The present paper included data from a number of sources. The main source of data was the Swedish traffic data acquisition system (STRADA), where information of road traffic crashes between passenger cars and pedestrians or bicyclists for the period January 2003–December 2022 was obtained. STRADA contains information of road traffic crashes reported by either the police and/or emergency care hospitals. STRADA contains a number of characteristics, such as information regarding sex, age, location, crash circumstances, weather conditions etc (Amin et al., 2022). When there is medical information about the crashes, such as injury severity, the reports from the two sources are linked together (Howard & Linder, 2014). The information contained in the register is pseudonymized (Transportstyrelsen, 2023).

AEB with pedestrian and/or cyclist detection is standard in some car models and optional on some others, depending on model and model year. To find this information, ethical approval was obtained in order to access individual vehicle information in form of the vehicle license registration number from STRADA. Cars with optional fitment of these systems were identified, and individual information in terms of manufacturer codes was identified through license registration numbers and individual car equipment lists via the website biluppgifter.se. Cars with uncertain or unknown AEB status were not included in the analysis.

To identify crashes that occurred on or near a traffic calming treatment, the coordinates for all crashes was sent to the STA, which provided a list of bicycle and pedestrian crossing facilities and any traffic calming infrastructure in connection to the crash sites. The data also included a report date of completion reported to STA. Crashes that had occurred before the report date of the treatment were identified, and for these, Google Street View was used to identify if the treatment was present at the time of the crash. This was, however only possible for crashes occurring after 2009 since that was the first year with available Street View.

Furthermore, crashes in which the AEB VRU would not be expected to be activated were excluded. Such crashes are when the pedestrian or cyclist had not been struck by the car front, for example if the passenger car was reversing, or if a cyclist had rear-ended a passenger car. Only crashes with available information from both the police and emergency care hospital, and where the vehicle identification number/license registration number was available were included. Fatalities as well as cases without fully diagnosed injuries were also excluded. The final sample consisted of 11 035 crashes involving 4 306 pedestrians and 6 902 cyclists.

The injury measure used in this study was the Risk of Permanent Medical Impairment (RPMI), which is the basis for the Swedish national target for serious injuries and is included in the STRADA database (Berg et al., 2016). RPMI estimates the risk of long-term medical impairment based on the diagnosed injury location and severity and criteria of the Swedish Insurance Companies (Försäkringsförbundet, 2004; Malm et al., 2008). A medical impairment is considered permanent when no further improvement in physical and/or mental function is expected with additional treatment. The assessment impairment degree is independent of cause and without regard to occupation, hobbies or other special circumstances of the injured person. An injury is given a degree of medical impairment between 1% and 99%. Some general examples are as follows: limited motion of shoulder 1–20%, amputation of tibia is set to an impairment of 19%, total loss of hearing 60%. While permanent medical impairment of at least one percent includes all levels of impairments, a permanent medical impairment of at least ten percent results in persistent symptoms affecting activities of daily living of a person. The present study used the mean RPMI (mRPMI) of PMI of at least 1% (mRPMI1%+) and at least 10% (mRPMI10%+).

Risk of permanent medical impairment of at least 1% and 10% can be found in the Supplements (Table S1, published separately), and these values are used to calculate an overall RPMI according to Equation 1:

See Malm et al. (2008) for a more detailed description of the method.

The relative difference between the mean values of RPMI (mRPMI1% and mRPMI10%+) was calculated and tested using an independent two sample t-test which was conducted for unequal sample sizes and variance, the same approach as used in (Ohlin et al., 2017; Rizzi et al., 2016; Strandroth et al., 2011). p-values are displayed to indicate significance level at 99%, 95% and 90%. Following are descriptions of the comparisons made and how comparison groups were created.

The data was analysed in order to match case and control groups of cars with and without AEB with pedestrian/cyclist detection with regards to car sizes, model years and pedestrian protection score according to the Euro NCAP tests. This meant that the following conditions were set and applied to both pedestrians and cyclists:

For injured pedestrians, this resulted in 135 cases involving AEB with pedestrian detection and 188 cars without. For injured cyclist the result was 169 cases involving AEB with cyclist detection and 500 without. This means that a variety of manufacturers, with various different AEB systems, are included. Hence, this study mirrors the variety of AEB systems, with different technical specifications and performance, present in the modern Swedish car fleet. A list of all car models and sizes included in the analysis can be found in the Supplements (Table S2).

The Euro NCAP pedestrian test score (i.e. without the score for AEB system) was used to estimate the effect of pedestrian protection. All car models and model years with a pedestrian test score were included, although cars with AEB with pedestrian and/or cyclist detection were excluded as this could potentially confound the results. Four groups of cars were created, based on the star rating previously used by Euro NCAP, where:

For this analysis, the impact of age of the struck pedestrian or cyclist was considered by estimating the effect of pedestrian protection for those aged ≤ 64 years in addition to all age groups. The cut-off at 65 years has been done in similar studies, e.g. (Lubbe et al., 2022; Wisch et al., 2017).

