The Slip and Fall Index: Assessing the risk of slipping and falling on ice

Abstract

Objectives

Canadians are at an increased risk of outdoor slip and fall accidents during periods of ice and snow. The aim of this study was to create an index to alert the public of slippery outdoor conditions and promote pedestrian safety.

Methods

Emergency department (ED) presentations from the four adult hospitals in Calgary, Alberta, Canada, over an 11-year period (January 2008‒December 2018) were extracted and filtered using the ICD-10 code W00 (fall due to ice and snow). Multivariable dispersion-corrected Poisson regression models were used to determine the variables most predictive of these presentations. Month of year, the presence of ice, snow on ground (per 10 cm), and interactions between ice and snow, all up to 3 days prior, were used to create the Slip and Fall Index (SFI).

Results

The dataset included 14,977 slip and fall on ice/snow ED presentations. Females (57.36%, n = 8591) accounted for more presentations than males (42.64%, n = 6386). All months had a significant effect, either being predictive or protective of slip and falls on ice/snow. Current-day ice, snow on ground, and ice up to 3 days prior were predictive of increased presentations. Month and measurements of ice and snow can be input into the SFI, which generates the level of daily risk.

Conclusion

The SFI is the first Canadian index with the purpose of measuring the risk of having a slip and fall accident on ice/snow.

Introduction

Falls are the leading cause of unintentional injuries requiring hospitalization in Canada for all age groups (Yao et al., 2020). The short- and long-term consequences of fall-related injuries significantly burden the Canadian healthcare system. Falls accounted for the highest cost of all injuries sustained in 2018 in Canada, amounting to $10.3 billion (Parachute, 2018). Those aged 65 and over are hospitalized at significantly higher rates due to fall-related injuries than all other age groups; these incidents are often followed by devastating physical and psychological consequences including functional decline, loss of autonomy, depression, and increased morbidity (Terroso et al. 2013; Yao et al., 2020).

The number one cause of sport and winter injuries is falls on ice, accounting for over a third of hospitalizations in this category (Canadian Institute for Health Information [CIHI], 2020). The prevalence of slip and fall on ice/snow injuries is variable across the country, with Albertans being affected at a disproportionate rate—hospitalizations due to falls on ice in Alberta are nearly three times more numerous than those in Ontario (Fletcher, 2017). The relationship between winter weather and increased outdoor fall risk has been well documented in many parts of the world (Gevitz et al., 2017; Unguryanu et al., 2020; Van den Brand et al., 2014; Vergouwen et al., 2021), and programs have been tested to reduce the rates of outdoor falls (Chippendale et al., 2023; Holmberg et al., 2021). However, there is no established system for capturing the risk of falling on ice on a quantitative or qualitative scale. The purpose of this study is to address this gap by creating a preliminary measurement tool for assessing the pedestrian risk of falling on ice/snow and suffering a severe orthopaedic injury resulting in an ED visit.

Objectives

The aim of this study was to generate a preliminary public health warning tool that measures the risk of slipping and falling on ice/snow and presenting with an orthopaedic injury. Alerts currently in place to inform the public of potentially dangerous outdoor conditions include the UV Index (UVI) (World Health Organization, 2002) and the Air Quality Health Index (AQHI) (Environment Canada, 2023). To the best of our knowledge, no mass-implemented alert system exists to warn civilians when the risk of falling on ice is high. In our previous research, ice and snow were related to a significant increase in overall trauma volumes in Calgary (Vergouwen et al., 2021). The present study aims to incorporate these observations into a predictive model for assessing risk. By communicating hazardous conditions to the public, appropriate pre-emptive actions for risk mitigation can be taken. Past emergency department (ED) presentation patterns and weather variables in Calgary, Alberta, were analyzed to generate a predictive model for measuring the daily risk of slipping and falling on ice and presenting with an orthopaedic injury. The current Slip and Fall Index (SFI) in this study is a prototype version of a warning system intended to be refined into a province-wide index.

Methods

ED presentations to the four major adult hospitals in Calgary, Alberta, Canada, were extracted from the National Ambulatory Care Reporting System (NACRS). Presentations over an 11-year period (January 2008 to December 2018) were filtered based on the ICD-10 diagnostic code for fall due to ice and snow (W00). Spine, head and neck, and non-trauma injuries were excluded to contain the dataset to orthopaedic trauma. Weather data were collected from the Environment Canada weather station at Calgary International Airport and Alberta Road Weather Information System (RWIS) sensors. Presence of ice was assessed using previously established criteria, outlined in Appendix 2 (Vergouwen et al., 2021).

