1. Introduction
Weather and climate affect many aspects of human health, whether directly, such as heat stress impacts induced by prolonged exposure to extreme temperature, or indirectly, for example, by influencing the habitat and life cycle of mosquitoes, ticks, and other infectious disease vectors (Smith et al. 2014). Included among the more direct effects are fall-related injuries—slips and trips on wet and slippery surfaces that routinely occur to mobile people in pursuit of everyday activities.
Falls are the second-greatest source of unintentional death in the world with the highest fatality rates experienced in the wealthiest nations of western Europe and North America (Peden et al. 2002). In Canada, unintentional falls accounted for over 1 million emergency department visits (EDV) and 128 000 hospitalizations in 2010, placing tremendous demands on public health care systems, with direct annual costs reported to be in excess of CAD 6 billion (Parachute 2015). In 2014, there were about 13.7 fatalities attributed to falls per 100 000 Canadians, as compared with 5.1 fatalities resulting from motor vehicle collisions (Parachute 2015; Transport Canada 2016). This difference suggests an answer to the question posed in the paper title and, more generally, supports calls to examine pedestrian falls that occur outdoors as an important complement to transportation safety and mobility studies focused on motor vehicle collisions (Elvik and Bjørnskau 2019; Methorst et al. 2017; Schepers et al. 2017).
Weather and climate are important environmental factors that affect the type and risk of injuries from falls. Researchers have examined the influence of particular storms or short periods of inclement weather, seasonal effects, and subseasonal-scale impacts of weather or walking-surface conditions. Studies have generally focused on one or more combinations of vulnerable populations (most often elderly cohorts), higher-risk occupations (e.g., mining, disaster response and recovery), and particular types of injuries (e.g., hip or wrist fractures). Most research has been completed in Europe and North America using a range of qualitative (e.g., focus groups; Nyman et al. 2013) and quantitative methods [see Schepers et al. (2017) for a thorough review]. Quantitative studies have relied on self-reports and/or a variety of medical records (e.g., hospital admissions, emergency room visitations, other healthcare service use, insurance claim information) to develop descriptive statistical accounts, conduct retrospective analyses, or to establish prospective study samples.
Several researchers have documented the incidence of injury during or immediately following individual winter storm events, either in absolute terms (e.g., Smith and Nelson 1998) or in comparison to periods void of significant weather but consistent in other respects to the event (e.g., same duration, time of year, days of the week, etc.) (Avery 1982; Marshall et al. 2016; Ràliš 1981; Ràliš et al. 1988; Stansbury et al. 1995). The number of injuries during an ice or snow event is generally several times greater than a corresponding fine-weather period, although results vary by the type of injury (Stansbury et al. 1995). Many studies have observed that fall-related injuries, most commonly hip or wrist fractures, increase during the winter season or in colder relative to warmer regions of specific countries (Arnold et al. 2016; Bamzar and Ceccato 2015; Bulajic-Kopjar 2000; Chiu et al. 1996; Crawford and Parker 2003; Elvik and Bjørnskau 2019; Gyllencreutz et al. 2015; Hagino et al. 2004; Hemenway and Colditz 1990; Jacobsen et al. 1999; Lauritzen et al. 1993; Levy et al. 1995). While activities unique to winter, such as removing snow from roofs (Bylund et al. 2016; Pipas et al. 2002), have been shown to lead to falls, these account for only a small fraction of the seasonal burden. Greater incidence of injury in winter has also been found in occupational studies focused on construction (Lipscomb et al. 2006) and mail delivery (Bentley and Haslam 2001).
Most pertinent to the current study is literature that examined relationships between fall-related injuries and weather, or walking-surface conditions, at subseasonal scales over longer periods of time that include many weather events. Retrospective studies have compared the presence or absence of hazardous weather conditions during days with a high prevalence of falls relative to other days (e.g., Gevitz et al. 2017). Most investigations, however, relate daily to weekly fall or fracture incident counts or rates to one or more independent weather variables, such as temperature, snowfall, freezing rain, rainfall, wind speed, and day length (e.g., Bell et al. 2000; Bobb et al. 2017; Hajat et al. 2016; Hassi et al. 2000; Levy et al. 1998; Luukinen et al. 1996; Mondor et al. 2015; Morency et al. 2012; Murray et al. 2011). Exposure-related factors (e.g., location, day of week, month, year, self-reported walking distance) and sociodemographic variables (e.g., age, sex) have been controlled to varying extents in these studies. Analytical methods have ranged widely from descriptive statistics (e.g., Berggård and Johansson 2010) to nonlinear time series modeling (e.g., Modarres et al. 2014) with each study adopting somewhat unique geographic boundaries, location, data sources, fall inclusion definitions, and sets of meteorological parameters. Despite these differences, common findings regarding precipitation effects are apparent:
most studies report increases in fall-related injuries associated with snowfall (generally less than 50%),
snowfall effects on injury counts are often lagged by 1 to several days following precipitation,
elderly (≥65 yr) and younger (school-age children) cohorts seem much less affected by precipitation events with researchers generally attributing this to exposure factors (e.g., ability to avoid travel in severe conditions, instituted school closures),
freezing rain or rain followed by dropping temperatures appear to have greater (up to 300% increase) and more immediate impacts on falls, and
risk of falls tends to rise as weather severity increases, except for snow events where risk levels taper off once accumulations become very large.
Weather warnings have the potential to affect related outcomes. While not normally considered in analyses of falls, two studies made novel use of government-issued weather warnings as a proxy indicator for injury-causing meteorological hazards to examine their association with fall-related injuries. Murray et al. (2011) found that severe weather warnings for icy roads issued by the Met Office for Edinburgh, Scotland, were associated with a 40% increase in fractures (95% confidence limits: 20%–52%). Mondor et al. (2015) conducted a study of a large elderly (≥65 yr) cohort in Montreal, Canada, over a 9-yr period (1998–2006) to evaluate the association between fall-related injury incident rates and public warnings issued by the federal government (Environment Canada) for freezing rain and snowstorms. They observed a significant rise in injury incident rate ratios (IRR) on days when freezing-rain warnings were issued [IRR 1.20; 95% confidence interval (CI): 1.08–1.33)] and a significant decrease on days with snowstorm warnings (IRR 0.89; 95% CI: 0.80–0.99) relative to days during the winter season without warnings. Except for the immediate day following freezing-rain warnings, injury rates stayed significantly higher for up to five days after events had ended (Mondor et al. 2015).
