Accid Anal Prev [JOURNAL]
Sun
200 papers
Interpretable spatiotemporal traffic crash risk prediction using DMD-based graph neural networks.
Xu W, Wang C, Chu Y
Accid Anal Prev · 2026 Jul · PMID 41911624 · Publisher ↗
Spatiotemporal prediction of urban traffic crashes is an important basis for proactive safety management, yet many existing models have difficulty capturing both temporal dynamics and spatial dependence in an interpretab...
Spatiotemporal prediction of urban traffic crashes is an important basis for proactive safety management, yet many existing models have difficulty capturing both temporal dynamics and spatial dependence in an interpretable way. This study develops a hybrid framework that integrates Hankel-based Dynamic Mode Decomposition (Hankel-DMD) with a spatiotemporal graph neural network (STGNN) to predict short-term neighborhood-level crash counts. Daily crash records from 2019 to 2021 for 78 neighborhoods in Denver, USA are aggregated into a neighborhood-day matrix. Hankel-DMD is applied to this matrix to extract low-rank spatiotemporal modes that describe dominant trends and recurrent fluctuations. A graph neural network defined on a distance-correlation-based neighborhood graph then learns nonlinear residuals that correct the linear DMD prior and transmit information along the urban network. The proposed model is evaluated in a multi-step prediction setting with horizons of 1, 3, 5, and 7 days, and is compared with statistical time-series models, tree-based machine learning models, a pure Hankel-DMD model, and deep learning baselines including a STGNN. Across all horizons, the hybrid model achieves the lowest mean absolute error and root mean squared error, with improvements of about 17 to 30% in mean absolute error and 13 to 24% in root mean squared error relative to the best deep learning benchmark. Performance gains are consistent across high-, medium-, and low-risk neighborhood groups. Hankel-DMD eigenvalues and spatial modes reveal stable temporal and spatial structures in 2019 and 2021, together with clear deviations in 2020 associated with disrupted mobility patterns. These results show that dynamics-informed graph learning can provide accurate and interpretable crash risk forecasts at the neighborhood scale and can support targeted urban safety interventions.
England-wide injury-severity analysis of e-scooter riders using a Bayesian spatial field model.
Zhao J, Konstantinoudis G, Heydari S
Accid Anal Prev · 2026 Jul · PMID 41905265 · Publisher ↗
As electric scooter (e-scooter) use has expanded, understanding the factors associated with e-scooter rider injury severity has become increasingly important for road safety policy. This study analyses 2,128 crashes invo...
As electric scooter (e-scooter) use has expanded, understanding the factors associated with e-scooter rider injury severity has become increasingly important for road safety policy. This study analyses 2,128 crashes involving e-scooters and motor vehicles across England (2020-2023) to identify factors associated with severe and fatal injuries to e-scooter riders. Using the geographic coordinates of crashes, we developed a Bayesian spatial field model implemented via the Stochastic Partial Differential Equation (SPDE) approach for fast Bayesian estimation. Our approach accounts for spatial unobserved heterogeneity (area-level "context" effects) often overlooked in injury severity studies. Results indicate that severe or fatal injuries are more likely among older riders, male riders, and in crashes occurring in darkness, on single carriageways, on roads with speed limits of 40 mph or higher, involving heavy vehicles, at night or early morning, or with e-scooter skidding/overturning, frontal impacts, e-scooters entering main roads, or opponent vehicles moving straight. Conversely, motor vehicles performing moving-off manoeuvres are linked to lower odds of severe injuries. Importantly, the presence of authorised e-scooter trials was not found to be associated with rider injury severity outcomes. Our spatial analysis reveals higher odds of severe injury in parts of north-western and south-eastern England relative to the national average. Our research highlights the importance of vehicle kinematics, road environment, and spatial context in shaping injury severity and support targeted, evidence-based interventions, including infrastructure measures and vehicle-based safety technologies such as blind-spot detection.
Behavioral adaptation in mixed traffic: the roles of AV penetration rate and human driving style.
Chen H, Ma Z, Zhang Y
Accid Anal Prev · 2026 Jul · PMID 41886865 · Publisher ↗
Automated vehicles (AVs) are designed to improve traffic safety, mobility, and driver comfort; however, their benefits depend on the widespread adoption of AVs. As full market penetration remains unlikely in the near ter...
