Intervention conceptualisation
We based our estimate of the effect of cold housing on cardio-vascular disease burden on two sources of evidence. First, the negative effect of cold indoor temperature on blood pressure (4, 5). Second, the effect of hightened blood pressure on cardiovascular disease risk (specifically ischemic heart disease and ischemic and haemorrhagic stroke) (13) (see supplementary file 1). Cardiovascular disease responds rapidly to change in risk factors; therefore, we assume little to no time lag in response to temperature change.
Model overview
We used a proportional multistate lifetable (pMSLT) simulation model to estimate health gains achievable through interventions on exposure to cold indoor temperature within a specified population (11). We simulated the 2016 Australian population through to 2126 (maximum potential lifetime of the living cohort in 2016 being set to 110 year of age) in annual time steps with transition probabilities for all-cause mortality, and incidence and case fatality rates of cardiovascular diseases in subsidiary lifetables. This model was applied once for Business as Usual (BAU), based on the current prevalence of cold housing (prevalence assumed unchanging into the future), and then for the intervention (hypothetical elimination of cold housing) by altering the CVD incidence rates given the shift in population average blood pressure. The two components of the model are linked by population impact fractions (PIFs), that summarise the proportion reduction in diseases associated with change in indoor temperature exposure for cohorts defined by age and sex.
We modelled relevant cardiovascular diseases (heart disease and stroke) as independent of each other in parallel lifetables. Estimated changes in morbidity and mortality rates were summed together in an overall lifetable at each annual cycle, adding the health gain (Intervention minus BAU) across the diseases and years. The ‘health adjustment’ to convert life years gained to HALYs gained was achieved by subtracting off each life year gained the proportion ‘lost’ due to morbidity, using years of life lived with disability (YLDs) from burden of disease studies divided by the population in each sex by age-group as a measure of proportionate morbidity.
Input parameters
We have presented data inputs with their sources in Table 1.
Table 1
Parameter
|
Data Source
|
Comments/ notes/ model and data assumptions
|
Value
|
Unhealthy indoor temperature prevalence at base year 2016
|
AHCD
|
Prevalence of people experiencing indoor cold temperature was obtained from the Australian Housing Conditions Dataset (AHCD) (14). The AHCD survey asked participants ‘Are you able to warm your house during winters’? Those responding ‘Yes’ were considered as experiencing indoor cold temperature. We accounted for age variations in the prevalence as estimated from the AHCD.
Uncertainty: Double of standard errors in age-specific prevalence obtained from AHCD with correlation of 1.
|
5.74%
(Refer to Table 2 for age variations)
|
Average temperature in cold houses
|
|
Average outdoor temperatures: Victoria (15.04°C), New South Wales (18.43°C), South Australia (20.19°C)
We assume average indoor cold temperature at 16 Celsius
|
|
All-cause mortality rates
|
GBD
|
Data on all-cause mortality rates by sex and age group for 2016 were obtained from the Global Burden of disease results tool and inputted directly (27).
|
Refer to Table 2 for age and sex variations
|
All-cause morbidity rates
|
GBD
|
Data on years of life lived with disability (YLD) were obtained from the Global Burden of Disease study for each sex and age group in 2016. No time trend was allowed, as YLD rates by age in the GBD have not changed much over time. Morbidity rates were directly inputted in the main life table to estimate HALYs (27).
|
Refer to Table 2 for age and sex variations
|
Disease specific incidence, prevalence and case fatality rates
|
GBD
|
We applied national disease-specific estimates from GBD (27) to the population of three states New South Wales, Victoria and South Australia. Comparison of disease specific morbidity across the three states and national estimates showed a maximum of 10% difference – therefore we applied Australian disease data to these three states. The disease-specific incidence rates, prevalence and mortality rates, and case fatality rates (mortality rate divided by prevalence) for ischemic heart disease and stroke were obtained from the GBD data (27). Stroke includes ischemic stroke and haemorrhagic stroke (subarachnoid and intracerebral). Disease specific rates for subarachnoid and intracerebral haemorrhagic stroke were summed and the ratio to ischemic stroke was included in the model for uncertainty analysis. All disease-specific epidemiological inputs were processed through DISMOD II and used to ensure coherence and smoothing for age (28).
Annual Percentage Changes: the annual percentage changes were estimated using Poisson regression on incidence rates and case fatality rates from 1990 to 2016 GBD data and included as inputs to the PMSLT.
Uncertainty: +/- 5% SD (log normal distribution for incidence), correlations 1.0 between sexes for all disease.
|
Refer to Table 2 for age and sex variations
|
Disease specific morbidity
|
IHME/GBD
|
The sex and age specific disability rates were calculated as disease’s YLD obtained from GBD (27)divided by the number of prevalent cases.
Uncertainty: +/- 10% SD
|
|
Relative risk from indoor cold to systolic blood pressure
|
Review of relative risks as part of the project
|
Using evidence and search terms from the WHO Housing and Health Guidelines, we reviewed the health effects of exposure to indoor cold. Our review found consistent evidence for the effect of indoor cold on hypertension. We performed risk of bias assessment using ROBINS-E and ROB tools on interventional and observational studies on the relationship between indoor cold and systolic blood pressure. Two studies (one cohort (7) and one randomised controlled trial(4)) were found to have low to moderate risk of bias. Relative risk from the randomised controlled trial was used.