Both of these investigations failed to acknowledge the potential role of weather warnings as an intervention of information that could directly or indirectly affect exposure, and other risk-mitigating behavior. Risk was evaluated by comparing injuries or incidence rates during days with warnings against all days in the study period without warnings. The research did not appear to control for the occurrence of conditions that did not meet the threshold for the issuance of a weather warning but were still capable of influencing injury incidents; as a result, relative increases in injury incidence may actually be higher than reported. While potential effects of festive seasons were addressed by Murray et al. (2011), no other measures to control for variation in exposure caused by human factors were evident.
More generally, none of the studies identified in the literature review reported on long-term shifts in weather-related injury risk from falls that might be expected due to improvements in healthcare, outdoor walkway construction and maintenance technology, and safety measures. As well, no analyses examined or accounted for subdaily-scale effects of weather on injury occurrence. This may be particularly important as the same storm, measured by daily precipitation accumulation, will present a different risk depending on whether the timing of heavy snowfall, freezing rain, and associated hazardous conditions are coincident with activity patterns that lead to exposure (e.g., overnight storm with walkways cleared, salted, and gritted before morning rush hour). In addition, storms that occur over multiple days may not be properly discretized and accounted for as they may be recorded as separate events with smaller accumulations when in fact they are experienced by the public as a single, continuous, and cumulative event.
This paper aims to address these shortcomings by applying a unique definition and classification of winter storm events and dry-weather control periods in a matched-pair design developed originally for an analysis of motor vehicle collision risk (Mills et al. 2019). In addition to providing a robust estimate of fall-related injury risk, the study facilitates comparisons of injury risks associated with winter storms across two primary modes of transportation—walking and automobile.
2. Study area, data, and methods
a. Study area
The study area is a Canadian midsized community located about 100 km west of Toronto, Ontario, Canada. It consists of the cities of Cambridge, Kitchener, and Waterloo (Ontario) that together form the continuous urban core of the Regional Municipality of Waterloo (RMW). The approximately 525 000 people inhabiting the study area (RMW 2018) experience a continental climate modified in all seasons by the Great Lakes. Frequent synoptic weather systems producing snow and mixed-precipitation events of varying intensities and durations affect the study area, thus making it an attractive place to study the effects of winter storms on injury risk. As shown in Fig. 1, below-freezing temperatures are typically observed from November through April with measurable snowfalls (≥0.2 cm) recorded about 61 days each winter [Environment and Climate Change Canada (ECCC) 2018]. Over the November 2009–March 2017 study period, 74 warnings for severe winter weather in the forecast region were issued by ECCC, including those for freezing rain (34), snow squalls (20), snowfall (12), winter storms (4), blowing snow (2), blizzards (1), and flash freezes (1).
Average daily climatic conditions (1981–2000) during the winter season for the Region of Waterloo International Airport (ECCC 2018).
Citation: Weather, Climate, and Society 12, 3; 10.1175/WCAS-D-19-0099.1
b. Data
1) EDV records
Disaggregated EDV data were obtained for the period April 2009 through March 2017 through the Canadian Institute for Health Information (CIHI). Records were drawn from the National Ambulatory Care Reporting System (NACRS) for all patients residing in the Cambridge, Kitchener, and Waterloo postal forward sortation areas and presenting to a hospital emergency department with a fall-related injury problem, as defined in the International Classification of Diseases ICD-10-CA. EDV data were sorted into two subsets: 1) falls on the same level involving ice and snow (ICD-10 code W00) and 2) all other types of falls (ICD-10 codes W01-W19). Lacking comprehensive information on the specific location where victims were injured, this distinction was made to isolate the subset of falls that were expected to be most sensitive to hazardous winter weather conditions (W00 falls on the same level involving ice and snow). The second data subset is composed of all other falls, including those that would be expected to occur primarily indoors (e.g., W06 fall involving bed), either indoors or outdoors (e.g., W11 fall on and from ladder), outdoors but with little direct connection to hazardous winter weather (e.g., W02 fall involving skates, skis, sport boards, and in-line skates) and where the activity or situation was unspecified (W19).
As presented in Table 1, over 44 500 EDVs for fall-related injuries spread over eight winter seasons (November–April) were included in the final dataset. Falls involving ice and snow peaked in January, when they accounted for over 22% of all fall-related EDVs, and were most prevalent during the morning, whereas visits for other types of falls were greatest during the late afternoon and early evening hours.
Types of fall-related emergency department visitation (EDV) counts by winter season (November–April).
2) Weather condition data
Mills et al. (2019, p. 189) provide a detailed description of the information sources used to define and characterize variable-length winter storm events and corresponding control periods. A combination of temperature, precipitation, and visibility data from two weather radar stations and four surface observation sites, supplemented with weather and road condition information from motor vehicle collision records, covering the period 2009–17 were obtained from ECCC, Grand River Conservation Authority, and the RMW. These sources were used to characterize winter precipitation events by type (e.g., snow, mixed precipitation), amount of precipitation, thermal environment, and presence of government-issued weather watch and warning bulletins. Weather radar imagery was the primary source of information for determining the selection of winter storm events and controls. When precipitation was observed, the following criteria were used to assign each hour into one of three classes of coverage within the RMW:
class 1: precipitation observed to be just entering RMW, restricted to a very small part of the region, or consisting of widely scattered precipitation of low intensity (<0.5 cm h−1);
class 2: precipitation observed to affect two of the main cities (Cambridge, Kitchener, and Waterloo); and
class 3: precipitation observed over most of RMW and all three main cities.
c. Approach and method
A matched-pair, retrospective cohort method, as detailed in Mills et al. (2019), was applied to estimate fall-related injury risk during winter storm events for the study area. An event was defined as a period of time during which a winter storm occurs and was paired with a dry-weather control matched by duration, hour, and day of week, either one week prior to or one week following the event. Matching periods allowed for some control of nonweather influences on exposure that exhibit regular patterns (e.g., work and recreational activities and trip routines) and are often difficult to measure directly or representatively across a large area or population. Criteria used to determine events and assign controls are elaborated in Table 2 below. For the purpose of the analysis, winter storms were not restricted to the criteria used to issue official ECCC warnings (https://www.canada.ca/en/environment-climate-change/services/types-weather-forecasts-use/public/criteria-alerts.html#winterStorm), but rather included a broader set of events consisting of at least some winter precipitation.