Automated vehicles (AVs) are designed to improve traffic safety, mobility, and driver comfort; however, their benefits depend on the widespread adoption of AVs. As full market penetration remains unlikely in the near term, AVs will continue to operate alongside human-driven vehicles (HDVs), making it essential to understand their interactions and the implications for traffic safety. This study investigated how different AV penetration rates (0%, 25%, 50%, 75%) influence HDV drivers' behavioral adaptation at the tactical level and subjective evaluations, while considering drivers' individual driving styles (aggressive, moderate, and defensive). Thirty-six drivers participated in a driving simulator experiment involving two driving scenarios (left-turn and lane-change scenarios) under varying AV penetration rates. Drivers' adaptive decision-making in response to AVs' defensive driving, as the AV penetration rate changed, was measured by the frequency of left turns executed without yielding and the distance maintained from surrounding vehicles during lane changes. Subjective evaluations were assessed through perceived safety and anxiety ratings collected after each trial. Results indicated that the influence of AV penetration rate was moderated by driving style. In the lane-change scenario, increased AV penetration rate resulted in more adaptive decision-making among aggressive and moderate drivers, whereas in the left-turn scenario, this effect emerged only for aggressive drivers. In contrast, AV penetration had no significant effect on defensive drivers' behavior in either scenario. These findings suggest that higher AV penetration may compromise safety in mixed traffic by provoking more aggressive decision-making and aggressive behavior among certain driver types, highlighting the need to account for driver adaptation patterns in AV deployment strategies.
Factors influencing traffic conflict severity in shared spaces.
Ghimire A, Payre W, Birrell SA … +1 more , Debnath AK
Accid Anal Prev · 2026 Jun · PMID 41871472 · Publisher ↗
While shared spaces offer many benefits for pedestrians, traffic conflicts between pedestrians and other road users are common and often lead to crashes. However, little is known about the factors influencing the safety...
While shared spaces offer many benefits for pedestrians, traffic conflicts between pedestrians and other road users are common and often lead to crashes. However, little is known about the factors influencing the safety of pedestrians in shared spaces. This paper aims to examine how pedestrian-vehicle (both four and two wheelers) conflicts are influenced by factors related to pedestrians' crossing behaviour, environmental conditions, and traffic characteristics. Through collecting real-world observational data using roadside video cameras, pedestrian-vehicle conflicts were measured in shared spaces in Victoria, Australia. The results from a series of generalised linear regression models showed that the severity of conflicts reduces when a vehicle yields to let pedestrians cross, the conflicting vehicle is a 4-wheeler, and there is an increase in the number of pedestrians near a pedestrian who is crossing a shared zone. Conversely, increased conflict severity was observed on weekdays than weekends, and during the AM peak hours than other times. Higher severity was also observed for conflicts involving pedestrians crossing between parked vehicles and when there is an increase in the proportion of stationary pedestrians near the kerbs. These new results on the factors influencing conflict severity and the probability distribution of these factors will be helpful in planning and designing safer shared spaces.
Traffic crash data augmentation with multi-type variables using hybrid VAE-Diffusion generative neural networks for enhancing crash frequency modeling.
Chen J, He Q, Liu P … +4 more , Ma W, Zheng N, Liu P, Pu Z
Accid Anal Prev · 2026 Jun · PMID 41871471 · Publisher ↗
Crash frequency modeling aims to analyze influential factors of crashes to enhance road safety. However, as crashes are inherently rare events, excessive zero observations in crash datasets undermine crash frequency mode...
Crash frequency modeling aims to analyze influential factors of crashes to enhance road safety. However, as crashes are inherently rare events, excessive zero observations in crash datasets undermine crash frequency models' ability to identify high-risk road segments. Existing statistical models are often limited by strict distributional assumptions, while resampling and deep generative methods often distort data representation or struggle with multi-type crash data (count, ordinal, nominal, and real-valued) since these heterogeneous variables follow different statistical distributions and require distinct encoding strategies. This study proposes TabSyn, a hybrid VAE-Diffusion model that transforms multi-type crash data into a continuous latent space through a tokenizer-based embedding and VAE encoding, enabling the model to preserve the inter-variable correlations and underlying data structure. To validate its effectiveness, TabSyn is compared with state-of-the-art generative models (CTGAN, TVAE, GReaT, and StaSy) in synthetic data quality, and integrated with XGBoost for crash frequency prediction against two statistical models (ZIP, GAM-Poisson). Results demonstrate that TabSyn outperforms benchmark methods by achieving the best synthetic data distributional and structural fidelity, and the lowest prediction error, particularly for Top-K high-risk segment identification. This study offers valuable insights for improving crash frequency modeling and traffic safety management through imbalanced data augmentation of multi-type crash data.
