2.0 Methods
2.1 Study context
The province of BC is located on the west coast of Canada. It spans
nearly one million km2 and had a 2021 population of
approximately 5.2 million people (Government of British Columbia, 2021).
The climate in BC is generally temperate, and most homes do not have air
conditioning, especially in more densely populated southern coastal
areas (BC Hydro, 2020).
The 2021 EHE was caused by a high-pressure system known as a ‘heat dome’
(Philip et al., 2022) and was characterized primarily by unprecedented
high temperatures, and secondarily by increased concentrations of
ground-level ozone (O3) and fine particulate matter
(PM2.5) in many regions (Table S1). Specifically, this
event resulted in daily high temperatures 10-20°C above seasonal norms
across the province (Figure 1; Table S1) and it happened following the
summer solstice, when daylight ranges from ~16 to
~19 hours, south to north (Henderson, McLean, et al.,
2022; Slattery, 2016; Philip et al. 2022). More than half of the
provincial population (~3.05 million) lives in the
greater Vancouver region, where more than half of the deaths occurred
(Figure 2). This area was under an air quality advisory during the EHE
because of high ground-level O3 concentrations
(Henderson, McLean, et al., 2022). We consider the overall impact of the
EHE period in this study, which includes the effects of both extreme
heat and increased air pollution.
2.2 Study design
This study compares adults who died during the EHE with adults who died
during more typical summer weather to identify differences between the
two groups with respect to 26 chronic diseases. The EHE period was
defined as 25 June – 02 July 2021 because this is when there was
statistically significant excess daily mortality (Figure 1). Statistical
significance was determined using the Public Health Intelligence for
Disease Outbreak (PHIDO) algorithm, which compares observed daily
mortality counts with the expected range based on a model using data
from the past five years (Henderson et al., 2021). The PHIDO method was
developed by BCCDC, and is similar to other anomaly detection techniques
such as the Farrington method (Salmon et al., 2016). The typical weather
period was defined as 25 June – 02 July from 2012-2020, so that
comparator deaths were drawn from the same dates as the EHE deaths. We
used data from 2012 onward to ensure that we included at least four
typical weather deaths for each EHE death in the analyses. Using an
approximate 4:1 ratio maximized statistical power while minimizing the
time between the EHE and typical weather groups (Foppa & Spiegelman,
1997; Gordis, 2014).
2.3 The COVID-19 Data Library
During the COVID-19 pandemic, BC established the COVID-19 Data Library
(Wilton et al., 2022) to support rapid public health informatics. This
platform facilitates individual-level linkage between multiple
administrative databases. BC has a single-payer healthcare system, and
health records are captured in different datasets for every individual
covered by the provincial Medical Services Plan. Datasets are linkable
by a unique and anonymous patient master key. The BCCDC obtained
authorization from the BC Ministry of Health to use the COVID-19 Data
Library to generate evidence about the public health impacts of the EHE.
2.4 Mortality
Deaths were extracted from BC vital statistics records in the COVID-19
Data Library. Each record includes one underlying cause of death, age,
sex, and geographic health unit of residence. The underlying cause of
death is coded by the BC Vital Statistics Agency (BCVSA) according to
the 10th revision of the International Classification
of Diseases (ICD-10), based on information from the certificate of
death. Deaths among children (<18 years) were excluded from
the study because the BC Coroners Service (BCCS) reported that all
heat-related deaths during the 2021 EHE occurred among adults (BC
Coroners Service, 2022a). Deaths missing information on age, sex, or
location were also excluded.
2.5 Chronic disease
registries
The BC Ministry of Health maintains 26 administrative chronic disease
registries based on individual patterns of healthcare usage. They
reflect the prevalence of conditions such as asthma, heart disease, and
diabetes. Each registry has different publicly available inclusion
criteria (BCCDC, 2022). All registries are updated annually at the end
of the BC fiscal year (31 March), with a 1-year lag period. As such,
data for this study were available to 31 March 2020. To ensure
comparability between the EHE and typical weather deaths, we matched all
deaths to the chronic disease registries that ended on 31 March of the
year prior to the death.
Individuals with multiple chronic diseases are included in multiple
registries. We linked each death in the EHE and typical weather groups
with all 26 registries to capture chronic conditions for each decedent
at the time of death. The underlying cause of death in the vital
statistics data is independent of the information derived from the
chronic disease registries. As such, the cause of death may not reflect
any of the chronic conditions of the decedent at the time of death.
Registries were excluded if they were associated with <1% of
EHE deaths or if they were collinear with another registry (≥ 85%
overlap).
2.6 Statistical analyses
We compared EHE deaths with typical weather deaths using conditional
logistic regression. Models were conditioned on geographic health unit
(N=16) of residence (Government of British Columbia, 2022) so EHE deaths
were compared with others in the same area (Figure 2). We report odds
ratios (ORs) to quantify the association between EHE mortality and each
chronic disease adjusted for age, sex, and all other chronic diseases
(Eq.1). The supplementary appendix includes results for each chronic
disease adjusted for age and sex, but not the other chronic diseases
(Table S2).
\begin{equation}
\mathbf{Eq.1.}\ \ EHE\ Death\ |\ Geographic\ Health\ Unit=\text{Chronic\ Disease}_{1}+\text{Chronic\ Disease}_{2}+\ldots+\text{Chronic\ Disease}_{21}+Age+Sex\nonumber \\
\end{equation}Decedents were also classified by their total number of chronic diseases
(maximum 10+) to assess the overall burden of disease effects.
Conditional logistic regression was used to estimate the OR for EHE
mortality associated with the number of chronic diseases, with 0 as the
reference category (Eq.2).
\begin{equation}
\mathbf{Eq.2.\ }\ EHE\ Death\ |\ Geographic\ Health\ Unit=Number\ of\ Comorbidities+Age+Sex\nonumber \\
\end{equation}2.7 Subgroup analysis
The BC Coroners Service (BCCS) investigates all unattended deaths,
public deaths, deaths among children, and deaths otherwise reported by
healthcare providers or the public. For all such deaths, the BC Vital
Statistics Agency (BCVSA) must receive the coroner’s certificate of
death before assigning the underlying cause of death in the vital
statistics data. There are often long reporting delays (months or years)
between the two agencies, and BCVSA assigns the ICD-10 code R99 as the
underlying cause of death while it is waiting to receive a coroner’s
certificate of death. BCCS investigated hundreds of deaths that occurred
during the EHE and reported that 562 were due to extreme heat from 25
June – 02 July 2021. The role of extreme heat was assessed using a
protocol to review the physical and circumstantial evidence available
for each case (BC Coroners Service, 2022a).
When BCVSA receives a coroner’s certificate indicating death due to
extreme heat, it codes X30 as the underlying cause of death. BCVSA
cannot code X30 for certificates of death completed by anyone other than
a coroner. When it receives a certificate of death indicating extreme
heat from another source, it forwards the case to the BCCS for
investigation. We extracted the vital statistics data for the study on
30 November 2022. At this time, BCVSA had not yet received all the EHE
death certificates from BCCS, so some of the heat-related deaths were
coded as X30 and others were still coded as R99. To assess how chronic
diseases varied among heat-related deaths (X30), deaths with a pending
or unknown underlying cause (R99), and non-heat-related deaths (not X30
or R99), we reran regression models separately for each of the three
subgroups.