Criteria used to define winter precipitation (storm) events and corresponding control periods.
Estimates of risk were based on odds ratios—the ratio of the probability of a fall-related injury happening during the event condition (i.e., a winter storm) relative to the probability of a fall-related injury occurring during the control condition (i.e., dry weather). The method is congruous with underlying theory established by Fleiss (1973, 110–111). Mills et al. (2011) note that each event–control pair produces four counts: A (injuries during the event period), B (injuries during the control period), C (an estimate of the number of safe outcomes during the event period), and D (an estimate of the number of safe outcomes during the control period). The odds ratio (OR) is calculated as below (Bland and Altman 2000):
Because thousands of safely negotiated walking motions or pedestrian trips occur every hour in large urban centers like the study region, C and D are very large and, for this investigation, were set at 50 000 minus recorded event or control injuries. This level of exposure was derived from a rounded-down survey-based estimate of average daily walking trips made by RMW residents in 2016 (Region of Waterloo staff 2019, personal communication). Sensitivity analysis revealed minimal differences (less than 1%) in final relative risk estimates using exposure values ranging from 1000 to 100 000. Given that we lacked access to continuous trip count (or time or distance walked) data for specific winter storms, exposure was kept constant for both event and control periods. The implications of this decision are discussed later in the paper.
A logarithmic transformation of the sample odds ratio was conducted to ensure that the predictions were approximately normally distributed. A statistical weight for each event–control pair wi based on a fixed-effects model for combining estimates of risk [verified as per Johansson et al. (2009, p. 812)], was calculated by taking the inverse of its variance υi as shown in Eqs. (2) and (3):
where
The weighted mean effect on a set of g event–control pairs
This procedure was followed to obtain estimates of injury risk for three general sets of winter storm events (all storm events, snowfall only, and mixed precipitation) and three fall categories:
emergency department visits for any type of fall (EDVFTOTAL),
emergency department visits for falls occurring on the same level involving ice and snow (EDVFIS), and
emergency department visits for all other types of falls (EDVFOTHER).
Where samples were sufficiently large, relative risks were also tabulated separately for the subset of mixed events that involved freezing rain.
3. Results
Throughout the remainder of the paper, relative risks (RR) are reported in two primary ways: 1) as a mean estimate accompanied by 95% confidence intervals (e.g., RR 1.50; 95% CI: 1.20–1.80), or 2) as a relative mean increase or decrease in fall-related injuries (e.g., 50% more). The analysis included 145 variable-length winter storm events that ranged from 12 to 80 h in duration. Suitable controls for all hours were confirmed for 59 of these events with the remaining 86 having at least half of their event hours matched with acceptable controls. This lower 50% threshold was adopted to ensure a sufficiently large dataset of event–control pairs and injury incidents for disaggregated analysis. A sensitivity analysis confirmed that the effects of this decision on RR estimates were minimal.
a. Overall relative risk of injury
Modest increases in EDVFTOTAL were found during winter storms (RR 1.06; 95% CI: 1.01–1.10). In other words, one may expect between 1% and 10% more injuries to occur during winter storms as defined in this study than during comparable but dry-weather conditions. The added risk was entirely attributable to more frequent same-level falls involving ice and snow (EDVFIS) (RR 1.50; 95% CI: 1.34–1.68), where the increased incidence rate is much higher. Indeed, a reduction in EDV was observed for other forms of falls (EDVFOTHER), several of which occur indoors (RR 0.95; 95% CI: 0.91–1.00).
b. Precipitation type
Contrasting effects on fall-related injury risk were observed for events consisting entirely of snowfall (i.e., snow events) as compared with those involving mixed precipitation (i.e., snowfall and one or more of rainfall, freezing rain, or ice pellets; freezing rain alone or in combination with rainfall or ice pellets). For EDVFIS, snow events produced a smaller increase in risk (RR 1.38; 95% CI: 1.18–1.61) than mixed-precipitation events as a whole (RR 1.65; 95% CI: 1.40–1.96). This difference was statistically significant for the subset of mixed events that involved freezing rain (RR 2.02; 95% CI: 1.64–2.53). No appreciable difference in relative risk for EDVFOTHER was observed for mixed-precipitation events; however, snow events were associated with significantly lower risks (RR 0.91; 95% CI: 0.86–0.98) than during comparable dry-weather control periods.
c. Precipitation amount
Four classes of storm event accumulation, estimated in millimeters of liquid-equivalent precipitation, were evaluated. Actual snow-to-liquid ratios can vary by storm type and temperature (from 3:1 to over 25:1) but are typically assumed to be 10:1 (1 cm or 10 mm of snow is equal to 1 mm of liquid water) where the specific liquid water content of frozen precipitation is not directly measured (Baxter et al. 2005). As indicated in Table 3, EDVFIS mean risk estimates increased with precipitation amount, although there was notable overlap in confidence intervals across all categories. Very light events (≤2 mm) exhibited minimal differences in EDVFIS as compared with dry-weather control conditions. No significant increase or decrease in relative risk was observed for EDVFOTHER for any accumulation class.
Relative risk of falls on the same level involving ice and snow (EDVFIS) during all winter storms.