Multivariable dispersion-corrected Poisson regression models were used to determine the weather and time of year variables predictive of slip and fall on ice/snow injuries. October was used as the reference month. A complete list of explored weather variables can be found in Appendix 1. Akaike’s Information Criterion (AIC) was used to guide selection of our model. The independent effect of ice up to 3 days prior, snow on ground (per 10 cm) up to 3 days prior, interactions between ice and snow up to 3 days prior, and month were used to create the preliminary SFI.

In a sensitivity analysis, we investigated a model which excluded month effects under the hypothesis that snow and ice could explain seasonal effects, and compared predictive utility versus the primary model. Finally, as described above, in the primary analysis, snow on ground effect was per 10 cm. In a sensitivity analysis, to investigate possible non-linear effects, we fit an additional model with a three-level categorical variable (none, ≤ 5 cm, > 5 cm), and compared significance and directionality of the snow on ground effects versus the primary model.

Results

The dataset included 14,977 slip and fall on ice/snow events. The age and sex distribution of the dataset are shown in Fig. 1. More females (57.36%, n = 8591) presented to the ED because of a slip and fall on ice/snow over the study period than males (42.64%, n = 6386). The annual distribution of these presentations is shown in Fig. 2. The three months with the most ED presentations were January (n = 3581), February (n = 2997), and March (n = 2954). These three months were predictive of significantly increased presentations and accounted for over half of all presentations throughout the year (p < 0.001) (Fig. 2, Table 1). April, November, and December were also significant predictors of slip and fall on ice/snow accidents (p < 0.001) (Table 1). May through September displayed a protective effect against slip and fall on ice/snow incidents (p < 0.001). Same-day ice and same-day snow on the ground were both independently predictive of increased accidents (p < 0.001), and their interaction on the same day was protective (p < 0.001). Ice 1 day prior (p = 0.011), 2 days prior (p = 0.018), and 3 days prior (p = 0.049) were also predictive of slip and fall on ice/snow incidents. Snow 1 day prior was mildly protective against these accidents, but this effect was not significant (p = 0.589) (Table 1).

Fig. 1.

Total number of emergency department (ED) presentations due to slip and fall on ice (ICD-10 code W00) from 2008 to 2018 in Calgary by age at time of presentation and sex

Fig. 2.

Annual distribution of ED presentations associated with slip and fall injuries on ice/snow during 2008–2018 in adult Calgary hospitals

Table 1.

Effect of selected variables on ED presentations due to slip and fall on ice and/or snow (ICD-10 code W00)

	Count ratio (95% CI)​	p-value
Same day
Snow on grounda	1.291 (1.098–1.518)	0.002*
Ice	1.537 (1.388–1.702)	< 0.001*
Snow and ice interaction	0.727 (0.618–0.855)	< 0.001*
1 day prior
Snow on grounda	0.938 (0.742–1.184)	0.589
Ice	1.159 (1.034–1.298)	0.011*
Snow and ice interaction	1.116 (0.937–1.328)	0.218
2 days prior
Snow on grounda	1.034 (0.820–1.303)	0.780
Ice	1.149 (1.024–1.288)	0.018*
Snow and ice interaction	1.172 (0.993–1.382)	0.061
3 days prior
Snow on grounda	1.094 (0.931–1.285)	0.278
Ice	1.109 (1.000–1.230)	0.049*
Snow and ice interaction	1.106 (0.955–1.281)	0.180
Time of year (October used as reference month)
January	7.313 (5.680–9.415)	< 0.001*
February	6.677 (5.190–8.589)	< 0.001*
March	5.667 (4.396–7.305)	< 0.001*
April	1.751 (1.314–2.334)	< 0.001*
May	0.161 (0.085–0.305)	< 0.001*
June–July–August	0.017 (0.006–0.049)	< 0.001*
September	0.160 (0.086–0.297)	< 0.001*
November	3.761 (2.899–4.878)	< 0.001*
December	6.506 (5.055–8.375)	< 0.001*

*Statistically significant (p < 0.05)

^aSnow on ground effect was measured per 10 cm

Although snow on ground across previous days was not statistically significant, its inclusion on all 4 days improved predictive utility per AIC compared to its exclusion, and therefore was included in the SFI model. Similarly, although only present-day ice-snow interaction was statistically significant, the AIC of the model was improved when all days’ interactions were retained. In the additional sensitivity analysis which excluded month effects, model AIC increased, indicating that there is more at play in seasonality effects on slip and fall on ice/snow incidents than only presence of ice and amount of snow, and reinforcing the importance of retaining this highly significant predictor. Finally, in the sensitivity analysis in which we fit additional models with a three-level categorical variable (none, ≤ 5 cm, > 5 cm), significance and directionality of the snow on ground effects did not change. We found no evidence of a non-linear relationship between snow on ground and risk of slip and fall.

