A decomposition of the effects of the COVID-19 pandemic on changes in the motor vehicle collision related mortality in Alabama

Abstract

Background/Objective

Motor vehicle collisions are the leading cause of unintentional injury death in Alabama and at various points during the COVID-19 pandemic there were documented increases in the following risk driving behaviors: speeding, driving under the influence, and seat belt citations. Thus, the objective was to characterize the overall motor vehicle collision (MVC)-related mortality rate in Alabama and the contribution of each component over the first two years of the pandemic compared to before the pandemic by three different road classes: urban arterials, rural arterials, and all other road classes.

Methods

MVC data were derived from the Alabama eCrash database, an electronic crash reporting system used by police officers across the state. Data on vehicle miles traveled each year were collected from the U.S. Department of Transportation’s Federal Highway Administration estimates of traffic volume trends. MVC-related mortality in Alabama was the primary outcome and year of MVC was the exposure. The novel decomposition method broke down population mortality rate into four parts: deaths per MVC injury, injury per MVC, MVC per vehicle miles traveled (VMT), and VMT per population. Poisson models with scaled deviance were used to estimate rate ratios of each component. Relative contribution (RC) of each component was calculated by taking the absolute value of the component’s beta coefficient and dividing by the sum of the absolute values of all components' beta coefficients. Models were stratified by road class.

Results

Across all road classes combined, there were no significant changes to the overall MVC-related mortality rate (per population) and its components when comparing 2020–2022 to 2017–2019; this was due to the increased case fatality rate (CFR) being offset by decreases in the VMT rate and MVC injury rate. In 2020, among rural arterials a non-significant increased mortality rate was offset by a decreased VMT rate (RR 0.91, 95% CI 0.84–0.98, RC 19.2%) and MVC injury rate (RR: 0.89, 95% CI: 0.82–0.97, RC: 22.2%) when compared to 2017–2019. For non-arterials, a non-significant decreased MVC mortality rate was observed in 2020 when compared to 2017–2019 (RR 0.86, 95% CI 0.71–1.03). When considering 2021–2022 versus 2020, the only significant component for any road class was a decreased MVC injury rate for non-arterials (RR: 0.90,95% CI: 0.89–0.93) but this was offset by an increased MVC rate and CFR, resulting in no significant change to the mortality rate (per population).

Conclusions

In a state with one of the highest MVC-related mortality rates in the country, despite decreases in VMTs per population and injuries per MVC, the MVC mortality rate per population did not change during the pandemic due in part to the contributions of an increase in the case fatality rate. Future research should determine whether the increase in CFR was associated with risky driving behaviors during the pandemic.

1. Introduction

In 2020, Alabama had the fourth highest motor vehicle collision (MVC) related mortality rate in the country (National Safety Council). MVCs are the leading cause of unintentional injury death in Alabama and account for approximately half of all unintentional injury related deaths in the state (Alabama Department of Public Health). In March 2020, the COVID-19 pandemic was declared a national public health emergency and many states, including Alabama, issued stay-at-home (SAH) orders to mitigate spread of the virus. Previous studies have documented the substantial drop due to COVID-19 related lockdowns in population mobility or vehicle miles travelled (VMTs) ranging from 43% in a state-level analysis done in Connecticut to 63% province-level analysis in Spain (Doucette et al., 2021, Saladié et al., 2020). In Pennsylvania, there was a 10% decrease in MVC incidence under SAH orders in 2020 (Kaufman et al., 2021); however, the significant drop in VMTs and MVC incidence during 2020 did not correspond to drops in MVC-related injury or fatalities. In the Pennsylvania-based study, the zip codes where MVC incidence decreased, there was an average 16% increase in MVC-related injury severity. Another study in United Arab Emirates noted a significantly higher mortality rate of hospitalized MVC patients during the pandemic versus before the pandemic (Yasin et al., 2021).

Similarly, according to the National Highway Traffic Safety Administration (NHTSA), an estimated 38,824 traffic fatalities occurred in the United States in 2020 which was a 7.2% increase from 2019. In 2021, MVC-related fatalities increased by 10.5% from 2020, marking the largest annual percentage increase ever recorded in the history of the Fatality Analysis Reporting System (FARS) (National Highway Traffic Safety Administration, 2022). The changes in travel habits that occurred as a result of the COVID-19 pandemic has been well documented in prior studies but much less is known about how the change in travel habits is associated with changes in injury incidence and mortality rates especially when comparing rates in 2021 and 2022 to rates in 2020.