Quantitative human takeover reliability assessment in MASS transitions: Enhancing safety via integrated system control and cognitive modelling.
Yang X, Zhou T, Zhao Y … +3 more , Zhang W, Wang S, Wu Y
Accid Anal Prev · 2026 Jun · PMID 41865729 · Publisher ↗
Maritime Autonomous Surface Ships (MASS) with multiple modes of operation are increasingly adopted to improve efficiency and address crew shortages. Safe mode transitions, especially dependable human takeovers in urgent...
Maritime Autonomous Surface Ships (MASS) with multiple modes of operation are increasingly adopted to improve efficiency and address crew shortages. Safe mode transitions, especially dependable human takeovers in urgent scenarios, are essential for the effective prevention of navigation accidents. However, existing human reliability analysis methods often overlook how information degradation and cognitive processes jointly affect takeover performance. To address this challenge, the information, decision, and action in crew context model (IDAC) is extended by introducing an external filtering stage that explicitly accounts for information loss or distortion before reaching the operator's cognition. Building on this, an integrated system control and cognitive perspective is proposed. The model combines System-Theoretic Process Analysis with the enhanced IDAC model, and embeds Bayesian Networks to identify causal chains of takeover failures, quantify Human Error Probability, identify critical factors. Through scenario-based reasoning, it further derives the key causal paths of takeover failure in urgent scenarios. A case study involving both remote and onboard takeover scenarios demonstrates the framework's applicability. Results indicate that the diagnosis and decision-making stage is the most critical, with fatigue, attention, available time and trust level emerging as dominant factors. Based on these findings, this study proposes a three-level pre-alert mechanism and targeted intervention strategies to enhance human reliability during MASS mode transitions. This study provides a scalable and behaviourally grounded framework that supports the development of takeover guidelines and safety standards, aiming to prevent navigation accidents caused by human takeover failures during the development of ship autonomy.
Corrigendum to "Injury severity analysis of e-bike crashes: An age-stratified study of riders aged 40 and above" [Accid. Anal. Prev. 228 (2026) 108392].
Jia J, Yue H, Yang S … +2 more , Jia X, Qiu Y
Accid Anal Prev · 2026 Jun · PMID 41864843 · Publisher ↗
Abstract loading — click title to view on PubMed.
Redefining ramp influence area for curved diverging and merging freeway segments using crash data.
Novat NK, Kwigizile V, Mwende SI … +3 more , Bitaliho UG, Novat N, Shita H
Accid Anal Prev · 2026 Jun · PMID 41864153 · Publisher ↗
PROBLEM: Freeway Ramp Influence Areas (RIAs) are commonly defined using fixed buffer distances despite evidence that driver behavior, roadway geometry, and traffic operations produce spatially heterogeneous crash pattern...
PROBLEM: Freeway Ramp Influence Areas (RIAs) are commonly defined using fixed buffer distances despite evidence that driver behavior, roadway geometry, and traffic operations produce spatially heterogeneous crash patterns near ramps. This study proposes a data-driven framework to empirically identify where ramp-related crash influence stabilizes rather than assuming a predetermined distance using crash data from the Washington State from 2018 to 2023. METHOD: This study developed an analytical framework to delineate ramp influence areas. First, Negative Binomial Generalized Additive Models (NBGAMs) were used to quantify how geometric design, traffic demand, and operational features influence crash frequency across curved and straight entry and exit ramps. Second, Negative Binomial gradient boosting (XGBoost) with grouped site-level cross-validation was used to generate out-of-fold predicted crash-risk curves, which were then analyzed using a piecewise change-point estimator with site-level spatial bootstrap resampling to identify the crash influence distance (τ) and its uncertainty for each ramp configuration. RESULTS: Results reveal a geometry-dependent hierarchy of influence distances. Curved entry ramps exhibited the longest downstream disturbance (τ ≈ 1,800 ft), substantially exceeding conventional thresholds. Straight entry ramps showed localized and demand-sensitive influence (τ ≈ 300-1,000 ft). Curved exit ramps produced compact influence zones (τ ≈ 500-900 ft), while straight exit ramps were highly localized (τ ≈ 300 ft). These findings demonstrate a directional asymmetry in crash propagation: merging disturbances diffuse downstream whereas diverging disturbances dissipate rapidly near the gore. IMPACT ON INDUSTRY: This research offers a data-driven framework for refining RIA definitions based on geometric alignment. By identifying geometry-specific thresholds, transportation agencies can develop more accurate safety assessments and improve freeway ramp design standards. The study's findings support the adoption of flexible, risk-based RIA delineation approaches to enhance crash mitigation efforts in complex ramp environments.
