Revisiting the importance of temperature, weather and air pollution variables in heat-mortality relationships with machine learning.

Extreme heat events have significant health impacts that need to be adequately quantified in the context of climate change. Traditionally, heat-health association methods have relied on statistical models using a single air temperature index, without considering other heat-related variables that may influence the relationship and their potentially complex interactions. This study aims to introduce and compare different machine learning (ML) models, which naturally consider interactions between predictors and non-linearities, to re-examine the importance of temperature, weather and air pollution predictors in modeling the heat-mortality relationship. ML approaches based on tree ensembles and neural networks, as well as non-linear statistical models, were used to model the heat-mortality relationship in the two most populated metropolitan areas of the province of Quebec, Canada. The models were calibrated using a comprehensive database of heat-related predictors including various lagged temperature indices, temperature variations, meteorological and air pollution variables. Performance was evaluated based on out-of-sample summer mortality predictions. For the two studied regions, models relying only on lagged temperature indices performed better, or equally well, than models considering more heat-related predictors such as temperature variations, weather and air pollution variables. The temperature index with the best performance differed by region, but both mean temperature and humidex were among the best indices. In terms of modeling approaches, non-linear statistical models were as competent as more advanced ML models for predicting out-of-sample summer mortality. This research validated the current use of non-linear statistical models with the appropriate lagged temperature index to model the heat-mortality relationship. Although ML models have not improved the performance of all-cause mortality modeling, these approaches should continue to be explored, particularly for other health effects that may be more directly linked to heat exposure and, in the future, when more data become available.

Machine and deep learning for modelling heat-health relationships.

Extreme heat events pose a significant threat to population health that is amplified by climate change. Traditionally, statistical models have been used to model heat-health relationships, but they do not consider potential interactions between temperature-related and air pollution predictors. Artificial intelligence (AI) methods, which have gained popularity for health applications in recent years, can account for these complex and non-linear interactions, but have been underutilized in modelling heat-related health impacts. In this paper, six machine and deep learning models were considered to model the heat-mortality relationship in Montreal (Canada) and compared to three statistical models commonly used in the field. Decision Tree (DT), Random Forest (RF), Gradient Boosting Machine (GBM), Single- and Multi-Layer Perceptrons (SLP and MLP), Long Short-Term Memory (LSTM), Generalized Linear and Additive Models (GLM and GAM), and Distributed Lag Non-Linear Model (DLNM) were employed. Heat exposure was characterized by air temperature, relative humidity and wind speed, while air pollution was also included in the models using five pollutants. The results confirmed that air temperature at lags of up to 3 days was the most important variable for the heat-mortality relationship in all models. NO2 concentration and relative humidity (at lags 1 to 3 days) were also particularly important. Ensemble tree-based methods (GBM and RF) outperformed other approaches to model daily mortality during summer months based on three performance criteria. However, a partial validation during two recent major heatwaves highlighted that non-linear statistical models (GAM and DLNM) and simpler decision tree may more closely reproduce the spike of mortality observed during such events. Hence, both machine learning and statistical models are relevant for modelling heat-health relationships depending on the end user goal. Such extensive comparative analysis should be extended to other health outcomes and regions.

Constrained groupwise additive index models.

In environmental epidemiology, there is wide interest in creating and using comprehensive indices that can summarize information from different environmental exposures while retaining strong predictive power on a target health outcome. In this context, the present article proposes a model called the constrained groupwise additive index model (CGAIM) to create easy-to-interpret indices predictive of a response variable, from a potentially large list of variables. The CGAIM considers groups of predictors that naturally belong together to yield meaningful indices. It also allows the addition of linear constraints on both the index weights and the form of their relationship with the response variable to represent prior assumptions or operational requirements. We propose an efficient algorithm to estimate the CGAIM, along with index selection and inference procedures. A simulation study shows that the proposed algorithm has good estimation performances, with low bias and variance and is applicable in complex situations with many correlated predictors. It also demonstrates important sensitivity and specificity in index selection, but non-negligible coverage error on constructed confidence intervals. The CGAIM is then illustrated in the construction of heat indices in a health warning system context. We believe the CGAIM could become useful in a wide variety of situations, such as warning systems establishment, and multipollutant or exposome studies.