Though previous studies have noted changes in crash severity, crash type and VMTs during the pandemic, no study to date has examined how these changes contributed to the overall mortality rate over the first three years of the pandemic and whether the contributions of each component changed over the course of the pandemic. Additionally, no study has examined whether mortality rate and its components varied between urban and rural roads. In light of this, the current study sought to characterize overall MVC-related mortality rate and the contribution of each component over the first three years of the pandemic compared to before the pandemic using a novel decomposition method. Secondary analyses examined the statewide MVC-related mortality rate and the relative contribution of each component by three different road classes: urban arterials, rural arterials, non-arterials. The non-arterial class category includes local and collector road types in both urban and rural areas.

2. Methods

2.1. Data source

The six years of data for this ecological study was derived from two publicly available sources: Alabama eCrash database and the U.S Department of Transportation’s Federal Highway Administration estimates of traffic volume trends. The former was developed by the Center for Advanced Public Safety at the University of Alabama in an effort to remove the need for paper crash forms; to move the entry of crash data as close to the crash scene as possible; and to ensure greater compliance with Model Minimum Uniform Crash Criteria, a minimum set of motor vehicle crash data elements and attributes that the National Highway Traffic Safety Administration (NHTSA) suggests states should collect and include in their state’s crash data system.

The eCrash system is used by police officers responding to MVCs across the state and includes features such as individual and customizable reports to reflect the different numbers of people or vehicles involved in a crash, required report sections and validation checks—which prevent reports from being submitted with incomplete or invalid data—and electronic submission, which has reduced duplicate reports and provides nearly instantaneous access to crash report data. The second source of data was NHTSA’s estimates of traffic volume trends. The traffic volume trends are monthly reports based on traffic data form the Highway Performance Monitoring System from about 5,000 continuous traffic counting locations across the nation. For the purposes of this analysis, traffic volume data from the monthly reports from January through December were summed for any given year to obtain annual traffic volume and trend data.

2.2. Variable definitions

Changes in MVC-related injuries and fatalities were the primary outcome of interest in this analysis and were defined as an injury or fatality obtained from a motor vehicle collision. In the Alabama eCrash database, any person that dies within 30 days as a result of the injuries they obtained in the collision are counted as a traffic fatality. Injury coding is based upon the KABCO scale in police crash reports, and non-fatal injuries could be classified as incapacitating, non-incapacitating, or possible injury (U.S. Department of Transportation Federal Highway Administration). The exposure of interest was year of MVC grouped into three categories for purposes of the current analysis: 2017–2019 (i.e., pre-pandemic), 2020 (early pandemic with stay-at-home orders), and 2021–2022 (latter pandemic).

Road class was defined as arterial (i.e., interstate, principal arterial, or minor arterial) or non-arterial (i.e., major collectors, minor collectors, local roads, or unclassified). Arterial roads offer the highest level of mobility and posted speeds and includes interstates, freeways, multilane highways as well as other roadways that support the interstate system. Collectors are roads that connect local roads with major arterials and have lower posted speeds than arterial roads. Local roads include streets in residential areas, businesses, farms and often have posted speeds between 20 and 45 miles per hour. Arterial classes were further categorized as urban or rural based on the classification in the eCrash system.

2.3. Decomposition method

Li and Baker proposed a novel decomposition method breaking down a cause-specific mortality rate into three parts: exposure rate, injury or disease rate, and case fatality rate (CFR). They utilized the method in Equation (1) to investigate male–female differences in the death rates from bicycling injuries and later to examine gender and age differences for fatal motor vehicle crash involvement (Li and Baker, 1996, Li et al., 1998).

$$\mathit{MVCMortalityRate}=\frac{\mathit{Deaths}}{\mathit{Injury}}\times \frac{\mathit{Injury}}{\mathit{Exposure}}\times \frac{\mathit{Exposure}}{\mathit{Population}}$$ (1)

