Social Deprivation Index and Pediatric Pedestrian-vs-Automobile Collisions in Ohio

Pedestrian-vs-automobile (PVA) collisions account for the highest number of traffic-related fatalities worldwide.^1–3 Automobile-related trauma is one of the leading causes of death among children, with PVA collisions accounting for 25% of these fatalities.^4,5 Among adults, the relationship between PVA collisions and socioeconomic measures has been studied at the city, county, state, and country levels.^6–12 Pediatric-specific data of geographic area-based socioeconomic impact on PVA collisions are more limited.^13,14 

The Social Deprivation Index (SDI) is a 7-variable, area-based socioeconomic measure in which higher SDI indicates worse social deprivation.¹⁵ SDI incorporates measures of poverty, kindergarten through twelfth-grade education, single-parent households, rented and overcrowded housing, vehicle ownership, and nonemployed adults, all of which are factors that may play a role in pediatric PVA collisions.¹⁵ 

This study sought to determine the incidence of pediatric PVA collisions by zip code and to evaluate whether pediatric PVA collision incidence differs by SDI and urbanicity. We hypothesized that pediatric PVA collision incidence would be higher in areas with higher SDI and greater urbanicity across Ohio. Secondary outcomes included collision characteristics and factors associated with collision-scene fatalities. By examining patterns in collision incidence related to geography and SDI, this study aimed to provide insight into directing future traffic safety measures and reducing the burden of injury from pediatric PVA collisions.

Pedestrians aged younger than 18 years who were struck by an automobile between January 2012 and January 2023 were identified from Ohio Department of Transportation (ODOT) collision reports. Pedestrians were included if they were using a slow-speed form of transportation, including ambulating, riding in a stroller, using a wheelchair, or riding on a skateboard or roller skates, at the time of the collision. They were excluded if SDI or rurality was undetermined for their zip code, if they had missing injury severity data, or if they were using a fast-speed form of transportation, including riding a bicycle, scooter, or any form of motorized vehicle, at the time of the collision.

Data elements collected from the collision reports included pedestrian age and sex, details of the collision conditions, collision zip code, and injury severity as determined at the scene. The study was exempt from institutional review board review given that it did not meet criteria for human subject research.

The primary outcome for this study was collision incidence and collision incidence per capita using American Community Survey (ACS) pediatric population estimates by the zip code in which the collision occurred. Secondary outcomes included variability in collision characteristics by SDI quintile and factors associated with collision-scene fatality.

Collision coordinates for pediatric PVA collisions from ODOT collision reports were spatially joined to 2019 zip code tabulation areas. SDI was merged at the zip code level with 2015 to 2019 ACS 5-year population estimates of residents aged 0 to 17 years. Urbanicity was determined by 2019 Rural-Urban Commuting Area (RUCA) codes such that zip codes in which at least 30% of workers commute to an urbanized area were considered urban, and the remainder were considered rural.¹⁶ 

Differences in collision incidence among SDI quintiles were tested with a Kruskal-Wallis test. Pediatric PVA collision incidence, defined as the number of collisions per 10 000 residents aged younger than 18 years in the zip code, was compared pairwise among SDI quintiles using Wilcoxon rank-sum tests with the Holm method to correct for multiple comparisons.

Regression models tested for associations between the number of collisions and zip code urbanicity, comparing rural and urban zip codes of the same mean-centered SDI. To account for the high number of zip codes with zero collisions during the study period, several count regression models were compared, including Poisson, zero-inflation Poisson, negative binomial, and zero-inflated negative binomial (ZINB). The ZINB model fit the data best according to Akaike Information Criteria and Bayesian Information Criteria.

The ZINB model is divided into the following 2 parts to account for different types of zero outcomes: the count model accounts for sampling zeroes, in which having a zero was by chance, whereas the zero-inflated model accounts for structural zeroes, in which having a zero was inevitable. Possible sources for inevitable zeroes are lack of pediatric residents or places within walking distance of homes or other scenarios that prevent automobiles and pediatric pedestrians from interacting. The zero-inflated model assesses the odds of being collision-free.

SDI quintiles at the census-tract level were used to compare collision characteristics to understand more granular details of collision circumstances. These variables are presented through summary statistics, using medians and IQR for continuous data and sums and proportions for categorical data. We compared collision-level variables based on SDI quintiles using Wilcoxon rank-sum, χ², and Fisher’s exact tests.