Evaluation of the factors governing dissolved organic carbon concentration in the soil solution of a temperate forest organic soil.

The widespread increase of dissolved organic carbon (DOC) in northern hemisphere surface waters have been generally attributed to the recovery from acidic deposition and to climatic variations. The long-term responses of DOC to environmental drivers could be better predicted with a better understanding of the mechanisms taking place at the soil level given organic forest soils are the main site of DOC production in forested watersheds. Here, we assess the long-term variation (25 years) of DOC concentration in the solution leaching from the soil organic layer (DOCOL) of a temperate forest. Our results show that DOCOL increased by 32 % (p < 0.001) during the period of study while the lake outlet DOC concentration did not show any changes. Weekly and annual models based on a simple set of explicative variables including throughfall DOC, throughfall precipitation, temperature, litterfall amounts and organic layer leachate calcium concentration (CaOL, taken as a proxy for soil solution ionic strength) explain between 17 and 58 % of the variance in DOCOL depending on model structures and temporal scales. Throughfall DOC and CaOL were both positively related to DOCOL in the models describing its variations at the weekly and annual scale. Temperature was positively correlated to DOCOL, probably due to increased microbial activity, while precipitation had a negative effect on DOCOL (only at the weekly scale), most probably due to a dilution effect. Contrary to our expectations, annual litterfall inputs had no impacts on annual DOCOL variations. Overall, the results shows that DOCOL control is a complex process implicating a set of environmental factors that are acting in different ways while no single variable alone can explain a large part of the variation in both, weekly or annual DOCOL variations.

Heat-related mortality prediction using low-frequency climate oscillation indices: Case studies of the cities of Montréal and Québec, Canada.

Heat-related mortality is an increasingly important public health burden that is expected to worsen with climate change. In addition to long-term trends, there are also interannual variations in heat-related mortality that are of interest for efficient planning of health services. Large-scale climate patterns have an important influence on summer weather and therefore constitute important tools to understand and predict the variations in heat-related mortality. Methods: In this article, we propose to model summer heat-related mortality using seven climate indices through a two-stage analysis using data covering the period 1981-2018 in two metropolitan areas of the province of Québec (Canada): Montréal and Québec. In the first stage, heat attributable fractions are estimated through a time series regression design and distributed lag nonlinear specification. We consider different definitions of heat. In the second stage, estimated attributable fractions are predicted using climate index curves through a functional linear regression model. Results: Results indicate that the Atlantic Multidecadal Oscillation is the best predictor of heat-related mortality in both Montréal and Québec and that it can predict up to 20% of the interannual variability. Conclusion: We found evidence that one climate index is predictive of summer heat-related mortality. More research is needed with longer time series and in different spatial contexts. The proposed analysis and the results may nonetheless help public health authorities plan for future mortality related to summer heat.

Data-Enhancement Strategies in Weather-Related Health Studies.

Although the relationship between weather and health is widely studied, there are still gaps in this knowledge. The present paper proposes data transformation as a way to address these gaps and discusses four different strategies designed to study particular aspects of a weather-health relationship, including (i) temporally aggregating the series, (ii) decomposing the different time scales of the data by empirical model decomposition, (iii) disaggregating the exposure series by considering the whole daily temperature curve as a single function, and (iv) considering the whole year of data as a single, continuous function. These four strategies allow studying non-conventional aspects of the mortality-temperature relationship by retrieving non-dominant time scale from data and allow to study the impact of the time of occurrence of particular event. A real-world case study of temperature-related cardiovascular mortality in the city of Montreal, Canada illustrates that these strategies can shed new lights on the relationship and outlines their strengths and weaknesses. A cross-validation comparison shows that the flexibility of functional regression used in strategies (iii) and (iv) allows a good fit of temperature-related mortality. These strategies can help understanding more accurately climate-related health.

Cardiovascular Health Peaks and Meteorological Conditions: A Quantile Regression Approach.

