A systematic review and meta-analysis of the impact of curbs on crash outcomes

Abstract

Introduction:

Road traffic crashes involving vertical curbs are commonly reported to occur on highways and expressways in India. We found a gap in terms of systematically assessing the evidence of the impact of curbs on road safety outcomes in the real world.

Method:

We conducted a systematic review and meta-analysis of the impact of curbs on the risk of road traffic injuries. We used keywords in a database of records prepared by an earlier evidence gap map (EGM). The EGM used a comprehensive search strategy including 6 academic database, 17 organisational websites, hand searching, contacting experts and back referencing.

Results:

We found 4 studies that evaluated impact of a curbed median or a curbed shoulder. We found that the presence of a curb on a median increases the risk for all crashes, all single-vehicle crashes, all median-related crashes and median-related injury crashes. The data also indicate that the severity of accident reduces for curbs on median while it increases for curbs on shoulder, though the latter effect is not statistically significant. All the epidemiological studies were conducted on rural highways and did not report effects for different traffic speeds or vehicle types. However, our review of crash tests and simulation studies indicates that the impact of a curb design may be highly sensitive to speed and vehicle types.

Conclusions:

The safety impacts of a curb depend on the context of the road. In an urban road, a curb should ensure safety of pedestrians from an errant vehicle. On high-speed rural roads, curbs should be avoided and treatments should facilitate safe departure of the vehicle from the roadway.

INTRODUCTION

Road traffic injuries contribute 1.2 million deaths worldwide annually. Approximately 90% of the global deaths and injuries occur in low- and middle-income countries (LMICs) (WHO, 2018). In many LMICs, new highways are being constructed and existing roadways are being widened at a fast pace (Chakrabarti, 2018). As a result, road death rates are rising or stable at high levels in most countries (Vos et al. 2020). In this context, it is imperative to understand the safety impact of highway design elements.

In this study, we are interested in assessing how curbs on the median and shoulders affect the risk of traffic crashes and injuries. Curbed medians and shoulders are common on high-speed roads in India (Tiwari et al. 2019). Our review of Google Street View imagery from other LMICs shows that they are commonly used on highways. In Abu Dhabi (UAE), for example, Albuquerque and Awadalla (2020) and Awadalla and Albuquerque (2021) reported a high prevalence of curbs greater than 150 mm in height even on highways with a design speed of 100 km/h or greater. Although the issue has not been systematically investigated, news reports from India commonly describe crashes that involve vehicles overturning after striking the median on highways and expressways (HT, 2019; PTI, 2020). This is notable because although curbs on high-speed roadways are rare in HICs, many of these news reports describe crashes on national highways or expressways. Furthermore, road safety audits have reported that the practice of constructing medians with a curb (as high as 200 mm) on high-speed roads may be common in India (Tiwari et al. 2019), though there is no systematic assessment of its prevalence.

According to the American Association of State Highway and Transportation Officials (AASHTO)’s “A Policy on Geometric Design of Highways and Streets” (AASHTO, 2011), a curb, by definition, incorporates some raised or vertical element. It can be located along the edge of shoulder or median. The purposes of curbs include drainage control, roadway edge delineation, right-of-way reduction, aesthetics, delineation of pedestrian walkways, reduction of maintenance operations, and assistance in orderly roadside development. AASHTO classifies curbs as vertical and sloping. Vertical curbs, as the name suggests, are vertical or nearly vertical and are intended to function as barriers and redirect straying vehicles back to the roadway (to protect pedestrians, for instance). In contrast, sloping curbs are intended to be mountable, designed so that vehicles can ride up and cross them if necessary. Barrier curbs are greater in height than mountable curbs and range from 150 to 255 mm, while mountable curbs have a maximum height of 150 mm (Lafond, 1997).

According to AASHTO, vertical curbs should never be used on high-speed facilities or roads with speeds greater than 70 km/h. If the use of a curb is necessary on such roads (e.g. due to drainage considerations, restricted right-of-way, or where there is a need for access control), AASHTO recommends using a sloping curb with a maximum height of 100 mm. At high-speed urban/suburban facilities with frequent access points and intersecting streets, a sloping curb of height up to 150 mm “may be considered”(AASHTO, 2011). All the design recommendations by AASTHO related to curbs are presented in Appendix A. Despite these recommendations, many states in the United States continue to construct vertical curbs either as a part of the shoulder (Jiang et al. 2013; Plaxico, 2005).