Univariable and multivariable logistic regression models evaluated potential associations among pediatric PVA fatality and pedestrian age; SDI; poor road, weather, and light conditions; weekday; and driver-level and site-level factors. Poor road conditions were defined by the presence of ice, sand, mud, dirt, gravel, slush, snow, or standing water. Driver-level factors were defined by collisions involving a senior citizen driver or distracted driver or a collision related to alcohol, marijuana, or recreational drug use.

Statistical and geospatial analysis was performed using R version 4.4.0 (R Foundation for Statistical Computing), STATA version 16 (StataCorp LLC), SAS Enterprise Guide version 8.1 (SAS Institute, Inc.), ArcGIS Pro version 3.2.2 (Esri), and Python version 3.11 (Python Software Foundation).

Using ODOT collision reports between January 2012 and January 2023, 5499 pediatric PVA collisions were identified. Across 1414 zip codes in Ohio, 1188 (84.0%) had data available on zip code-level SDI, estimated population younger than age 18 years, and RUCA codes. Among these, 603 (50.1%) zip codes had zero pediatric PVA collisions during the study timeframe. Zero collisions were found in 71% (384/544) of rural zip codes but only 34% (219/644) of urban zip codes. For collision incidence, data from 5521 collisions with SDI information available among 1188 zip codes were used. For collision-level characteristics, data from 5499 collisions with complete data among 1188 zip codes were used.

Pedestrian age for those in PVA collisions varied based on SDI quintile, with the youngest children in higher SDI quintiles (age 11 years [IQR, 6–14] in quintiles 4–5 vs age 13 years [IQR, 8–15] in quintiles 3 vs age 13 years [IQR, 10–15] in quintiles 1–2; P < .001), meaning that areas with the highest degree of deprivation more often had younger children involved in PVA collisions (Table 1). In all quintiles, pedestrians were more often male (59.0%) rather than female, and collisions typically occurred in the afternoon (16:00). There were a greater proportion of serious injuries and fatalities in lower SDI quintiles than higher SDI quintiles (23.7% in quintiles 1–2 vs 19.8% in quintiles 3 vs 17.6% in quintiles 4–5; P < .001).

Assuming that collisions are equally likely on each day of the week, we would expect that 5 out of 7 (71.4%) would occur on weekdays; however, more collisions (79.5%) occurred on weekdays (P < .001). This effect was amplified in lower SDI quintiles (82.0% in quintiles 1–2 vs 79.3% in quintile 3 vs 78.0% in quintiles 4–5; P = .004). Collisions were more common in lower-SDI school zones than higher-SDI school zones (5.0% in quintiles 1–2 vs 4.0% in quintile 3 vs 2.3% in quintiles 4–5; P < .001).

In all quintiles, collisions usually occurred during daylight (76.8%) and on days with clear weather conditions (66.6%). Very few collisions occurred on an interstate (0.4%) or in a road work zone (0.4%). There was no difference in these environmental factors between SDI quintiles (both P > .05). Distracted drivers were more common in low SDI quintiles (3.0% in quintiles 1–2 vs 1.6% in quintile 3 vs 1.8% in quintiles 4–5; P = .008), as were collisions involving senior citizen drivers (12.0% in quintiles 1–2 vs 9.6% in quintile 3 vs 7.8% in quintiles 4–5; P < .001).

Adjusting for all significant covariates, factors independently associated with a collision-scene fatality were younger pedestrian age, clear road conditions, light conditions, and driver-level factors. SDI quintile was not predictive of collision-scene fatality, nor were poor weather conditions, weekday compared with weekend collision, or site-level factors (Table 2). For each 1-year increase in pedestrian age, the odds of fatality at the scene decreased by 10% (adjusted odds ratio [aOR], 0.90, 95% CI: 0.86–0.93). Collisions occurring in poor road conditions had 0.40 (95% CI: 0.19–0.83) times the odds of pedestrian fatality at the scene compared with collisions occurring in clear road conditions. Collisions occurring during the night had 3.05-times (95% CI: 1.97–4.70) higher odds of pedestrian fatality at the scene compared with those occurring during the day, and there was no statistically significant difference in the odds of pedestrian fatality at the scene for collisions occurring at dawn or dusk compared with daytime (P = .88). Collisions with driver-level factors had 2.36-times (95% CI: 1.59–3.51) higher odds of pedestrian fatality at the scene compared with collisions involving no driver-level factors.