The current study expanded Equation (1) to include the following components in Equation (2) based on the analysis performed by Griffin et al. (Griffin et al., 2018) on vehicle safety technologies’ effect on the mortality rate decomposition: travel density [vehicle miles traveled (VMT) per population], collision density (MVC per VMT), injury incidence (injuries per MVC), and case fatality rate (deaths per MVC injury).

$$\mathit{MVCMortalityRate}=\frac{\mathit{Deaths}}{\mathit{MVCInjuries}}\times \frac{\mathit{MVCInjuries}}{\mathit{MVC}}\times \frac{\mathit{MVC}}{\mathit{VMT}}\times \frac{\mathit{VMT}}{\mathit{Population}}$$ (2)

2.4. Statistical analysis

The relative contribution (RC) of each component factor identified in Equation (2) was estimated using Equation (3). Rate ratios (RRs) for each year were estimated from a Poisson regression model for each mortality rate component. Scaled deviance was used in the Poisson models to account for over dispersion. The model offset in each component-specific Poisson model was the natural logarithm of the denominator value. For example, the offset in both the overall mortality rate models and VMTs per population models was natural log of the population. Offsets in other component-specific models included the natural log of VMTs, MVCs, and injuries due to MVCs.

To calculate the relative contribution (RC) of each component rate ratio, per equation (3), the beta coefficients from the Poisson model (i.e., the natural log of the rate ratio) were summed; each individual component’s beta coefficient was then divided by this sum.

$$\mathit{RC}=\frac{\mathit{abs}(\ln \left[RateRatio\right])}{{\sum }_1^{x}abs(\ln [{\mathit{RateRatio}}_x])}$$ (3)

Models were created for the overall comparison of the pandemic vs non-pandemic years (i.e., 2020–2022 vs. 2017–2019) and comparing the three categories to the prior category (i.e., 2020 vs. 2017–2019 and 2021–2022 vs. 2020). In a secondary analysis, models were stratified by road class and urbanality.

3. Results

For non-arterials, there was a general downward trend in MVC-related mortality rate per 100,000 persons from 2017 to 2020, but then it began trending upward from 2020 to 2022 (Fig. 1 , part A). Rural and urban arterials saw increases in the mortality rate per 100,000 persons from 2017 to 2020. VMTs per person, MVCs per 100,000 VMTs, and injuries per 100,000 MVCs decreased or remained stable between 2019 and 2021, but deaths per 100 MVC injuries rose sharply during this same time frame (Fig. 1, part B). The MVC mortality rate per 100,000 people decreased from 2017 to 2020 for non-arterials. Injuries per MVC trended down across the whole study period for all road classes; however, this did not correspond to a similar trend in deaths per injury which increased for all road classes during the same time frame (Fig. 1, part B).

Fig. 1.

Trend in rates of component factors of motor vehicle collision (MVC)-related mortality rate in Alabama by road class category for A) MVC mortality rate and B) components of MVC mortality rate.

When comparing MVC-related injury and mortality trends during the COVID-19 pandemic (2020–2022) to before COVID (2017–2019), the overall MVC-related mortality rate did not significantly change; however, deaths per MVC injury (case fatality rate[CFR]) was significantly higher in 2020–2022 than 2017 to 2019 and was the largest contributor to the overall mortality rate (RR: 1.26, 95% CI: 1.19–1.33, RC = 48.3%)). The increased case fatality rate was offset by decreased rate ratios for other components (Table 1 ). This pattern also occurred when broken down by road class for urban arterials, rural arterials, and non-arterials. Rural arterials had the largest increase in CFR at 31%, accounting for over half of the MVC-related mortality rate (RR: 1.31, 95% CI: 1.19–1.43, RC = 56.9%). For rural arterials, the significant 12% drop in injury incidence was the largest decreased contributor to mortality rate (RR 0.88, 95% CI 0.84–0.92, RC = 27.5%), but a 12% decreased collision density (RR 0.88, 95% CI 0.77–1.01, RC = 25.1%) was the largest contributing decreased factor for non-arterials.

Table 1.

Risk ratios (RRs), associated 95% confidence intervals (CIs) and relative contributions (RCs) of component factors of motor vehicle collision (MVC)-related mortality rate by road class.