Cardiovascular morbidity and mortality are influenced by meteorological conditions, such as temperature or snowfall. Relationships between cardiovascular health and meteorological conditions are usually studied based on specific meteorological events or means. However, those studies bring little to no insight into health peaks and unusual events far from the mean, such as a day with an unusually high number of hospitalizations. Health peaks represent a heavy burden for the public health system; they are, however, usually studied specifically when they occur (e.g., the European 2003 heatwave). Specific analyses are needed, using appropriate statistical tools. Quantile regression can provide such analysis by focusing not only on the conditional median, but on different conditional quantiles of the dependent variable. In particular, high quantiles of a health issue can be treated as health peaks. In this study, quantile regression is used to model the relationships between conditional quantiles of cardiovascular variables and meteorological variables in Montreal (Canada), focusing on health peaks. Results show that meteorological impacts are not constant throughout the conditional quantiles. They are stronger in health peaks compared to quantiles around the median. Results also show that temperature is the main significant variable. This study highlights the fact that classical statistical methods are not appropriate when health peaks are of interest. Quantile regression allows for more precise estimations for health peaks, which could lead to refined public health warnings.

A heat-health watch and warning system with extended season and evolving thresholds.

BACKGROUND: Many countries have developed heat-health watch and warning systems (HHWWS) or early-warning systems to mitigate the health consequences of extreme heat events. HHWWS usually focuses on the four hottest months of the year and imposes the same threshold over these months. However, according to climate projections, the warm season is expected to extend and/or shift. Some studies demonstrated that health impacts of heat waves are more severe when the human body is not acclimatized to the heat. In order to adapt those systems to potential heat waves occurring outside the hottest months of the season, this study proposes specific health-based monthly heat indicators and thresholds over an extended season from April to October in the northern hemisphere. METHODS: The proposed approach, an adoption and extension of the HHWWS methodology currently implemented in Quebec (Canada). The latter is developed and applied to the Greater Montreal area (current population 4.3 million) based on historical health and meteorological data over the years. This approach consists of determining excess mortality episodes and then choosing monthly indicators and thresholds that may involve excess mortality. RESULTS: We obtain thresholds for the maximum and minimum temperature couple (in °C) that range from (respectively, 23 and 12) in April, to (32 and 21) in July and back to (25 and 13) in October. The resulting HHWWS is flexible, with health-related thresholds taking into account the seasonality and the monthly variability of temperatures over an extended summer season. CONCLUSIONS: This adaptive and more realistic system has the potential to prevent, by data-driven health alerts, heat-related mortality outside the typical July-August months of heat waves. The proposed methodology is general and can be applied to other regions and situations based on their characteristics.

A cold-health watch and warning system, applied to the province of Quebec (Canada).

CONTEXT: A number of studies have shown that cold has an important impact on human health. However, almost no studies focused on cold warning systems to prevent those health effects. For Nordic regions, like the province of Quebec in Canada, winter is long and usually very cold with an observed increase in mortality and hospitalizations throughout the season. However, there is no existing system specifically designed to follow in real-time this mortality increase throughout the season and to alert public health authorities prior to cold waves. OBJECTIVE: The aim is to establish a watch and warning system specifically for health impacts of cold, applied to different climatic regions of the province of Quebec. METHODOLOGY: A methodology previously used to establish the health-heat warning system in Quebec is adapted to cold. The approach identifies cold weather indicators and establishes thresholds related to extreme over-mortality or over-hospitalization events in the province of Quebec, Canada. RESULTS AND CONCLUSION: The final health-related thresholds proposed are between (-15 °C, -23 °C) and (-20 °C, -29 °C) according to the climatic region for excesses of mortality, and between (-13 °C, -23 °C) and (-17 °C, -30 °C) for excesses of hospitalization. These results suggest that the system model has a high sensitivity and an acceptable number of false alarms. This could lead to the establishment of a cold-health watch and warning system with valid indicators and thresholds for each climatic region of Quebec. It can be seen as a complementary system to the existing one for heat warnings, in order to help the public health authorities to be well prepared during an extreme cold event.

Predicting seismic-induced liquefaction through ensemble learning frameworks.