The safety aspects of curbs on high-speed roads have been studied in detail through computer simulations, full-scale crash tests, and analytical techniques (Dunlap, 1973; Navin & Thomson, 1997; Plaxico, 2005). These studies focused on understanding the mechanistic aspects of the vehicle-curb interaction. They assess how variation in curb design (height, slope, shape) as well as variation in crash characteristics such as speed of impacting vehicle, angle of impact, or vehicle type, affect vehicle dynamics, including vehicle trajectory and the likelihood of overturning. From these simulated trajectories and damage to the vehicle, the outcome of a crash can be inferred. For example, a vehicle redirected into the roadway after crashing with a curb is a more desired outcome than the one where the vehicle is overturned. Since curbs in HICs are now typically used in combination with barriers, most recent studies focus on assessing how the placement of curbs relative to barriers affects vehicle dynamics (Plaxico, 2005). Much of this work aims to inform road design standards.

There are a few studies that have reported evaluations of the impact of curbs on road safety outcomes in real-world crashes (Elvik et al. 2009). Since curbs are fixed objects, studies have also focussed on single-vehicle crashes in addition to all crashes combined (Billion et al., 1958; Jiang et al. 2013; Plaxico, 2005). However, in this literature, we found a gap in terms of systematically assessing the evidence of the impact of curbs on road safety outcomes in the real world.

In this review we aim to fill this gap and answer the following two specific questions:

1. What effect does the presence of curbs (median or shoulder) have on the number of all crashes, single-vehicle crashes, and median-related crashes?

2. What effects does the presence of curbs have on the severity of crashes?

3. How do the effects of curbs vary with vehicle type and speed?

METHODS

Database Search

We used the database of road safety related publications developed by an existing evidence and gap map (EGM) of effectiveness of road safety interventions (Mohan et al. 2020). The EGM database is extensive as the search strategy contains all road safety interventions related to highway design and includes publications with no time limit. The EGM used a comprehensive search strategy to identify all the relevant impact evaluation studies, systematic reviews and grey literature by searching 6 academic databases (SafetyLit, PubMed, Web of Science, EASTS, TRID, EMBASE, TRANSPORT) and 17 organisational websites (all listed in the Appendix B), hand searching, contacting experts and back referencing.

We used EPPI-Reviewer 4 software (Thomas & Brunton, 2007) to search all the records in the EGM database. We searched for records that included any of the following three keywords in the title or abstract—’curb’, ‘kerb’ and ‘raised median’. Since the EGM included studies conducted up to December 2019, we repeated the search for studies published from January 2020 to September 2020. The details of the search such as date, search string, filters, website, and results obtained are presented in Appendix B. The search did not limit by language, date, or status of publication. Next, we screened the records to include those studies for full-text screening that included evaluation of a curb. Among those selected for full-text screening, we used backreferencing to identify additional studies from the list of cited references. We included studies for full-text screening which had curb (vertical or slope) as a design element and reported traffic crashes as one of the outcomes. We excluded studies that included a curb in combination with a guard rail. Several studies compared the effects of including a curbed median on an undivided roadway, often with two-way left-turn lanes (TWLTL). Since these studies primarily assess the benefits of physical separation of two-way traffic, we excluded these studies from our review.

Calculation of Odds Ratio

To estimated odds ratios for median-curbs, we compared medians that provide a physical separation and whose only difference was the presence of a curb. For example, if a curbed median included a grassy median with a barrier in the middle, we compared it to a non-curbed grassy median also with a barrier in the middle. Similarly, we compared curbed shoulders with shoulders without a curb. We also compared a curb with a higher height to a curb with lower heights. We used the log odds ratio method with a fixed-effects model to calculate meta-summary as explained in (Elvik, 1995). We tested if there is systematic variation between the studies for each of the summary estimates and calculated odds ratios with a random-effects model using the method reported in Elvik (2011), Papadimitriou and Theofilatos (2017) and Elvik and Goel (2019).

We extracted data for all crashes, single-vehicle crashes, and median-related crashes. Both single-vehicle and median-related crashes can be expected to be directly affected by the presence of curbs in the median. For these crash configurations, we extracted data for crash outcomes classified as all crashes, which includes crashes of all severity levels (fatal, injury, and property-damage only), injury plus fatal crashes, and property-damage only (PDO) crashes. Not all studies reported all the combinations of crash configurations and severity levels. In addition to odds ratios of different crash types, we calculated the severity of crashes, which is defined as the conditional probability of an injury or deaths given that there was a crash. We do not present an odds ratio specific to PDO crashes, but we use PDO crashes in the calculations of severity. In total, we present the following outcomes—

1. odds ratio of rate of all crashes

2. odds ratio of rate of injury and fatal crashes

3. odds ratio of rate of median-related crashes

4. odds ratio of rate of median-related injury and fatal crashes

5. odds ratio of rate all single vehicle crashes

6. odds ratio of rate of single vehicle injury and fatal crashes

7. severity of all crashes

8. severity of median-related crashes

All outcomes, except median-related crashes, are presented for curbs on median and shoulder, and for comparing two median curbs of different heights. All the calculations for these outcomes using fixed-effects and random-effects methods are presented in the Appendix C.