To compare results by event precipitation type while maintaining sufficient sample size, accumulation thresholds were chosen to split each sample into roughly equal groups, 5 mm for snowfall events and 10 mm for mixed events. Modest (30%–45%) mean increases in EDVFIS were observed during snowfall events whether above or below 5 mm liquid-equivalent accumulation. Significantly greater EDVFIS were found during mixed events with 10 mm or more of precipitation (RR 2.17; 95% CI: 1.77–2.67) when compared with those with less than 10 mm (RR 0.93; 95% CI: 0.69–1.26). For EDVFOTHER, only risks during snow events with less than 5 mm of precipitation were significant, with a mean estimated reduction of 10% (RR 0.90; 95% CI: 0.82–0.98).
d. Temperature effects
Mean event temperatures Tavg were used to distinguish two groups each for snowfall events (warm, with Tavg > −5.5°C, and cold, with Tavg ≤ −5.5°C) and mixed-precipitation events (warm, with Tavg > 0°C, and cold, with Tavg ≤ 0°C). Temperature thresholds were set to maintain similar event sample sizes in each grouping. Mean estimates for same-level falls involving ice and snow during cold snowfall events (RR 1.23; 95% CI: 1.00–1.52) were much lower than for warm snowfalls (RR 1. 60; 95% CI: 1.26–2.03), but again considerable overlap was observed in confidence intervals. Results for the two mixed-precipitation groups were almost identical. For EDVFOTHER, modest reductions in risks were found during warm snowfall events (RR 0.90; 95% CI: 0.82–0.98). Otherwise fall risks did not vary significantly for other temperature classes and event types.
e. Temporal and spatial variation
Relative risks of injury during winter storms were calculated to assess long-term trends, subseasonal patterns, day-of-week variations, and differences across the three cities. Mean estimates of relative risk for EDVFIS fell over the course of the 2009–17 study period (Fig. 2) while risks for EDVFOTHER showed no consistent pattern. Linear trends in mean relative risk estimates for EDVFIS (significance level p = 0.021) and EDVFTOTAL (p = 0.018) were significant. No significant trends in EDVFIS were observed for snow or mixed events when analyzed separately.
Two-season mean risk estimates for same-level falls involving ice and snow (EDVFIS).
Citation: Weather, Climate, and Society 12, 3; 10.1175/WCAS-D-19-0099.1
Monthly fall risks were grouped into early winter (December–January), late winter (February–March), and shoulder (November, April) months to evaluate within-season variations. Mean risk estimates for EDVFIS peaked during winter storms that occurred in the shoulder months (RR 2.83; 95% CI: 1.77–4.53). EDVFIS risks lowered steadily through the early winter months of December–January (RR 1.65; 95% CI: 1.40–1.94) and late winter months of February–March (RR 1.24; 95% CI: 1.05–1.48). The reduction in EDVFOTHER was most prominent in early winter (RR 0.93; 95% CI:0.87–1.00).
Minor differences were apparent by day of the week and city. EDVFIS risks increased by 40% during winter storm events that occurred during weekday hours and by 65% for events that involved any weekend hours. A comparison of the three cities showed that EDVFIS mean risk estimates for Kitchener (1.60) were higher than either Cambridge (1.34) or Waterloo (1.36), but the CIs showed considerable overlap. Reductions in EDVFOTHER during winter storms were significant in Cambridge (RR 0.89; 95% CI: 0.82–0.97) but not for the other cities.
f. Sociodemographic factors
The CIHI NACRS EDV data included information about the gender and age of patients, and analyses were undertaken to discern any implications for relative injury risks. Modest differences were observed across gender, with higher EDVFIS relative risks for females as compared with males during winter storms involving mixed precipitation. EDVFOTHER risks for females were consistently less than 1.0 during winter storms relative to dry-weather controls. In terms of age, little variation in EDVFIS or EDVFOTHER risk was observed between the 18–54 and over-55 groups (Table 4). The youngest cohort (less than 18 yr old) experienced lower risks than older cohorts in every injury category, with EDVFOTHER slightly but significantly less during winter storms than corresponding controls (RR 0.89; 95% CI: 0.81–0.97).
Mean relative risk estimates (with 95% CI in parentheses) of emergency department visits for falls (EDVF) during winter storm events for different age cohorts.
g. Lag impacts
As noted in previous research, fall-contributing snow and ice conditions on walking surfaces may endure for extended periods beyond a winter storm, even considering the three-hour lag built into the standard storm event definition criteria (Table 2). The analysis process described for winter storms was repeated for 6-, 12-, 24-, and 48-h lag periods immediately following the last event hour. Controls were obtained to match the lag-event periods by hour and day of the week either 1 week prior to or following the event. Since the authors were less interested in extensive disaggregation of results and wanted the option of comparing individual event and event-lag risks, only events that had 90% or greater hours with matching controls were examined for lag effects. This provided 76 lag event–control pairs for analysis.
The net effect on EDVFTOTAL was negligible for all four lag periods evaluated relative to dry control periods. Table 5 compares relative risks during winter storm events and the first and second 24-h lag periods, days 1 and 2, respectively, across each of the fall injury groups and storm types. Mean lag EDVFIS risk estimates for “all events,” and the mixed-precipitation subset remained elevated on day 1 but was consistently lower than those observed during winter storms. While the mean EDVFIS risk estimate remained above 1.0, no significant lag effect was found for snowfall events. For EDVFOTHER, the opposite was observed, with significantly reduced risks estimated during day-1 lag periods following snowfall events but no appreciable effect following mixed-precipitation events. No meaningful difference was observed between day-2 lag periods and corresponding dry-weather controls.
Mean relative risk estimates (with 95% CI in parentheses) of EDVF during lag periods immediately following winter storm events. Boldface type indicates results that are significantly greater or less than 1.0.
h. Impact of weather watches and warnings
As mentioned in the introduction, government-issued communications about expected (watch bulletins) and imminent or occurring (warning bulletins) severe winter weather have been used in previous research as proxy indicators of dangerous, impactful events. The effects of these communications were evaluated in this study by comparing relative risks in events for which a watch or warning was in place during any hour with those for events during which no bulletin was activated. Substantive differences were observed only for EDVFIS. As shown in Table 6, mean relative risk estimates were about 40% lower during snow events and about 20% higher during mixed events for which watches or warnings were issued as compared with events lacking such bulletins. It is noted that the sample of snow events with watches or warnings is small.
Relative risk of EDVFIS during winter storms differentiated by precipitation type and presence of weather watch or warning.
i. Comparisons with motor vehicle collision injury risks
Mills et al. (2019) describe a parallel analysis of injury and property-damage-only motor vehicle collisions (MVCs) during winter storms. The collision data from that study were used to extract injury counts for an overlapping time frame (2009–16) and common set of 96 winter storm events and control periods. This facilitated a direct comparison and union of relative injury risks and absolute outcomes for two dominant modes of mobility in the region. Results presented in Table 7 indicate that winter storms involving any type of precipitation present higher relative risks of injury for EDVFIS than for injuries associated with MVC. Differences were greatest for mixed-precipitation storms involving either freezing rain or with accumulations of 10 mm or more, with EDVFIS increasing by 108% and 139%, and MVC increasing by 43% and 45%, respectively, during the same set of winter storms. In absolute terms, EDVFIS and MVC accounted for 62% and 38% of the total injury burden attributable to winter storms, respectively.