	2020–2022 vs. 2017–2019
	RR (95% CI)	P value	RC (%)
OVERALL
Mortality rate	0.98 (0.83–1.16)	0.8075	–
VMT per pop	0.93 (0.77–1.12)	0.4354	15.4
MVC per VMT	0.91 (0.69–1.20)	0.4968	19.8
Injuries per MVC	0.92 (0.81–1.06)	0.2485	16.5
Deaths per injury	1.26 (1.19–1.33)	<0.0001	48.3
URBAN ARTERIALS
Mortality rate	0.97 (0.85–1.10)	0.6019	–
VMT per pop	0.92 (0.89–0.96)	<0.0001	18.7
MVC per VMT	0.92 (0.82–1.02)	0.1268	19.9
Injuries per MVC	0.94 (0.90–0.98)	0.0054	15.0
Deaths per injury	1.22 (1.10–1.35)	0.0001	46.4
RURAL ARTERIALS
Mortality rate	1.06 (0.94–1.20)	0.3642	–
VMT per pop	0.95 (0.90–0.99)	0.0311	11.9
MVC per VMT	0.98 (0.91–1.06)	0.6553	3.7
Injuries per MVC	0.88 (0.84–0.92)	<0.0001	27.5
Deaths per injury	1.31 (1.19–1.43)	<0.0001	56.9
NON-ARTERIAL
Mortality rate	0.92 (0.81–1.04)	0.1986	–
VMT per pop	0.93 (0.89–0.96)	<0.0001	15.6
MVC per VMT	0.88 (0.77–1.01)	0.0687	25.1
Injuries per MVC	0.92 (0.86–0.98)	0.0085	17.2
Deaths per injury	1.23 (1.13–1.34)	<0.0001	42.1

*Estimated from Poisson regression with scaled deviance.

VMT: Vehicle miles travelled.

When considering only 2020 compared to before the pandemic, among all road classes combined, there were no significant changes in the MVC-related mortality rate or any of its components (Table 2 ). Increases in the case fatality rate were offset by the increased relative contribution of decreased collision density (RC = 30.6%) and travel rates (RC = 19.7%). During the post stay-at-home order period (2021–2022) compared to 2020, there was a non-significant 7% increase in the overall mortality rate (RR: 1.07, 95% CI: 0.82–1.39) that can be partially attributed to increased CFR (20.6%) and collision density (RC = 35.5%).

Table 2.

Risk ratios (RRs), associated 95% confidence intervals (CIs) and relative contributions (RCs) of component factors of motor vehicle collision (MVC)-related mortality rate by year category and road class.

	2020 vs 2017–2019			2021–2022 vs 2020
	RR (95% CI)	P value	RC (%)	RR (95% CI)	P value	RC (%)
OVERALL
Mortality rate	0.93 (0.73–1.20)	0.5957	–	1.07 (0.82–1.39)	0.6088	–
VMT per pop	0.91 (0.68–1.22)	0.5357	19.7	1.02 (0.75–1.40)	0.8922	10.6
MVC per VMT	0.87 (0.57–1.32)	0.4991	30.6	1.08 (0.69–1.68)	0.7512	35.5
Injuries per MVC	0.97 (0.79–1.18)	0.7532	6.8	0.93 (0.75–1.16)	0.5424	33.2
Deaths per injury	1.22 (0.62–2.43)	0.5631	42.9	1.04 (0.51–2.14)	0.9090	20.6
URBAN ARTERIALS
Mortality rate	0.92 (0.75–1.14)	0.4657	–	1.07 (0.85–1.33)	0.5750	–
VMT per pop	0.90 (0.84–0.95)	0.0003	26.2	1.04 (0.97–1.11)	0.2467	22.5
MVC per VMT	0.90 (0.74–1.08)	0.2479	26.1	1.03 (0.85–1.26)	0.7385	20.2
Injuries per MVC	0.97 (0.92–1.03)	0.2994	7.0	0.95 (0.89–1.01)	0.0898	31.1
Deaths per injury	1.19 (1.01–1.39)	0.0352	40.7	1.04 (0.88–1.24)	0.6082	26.2
RURAL ARTERIALS
Mortality rate	1.04 (0.85–1.27)	0.7310	–	1.03 (0.84–1.28)	0.7577	–
VMT per pop	0.91 (0.84–0.98)	0.0160	19.2	1.05 (0.97–1.14)	0.2361	54.5
MVC per VMT	0.97 (0.86–1.11)	0.6905	5.1	1.01 (0.89–1.16)	0.8563	13.3
Injuries per MVC	0.89 (0.82–0.97)	0.0079	22.2	0.98 (0.89–1.07)	0.5864	26.7
Deaths per injury	1.31 (1.21–1.42)	<0.0001	53.5	0.99 (0.91–1.09)	0.9083	5.5
NON-ARTERIAL
Mortality rate	0.86 (0.71–1.03)	0.0915	–	1.11 (0.92–1.35)	0.2729	–
VMT per pop	0.93 (0.87–0.99)	0.0316	15.7	0.99 (0.92–1.06)	0.7287	3.7
MVC per VMT	0.80 (0.67–0.95)	0.0132	47.9	1.15 (0.96–1.39)	0.1355	42.7
Injuries per MVC	0.99 (0.96–1.01)	0.2575	3.2	0.90 (0.88–0.93)	<0.0001	30.4
Deaths per injury	1.17 (1.02–1.34)	0.0238	33.3	1.08 (0.94–1.25)	0.2835	23.2

