of indoor origin, and Figure 18, which shows the increase in radon concentrations from (in this case) solid wall insulation. (The effect of volatile organic compounds has not been quantified in any of the results shown here because of the lack of clear evidence of exposure–response relationships.) Note the relatively large impacts in converted flats and selected other dwelling types.

FIGURE 15 Changes in indoor concentration (µg/m3) of PM2.5 of outdoor origin as a result of different forms of energy efficiency intervention. (a) Loft insulation; (b) cavity wall insulation; (c) improved glazing; and (d) solid wall insulation. The dashed line indicates a reference point of a change of –1 µg/m3. (Figure not reproduced here; see the original article.)

FIGURE 16 Changes in indoor concentration (µg/m3) of PM2.5 of indoor origin as a result of different forms of energy efficiency intervention. (a) Loft insulation; (b) cavity wall insulation; (c) improved glazing; and (d) solid wall insulation. The dashed line indicates a reference point of a change of 1 µg/m3. (Figure not reproduced here; see the original article.)

FIGURE 17 Modelled changes in the concentration of indoor PM2.5 of outdoor (y-axis) and indoor (x-axis) origin following selected energy efficiency interventions. (a) Indoor concentrations of PM2.5 of outdoor and indoor origin; (b) changes with loft insulation; (c) changes with improved glazing; and (d) changes with cavity wall insulation. (Figure not reproduced here; see the original article.)

FIGURE 18 Distribution of changes in indoor radon concentration consequent to the installation of solid wall insulation. (Figure not reproduced here; see the original article.)

Unfortunately, the available data do not allow for the change in ventilation characteristics and, consequently, for indoor air quality to be modelled with precision and certainty. It is, for example, a matter of choice whether or not trickle vents are installed at the same time as other energy efficiency measures and whether or not householders subsequently choose, or even know how, to use them. Therefore, all quantified changes in indoor air quality are based on ‘best reasonable’ assumptions, informed by published empirical data. However, different assumptions about installed measures and household behaviour, as well as about local environment conditions (e.g. outdoor PM_2.5, radon levels), can appreciably alter the balance of positive and negative effects from alterations to the ventilation characteristics of a dwelling.

Energy use

Article extract reproduced from Hamilton et al.25 Copyright © 2016 The Authors. Published by Elsevier B.V. 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/. The text below includes minor additions and formatting changes to the original text.

The results for the 168,998 English dwellings included in the analysis are shown in Figure 19 and Table 5. Overall, 39% received a major energy efficiency measure, 36% received a fabric (i.e. insulation) measure and 9% received a heating measure. The annual average change in energy demand across the stock over the period of analysis was –810 kWh/year (with individual year changes of –740 kWh/year in 2004/5, –860 kWh/year in 2005/6 and –830 kWh/year in 2006/7). These are equivalent to percentage changes of –3.6% for 2004/5, –4.4% for 2005/6 and –4.4% for 2006/7 (see Figure 19).

FIGURE 19 Distribution of annual gas demand (kWh/year) per dwelling in study sample (n = 145,885), 2004–7. (Figure not reproduced here; see the original article.)

In regression analyses that allow for differences in dwelling type, age, tenure, size, region and median neighbourhood income, we estimate that the presence of an installed fabric energy efficiency retrofit, on average, was associated with a –790 kWh/year change in energy demand, or a 3.9% reduction from the stock mean gas demand in 2006. The presence of a heating energy efficiency retrofit was on average associated with a –1950 kWh/year (10.4%) change. When installed together, the average change in energy demand was –2290 kWh/dwelling/year, or a 11.7% reduction.

The proportional change in gas demand in 2005–7 associated with specific forms of energy efficiency intervention adjusted for dwelling type, age, tenure, size, region and neighbourhood income is shown in Table 5. (See Appendix 1 for further details by energy efficiency measure.)

Cavity wall insulation and condensing boiler installations were associated with reductions in demand of 4.9% and 5.5%, respectively, whereas loft insulation and double-glazing installation showed almost no associated change. Larger reductions were achieved with combinations of measures, especially those that included condensing boiler and cavity insulation installations (e.g. cavity wall and loft insulation combined with a condensing boiler was associated with an estimated –11.2% change in demand).