The SFI output represents the daily expected number of ED presentations due to a fall on ice/snow. Daily ice (present or not present), snow (centimetres), and month of the year can be input into the SFI (Fig. 3). The index then computes the risk of falling due to ice/snow conditions and having an orthopaedic injury on a 40-point scale based on the count ratio for each independent variable and interaction term. Lower ranges indicate a lower risk of slipping and falling on ice, and higher ranges indicate a high risk (Fig. 3). Upon testing the index, average index output showed a similar trend to same-day snow on ground and same-day presence of ice, as well as average ED presentations, but we noted that the average index even in “high-risk” months did not near 40 (Fig. 4, Table 2).

Fig. 3.

The Slip and Fall Index

Fig. 4.

Monthly average calculated index output, average snow on ground (cm), average ice (yes = 1; no = 0), and average ED presentations due to fall on ice and/or snow in Calgary across 2008–2018

Table 2.

Monthly average ED presentations due to slip and fall injuries on ice/snow during 2008–2018 in adult Calgary hospitals and mean calculated Slip and Fall Index output scores

Month	ED presentations	Mean index output
January	3581	10.357
February	2997	9.404
March	2954	8.153
April	820	1.868
May	58	0.132
June	7	0.014
July	0	0.014
August	10	0.014
September	41	0.130
October	400	0.953
November	1541	4.733
December	2568	7.894

Discussion

The relationship between winter weather and increased rates of outdoor fall accidents has been well documented in literature, but a gap exists in measuring and communicating this increased fall risk. A single report of a fall-risk assessment tool was presented in 2004, Finland, that generated 5–10 warnings per year on the most slippery days (Ruotsalainen et al., 2004). The significant impact of winter falls on Canadians and specifically Albertans presents the urgent need for new injury-prevention strategies. We determined the variables most predictive of slip and fall on ice/snow incidents leading to an ED presentation using data from Calgary, Alberta, Canada. A data-driven tool that measures the risk of slipping and falling on ice and snow was created to translate these observations into practice. The size of this dataset and the traditional statistics used are an ideal foundation for establishing a machine learning–driven index that can be scaled up for use in all of Alberta and eventually across Canada.

The sample used to construct the SFI consists of the population that presented to the ED due to falling on ice and/or snow in Calgary over an 11-year period. The weather conditions significantly predictive of falls due to ice/snow resulting in ED presentations were same-day ice and snow on ground, ice 1 day prior, 2 days prior, and 3 days prior (Table 1). Over the course of the year, winter months were the most predictive of increased presentations. These findings align with current evidence that suggests winter season, snow, and ice are significant predictors of orthopaedic trauma (Gevitz et al., 2017; Van den Brand et al., 2014). In this sample, females and males presented at similar rates until age 50. After 50 years of age, females accounted for an average of 25.48% more ED presentations due to slip and fall on ice/snow than males (Fig. 1). One potential explanation for this difference may be that men and women fall at similar rates, but post-menopausal women experiencing accelerated levels of bone loss and increased risk of osteoporosis-related fragility fractures are more likely to sustain an injury severe enough to present to the hospital (Finkelstein et al., 2008; Gärdsell et al., 1991). Differences in behaviour and in the utilization of ED services between men and women may also contribute to our observation (Petrie et al., 2022). Regardless of the root cause, the results of this study highlight that women over age 50 account for a large proportion of serious injury presentations as a result of slip and fall on ice/snow accidents; therefore, preventive interventions and risk-mitigation strategies should be targeted towards this cohort. Intrinsic and extrinsic factors such as visual impairments, medications, footwear, and gait have been associated with increased risk of falls (Ambrose et al., 2013). These are all potential interacting factors that may contribute to the falls and ultimate ED presentations that we captured in this study. To characterize our target population more accurately, prospective data collection of patients falling on ice and their associated risk factors will be of value.