PVA incidence was higher in zip codes with high or very high (quintiles 4–5) SDI compared with moderate (quintile 3) SDI (P < .001). PVA incidence was also higher in zip codes with moderate SDI compared with low or very low (quintiles 1–2) SDI (P = .02). Figure 1 presents pediatric PVA collision incidence across Ohio, demonstrating the clustering of PVA collisions in urban areas with high SDI.

In the count model, among zip codes with a nonzero number of collisions, the average number of collisions in urban zip codes with average SDI during the study period was 5.52 (95% CI: 4.77–6.39) compared with 1.71 (incident risk ratio [IRR], 0.31, 95% CI: 0.24–0.39) in rural zip codes with average SDI (Table 3). For each unit increase in SDI, the overall number of collisions increased by 3% (IRR, 1.03, 95% CI: 1.03–1.03). In the zero-inflated model, the baseline odds of an urban zip code being collision-free was 0.16 (95% CI: 0.08–0.32) compared with a rural zip code. Rural zip codes had 6.24-times (IRR, 6.24, 95% CI: 3.30–11.79) higher odds of being collision-free than urban zip codes.

This study investigated the relationship among pediatric PVA collisions and social deprivation, as measured by the SDI, as well as urbanicity across Ohio. This represents the largest pediatric-specific study of PVA collisions as they relate to a geographic measure of socioeconomic status, which can be a proxy for safety infrastructure. Pediatric PVA collisions were more common in urban areas and impoverished areas with higher social deprivation. Collisions often occurred on weekdays, and higher SDI areas more commonly had collisions in school zones. Collisions in lower SDI areas more commonly had associated serious injury or death. The strongest factors associated with collision-scene fatality were pedestrian age, road conditions, light conditions, and driver-related factors.

Our study supports the findings of a previous study of traffic safety data and Area Deprivation Index (ADI) in San Diego County that found a relationship between ADI and the likelihood of involvement in PVA collisions among children.¹³ We demonstrate a relationship between SDI and pediatric PVA collisions, with a higher crash incidence rate observed in zip codes with greater degrees of deprivation. Because vehicle ownership is lower in these same areas, the increase in collisions is probably not solely attributable to having more vehicles on the road but may also be explained by driving conditions or driver behaviors that are more dangerous to pedestrians or by a greater amount of foot traffic. This is consistent with prior literature demonstrating greater numbers of pedestrian casualties in lower-income groups with lower vehicle ownership.⁸. Driver-level factors, including substance use, senior age, or distracted driving, were independently associated with collision-scene fatality, which may indicate that distracted driving awareness campaigns, increasing substance-use patrols, and driver-level interventions such as stricter regulations for license renewal or postlicense driver education may benefit pediatric pedestrians.^17–29 

Distracted driving may present a unique opportunity for intervention. Because the frequency of distracted driving was lower in higher SDI areas, it may not be the primary cause of serious injuries and collision-scene fatalities in these zip codes, highlighting the need for improvement on built safety infrastructure such as streetlights and cross-walks. Previous studies have found that driver inattention or distraction are common in collisions with serious injuries or fatalities and that strictly enforced distracted driving laws improve injury and fatality rates.^20,21 Our findings of driver-level factors such as distracted driving predicting collision-scene fatality combined with higher frequencies of distracted drivers, serious injuries, and collision-scene fatalities in lower SDI census tracts suggest that there may be an inverse relationship between degree of social deprivation and the likelihood of engaging in potentially lethal distracted driving behavior.

PVA collisions are influenced by socioeconomic factors and road conditions. Previously identified risk factors include family income, parental education level, pedestrian age, and population density, with low income being the strongest predictor of pedestrian injuries.^14,30 Even nonfatal collisions can have devastating consequences on children and their families and communities. Reducing deprivation or enhancing protections in areas of higher deprivation may help mitigate risk difference. The overall fatality rate among our population was low, but the factors with the strongest association with collision-scene fatality were younger pedestrian age, clear road conditions, nighttime collisions, and driver-level factors. In our population, collisions most commonly occurred on roads with clear conditions, which makes this relationship with fatalities unsurprising, although previous studies have more commonly described the impact of adverse weather conditions on fatalities.^31–33 Prior studies have demonstrated more significant injury patterns in younger children, even in lower-speed PVA collisions, which may explain the higher odds of collision-scene fatality among younger pedestrians.^34–36 Improved visibility through visibility aids and street lighting is known to be associated with lower rates of collisions, injuries, and fatalities and may reduce the number or severity of nighttime collisions.^37–39 