These findings are very similar to those of a study by Hong et al.46 on the impact that the Warm Front energy efficiency scheme had on British houses, which found that loft and full cavity wall insulation reduced demand during a 1-year period by 10–17%. Although these changes in energy demand are lower than notional ‘savings’ or those of the Hong et al. study,46 it is important to bear in mind that they are based on estimates that control for physical, household and area factors. Controlling for physical factors, for example, means that the effect of number and area of exposed walls or the age of the dwelling is removed, whereas control of household effects adjusts for such factors as: ability to afford larger areas to heat and demand for greater thermal comfort.

It is noteworthy that in our adjusted estimates neither loft insulation nor double glazing was associated with appreciable energy savings over the 3-year period, suggesting that the effect of these interventions is, on average, fairly small and/or cannot easily be detected using annualised energy data. Glazing is one of the thermally weakest elements of the building fabric, and replacement with more thermally efficient glazing should have an effect on energy use. However, because of the way the data were reported, it was not possible to account for the area of glazing that was replaced, so the small estimate of impact may, in part, reflect the fact that, in many cases, only a small proportion of the glazed area of the dwelling was replaced. For loft insulation, many installations were top-ups of around 5–75 mm and the expected change in gas demand would be minimal.

There were variations in the savings associated with certain household/dwelling characteristics (see Appendix 1). For example, dwelling age appears to have an inconsistent effect on changes in gas demand, with the 1967–75 age group having much greater reduction in demand for all single retrofit measures than other age bands. It is not simply that older dwellings have greater energy savings than newer dwellings. Such variation may, in part, be explained by the eligibility and type of retrofits installed. Cavity wall filling is mostly relevant to dwellings built in the mid-century: earlier builds (pre-1950) tend not to have a wall cavity, and those built after 1990 tend to have higher specification. The impact that boilers among this mid-century group had was also greater (after controlling for size), perhaps reflecting a number of building design features, such as the prevalence of gas central heating.47

Changes in energy demand were also generally lower among renting households or those living in areas of lower income, supporting the hypothesis that those on lower incomes are less likely than higher-income households to realise energy savings. This observation may in part be because lower-income households have higher levels of energy utility (i.e. greater need for the amount used).48 Both social renting and living in lower-income neighbourhoods were associated with reduced energy demand, even after controlling for type and age of dwelling,6 which may also suggest that these households have a greater potential for increasing demand if energy efficiency retrofits are made. The differences could also be construed as ‘comfort taking’, whereby these households in areas of lower income reduce the potential ‘energy savings’ by taking the savings in the form of temperature increases, an effect that has been shown in a study of vulnerable households in England.29

The impact of energy efficiency retrofits demonstrates a broad dose–response effect in which combined packages of retrofits are associated with greater (negative) changes in energy demand. Larger increases in reductions in gas demand were associated with boilers and cavity wall insulation, with only minor additional effects from lofts and glazing. The largest change in gas demand was associated with the combined installation of a condensing boiler, and cavity and loft insulation at –10.8%.

In some cases, the effect of combinations of measures appear ‘super-additive’, that is, to have an effect greater than the simple sum of the component measures. For example, the individual change attributable to cavity insulation (–3.8%), condensing boiler installation (–5.2%) and loft insulation (1.2%) ought to result in a change of –7.8%, but the combined estimate was a reduction of around 11%. This may reflect true advantages of combining retrofits into single package (which may have benefits in achieving energy demand reduction and potential cost savings of installation, e.g. wall scaffolding is set up only once), or that there is a ‘take-back’ threshold after which rebound related to thermal comfort is lessened (i.e. the potential rebound has been met).