However, enacting behavioural change in a target population is a challenge that existing indexes such as the AQHI and UVI face. An index alone has been shown insufficient to drive meaningful change on a population level (Bränström et al., 2003; Dixon et al., 2007; Mirabelli et al., 2020). It is crucial that strategies to increase risk awareness, such as public health indexes, are paired with techniques that support individual motivation and self-efficacy in a comprehensive intervention plan (Gies et al., 2018; McCarron et al., 2023). The SFI can accomplish this by communicating the risk of slipping and falling on ice and/or snow alongside prevention strategies for vulnerable individuals to incorporate into their daily lives, such as the use of ice-cleats on shoes (Gard et al., 2018; Holmberg et al., 2021). Work has gone into the development and testing of an outdoor fall-prevention program for older adults which successfully increased participants’ knowledge of risk and engagement in fall-prevention strategies. Some strategies implemented by participants include holding rails on stairs, taking alternative routes, and choosing different footwear (Chippendale et al., 2023). The most efficient fall-prevention method would be remaining indoors on a particular day, if possible. The implementation of such an education program in combination with the SFI would accomplish the two-part goal of increasing awareness and supporting individuals in how to take preventive action.

The preliminary SFI presented in this paper is limited by the quality and availability of weather and patient data. Generating real-time warnings is challenging because the current model relies on present-day weather variables to calculate risk. Modifying the SFI to use a combination of forecasted and current weather data may alleviate this problem. Combining existing weather variables in the SFI model with forecasted data may also improve its generalizability by facilitating validation of the SFI in smaller towns and cities without Environment Canada and Alberta RWIS sensors. The generalizability of this index to other Canadian cities and rural regions with different climates, road conditions, snow-clearing policies, and pedestrian outdoor activities is unknown. Edmonton, Alberta, being the closest population centre and experiencing similar weather patterns (Environment and Climate Change Canada, 2023) would be the ideal first location to pilot the index in addition to Calgary. Non-urban locations across Alberta should also be included for piloting. Additionally, the index provides a daily risk rather than an hourly or continuous risk. Therefore, the risk reported by the present version of the SFI may not reflect the real-time weather conditions, and the daily warning could over- or under-estimate fall risk at a given time of day.

Another limitation is the quality of patient data coding. Patients were included in the analysis if their assigned ICD-10 mechanism of injury code, as determined by ED staff, was slip and fall due to ice and/or snow (W00). This system is likely imperfect, and the collection method may exclude falls on ice/snow coded more broadly or may include other unrelated falls. Based on research of other ICD-10 codes in Alberta, we suspect that W00 coding is highly specific but not necessarily highly sensitive (Quan et al., 2008). The patient data included in the current analysis are likely an under-representation of all slip and fall incidents on ice over the study period. A chart review could be performed on W00 and other ICD-10 codes, including W01 (fall on same level from slipping, tripping, or stumbling), W19 (unspecified fall), and W10 (fall on and from stairs and steps), which may help capture mistakes in coding and assess general accuracy of the system locally. Despite this lack of sensitivity, the distribution of slip and fall accidents due to ice and snow throughout the year (Fig. 2) is reassuring, with a relative lack of incidents in the summer months.

Future directions for this research in the short term should be focused on improving the quality of the data used. This model includes only the falls on ice/snow that resulted in an ED visit with an associated orthopaedic injury. To achieve a more accurate assessment of general rates of falls in the city of Calgary, a survey could be implemented to determine when and where people are falling regardless of an accompanying presentation to the ED. Additionally, the model does not include patient risk factors. Prospective data collection of patients presenting to Calgary hospitals under ICD-10 code W00 would allow collection of additional information and refinement of our model.

The quality and utility of the SFI would benefit most from a machine learning approach. Machine learning is known to have three major strengths: management of complex large data, pattern recognition, and predicting outcomes (Fischer et al., 2021). Refining the SFI using this method has remarkable potential to establish accurate real-time fall risk using location-specific weather data in any geographic location. The current version of the SFI presented here uses traditional statistical models (multivariable dispersion-corrected Poisson regression). This approach requires clearly defined independent predictors of slip and fall on ice/snow accidents prior to statistical analysis and limits the ability of the model to explore complex interactions between weather, time of year, and any potential patient intrinsic and extrinsic factors not accounted for in this model. Machine learning can explore the interaction between many complex variables; in Calgary, this may have a notable role in capturing the effect of weather phenomena like chinooks, suggested to impact trauma volumes (Yeung et al., 2020). In addition, machine learning strategies allow for continuous refinement of model-predicted risk and will result in increased accuracy of the SFI over time.