Relative risks of MVC and EDVFIS during winter storms differentiated by precipitation type (2009–16).
Combining the two sets of weather-sensitive injuries provides a multimodal estimate of mobility-related winter storm injury risks (RR 1.59; 95% CI: 1.43– 1.76). As expected, mixed-precipitation storms (RR 1.79; 95% CI: 1.54–2.09) were more problematic than storms involving only snowfall (RR 1.43; 95% CI: 1.24–1.65).
4. Discussion
The study provides the first comparable citywide estimates of fall and motor vehicle injury collision risk associated with variable-length winter storms. Winter storms, as defined in this investigation, were found to significantly increase the incidence of EDV for same-level falls involving ice and snow. Common sense alone suggests that such falls should occur more frequently as snow and/or mixed precipitation coats surfaces, compacts or freezes, becomes very slippery, and therefore presents challenges to pedestrians. The greater risks estimated for mixed-precipitation events than for storms consisting only of snowfall are consistent with previous research that associates slipperiness with surface temperatures near 0°C (Andersson and Chapman 2011). Snowfall at colder temperatures requires considerable mechanical packing (i.e., through foot or vehicular traffic) and/or warming to become slippery—conditions that may not always materialize before routine winter maintenance removes accumulated snow from walkways, streets, and parking areas and applies salt or sand to improve traction. This reasoning may also serve to explain in part why lag effects for same-level falls on ice and snow were weak to nonexistent for snowfall events and observed to only last for the first 24 h following mixed-precipitation winter storms. Mixed precipitation, especially events with freezing rain or falling temperatures, leads to ice accumulations that are likely more difficult to physically or chemically remove than snowfall. In most cases, however, these are likely mitigated within a day in regions like the study area that have considerable resources dedicated to winter maintenance and municipal bylaws setting minimum standards for sidewalk clearance (e.g., City of Waterloo 2009).
The study results suggest that same-level falls involving ice and snow likely contribute more to the total winter-storm-related injury burden than motor vehicle collisions. Estimated risk of same-level falls involving ice and snow during winter storms was 24% higher than for injuries associated with motor vehicle collisions relative to dry-weather control periods. Because identical sets of event and control periods were used in both analyses, it was also possible to compare absolute effects and combine counts to develop an integrated injury risk estimate. Same-level falls on ice and snow accounted for 64% more of the injury burden attributable to winter storms than motor vehicle collision injuries over the 96 storms examined. Slight variations in minimum injury severity definitions (e.g., the ED fall injury definition is more strict) and jurisdictional coverage (e.g., MVC data include townships surrounding cities) result in a net underrepresentation of falls relative to MVC injuries, thus the importance of winter storms on falls noted in this paper is likely conservative. Therefore, returning to the title of this paper and within the confines of this particular study, driving appears to be a safer mode of travel than walking during winter storms. Although this finding demands additional research (e.g., injury severity and exposure aspects) and corroboration from studies in other locations, it supports greater consideration of pedestrian injuries when developing and evaluating policies to promote active transportation in winter cities. This includes greater investment in actions to reduce fall-related injuries during winter storms, such as improvements to winter maintenance servicing of sidewalks, paths, or walkways. More generally, it calls for an integrated look at transportation safety that extends beyond motorized vehicle–pedestrian interactions that comprise much of the literature. Toward this end, combining MVC and fall injury data into an aggregate analysis allowed the authors to provide a more complete picture of mobility-related injury risk during winter storms (59% increase in injuries) thereby addressing a need raised by the transportation safety research community (e.g., Methorst et al. 2017; Schepers et al. 2017).
While this study adds further evidence demonstrating that winter storms lead to increased risk of injury, both from falls and motor vehicle collisions, it also uncovered features that suggest these risks are variable and dynamic. Several stratified analyses were used to attempt to identify patterns or changes in risk that might be linked to societal factors including intentional interventions to influence exposure or sensitivity to winter weather hazards. A statistically significant linear downward trend of about 10% per season was observed in relative winter storm risk for same-level falls on ice and snow. No such trend was found for other types of falls. However, Mills et al. (2019) observed a similar trend over a longer time frame (2002–16) for motor vehicle injury collisions, although at roughly one-fifth of the rate estimated for EDVFIS in the current analysis. No obvious explanation for the trend in EDVFIS injuries was apparent to the authors, but some combination of shifts in activity patterns, decisions to seek treatment, emergency response and advances in medicine, awareness and protective actions on the part of the public, and improvements in winter maintenance practices or transportation infrastructure might be interacting to produce a safer environment as indicated by better injury outcomes. The lack of significant trends in relative risk for either snowfall or mixed-precipitation events, when analyzed separately, suggest a more complex interactive effect that requires a longer study time frame in order to be resolved.
Variation in risk was evident at shorter time scales as well, with same-level falls involving ice and snow up to 183% greater during the shoulder season months of November and April as compared with 65% during early winter (December–January) and 24% in late winter (February–March) months. This finding lends support to the theory that pedestrians, much like drivers, take greater precautions to reduce their risks as they once again encounter and become familiar with winter conditions, with the reverse process taking place in April (i.e., surprise storms). The results might also be explained, however, by the greater chance of having ice and snow on walking surfaces—and therefore EDV for same-level falls involving ice or snow—during colder midwinter control periods as compared with shoulder months when snow and ice typically completely melt away between precipitation events. This would have the effect of reducing relative risks outside of the shoulder months.
The analysis also revealed interesting relationships between winter storm events and the risk of other types of falls. In several cases, winter storm events were associated with a significant decrease in EDV for injuries other than those occurring on the same level involving ice and snow, many of which occur indoors. Estimated mean decreases ranged from 8% to 15% during winter storms relative to dry control periods. Given that reductions were only prevalent for particular storm types, age cohorts and months, the results cannot be simply interpreted as a shift from EDVFOTHER to EDVFIS whereby it is assumed that people experiencing same-level falls involving ice and snow would have suffered another type of fall had there been no winter storm. It is not clear whether the findings result from reduced exposure (e.g., by encouraging people to stay indoors, walk less, or take greater precautions) or deferred visits to an emergency department (i.e., falls still occur but treatment is not sought immediately).