Uncertainty: As provided by Saeki, Obayashi (4)
|
5.8 mmHg (95% CI (-9.3, -2.4))
More detailed review results presented in Table 1 in Appendices
|
Systolic blood pressure distribution
|
ABS
|
Data on systolic blood pressure by age and sex was obtained from the National Health Survey 2017-18 from the Australian Bureau of Statistics (ABS) (29). Mean and standard deviations of systolic blood pressure were included as input to the pMSLT simulation model.
Uncertainty: As provided by the National Health Survey (29)
|
Refer to Table 4 in Appendices for age and sex variations
|
Relative risk from systolic blood pressure to ischemic heart disease and stroke
|
Forouzanfar, Liu (18)
|
Rate ratios for systolic blood pressure to ischemic heart disease, ischemic stroke and haemorrhagic stroke were taken from IHME GBD (27).
Uncertainty: As provided by Forouzanfar, Liu (18)
|
Refer to Table 5 in Appendices for age variations
|
Base year and BAU parameters
Estimates of the number of people exposed to inadequate indoor temperature by age and sex was obtained from the Australian Housing Conditions Dataset survey (14, 15) and assumed to be constant into the future. Data on the age and sex distribution of the Australian population was obtained from the Australian Population Census 2016. Disease-specific incidence, prevalence and case fatality rates were obtained from IHME Global Burden of Disease for Australia. We checked for coherence between epidemiological parameters derived from this array of data sources (i.e., incidence, case fatality and prevalence) for each disease by examining plotted trends and further processed them through epidemiological tool DISMOD II to use as data inputs (16).
Intervention specification
The relative risk of high blood pressure from exposure to indoor cold was estimated from a randomised controlled trial (4). This study reported a difference of 5.8 mmHg (95% CI (-9.3, -2.4)) between an intervention group, who occupied a room heated to 22°C, and the control group who occupied a room kept stable at 12°C. Both groups were given sufficient clothing and bedclothes to be warm. Both groups were exposed to the respective interventions for 11 hours during the night and blood pressure was measured in 15 minutes intervals at night time and in the morning after rising. The 5.8mmHg difference was that when awake (as there was no difference when awake due to compensation using more bedding in the experimentally colder group). This effect estimate was converted into an absolute change achievable in systolic blood pressure per 1°C temperature increase.
Data on the prevalence of people experiencing indoor cold was obtained from the Australian Housing Conditions Dataset that representatively sampled housing from 4,500 households' condition across three Australian States (Victoria, New South Wales and South Australia) (14). The measure comprised a self-reported assessment of ability to keep warm indoors at home during cold winter weather. The average outdoor temperature for Victoria in 2016 was 15.0°C, for New South Wales was 18.4°C and for South Australia was 20.2°C (17). To account for seasonal variations in temperature and time spent outside homes we assumed that our simulated population is exposed to indoor cold ranging from one-third (the awake 2/3rds of the day for the colder half of the year for those people at home most of the day) to one-sixth (same logic, but for people working or out of the home for approximately half of waking hours) of the time. A beta distribution for uncertainty in the intervention effect estimate was applied to account for variability in this exposure time. To simulate the effect of indoor cold on blood pressure we estimated the difference between cold houses (average temperature of 16°C) and adequately warmed houses (average temperature of 20°C). Assuming a short latency of cold to blood pressure, the change in average blood pressure of the year was calculated within each iteration of the simulation as: this proportion of the year exposed ranging from 1/6 to 1/3; multiplied by the difference in temperature (4 degrees, by lifting average cold housing from 16 to 20 degrees); multiplied by the RCT-based estimate of change in systolic blood pressure per 1 degree celsius.
Relative risks for the causal relationship between systolic blood pressure and ischemic heart disease, ischemic stroke and haemorrhagic stroke were obtained from the Global Burden of Disease (GBD) study (18) (see supplementary Table 5). The intervention was simulated on the 2016 population for the same jurisdictions covered by the Australian Housing Conditions datasets (i.e., Victoria, New South Wales and South Australia). The intervention (eradication of cold housing) was modelled as lifelong. BAU exposure to cold housing was based on exposures in 2016 for cohorts defined by age-group and sex.
Analyses
Probabilistic uncertainty analyses using a Monte Carlo simulation method was conducted on input parameters (see Table 1). More generous uncertainty was applied where we were less confident on input parameter (for example the subjective measurement of indoor cold in Australia). Simulations were run using the ERSATZ add in to Excel with 2000 iterations used to generate 95% uncertainty intervals (UI) for the HALY estimates.
Our outputs included HALYs gained by indoor cold eradication within a life-time and ten- and -twenty-year time horizons. Outcomes were reported both with 0% and 3% discount rates and also per 1000 persons.
We compared the estimated HALYs gained from cold housing eradication with other CVD-related interventions described in interactive league tables (12).