In the longer term, future research should be oriented towards a small-scale pilot of the SFI to understand the user experience, potential barriers to behaviour change, and effectiveness of the SFI at preventing falls in Calgary. This pilot study could include various communication strategies such as social media platforms, local television, news, and weather apps, to determine which methods are preferred by the public. Phone applications that combine weather reports with personal factors on a user profile are creating a novel method for accessible, individualized communication with the public (Kingma et al., 2021). Our observation about variable ED presentation volumes between the sexes provides the opportunity for such personalization when assessing and communicating risk. Focus groups and other qualitative research strategies could be used to understand the needs and feasible behavioural changes of the target population (Chippendale et al., 2023). Although the primary role of the SFI would be to warn and advise pedestrians, a potential benefit of a comprehensive index is notifying ED healthcare staff of likely trauma surges on high-risk days.

Conclusion

Despite limitations, this is the first Canadian study to generate a risk-assessment index for pedestrians experiencing an orthopaedic trauma injury due to slipping and falling on ice and/or snow. The Slip and Fall Index was created as the launch point for a more comprehensive tool. Improvements in weather and patient data quality in addition to machine learning have an unprecedented ability to optimize this predictive model and provide location-specific public health messaging to Canadians in real time.

Contributions to knowledge

What does this study add to existing knowledge?

• The SFI is the first Canadian predictive index created for the purposes of measuring the risk of slipping and falling on ice and sustaining an orthopaedic injury.

• This is a preliminary index intended to be utilized as a starting point for the generation of a more accurate model using improved/additional variables and machine learning for measuring outdoor fall-risk due to winter conditions.

What are the key implications for public health interventions, practice, or policy?

• If the SFI is implemented with adequate communication strategies, it has the potential to create pedestrian behavioural changes that would lower the rates of orthopaedic trauma presentations, reduce the number of hospitalizations due to slip and fall on ice/snow injuries and the long-term consequences, and lower associated healthcare costs.

• Implementing the SFI could additionally help healthcare staff prepare accordingly for any potential trauma surges on a high-risk day.

Appendix 1. Weather variables explored in original analysis

Daily max air temperature (°C)b Daily minimum air temperature (°C)b Daily minimum surface temperature (°C) Daily maximum surface temperature (°C) Daily total rain (mm)b Daily total snow (cm)b Daily total precipitation (cm)b Snow on ground (cm)b Presence of wind gust Speed of max wind gust (km/h)b Presence or absence of ice (see Appendix 2)

^bDenotes a continuous variable where effect represents the effect of a 10-unit change

Appendix 2. Ice conditions

Ice is likely to occur if at least one of three conditions are met:

1. Daily total rain is greater than zero (mm) and daily minimum surface temperature is less than zero (degrees C). This is to account for freezing rain.

2. Daily snow on ground is greater than zero (cm) or daily total precipitation is greater than zero (mm) and daily maximum surface temperature is greater than three (degrees C) and daily minimum surface temperature is less than zero (degrees C). This is to account for any snow on ground or precipitation that would melt on the ground and then freeze.

3. Snow on ground 1 day prior is greater than zero (cm) or total precipitation 1 day prior is greater than zero (mm) and maximum surface temperature 1 day prior is greater than three (degrees C) and daily minimum surface temperature today is less than zero (degrees C). This is to account for overnight freeze events.

Author contributions

Tarcea was responsible for writing the original draft, review and editing, and visualization. Vergouwen had the roles of conceptualization, data curation, project administration, and writing — review and editing. Sayre was involved in the methodology and performed the formal analysis. White had the roles of supervision, conceptualization, and writing — review and editing.

Funding

This work received funding from the University of Calgary, the Canadian Orthopaedic Foundation, and the Alberta Strategic Clinical Network. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Availability of data and material

Additional details about the model are available on request from the authors. Patient data used are not available for external sharing due to ethical restrictions. Any data-specific inquiries can be directed to the corresponding author.

Code availability

Not applicable

Declarations

Ethics approval

All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee (The Conjoint Health Research Ethics Board; REB18-0777) and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards.

Consent to participate

Not applicable

Consent for publication

Not applicable

Conflict of interest

The authors have no conflict of interest to declare.