Office of the Provincial Health Officer, BC

In the summer of 2021, the province of British Columbia (BC) experienced a weather phenomenon called a heat dome, a high atmospheric pressure system acts like a “lid” trapping hot air underneath (Philip et al., 2021). Over 600 deaths were attributed to this extreme weather event, with the greatest percentage of fatalities occurring among the most vulnerable populations such as the elderly, those with chronic illnesses, those that lived alone and lower income neighbourhoods, according to the BC Coroner’s Report (2022). If an early warning system had been in place, the resulting loss of life due to extreme heat could have been significantly minimized.

To reduce the risk of excess mortality due to extreme weather events such as the 2021 heat dome, the Office of the BC Provincial Health Officer (OPHO) partnered with a team of UBC Master of Data Science (MDS) students on a capstone project. The BC Provincial Health Officer is the senior public health official for BC and its office is responsible for monitoring the health of the population of BC and providing independent advice to the ministers and public officials on public health issues.

The goal of the capstone project was to estimate mortality risk due to extreme temperatures and forecast the estimated excess deaths in BC communities. These efforts will help inform provincial climate mitigation policies, disaster preparedness, and response strategies.

Over an 8-week period the team developed several models to predict excess mortality for BC regional health authorities (Northern Health, Vancouver Coastal Health, Island Health, Fraser Health, and Interior Health) using historical temperature, demographic, and mortality data. Models explored for this project included Random Forest, Neural Network, XGBoost, LSTM, and Negative Binomial. While the students developed their models using synthetic mortality data, the OPHO trained the models in the OPHO secured data analytic space and validated the models using the Coroner’s 2021 heat dome report which provided the only temperature-related mortality data available for BC. Although the Negative Binomial model was the best performing model, it severely underestimated temperature-related deaths. Therefore, the students pivoted to an OPHO-adapted Distributed Lag Non-Linear Model (DLNM), originally developed by Mistry and Gasparrini (2024), that captured both the lagged and non-linear effects of temperature on mortality over time and produced the best predictive effects in comparison to other models.

To make the results of the DLNM easier to understand at a glance, the students created an interactive R Shiny dashboard which displays estimated temperature-related excess mortality for the current and following days across each regional health authority. Predicted mortality is depicted by a range in colour shades (lightest shades representing fewer deaths, darkest shades representing more deaths), and mean daily temperature forecast is displayed in a bubble overlaying each region. For more specific results, users can apply filters by non-optimal temperature category (i.e., heat, cold, or both) and by forecast day (current or following). In addition to mapped results, the dashboard also includes a chart and table of estimated trends in temperature-related mortality over the past 30 days and for the following two days, which can be filtered by health authority.

Although the environmental data used for this project was publicly available, the mortality data was restricted to protect the privacy of the individuals and accessible only by OPHO staff. As such, the students had to send their model code, built on synthetic data, to the OPHO for training and validation thereby adding extra steps in the development process. However, the OPHO was available for frequent and regular communications which helped reduce any potential delays. Additional challenges include the short time frame of the capstone project, limiting opportunities to explore alternative modeling strategies and exploration of other climate (e.g., humidity) and a broader set of region-specific data, which may improve predicted temperature-related deaths.

Despite the challenges faced and extensive scope of the project, the UBC MDS team was successful in building the groundwork for a forecasting model and visual display which can support public health officials in taking preventative measures against temperature-related mortality. The future goal is to calibrate the DLNM model so that its predictive outputs match the findings of a previously published OPHO study, thereby making it a reliable real-time predictive tool to aid public health planning and responses to extreme temperatures, with the aim of reducing future loss of life.

British Columbia Coroners Service. (2022). Extreme heat death review panel report. https://www2.gov.bc.ca/assets/gov/birth-adoption-death-marriage-and-divorce/deaths/coroners-service/death-review-panel/extreme_heat_death_review_panel_report.pdf

Philip, S. Y., Kew, S. F., van Oldenborgh, G. J., et al. (2021). Rapid attribution analysis of the extraordinary heatwave on the Pacific Coast of the US and Canada June 2021. Earth System Dynamics Discussions. https://doi.org/10.5194/esd-2021-90

Mistry, M. N., & Gasparrini, A. (2024). Real-time forecast of temperature-related excess mortality at small-area level: Towards an operational framework. Environmental Research: Health, 2(3), 035011. https://doi.org/10.1088/2752-5309/ad5f51

Explore our Data Science Programs Explore Other Data in Action Stories