RESULTS

We retrieved a total of 858 studies through the search of the EGM database and through backreferencing of those that were screened for title and abstract. After screening of title and abstract, we identified 43 articles for full-text screening. Among these, only 4 studies met our inclusion criteria for data extraction and meta-analysis (see Table 1). The reasons for exclusions included crashes not reported as outcome (n=20), cross-sectional analysis (n=13), and comparison with TWLTLs (n=6). All four studies were from the USA, and two of them were more than 50 years old. Two studies used a before-after study design, three used a case-control design, and one study used both study designs. The test for heterogeneity suggests large systematic variations across all the estimates. Therefore, we present random-effects along with fixed-effects estimates.

Table 1:

Studies used for data extraction

Study	Study type	Location of curb	Type of roads	Outcomes	Comparisons
(Billion, 1956)	Before-after with control—one site converted from painted lines to concrete parabolic barrier and one site from painted lines to concrete barrier with pipe	Median	4-lane divided parkway (rural)	-Total number of crashes and by severity: injury and PDO	431.8 mm (17-inch) concrete parabolic barrier vs 381 mm (15-inch) concrete barrier with pipe (odds ratio calculated as OR1/OR2, where OR1 is the odds ratio of parabolic barrier to painted lines and OR2 is the odds ratio of barrier with pipe to painted lines)
(Billion & Parsons, 1962)	Case-control: case and control sections are stratified by the type of median	Median	Urban, divided, multi-lane (4–6), free-access highways	-Total number of crashes and by severity: fatal, injury and PDO -Number of crashes by severity type: approach, single vehicle, overtaking	127 mm (5-inch) sloped curb of a 1.8 m(6-ft) median vs 3 m (10-ft) flush grass median 76.2 mm (3-inch) sloped curb of a 9.1 m (30-ft) grass median vs 3 m (10-ft) flush grass median 127 mm (5-inch) sloped curb of a 1.8 m (6-ft) median vs 76.2 mm (3-inch) sloped curb of a 9.1 m (30-ft) grass median 152.4 mm (6-inch) sloped curb in a 12-ft median with guide rail vs flush median with single beam guide rail in the middle 152.4 (6-inch) vertical curb with wood rail vs flush median with single beam guide rail 127 mm (5-inch) sloped curb of a 1.8 m (6-ft) median vs 76.2 mm (3-inch) sloped curb of a 9.1 m (30-ft) grass median
(Garner & Deen, 1972)	Case-control: case and control sections are stratified by the type of median	Median	Rural highways	-Total number of crashes and by severity: fatal, injury and PDO -Number of crashes by severity type: approach, single vehicle, overtaking	Unknown height sloped curb in a 6.7 m (22-ft) raised median vs 60-ft depressed median 228.6 mm (9 inch) sloped curb in a 6.7 m (22 ft) raised median vs 60-ft depressed median
(Lienau, 1996)	Simple before-after (10 sites converted from parallel drainage ditch to curbed shoulder) and case-control (9 sites with parallel drainage ditch matched to 9 sites without)	Shoulder	High-speed suburban multilane highways	-Total number of crashes and by severity: fatal, injury and PDO	Curbed (6 to 9 inch/152.4 to 228.6 mm) vs parallel drainage ditch using before-after Curbed (6 to 9 inch) vs parallel drainage ditch using case control

Within the limited number of studies included in the review, we found a large variation in the design of curbs. The height of the curb varied from a minimum of 76 mm to a maximum of 432 mm. Only one study included curbs with a height greater than 350 mm (Billion, 1956), while the rest of the studies included a maximum height of 230 mm. The curbs were vertical, sloping, and parabolic in shape. The design specifications of curbs are presented in appendix D.

Odds ratios for all comparisons using fixed-effects and random-effects models are presented in Table 2. Odds ratios for individual studies are presented in Appendix F. The confidence intervals (CI) for all estimates are larger for random-effects model than for fixed-effects because the former accounts for systematic variation between the studies in addition to the random variation within the study.