From an intervention perspective, it would be important to ensure that any strategies or interventions designed to reduce the risks of same-level falls on ice and snow do not incidentally remove whatever protective effect winter storms seem to exert on risks for other types of falls. One such intervention involves the issuance of official warnings to alert the public about severe winter weather, possible impacts, and calls for risk-reducing actions, for example, suggesting a higher likelihood of slippery roads and recommending restriction of nonessential travel. The intent is to provide information that could directly or indirectly affect exposure, other risk-mitigating behavior, and weather-sensitive outcomes such as the incidence of same-level falls involving ice and snow. A simple stratification of winter storms with and without watches or warnings was analyzed to assess any effects. Risks were observed to be lower during snowfall events with watches or warnings than those without any alerts, though the opposite was found for mixed-precipitation events. While these results were not statistically significant, they still imply that people may respond by adjusting their exposure or by taking other actions to reduce their sensitivity to snowfall events. They might also do this in response to mixed-precipitation storms, but less effectively. Unfortunately, it was not possible to discern from the analysis the extent to which the responses were simply coincidental, habitual, due to intentional consideration and behavior in reaction to weather warning information, or a function of seeing, experiencing, and reacting to the storm as it occurred. In the case of snowfall, a warned event could simply indicate a very heavy snowfall in which case one might interpret lower risks as a sign that people decided to remain inside, attempted to travel but altered plans because of deep snow, or experienced a fall but decided not to risk travel to an emergency department—the warning message may have had little impact on the behavior and outcomes. For mixed-precipitation events, in particular freezing rain, it is plausible that hazards initially appear less dangerous to people and decisions are thus made to venture out.
With respect to exposure, a few observational and stated adaptation survey studies have demonstrated that walking participation may decrease by a few to several percent during snowfall (e.g., Aultman-Hall et al. 2009; Böcker et al. 2013; Clarke et al. 2017), though the effect is likely not linear nor consistent across different types of winter storms, amounts of precipitation, or during the course of individual events. Lacking the data necessary to resolve trip counts at the scale of individual winter storms, exposure was held constant in the current investigation for both events and corresponding control periods. The primary effect of this decision was to generate more conservative RR estimates. For illustration, after reanalyzing the overall relative risk of EDVFIS assuming a 5% drop in exposure during all winter storm events, the mean estimate rose from 1.50 to 1.58. Reduction in event exposure would need to exceed 10% for these differences to be significant and ensure that the revised RR falls outside the confidence intervals of the original estimate (RR 1.50; 95% CI: 1.34–1.68).
5. Conclusions
This study examined relationships between winter storms and the occurrence of fall-related injuries for a midsized urban community in Ontario. Main findings are as follows:
Winter storms, in particular those involving mixed precipitation, were associated with significant increases in the incidence of same-level falls that involve ice and snow.
Significant reductions in mean relative risk estimates were apparent over the 2009–17 study period, suggesting possible long-term shifts in exposure, sensitivity, and/or risk-mitigating decisions, actions, and behavior.
Lag effects for same-level falls involving ice and snow were weak to nonexistent for snowfall events and were observed to only last for the first 24 h following mixed-precipitation winter storms, most likely because of effective winter maintenance in the study region.
Same-level falls on ice and snow likely contribute more to the total winter-storm-related injury burden than motor vehicle collisions do.
Clear, consistent, and significant effects of government-issued watch and warning communications on risk outcomes were not found.
Additional sources of injury data (e.g., physician or clinic visitation) may have provided a more comprehensive estimate of fall-related risk but would not likely have led to different associations with winter storms. Such data would be useful in conducting research on the severity of fall injuries and longer-term patient outcomes, something worthy of study but beyond the scope of the current investigation. In addition to estimating overall risks, retrospective population-based analyses of secondary data as conducted in this study seem best suited to raising or pointing to questions about why or how specific interventions such as weather warnings influence behavior and risk outcomes, but they are less useful in explaining processes operating at finer social scales within winter storm events.
It is recommended that those engaged in developing injury prevention strategies and related public risk messaging, in particular winter weather warnings and advisories, should place additional emphasis on falls and multimodal injury risks in future communications related to winter storm hazards. Strategies or interventions developed to reduce the risks of same-level falls on ice and snow should be designed not to incidentally remove the protective effect winter storms were found to have on risks for other types of falls. Additional research is required to assess the generalizability of the findings presented in this investigation, more precisely monitor pedestrian exposure, and to explore in greater detail the mechanisms through which weather, weather-related risk information, and associated interventions influence behavior and risk outcomes.
Acknowledgments
The authors acknowledge Environment and Climate Change Canada and the Grand River Conservation Authority for providing weather data used in the study. Injury data and information were obtained from the Canadian Institute for Health Information. The analyses, conclusions, opinions, and statements expressed herein are those of the authors and not those of any of these organizations. The authors also thank the reviewers for their constructive comments. This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.
REFERENCES
Andersson, A., and L. Chapman, 2011: The use of a temporal analogue to predict future traffic accidents and winter road conditions in Sweden. Meteor. Appl., 18, 125–136, https://doi.org/10.1002/met.186.
Arnold, C. M., V. P. M. Dal Bello-Haas, J. P. Farthing, K. L. Crockett, C. R. A. Haver, G. Johnston, and J. Basran, 2016: Falls and wrist fracture: Relationship to women’s functional status after age 50. Can. J. Aging, 35, 361–371, https://doi.org/10.1017/S0714980816000337.
Aultman-Hall, L., D. Lane, and R. R. Lambert, 2009: Assessing impact of weather and season on pedestrian traffic volumes. Transp. Res. Rec., 2140, 35–43, https://doi.org/10.3141/2140-04.
Avery, J. G., 1982: Fractures during ice and snow. Br. Med. J., 284, 270, https://doi.org/10.1136/bmj.284.6311.270.
Bamzar, R., and V. Ceccato, 2015: The nature and the geography of elderly injuries in Sweden. GeoJournal, 80, 279–299, https://doi.org/10.1007/s10708-014-9552-z.