Realising the potential energy savings set out in the DECC energy efficiency strategy would require a ‘whole-house’ retrofit package for every home in England. If an average energy saving of 10% (e.g. ≈2300 kWh reduction) was achieved from the average UK dwelling, it would take approximately 9,565,000 ‘whole-house’ retrofits (40% of the whole UK stock) to achieve the estimated 22 terawatt hours (TWh) of energy savings by 2020; to achieve a 10% reduction in 2006 levels by 2020 (i.e. 54 TWh) through energy efficiency alone would take the equivalent of every home in the UK being refurbished (i.e. 23,500,000 dwellings). Although further efficiencies might be gained from water heating and appliances, space-heating accounts for the bulk of residential demand. Thus, achieving such savings (to meet climate change targets and to realise the ancillary public health benefits) would require substantial acceleration of the historical rate of retrofit uptake, which is challenging but achievable using widely available technologies and insulating techniques that rely on an existing deployment system and skill base.

In summary, energy efficiency retrofits (although not clearly for loft insulation or double glazing) result in observable reductions in space-heating energy demand (with resultant cost savings for the householders). These savings are part of the gain from energy efficiency measures that are taken instead of increases in temperature. They may have indirect (but unquantified) benefits for the health of householders, especially those on low incomes, because of change in disposable income, and are additional to the health impacts relating to change in the indoor environment (changes in winter indoor temperature and air quality).

Discussion

Several features of these results are noteworthy. First is the relatively small size of the effect of energy efficiency interventions on indoor temperatures and, consequently, on health. This appears to be a result of two principal factors: (1) the modest nature of most individual energy efficiency upgrades and (2) the shape of the function relating to indoor temperature to dwelling E-value. As shown in Figure 20, not only is there a relatively shallow gradient of increasing temperature with improving energy efficiency at E-values above around 300 W per Kelvin, but that gradient flattens to close to zero (i.e. plateaus) at values to the left of this point. This point is very close to the middle of the distribution of dwelling E-values for the English stock (as shown in the bottom part of Figure 20). In other words, around half of the housing stock is already at a point where further increases in energy efficiency would not, on average, result in any increase in indoor temperatures (as represented by the SIT). It is also worth noting that this SIT versus E-value relationship is based on empirical data from two independent data sets (EFUS 2010/1130 and Warm Front 2006/527) that appear to generate very similar functional forms (Figure 21). The broad agreement of the results from these different data sets gives greater confidence that the underlying relationship is essentially correct.

FIGURE 20 Relationship between SIT and dwelling E-value above the distribution of E-values. The dotted line indicates the approximate location of the inflexion point at which further improvement in energy efficiency (positions to the left on the x-axis) result in no additional gain in SIT. (Figure not reproduced here; see the original article.)

FIGURE 21 Relationship between SIT and dwelling E-value (living room): results from two independent studies. (a) EFUS 2010/1130; and (b) Warm Front 2004 to 05.27 ‘Meta-analyses’ of SIT results from multiple dwellings. (Figure not reproduced here; see the original article.)

A second important observation is the fact that impacts on health related to air quality may be just as great as, and probably greater than, those from improvements in winter temperatures. Even though there are uncertainties in quantifying these impacts because of limited empirical data, the broad patterns of change are predictable given reasonable assumptions about intervention types, householder behaviour and other factors. These impacts are especially important because they arise over the medium to longer term and hence are not usually observable or even measured in short-term evaluation studies. Therefore, it seems probable that they have not been given sufficient attention in the development of policies relating to energy efficiency interventions and in building regulation. This is an area of research need.

As discussed at the outset of this report, there are pathways of health impact that are additional to those operating through changes to the indoor environment (i.e. temperature and air quality). The analysis here confirms that improvements in energy efficiency as expected are accompanied by observable changes in energy demand. It is a simple fact that the benefits of energy efficiency can be taken either as reduced energy consumption or as increased temperature, or as some combination of the two. It is this balance that lies behind the shape of the relationship between SIT and dwelling E-value, with relatively more of the gain being taken as increased temperature when the energy efficiency is low and almost none being taken as increased temperature when energy efficiency is already good. What individual families choose to do will be determined by their own preferences and circumstances and is almost certainly influenced in part by their disposable income. We do not yet have a method that allows even a broad quantification of the impacts had on health of energy and hence cost savings, especially for those in or near to FP.