The regional nature of liquefaction records and limited information available for a certain set of explanatories motivate the development of complex prediction techniques. Indirect methods are commonly applied to incidentally derive a hyperplane to this binary classification problem. Machine learning approaches offer evolutionary prediction models which can be used as direct prediction methods to liquefaction occurrence. Ensemble learning is a recent advancement in this field. According to a predefined ensemble architecture, a number of learners are trained and their inferences are integrated to produce stable and improved generalization ability. However, there is a need to consider several aspects of the ensemble learning frameworks when exploiting them for a particular application; a comprehensive evaluation of an ensemble learner's generalization ability is required but usually overlooked. Also, the literature falls short on work utilizing ensemble learning in liquefaction prediction. To this extent, this work examines useful ensemble learning approaches for seismic-induced liquefaction prediction. A comprehensive analysis of fifteen ensemble models is performed. The results show improved prediction performance and diminishing uncertainty of ensembles, compared with single machine learning models.

Toward an Improved Air Pollution Warning System in Quebec.

The nature of pollutants involved in smog episodes can vary significantly in various cities and contexts and will impact local populations differently due to actual exposure and pre-existing sensitivities for cardiovascular or respiratory diseases. While regulated standards and guidance remain important, it is relevant for cities to have local warning systems related to air pollution. The present paper proposes indicators and thresholds for an air pollution warning system in the metropolitan areas of Montreal and Quebec City (Canada). It takes into account past and current local health impacts to launch its public health warnings for short-term episodes. This warning system considers fine particulate matter (PM2.5) as well as the combined oxidant capacity of ozone and nitrogen dioxide (Ox) as environmental exposures. The methodology used to determine indicators and thresholds consists in identifying extreme excess mortality episodes in the data and then choosing the indicators and thresholds to optimize the detection of these episodes. The thresholds found for the summer were 31 µg/m3 for PM2.5 and 43 ppb for Ox in Montreal, and 32 µg/m3 and 23 ppb in Quebec City. In winter, thresholds found were 25 µg/m3 and 26 ppb in Montreal, and 33 µg/m3 and 21 ppb in Quebec City. These results are in line with different guidelines existing concerning air quality, but more adapted to the cities examined. In addition, a sensitivity analysis is conducted which suggests that Ox is more determinant than PM2.5 in detecting excess mortality episodes.

A new look at weather-related health impacts through functional regression.

A major challenge of climate change adaptation is to assess the effect of changing weather on human health. In spite of an increasing literature on the weather-related health subject, many aspect of the relationship are not known, limiting the predictive power of epidemiologic models. The present paper proposes new models to improve the performances of the currently used ones. The proposed models are based on functional data analysis (FDA), a statistical framework dealing with continuous curves instead of scalar time series. The models are applied to the temperature-related cardiovascular mortality issue in Montreal. By making use of the whole information available, the proposed models improve the prediction of cardiovascular mortality according to temperature. In addition, results shed new lights on the relationship by quantifying physiological adaptation effects. These results, not found with classical model, illustrate the potential of FDA approaches.

Aggregating the response in time series regression models, applied to weather-related cardiovascular mortality.

In environmental epidemiology studies, health response data (e.g. hospitalization or mortality) are often noisy because of hospital organization and other social factors. The noise in the data can hide the true signal related to the exposure. The signal can be unveiled by performing a temporal aggregation on health data and then using it as the response in regression analysis. From aggregated series, a general methodology is introduced to account for the particularities of an aggregated response in a regression setting. This methodology can be used with usually applied regression models in weather-related health studies, such as generalized additive models (GAM) and distributed lag nonlinear models (DLNM). In particular, the residuals are modelled using an autoregressive-moving average (ARMA) model to account for the temporal dependence. The proposed methodology is illustrated by modelling the influence of temperature on cardiovascular mortality in Canada. A comparison with classical DLNMs is provided and several aggregation methods are compared. Results show that there is an increase in the fit quality when the response is aggregated, and that the estimated relationship focuses more on the outcome over several days than the classical DLNM. More precisely, among various investigated aggregation schemes, it was found that an aggregation with an asymmetric Epanechnikov kernel is more suited for studying the temperature-mortality relationship.