Baxter, M. A., C. E. Graves, and J. T. Moore, 2005: A climatology of snow-to-liquid ratio for the contiguous United States. Wea. Forecasting, 20, 729–744, https://doi.org/10.1175/WAF856.1.
Bell, J., L. Gardner, and D. P. Landsittel, 2000: Slip and fall-related injuries in relation to environmental cold and work location in above-ground coal mining operations. Amer. J. Ind. Med., 38, 40–48, https://doi.org/10.1002/1097-0274(200007)38:1<40::AID-AJIM5>3.0.CO;2-F.
Bentley, T., and R. Haslam, 2001: Identification of risk factors and countermeasures for slip, trip and fall accidents during the delivery of mail. Appl. Ergon., 32, 127–134, https://doi.org/10.1016/S0003-6870(00)00048-X.
Berggård, G., and C. Johansson, 2010: Pedestrians in wintertime—Effects of using anti-slip devices. Accid. Anal. Prev., 42, 1199–1204, https://doi.org/10.1016/j.aap.2010.01.011.
Bland, J. M., and D. G. Altman, 2000: Statistics notes: The odds ratio. BMJ, 320, 1468, https://doi.org/10.1136/bmj.320.7247.1468.
Bobb, J. F., K. K. L. Ho, R. W. Yeh, L. Harrington, A. Zai, K. P. Liao, and F. Dominici, 2017: Time-course of cause-specific hospital admissions during snowstorms: An analysis of electronic medical records from major hospitals in Boston, Massachusetts. Amer. J. Epidemiol., 185, 283–294, https://doi.org/10.1093/aje/kww219.
Böcker, L., M. Dijst, and J. Prillwitz, 2013: Impact of everyday weather on individual daily travel behaviours in perspective: A literature review. Transp. Rev., 33, 71–91, https://doi.org/10.1080/01441647.2012.747114.
Bulajic-Kopjar, M., 2000: Seasonal variations in incidence of fractures among elderly people. Inj. Prev., 6, 16–19, https://doi.org/10.1136/ip.6.1.16.
Bylund, P.-O., J. Johansson, and P. Albertsson, 2016: Injuries sustained during snow removal from roofs resulting in hospital care. Int. J. Inj. Control Saf. Promot., 23, 105–109, https://doi.org/10.1080/17457300.2014.992349.
Chiu, K. Y., T. P. Ng, and S. P. Chow, 1996: Seasonal variation of fractures of the hip in elderly persons. Injury, 27, 333–336, https://doi.org/10.1016/0020-1383(95)00232-4.
City of Waterloo, 2009: Snow & ice removal by-law. Corporation of the City of Waterloo By-law 09-156, 7 pp., https://www.waterloo.ca/en/government/resources/Documents/By-law/Snow-removal-bylaw.pdf.
Clarke, P., J. A. Hirsch, R. Melendez, M. Winters, J. Sims Gould, M. Ashe, S. Furst, and H. McKay, 2017: Snow and rain modify neighbourhood walkability for older adults. Can. J. Aging, 36, 159–169, https://doi.org/10.1017/S071498081700006X.
Crawford, J. R., and M. J. Parker, 2003: Seasonal variation of proximal femoral fractures in the United Kingdom. Injury, 34, 223–225, https://doi.org/10.1016/S0020-1383(02)00211-5.
ECCC, 2018: Canadian Climate Normals. Environment and Climate Change Canada, accessed 10 October 2018, http://climate.weather.gc.ca/climate_normals/index_e.html.
Elvik, R., and T. Bjørnskau, 2019: Risk of pedestrian falls in Oslo, Norway: Relation to age, gender and walking surface condition. J. Transp. Health, 12, 359–370, https://doi.org/10.1016/j.jth.2018.12.006.
Fleiss, J. L., 1973: Statistical Methods for Rates and Proportions. John Wiley and Sons, 223 pp.
Gevitz, K., R. Madera, C. Newbern, J. Lojo, and C. C. Johnson, 2017: Risk of fall-related injury due to adverse weather events, Philadelphia, Pennsylvania, 2006-2011. Public Health Rep., 132, 53S–58S, https://doi.org/10.1177/0033354917706968.
Gyllencreutz, L., J. Björnstig, E. Rolfsman, and B.-I. Saveman, 2015: Outdoor pedestrian fall-related injuries among Swedish senior citizens—Injuries and preventive strategies. Scand. J. Caring Sci., 29, 225–233, https://doi.org/10.1111/scs.12153.
Hagino, H., and Coauthors, 2004: Nationwide survey of hip fractures in Japan. J. Orthop. Sci., 9, 1–5, https://doi.org/10.1007/s00776-003-0741-8.
Hajat, S., Z. Chalabi, P. Wilkinson, B. Erens, L. Jones, and N. Mays, 2016: Public health vulnerability to wintertime weather: Time-series regression and episode analyses of national mortality and morbidity databases to inform the Cold Weather Plan for England. Public Health, 137, 26–34, https://doi.org/10.1016/j.puhe.2015.12.015.
Hassi, J., L. Gardner, S. Hendricks, and J. Bell, 2000: Occupational injuries in the mining industry and their association with statewide cold ambient temperatures in the USA. Amer. J. Ind. Med., 38, 49–58, https://doi.org/10.1002/1097-0274(200007)38:1<49::AID-AJIM6>3.0.CO;2-3.
Hemenway, D., and G. A. Colditz, 1990: The effect of climate on fractures and deaths due to falls among white women. Accid. Anal. Prev., 22, 59–65, https://doi.org/10.1016/0001-4575(90)90007-8.
Jacobsen, S., D. Sargent, E. Atkinson, W. O’Fallon, and L. Melton III, 1999: Contribution of weather to the seasonality of distal forearm fractures: A population-based study in Rochester, Minnesota. Osteoporosis Int., 9, 254–259, https://doi.org/10.1007/s001980050145.
Johansson, O., P. Wanvik, and R. Elvik, 2009: A new method for assessing the risk of accident associated with darkness. Accid. Anal. Prev., 41, 809–815, https://doi.org/10.1016/j.aap.2009.04.003.
Lauritzen, J. B., P. Schwartz, P. McNair, B. Lund, and I. Transbol, 1993: Radial and humeral fractures as predictors of subsequent hip, radial or humeral fractures in women, and their seasonal variation. Osteoporosis Int., 3, 133–137, https://doi.org/10.1007/BF01623274.