Table 2:

Summary results of all studies with mean and 95% confidence interval range of odds ratios

Curb location	Comparison	Crash type	Odds ratios: Mean (95th CI)
			Fixed-effects model	Random-effects model
Median	Curb vs no curb&	All	1.70 (1.61, 1.80)	1.53 (1.00, 2.61)
Median	Curbs of different heights#	All	1.74 (1.42, 2.28)	1.76 (0.80, 3.86)
Shoulder	Curb vs no curb	All	1.32 (1.26, 1.38)	1.18 (0.82, 1.68)
Median	Curb vs no curb	Injury	1.35 (1.24, 1.48)	1.23 (0.91, 1.68)
Median	Curbs of different heights	Injury	1.12 (0.82, 1.53)	--
Shoulder	Curb vs no curb	Injury	1.44 (1.35, 1.53)	1.29 (0.81, 2.05)
Median	Curb vs no curb	All single vehicle	1.47 (1.35, 1.59)	1.57 (1.06, 2.33)
Median	Curbs of different heights	All single vehicle	2.20 (1.12, 4.33)	--
Shoulder	Curb vs no curb	All single vehicle	1.70 (1.51, 1.91)	1.46 (0.76, 2.83)
Median	Curb vs no curb	Injury single vehicle	1.56 (1.11, 2.20)	1.48 (0.80, 2.74)
Median	Curbs of different heights	Injury single vehicle	1.00 (0.46, 2.18)	--
Shoulder	Curb vs no curb	Injury single vehicle	1.76 (1.43, 2.16)	1.60 (0.76, 3.36)
Median	Curb vs no curb	All median-related	1.54 (1.35, 1.75)	1.65 (1.20, 2.10)
Median	Curb vs no curb	Injury+fatal median-related	1.85 (1.51, 2.28)	2.00 (1.19, 3.38)
Conditional probability of injury or fatal crash given that a crash has occurred
Median	Curb vs no curb	Severity$	0.66 (0.58, 0.74)	0.66 (0.46, 0.94)
Median	Curbs of different heights	Severity	0.48 (0.28, 0.82)	0.47 (0.12, 1.99)
Shoulder	Curb vs no curb	Severity	1.18 (1.04, 1.21)	1.19 (0.96, 1.38)
Median	Curb vs no curb	Severity of median-related	1.17 (0.89, 1.54)	1.17 (0.88, 1.55)

^&curb heights from 76.2 mm to 228.6 mm

^#includes comparisons of curbs with higher height to curbs with lower height: 127 mm vs 76.2 mm and 432 mm vs 381 mm

We first discuss the findings that are statistically significant at 95% CI in both models. For medians, we found that curbs (sloping or vertical) increased the number of all crashes by 53%–70% (depending on the model), and all single-vehicle crashes by 47%–57%. Curbs also increased median-related all crashes and median-related injury and fatal crashes by 54%–65% and 85%–100%, respectively. Curbs on medians reduced the severity of all crashes by 34% in both models. The only effect estimates for shoulders that was nearly statistically significant was the severity of crashes, which shows that curbs on shoulders increased the severity of crashes.

Among the effects that were not statistically significant, the mean estimates were of the same general magnitude and direction as those that were statistically significant. The results for curbs on shoulders were the same as those for curbs on medians, except in the case of severity of crashes which seems to increase with the former and decrease with the latter.

Finally, we compared two median vertical curbs of different heights, however, due to lack of estimates, the random-effects estimate is only for all accidents. Curbs with higher height compared with curb of lower height resulted in an increase in the number of crashes by 74% to 76%.

DISCUSSION

Findings from the Review of the Published Epidemiological Literature

We found that the presence of curbs on medians or shoulders increases the risk of all crashes and all single-vehicle crashes. We found the same result for median-related all crashes and median-related injury/fatal crashes. We found that curbed medians decreased the severity of all crashes while curbed shoulders increased the severity (see Table 2).

It is not clear why the severity of all crashes reduces in the presence of a curb on the median. A possible explanation for reduced severity is that curbed medians prevent vehicles from crossing over to the opposite side and subsequently engaging in a side or head-on crash, or crashing into trees or other hazardous structures within the median. It is noteworthy that in one of the reviewed studies, we are comparing two grass medians where one is curbed and the other is flush, and the curbed median is narrower than the flush median (1.8 m and 3 m, respectively). Even here, the reduction in severity is by 65%, indicating that a curb may compensate for narrow median width as far as the severity of crashes is concerned. However, this explanation fails when we look at the results when both the median types (curbed and flush) also have a guide rail or wood rail in the middle. Given that a guide rail prevents vehicles from crossing over, we would expect the changes in severity levels to be moderate. But even for these cases, the severity of crashes is reduced by 87 to 90 percent.