EMD-regression for modelling multi-scale relationships, and application to weather-related cardiovascular mortality.

In a number of environmental studies, relationships between nat4ural processes are often assessed through regression analyses, using time series data. Such data are often multi-scale and non-stationary, leading to a poor accuracy of the resulting regression models and therefore to results with moderate reliability. To deal with this issue, the present paper introduces the EMD-regression methodology consisting in applying the empirical mode decomposition (EMD) algorithm on data series and then using the resulting components in regression models. The proposed methodology presents a number of advantages. First, it accounts of the issues of non-stationarity associated to the data series. Second, this approach acts as a scan for the relationship between a response variable and the predictors at different time scales, providing new insights about this relationship. To illustrate the proposed methodology it is applied to study the relationship between weather and cardiovascular mortality in Montreal, Canada. The results shed new knowledge concerning the studied relationship. For instance, they show that the humidity can cause excess mortality at the monthly time scale, which is a scale not visible in classical models. A comparison is also conducted with state of the art methods which are the generalized additive models and distributed lag models, both widely used in weather-related health studies. The comparison shows that EMD-regression achieves better prediction performances and provides more details than classical models concerning the relationship.

Effects of climate and fine particulate matter on hospitalizations and deaths for heart failure in elderly: A population-based cohort study.

BACKGROUND: There are limited data on the effects of climate and air pollutant exposure on heart failure (HF) within taking into account individual and contextual variables. OBJECTIVES: We measured the lag effects of temperature, relative humidity, atmospheric pressure and fine particulate matter (PM2.5) on hospitalizations and deaths for HF in elderly diagnosed with this disease on a 10-year period in the province of Quebec, Canada. METHODS: Our population-based cohort study included 112,793 elderly diagnosed with HF between 2001 and 2011. Time dependent Cox regression models approximated with pooled logistic regressions were used to evaluate the 3- and 7-day lag effects of daily temperature, relative humidity, atmospheric pressure and PM2.5 exposure on HF morbidity and mortality controlling for several individual and contextual covariates. RESULTS: Overall, 18,309 elderly were hospitalized and 4297 died for the main cause of HF. We observed an increased risk of hospitalizations and deaths for HF with a decrease in the average temperature of the 3 and 7days before the event. An increase in atmospheric pressure in the previous 7days was also associated with a higher risk of having a HF negative outcome, but no effect was observed in the 3-day lag model. No association was found with relative humidity and with PM2.5 regardless of the lag period. CONCLUSIONS: Lag effects of temperature and other meteorological parameters on HF events were limited but present. Nonetheless, preventive measures should be issued for elderly diagnosed with HF considering the burden and the expensive costs associated with the management of this disease.

Association between Floods and Acute Cardiovascular Diseases: A Population-Based Cohort Study Using a Geographic Information System Approach.

BACKGROUND: Floods represent a serious threat to human health beyond the immediate risk of drowning. There is few data on the potential link between floods and direct consequences on health such as on cardiovascular health. This study aimed to explore the impact of one of the worst floods in the history of Quebec, Canada on acute cardiovascular diseases (CVD). METHODS: A cohort study with a time series design with multiple control groups was built with the adult population identified in the Quebec Integrated Chronic Disease Surveillance System. A geographic information system approach was used to define the study areas. Logistic regressions were performed to compare the occurrence of CVD between groups. RESULTS: The results showed a 25%-27% increase in the odds in the flooded population in spring 2011 when compared with the population in the same area in springs 2010 and 2012. Besides, an increase up to 69% was observed in individuals with a medical history of CVD. CONCLUSION: Despite interesting results, the association was not statistically significant. A possible explanation to this result can be that the population affected by the flood was probably too small to provide the statistical power to answer the question, and leaves open a substantial possibility for a real and large effect.

Health impacts of the July 2010 heat wave in Québec, Canada.