Levy, A., N. E. Mayo, and G. Grimard, 1995: Rates of transcervical and pertrochanteric hip fractures in the Province of Quebec, Canada, 1981-1992. Amer. J. Epidemiol., 142, 428–436, https://doi.org/10.1093/oxfordjournals.aje.a117651.
Levy, A., D. Bensiom, N. Mayo, and H. Leighton, 1998: Inclement weather and the risk of hip fracture. Epidemiology, 9, 172–177, https://doi.org/10.1097/00001648-199803000-00012.
Lipscomb, H. J., J. E. Glazner, J. Bondy, K. Guarini, and D. Lezotte, 2006: Injuries from slips and trips in construction. Appl. Ergon., 37, 267–274, https://doi.org/10.1016/j.apergo.2005.07.008.
Luukinen, H., K. Koski, and S.-L. Kivelä, 1996: The relationship between outdoor temperature and the frequency of falls among the elderly in Finland. J. Epidemiol. Community Health, 50, 107, https://doi.org/10.1136/jech.50.1.107.
Marshall, E. G., L. Shou-En, S. Zhengyang, J. Swerdel, M. Borjan, and M. E. Lumia, 2016: Work-related unintentional injuries associated with Hurricane Sandy in New Jersey. Disaster Med. Public Health Prep., 10, 394–404, https://doi.org/10.1017/dmp.2016.47.
Methorst, R., P. Schepers, N. Christie, M. Dijst, R. Risser, D. Sauter, and B. Wee, 2017: ‘Pedestrian falls’ as necessary addition to the current definition of traffic crashes for improved public health policies. J. Transp. Health, 6, 10–12, https://doi.org/10.1016/j.jth.2017.02.005.
Mills, B., J. Andrey, and D. Hambly, 2011: Analysis of precipitation-related motor vehicle collision and injury risk using insurance and police record information for Winnipeg, Canada. J. Safety Res., 42, 383–390, https://doi.org/10.1016/j.jsr.2011.08.004.
Mills, B., J. Andrey, B. Doberstein, S. Doherty, and J. Yessis, 2019: Changing patterns of motor vehicle collision risk during winter storms: A new look at a pervasive problem. Accid. Anal. Prev., 127, 186–197, https://doi.org/10.1016/j.aap.2019.02.027.
Modarres, R., T. B. M. J. Ouarda, A. Vanasse, M. G. Orzanco, and P. Gosselin, 2014: Modeling climate effects on hip fracture rate by the multivariate GARCH model in Montreal region, Canada. Int. J. Biometeorol., 58, 921–930, https://doi.org/10.1007/s00484-013-0675-6.
Mondor, L., K. Charland, A. Verma, and D. L. Buckeridge, 2015: Weather warnings predict fall-related injuries among older adults. Age Ageing, 44, 403–408, https://doi.org/10.1093/ageing/afu199.
Morency, P., C. Voyer, S. Burrows, and S. Goudreau, 2012: Outdoor falls in an urban context: Winter weather impacts and geographical variations. Can. J. Public Health, 103, 218–222, https://doi.org/10.1007/BF03403816.
Murray, I. R., C. R. Howie, and L. C. Biant, 2011: Severe weather warnings predict fracture epidemics. Injury, 42, 687–690, https://doi.org/10.1016/j.injury.2010.12.012.
Nyman, S. R., C. Ballinger, J. E. Phillips, and R. Newton, 2013: Characteristics of outdoor falls among older people: A qualitative study. BMC Geriatr., 13, 125, https://doi.org/10.1186/1471-2318-13-125.
Parachute, 2015: The cost of injury in Canada. Parachute Rep., 177 pp., https://parachute.ca/wp-content/uploads/2019/06/Cost_of_Injury-2015.pdf.
Peden, M., K. McGee, and G. Sharma, 2002: The Injury Chart Book: A Graphical Overview of the Global Burden of Injuries. World Health Organization, 76 pp.
Pipas, L., N. Schaefer, and L. Brown, 2002: Falls from rooftops after heavy snowfalls: The risk of snow clearing activities. Amer. J. Emerg. Med., 20, 635–637, https://doi.org/10.1053/ajem.2002.35494.
Ràliš, Z. A., 1981: Epidemic of fractures during period of snow and ice. Br. Med. J., 282, 603–605, https://doi.org/10.1136/bmj.282.6264.603.
Ràliš, Z. A., E. A. Barker, I. J. Leslie, W. J. Morgan, A. C. Ross, and S. H. White, 1988: Snow-and-ice fracture in the UK, a preventable epidemic. Lancet, 331, 589–590, https://doi.org/10.1016/S0140-6736(88)91383-9.
RMW, 2018: Year-end 2017 population and household estimates for Waterloo region. Region of Waterloo Planning, Development and Legislative Services, Community Planning Rep. PDL-CPL-17-12, 2 pp.
Schepers, P., B. den Brinker, R. Methorst, and M. Helbich, 2017: Pedestrian falls: A review of the literature and future research directions. J. Safety Res., 62, 227–234, https://doi.org/10.1016/j.jsr.2017.06.020.
Smith, K. R., and Coauthors, 2014: Human health: Impacts, adaptation, and co-benefits. Climate Change 2014: Impacts, Adaptation, and Vulnerability, Part A: Global and Sectoral Aspects, C. B. Field et al., Eds., Cambridge University Press, 709–754.
Smith, R., and D. Nelson, 1998: Fractures and other injuries from falls after an ice storm. Amer. J. Emerg. Med., 16, 52–55, https://doi.org/10.1016/S0735-6757(98)90065-1.
Stansbury, L. G., A. A. Swinson, J. M. Schmitt, M. B. Stevens, K. R. Kobayashi, C. P. Vangellow, and D. Lampkins, 1995: Work-related injuries associated with falls during ice storms—National Institutes of Health, January 1994. MMWR Wkly., 44, 920–922.
Transport Canada, 2016: Canadian motor vehicle traffic collision statistics 2014 collected in cooperation with the Canadian Council of Motor Vehicle Administrators. Transport Canada Doc. T45-3E-PDF, 6 pp., https://www.tc.gc.ca/media/documents/roadsafety/cmvtcs2014_eng.pdf.