A separate analysis was undertaken to understand if the active bonnet influences the injury severity compared to cars without active bonnet, but with a similar (high) score in the Euro NCAP pedestrian protection test. Included in the analysis was all car models with Model Year ≥ 2009, scoring 19–36 points, without AEB with pedestrian/cyclist detection.

A list of cars equipped with active bonnet included in the analysis can be found in the Supplements (Table S2).

A cut-off of at ± 20 meters from the collision point was set, which meant that any traffic calming treatment within 20 meters distance from the crash site assumed the crash to have occurred in a treated area. The STA classification of pedestrian and bicyclist crossings was used (Lundberg & Ekman, 2016), and is defined as:

Good safety standard (green):

Medium safety standard (yellow):

Poor safety standard (red):

The empirical combined effect was calculated, where a crash population with different treatments included at the same time was compared to another crash population without the treatments. Specifically, the following groups were created and compared:

As reference, the theoretical combined effect of the above-mentioned treatments was also calculated according to the method of common residuals (Elvik, 2009):

This approach assumes independence between treatments. The effect of helmet to reduce mRPMI10%+ among bicyclists was derived from Rizzi et al. (2013), where the effect of helmet to reduce serious head injuries (64%) was multiplied by the proportion of serious head injuries (42%).

The effect of AEB and the Euro NCAP pedestrian test score are confounding factors for one another, thereby, as described earlier, the effect the Euro NCAP pedestrian test score is calculated only for cars without AEB. Vice versa, the effect of AEB can be influenced by the proportion of cars with a high score in the Euro NCAP pedestrian test. The proportion between different groups of points split by AEB fitment can be fount in the Supplements (Table S4).

Another obvious confounder is speed. As we do not have information of involved vehicles travel speeds, the speed limit as well as the level of traffic calming is regarded as a proxy for speed, and results are calculated for different speed limits and levels of traffic calming (STA safety rating of crossings).

Another confounding factor is helmet wearing among cyclists and the rate of helmet wearing is presented in the text throughout the result section. Other factors related to the individuals such as age and gender are presented in the Supplements.

The characteristics of the comparison groups are shown in the Supplements (Tables S6–E11), along with the injury distributions (Tables S12–S13).

Table 1 shows the effect of AEB with pedestrian detection on pedestrians mRPMI10%+ and mRPMI1%+ by speed limit and STA safety rating of pedestrian and bicycle crossings. For mRPMI10%+ a reduction of 44% can be observed for speed zones ≤ 50 km/h, as well as speed zones 40–50 km/h, between the group struck by cars equipped with AEB with pedestrian detection compared with the group struck by cars without the system. However, the reductions are only significant at 90% level. In speed zones of ≥ 60 km/h no reduction can be observed. For mRPMI1%+ only the reduction of 21% in speed zones 40–50 km/h is statistically significant of at least 90%.

For pedestrian crossings, the effect of AEB with pedestrian detection shows a significant reduction of 60% for mRPMI10%+ (p = 0.05) and 34% for mRPMI1%+ (p = 0.02) at crossings of less good or poor quality. At crossings of good quality, no significant reductions can be observed either for mRPMI10%+ or mRPMI1%+.

Table 2 shows the effect of AEB with cyclist detection on mRPMI10%+ and mRPMI1%+ divided by speed limit and STA safety rating of pedestrian/bicycle crossings. In ≤ 50 km/h speed zones, a reduction of 35% (p = 0.05) is observed comparing the group struck by cars equipped with AEB with pedestrian detection compared with the group struck by cars without the system. In ≤ 50 km/h speed limit, the rate of helmet wearing was 45% in the group of cyclists struck by cars with AEB with cyclist detection and 46% in the group struck by cars without the system (Table 3). In speed zones of ≥ 60 km/h no reduction can be observed.

As regards the effect of AEB across the TS safety rating of pedestrian/bicycle crossings, no significant reduction of bicyclists mRPMI10%+ or mRPMI1%+ was observed. The rate of helmet wearing was lower or the same for cyclists struck by cars with or without AEB with cyclist detection (Table 3).

Table 3

Speed limit	% cyclists with helmet (AEB with cyclist detection)	% cyclists with helmet (AEB without cyclist detection)
≤ 30 km/h	44%	59%
40–50 km/h	47%	41%
≤ 50 km/h	45%	46%
≥ 60 km/h	33%	74%
TS safety rating of crossings
Green	19%	49%
Yellow	44%	47%
Red	52%	52%

The effect of Euro NCAP pedestrian protection score on pedestrians mPRMI1%+ and mRPMI10%+, divided by speed limit is shown in Table 4. Only cars not equipped with AEB with pedestrian detection are included. Comparing pedestrians struck by cars with a low score (1–9 points) or a high score (28–36 points) the total reduction in all speed limits is 48% on mRPMI10%+ (p = 0.01). In low speeds (≤ 30 km/h) the reduction is 98% (p < 0.01), however, there are only six pedestrians struck by a 28–36 point car in ≤ 30 km/h speed zones. Looking only at pedestrians aged ≤ 64 years, the reduction between low scoring cars and high scoring cars is 64% (p < 0.01). For mRPMI1%+ similar reductions, but smaller, are observed with 67% reduction in ≤ 30 km/h (p < 0.01) and 23% reduction across all speed limits (p = 0.07).