One advantage that this study offers is the wider scope of analysis through assessment of state-level collision patterns. Through examining all collisions in the seventh most populous state in the United States, we maximize the generalizability and external validity of our findings while addressing the nuances of collision characteristics that are specific to children.⁴⁰ Overall, fewer rural compared with urban zip codes had one or more collisions during our decade-long study window, and higher SDI urban zip codes had the highest likelihood of pediatric PVA collision occurrence. It is challenging to determine whether this increase in collision incidence is due to a greater population density in urban areas or differences in traffic infrastructure, particularly in areas with greater degrees of deprivation, which may have less funding for traffic safety measures.⁴¹ Our study aimed to address the impact of population density by focusing on the overall pediatric population within a zip code as our reference group. Prior research has demonstrated a higher likelihood of adult and pediatric pedestrian fatalities among lower-income areas, although the cause of this is likely multifactorial.^8,42–44 Rural areas are much less likely to have walkable areas with high foot traffic, whereas urban areas are designed for high pedestrian usage.^45,46 High usage combined with fewer safety interventions may explain the lower odds of urban areas being collision-free compared with rural areas.⁴⁷ This highlights the importance of traffic calming measures, particularly in urban areas with the greatest degree of social deprivation.

This study highlights important opportunities for policy change and resource allocation. Consistent with prior research, our study found that the frequency of collisions in school zones was greater in areas with a higher level of deprivation.^48,49 School-based interventions to enhance safety during school commuting hours may help reduce the number of pediatric PVA collisions, particularly because a disproportionate number of collisions occurred on weekdays, and most collisions occurred in the hours right after school ends.^48,50–52 Furthermore, traffic calming measures, especially for locations with high collision volume or dense traffic, may help reduce pediatric PVA collisions.^11,53,54 Finally, expansion of pedestrian safety education programs for children can equip them with critical skills for improving their safety behaviors.^55–60 Future research should include qualitative studies in collision hotspots, such as focus groups and interviews with community members to discuss local context and potential commonalities among hotspots to help target traffic safety interventions.

There are a few notable limitations to this study. First, our data come from police collision reports, which inconsistently record some valuable covariates that were excluded from analysis, such as vehicle size or events/activities that may have altered normal traffic patterns. Additional granularity regarding environmental conditions that facilitate pediatric PVA collisions could help target interventions or regulations to prevent these collisions. Second, this study was not able to address whether the likelihood of pediatric PVA collision involvement varied with pedestrian race and ethnicity. Future research using the injury equity framework could include an interrogation of the relationship between collision hotspots and racially driven policies such as redlining that may lead to discrepant funding for safety infrastructure. Due to limitations in the level of detail of available police reports, specific information regarding safety infrastructure and neighborhood characteristics, particularly as related to accessibility, investment and divestment, and community social norms, could not be ascertained for this study.⁶¹ Finally, because this study examines all PVA collisions in the state through collision reports instead of the electronic medical record, the impact of injuries sustained in these collisions is unknown. Injury severity could therefore be underestimated in patients who decompensate after transportation from the scene or patients who face long-term morbidity due to injuries sustained in this collision. Fatalities identified in this study are collision-scene fatalities, and those that occurred in transit to or at medical facilities are not captured in this data set.

This study identified a relationship between pediatric PVA collisions and SDI. Children from the most disadvantaged and impoverished areas were more likely to be hit by an automobile, and children from the poorest areas suffered the most serious injuries. Rurality appears to be protective against pediatric PVA collisions. Pediatric PVA collisions can be devastating to children, families, and communities, and although no panacea exists to address this complicated, multifactorial issue, a variety of interventions may help improve the safety of pediatric pedestrians. These data will be used to inform traffic calming measures and educational campaigns.

aOR

adjusted odds ratio;

ACS

American Community Survey;

ADI

Area Deprivation Index; IRR, incident risk ratio;

ODOT

Ohio Department of Transportation;

PVA

pedestrian-vs-automobile;

RUCA

Rural-Urban Commuting Area;

SDI

Social Deprivation Index;

ZINB

zero-inflated negative binomial.