*Estimated from Poisson regression with scaled deviance.

VMT: Vehicle miles travelled.

By road class, for urban arterials in 2020 a significantly increased CFR (RR 1.19, 95% CI 1.01–1.39, RC = 40.7%) was offset by decreases in travel density (RR 0.90, 95% CI 0.84–0.95, RC = 26.2%) and collision rates (RR 0.90, 95% CI 0.74–1.08, 26.1%). For rural arterials, the increased CFR (RR 1.31, 95% CI 1.21–1.42, RC 53.5%) was mostly offset by decreased travel density (RR 0.91, 95% CI 0.84–0.98, 19.2%) and injury rates (RR 0.89, 95% CI 0.82–0.97, RC = 22.2%). For non-arterial roads, the mortality rate decreased due to decreases in travel density (RR 0.93, 95% CI 0.87–0.99, RC = 15.7%) and collision rate (RR 0.80, 95% CI 0.67–0.95, RC = 47.9%) despite a 17% increased CFR (RR 1.17, 95% CI 1.02–1.34, RC = 33.3%). Comparing 2021–2022 vs 2020, the only significant change was a 10% decreased injury rate among non-arterial collisions (RR 0.90, 95% CI 0.88–0.93).

4. Discussion

The current ecological study found that, while the MVC mortality rate did not change significantly during COVID, the individual components of the mortality rate did change. Travel density decreased significantly in 2020 for all road classes while the injury density per MVC decreased significantly (relative to the pre-pandemic years) for all road classes and particularly non-arterial road class. Additionally, all road classes had an increased CFR during the early pandemic period.

The significant drop in vehicle miles traveled in 2020 for urban arterials, rural arterials, and other road classes is consistent with findings from previous studies that examined how vehicular travel changed in response to the pandemic (Sutherland et al., 2020, National Highway Traffic Safety Administration, 2020). In this study, the drop in VMTs per population in 2020 ranged from 7 to 10% depending on road class. Estimates from a prior state-level study that examined mean daily VMTs during active SAH orders found that mean daily VMTs decreased by 43% which is substantially larger than the decrease reported in this study (Doucette et al., 2021). This difference in findings could be attributed to being conducted in a smaller state with more urban areas than Alabama. It could be that their analysis only covered the two-month period where an active SAH order was in place whereas the current study’s unit of analysis was the entire year over which travel density likely varied as SAH orders were lifted or levels of viral transmission varied. The 7–10% drop in VMTs per population found in this study are more in line with NHTSA estimates from 2020 which show a 13.2 % decrease in VMTs in the United States (National Highway Traffic Safety Administration, 2020). When comparing 2021 to 2020, NHTSA reported an 11% increase in VMTs (National Highway Traffic Safety Administration, 2020). This is slightly more than the 4–5% increases noted in our study for rural and urban arterials when comparing 2021–2022 to 2020 (Table 2).