A significant reduction of pedestrian mRPMI10%+ comparing cars with 1–9 points and cars with 26–38 points was found at crossings with medium safety level according to the STA classification, -58% (p < 0.01). For mRPMI1%+, a significant reduction was observed at crossings with good safety level (green), -67% (p < 0.01).

Table 5 shows the effect of Euro NCAP pedestrian protection score on cyclists mRPMI10%+ and mRPMI1%+. Only cars without AEB with cyclist detection are included. A significant reduction of cyclist mRPMI10%+ can be seen comparing low scoring cars to high scoring cars in ≤ 30 km/h speed limit (-73%, p = 0.02) and across all speed limits (-36%, p = 0.06). Including only those aged ≤ 64 years, the reduction is 49% (p < 0.01). However, these results are irrespective of helmet wearing. The proportion of helmet wearing was higher in the group with cars 28–36 points, 46% compared with 33% in car group 1–9 points (Table 6). Looking at helmeted and non-helmeted cyclists separately, the difference between cars in group 1–9 and 28–36 points was not statistically significant for all age groups. However, looking at age groups ≤ 64 years, the relative difference in mRPMI10%+ between the cars in group 1–9 and 28–36 was 45% (p = 0.04) for helmeted cyclists and 44.2% (p = 0.09) for non-helmeted cyclists. A corresponding difference was not found for mRPMI1%+.

A significant reduction of cyclists mRPMI10%+ comparing cars with 1–9 points and cars with 26–38 points was found at crossings with good safety level (green) and medium safety level according to the STA classification, -67% (p < 0.01) and -55% (p = 0.02) respectively. This observed difference could, however, be a result of helmet wearing, as the proportion of helmet wearing was higher in the groups with high scoring cars (28–36 points) (Table 6). In difference to speed limit, it was not possible to divide between helmeted/non helmeted cyclists or older/younger age groups due to the limited size of the material.

Table 6

	%with helmet	%with helmet ≤64 years	%with helmet Green STA safety rating	%with helmet Yellow STA safety rating	%with helmet Red STA safety rating
1-9 points	33%	34%	26%	29%	33%
10-18 points	34%	35%	29%	34%	32%
19-27 points	42%	44%	43%	38%	39%
28-36 points	46%	43%	30%	39%	61%
Total	36%	37%	33%	35%	35%

The effect of active bonnet on pedestrians and bicyclists mRPMI10%+ is shown in Table 7. Only cars without AEB with pedestrian/cyclist detection is included. Reductions between 12–54% were observed, however, no results were significant at 95% confidence level. For mRPMI1%+, there was a difference observed across all speed limits with a 24% reduction (p = 0.01). However, the rate of helmet wearing was higher in the group with active bonnet (Table 8).

Table 8

Speed limit	% with helmet with Active bonnet	% with helmet without Active bonnet
≤ 30 km/h	65%	51%
40–50 km/h	42%	38%
≥ 60 km/h	100%	62%
TOTAL	51%	42%

The effect of traffic calming (TS rating) on pedestrians and cyclists mRPMI1%+ and mRPMI10%+ is shown in Table 9. Only cars without AEB with pedestrian/cyclist detection are included. Results show that pedestrians mRPMI1%+ is reduced by 15% comparing crossings of good quality (green) with poor quality (red) (p = 0.06) and cyclists mRPMI10%+ is reduced by 32% (p = 0.01). Cyclist helmet wearing rate was 30% in the group with good quality crossings and 34% in the group with poor quality crossings.

The empirical combined effect of pedestrian protection, traffic calming (STA safety rating of crossings), AEB with pedestrian/cyclist detection and helmet use among cyclists is shown in Table 10. Comparisons are made between groups of low scoring cars (1–9 points), poor quality (red) crossings, and non-helmeted cyclists and groups of high scoring cars (≥ 19 points) with AEB for pedestrian/cyclist, good quality (green) crossings, and helmeted cyclists. For pedestrians, a reduction of mRPMI10%+ by 69% (p < 0.01) is observed, and a corresponding reduction of 64% (p = 0.02) of cyclists mRPMI10%+. The total reduction for pedestrians and cyclists mRPMI10%+ together is 69% (p < 0.01). No significant reductions of mRPMI1%+ were found.

The corresponding theoretical effect of the individual treatments were estimated to a 48% reduction of pedestrian mRPMI10%+, and 63% reduction of bicyclists mRPMI10%+ (Table 11).