BACKGROUND: One of the consequences of climate change is the increased frequency and intensity of heat waves which can cause serious health impacts. In Québec, July 2010 was marked by an unprecedented heat wave in recent history. The purpose of this study is to estimate certain health impacts of this heat wave. METHODS: The crude daily death and emergency department admission rates during the heat wave were analyzed in relation to comparison periods using 95% confidence intervals. RESULTS: During the heat wave, the crude daily rates showed a significant increase of 33% for deaths and 4% for emergency department admissions in relation to comparison periods. No displacement of mortality was observed over a 60-day horizon. CONCLUSIONS: The all-cause death indicator seems to be sufficiently sensitive and specific for surveillance of exceedences of critical temperature thresholds, which makes it useful for a heat health-watch system. Many public health actions combined with the increased use of air conditioning in recent decades have contributed to a marked reduction in mortality during heat waves. However, an important residual risk remains, which needs to be more vigorously addressed by public health authorities in light of the expected increase in the frequency and severity of heat waves and the aging of the population.

A general and flexible methodology to define thresholds for heat health watch and warning systems, applied to the province of Québec (Canada).

Several watch and warning systems have been established in the world in recent years to prevent the effects of heat waves. However, many of these approaches can be applied only in regions with perfect conditions (e.g., enough data, stationary series or homogeneous regions). Furthermore, a number of these approaches do not account for possible trend in mortality and/or temperature series, whereas others are generally not adapted to regions with low population densities or low daily mortality levels. In addition, prediction based on multiple days preceding the event can be less accurate if it attributes the same importance to each of these days, since the forecasting accuracy actually decreases with the period. The aim of the present study was to identify appropriate indicators as well as flexible and general thresholds that can be applied to a variety of regions and conditions. From a practical point of view, the province of Québec constitutes a typical case where a number of the above-mentioned constraints are present. On the other hand, until recently, the province's watch and warning system was based on a study conducted in 2005, covering only the city of Montreal and applied to the whole province. The proposed approach is applied to each one of the other health regions of the province often experiencing low daily counts of mortality and presenting trends. The first constraint led to grouping meteorologically homogeneous regions across the province in which the number of deaths is sufficient to carry out the appropriate data analyses. In each region, mortality trends are taken into account. In addition, the proposed indicators are defined by a 3-day weighted mean of maximal and minimal temperatures. The sensitivity of the results to the inclusion of traumatic deaths is also checked. The application shows that the proposed method improved the results in terms of sensitivity, specificity and number of yearly false alarms, compared to those of the existing and other classical approaches. An additional criterion based on the Humidex is applied in a second step and a local validation is applied to historical observations at reference forecasting stations. An integrated heat health watch and warning system with thresholds that are adapted to the regional climate has thus been established for each sub-region of the province of Quebec and became operational in June 2010.

Spatial variability of climate effects on ischemic heart disease hospitalization rates for the period 1989-2006 in Quebec, Canada.

BACKGROUND: Studies have suggested an association between climate variables and circulatory diseases. The short-term effect of climate conditions on the incidence of ischemic heart disease (IHD) over the 1989-2006 period was examined for Quebec's 18 health regions. METHODS: Analyses were carried out for two age groups. A GAM statistical model, that blends the properties of generalized linear models with additive models, was used to fit the standardized daily hospitalization rates for IHD and their relationship with climatic conditions up to two weeks prior to the day of admission, controlling for time trends, day of the season and gender. RESULTS: Results show that, in most of Quebec's regions, cold temperatures during winter months and hot episodes during the summer months are associated with an increase of up to 12% in the daily hospital admission rate for IHD but also show decreased risks in some areas. The risk of hospitalization is higher for men and women of 45-64 years and varies spatially. In most regions, exposure to a continuous period of cold or hot temperature was more harmful than just one isolated day of extreme weather. Men aged 45-64 years showed higher risk levels of IHD than women of the same age group. In most regions, the annual maximum of daily IHD admissions for 65 years old was reached earlier in the season for both genders and both seasons compared to younger age groups. The effects of meteorological variables on the daily IHD admissions rate were more pronounced in regions with high smoking prevalence and high deprivation index. CONCLUSION: This study highlights the differential effects of cold and hot periods on IHD in Quebec health regions depending on age, sex, and other factors such as smoking, behaviour and deprivation levels.