While travel density significantly dropped in 2020 compared to 2017–2019, this did not correlate to a significant drop in MVC-related mortality. This is in line with previous studies that reported declines in VMTs, MVCs and MVC-related injuries but an increase in injury severity or fatalities (Doucette et al., 2021, Kaufman et al., 2021, Yasin et al., 2021, Waseem et al., 2021, Meyer, 2020). In Connecticut, VMTs dropped sharply in March and April of 2020, but single vehicle crash rate increase by 2.29 times while single vehicle fatal crash rate increase by 4.1 times (Doucette et al., 2021). In Pennsylvania, one study that utilized a state-wide trauma registry noted decreases in MVCs under SAH orders but there was a 16% increase in MVC injury severity (Kaufman et al., 2021).

The current study reported increased CFR for all road classes when considering only 2020 as well as when comparing the first three years of the pandemic to the three years before the pandemic. One possible explanation could be due to changes in the prevalence of risky driving behaviors during the pandemic. Data from the National Security Council and NHTSA indicate that while fewer people were on the roads, those that were on the roads exhibited higher rates of risky driving behavior such as speeding, lack of seat belt use, driving under the influence (DUI) of alcohol or other substances (National Highway Traffic Safety Administration, 2022, Alabama Highway Safety Plan Annual Report Fiscal Year, 2021). In Alabama, the number of speeding and seat belt citations was higher in 2020 compared to 2019 (Alabama Highway Safety Plan Annual Report Fiscal Year, 2021). Additionally, the number of DUI arrests increased 24.4% in 2021 compared to 2020 (Alabama Highway Safety Plan Annual Report Fiscal Year, 2021). Similar increases in risky driving behaviors were noted in other states. The proportion of speeding violations as percentage of total violations increased from 17.3% in March 2020 to 61.8% in April 2020 in New York City, and similar increases in speeding violations were also observed in Massachusetts (Meyer, 2020). Moreover, ejection rate per MVC, which is a proxy for seat belt use, increased significantly in April 2020 (National Highway Traffic Safety Administration, 2020). Finally, the percentage of drivers testing positive for at least one substance significantly increased in the second quarter of 2020 (National Highway Traffic Safety Administration, 2020).

The current study has strengths and limitations that should be noted. The main strength of this study is that it is the use of a novel decomposition method to examine how MVC-related mortality rate and its components changed during the pandemic compared to before the pandemic. It is the only study to examine rates during the whole pandemic from 2020 through 2022 compared to 2017–2019 and to examine how the mortality rate and its components changed each year of the pandemic by comparing 2020 to 2017–2019 and 2021–2022 to 2020. Analyzing 2021and 2022 together relative to 2020 may have resulted in a bias towards the null because travel patterns in 2021 may be closer to 2020 while travel patterns in 2022 were closer to pre-pandemic levels. One potential weakness of this study is that the state level data is more influenced by the more populous, urban areas and might not be reflective of rural areas; however, we accounted for this by examining trends in mortality rate and its components by road class. Additionally, the results of this Alabama-based study may not generalize to other states with different length SAH orders or states with more urban areas with more developed public transit systems. Finally, it is possible that associations could differ at a more granular (e.g., county, person) level, which would allow for the inclusion of factors that are best measured at the county- (e.g., social vulnerability index) or person-level (e.g., risky driving behaviors). Thus, we were not able to identify the cause of the increased case fatality and injury rates.

5. Conclusion

This study noted that despite a significant decrease in exposure (VMTs per population) and injuries per MVC for some road classes, the mortality rate did not change over the study period likely due to the increased case fatality rate. Future research should examine crashes that occur on non-arterial roads as the trends were more pronounced for this road class and person-level risk factors (e.g., risky driving behaviors) that could have resulted in the observed increases in injury rate and case fatality rate.

CRediT authorship contribution statement

LR: Conceptualization, Formal analysis, Methodology, Writing-original draft, Writing-reviewing & editing. RG: Conceptualization, Data curation, Formal analysis, Methodology, Writing-review & editing.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. This study was not funded.

Data availability

Data will be made available on request.