The effect of curbs on severity is in the opposite direction when they are located on shoulders compared to when they are located on medians. Crashes are more severe in the presence of curbed shoulders compared to shoulders without curbs, though this effect is only nearly statistically significant. This finding, unlike that for medians, however, can be explained. After crashing into a curbed shoulder, a vehicle is likely to overturn or attain a trajectory that could result in much more severe injuries. This is less likely to occur in the absence of a curb. The literature doesn’t support this finding. Jiang et al. (2013), using a cross-sectional approach, reported that while curbed shoulders increase the probability of injury crashes, they reduce the severity of crashes. The authors argued that “curbs may be associated with higher energy crashes but may absorb more energy on average leading to lower event severity”. They supported this argument using data according to which crashes on roadways with curbs were less than half as likely to result in overturning as crashes on roadways without curbs.

The findings on the severity of crashes should be interpreted with caution. There are two potential problems. First, the calculation for severity does not include exposure and is only a function of the observed number of crashes— i.e. it calculates, given a crash, how likely is it to result in a severe injury or death. Second, the calculation for severity ratio includes the number of PDO crashes and such crashes are most likely to be underreported. Let us suppose that we are comparing the severity of crashes of a curbed median with the severity of crashes of a flush grass median. It is likely that in the presence of a curb, a vehicle veering off the roadway may crash into it without any injury. However, the crash may lead to some damage to the vehicle and may be reported as a PDO. However, in a grassy median, a vehicle that veers off may just drive back onto the roadway and continue without reporting a crash. In other words, the underreporting bias may work in opposite directions for the two comparing units. A greater reporting of PDO in a curbed median compared to a flush median implies that for the former, the severity may be artificially reduced, and that is what we also found as the mean effect for curbed medians.

According to AASHTO guidelines, if necessary, high-speed roads can use sloping curbs with a maximum height of 100-mm to 150-mm. Of the four studies reviewed in our systematic review, Billion & Parsons (1962) reported the effects of 75 mm and 130 mm sloping curbs on medians, both within the range of AASTHO’s recommended heights. We found that the odds ratios of injury risk for these specific curbs were also significantly greater than 1 (see Appendix F), indicating that there may be no threshold curb height below which there is no added injury risk compared to a flush median. A policy recommendation would be that when curbs must be used on high-speed roads, it should always be accompanied with reducing speed limits to moderate the effect of curb. However, our review does not provide us with a speed threshold below which curbs do not pose greater injury risk. Some states in the United States have used this strategy of reducing speed limits for highways where curbs are installed, though it has been argued that it could result in confusion among drivers around appropriate driving speed (Baek et al., 2006; Yang et al., 2014).

Findings from Laboratory Experiments, Computer Simulations, and Analytical Studies

The main limitation of this review is that we found only four studies that were conducted a long time ago. This indicates that the real-world epidemiological evidence on the impact of curbs on crashes is weak. Further, all the studies were limited to rural highways and did not report effects for different traffic speeds or types of vehicles, which is one of our three research questions. Therefore, we summarized the findings from studies that have used laboratory tests, computer simulations, and analytical methods to assess the impact of a curb on crash outcomes. This summary is presented in Appendix E. Due to their experimental nature, these studies allow a more diverse specification of parameters, including curb design, vehicle type and speed of impacting vehicle, allowing a more nuanced understanding of the mechanism behind crashes involving a curb.

The risks that curbs pose to straying vehicles on high-speed roads have been understood since the 1950s. As a result, curbs are now rarely used on highways in HICs without additional mitigation measures like appropriately positioned deformable guardrails or reduced speed limits. And, since most impact evaluations of highway infrastructure are conducted in high-income countries, curbs are now rarely studied, and the issue receives little policy attention. Nevertheless, curbs are commonly used on high-speed highways in LMICs, with heights often greater than 150 mm. Our review strongly suggests that the use of such curbs results in greater injury risk. Furthermore, it should be noted that there are substantial differences between the traffic environments in LMICs and HICs. Roads in LMIC often have higher traffic volumes and a wider range of vehicle speeds, with slow-moving indigenous vehicles (e.g. auto-rickshaws/tuk-tuks that are often locally designed) sharing the road with new high-performance cars, resulting in more frequent overtaking maneuvers than in HICs. In such an environment, the risks posed by curbs are poorly understood and may be substantially higher. In our review, all the studies were conducted in the US. Similarly, the risks posed by curbs to motorcyclists, which comprise over half the fatalities in many LMICs (WHO, 2018), has not been studied. It is important for LMICs to establish research programs that use local empirical data to assess how road designs affect injuries in the local traffic environment.

Figure 1:

Literature search and study screening