Modified MRBQ for the gig economy: linking risky riding behaviors to self-reported safety incidents among food delivery riders.
Pervez A, Nguyen-Phuoc DQ, Quang Vo ND … +1 more , Lee JJ
Accid Anal Prev · 2026 Jun · PMID 41861611 · Publisher ↗
The rapid growth of the food delivery sector has created job opportunities, particularly for young and low-income workers, but has also introduced new road safety challenges. To examine these risks, this study adapted an...
The rapid growth of the food delivery sector has created job opportunities, particularly for young and low-income workers, but has also introduced new road safety challenges. To examine these risks, this study adapted and extended the Motorcycle Rider Behavior Questionnaire (MRBQ) to reflect the unique behaviors of food delivery riders. A 62-item MRBQ was developed through literature review and focus group discussions, and data were collected via an offline survey from 524 riders in Hanoi, Vietnam. The adapted MRBQ was analyzed using exploratory factor analysis to identify key dimensions of risky riding behavior, followed by a multivariate probit model to examine their associations with self-reported crashes, near-crashes, and fines. The analysis revealed six behavioral dimensions: traffic errors, safety violations, distractions, traffic violations, speeding violations, and control errors. Notably, the emergence of 'distractions' and 'traffic violations', linked to mobile phone use and frequent signal and lane breaches, highlights risk specific to app-based delivery work. The absence of the stunts factor, characterized by thrill-seeking behaviors, among food delivery riders further differentiates them from recreational or non-occupational riders, highlighting their more work-oriented and goal-driven riding style. The results further showed that older age, lower education, higher income, and lack of formal training were significantly associated with increased self-reported safety incidents, including crashes, near-crashes, and fines. Key behavioral predictors of these incidents were distractions, speeding, and traffic violations. The findings reinforce existing road safety recommendations by providing context-specific empirical evidence for food delivery riders, highlighting the continued importance of rider training and delivery platform designs that prioritize safety alongside efficiency in urban gig-economy settings.
Behavioral adaptation of human drivers in car-following interactions with automated vehicles.
Saljoqi M, Orsini F, Rossi R … +1 more , Gastaldi M
Accid Anal Prev · 2026 Jun · PMID 41861610 · Publisher ↗
The emergence of Automated Vehicles (AVs) will result in mixed traffic with Human-Driven Vehicles (HDVs), creating a complex traffic environment. While literature supports the effect of AVs on human driving behavior, the...
The emergence of Automated Vehicles (AVs) will result in mixed traffic with Human-Driven Vehicles (HDVs), creating a complex traffic environment. While literature supports the effect of AVs on human driving behavior, the influence of different AV driving styles in this context remains largely unexplored. This study investigated interactions between HDVs and AVs varying in driving style (cautious vs. aggressive) and appearance (recognizable vs. unrecognizable) at Market Penetration Rates (MPR) of 0% to 75% in 25% increments within a highway context. Car-following behavior of 160 participants (56 females, age range 19-65) in a driving simulator experiment was analyzed using average time headway (THW) and standard deviation of relative speed (SDRS) as key metrics. Linear Mixed Models (LMMs) and Structural Equation Modelling (SEM) were applied. Results showed that compared to the baseline (no AVs), drivers' THW decreased when interacting with aggressive AVs by 11.1%, 8.6%, and 13.2% at MPRs of 25%, 50%, and 75%, respectively. Interactions with cautious AVs led to even larger reductions in THW by about 9.9%, 16.6%, and 17.3% at the same MPRs. SDRS improved across MPRs by 4.5%, 5.2%, and 6.2%, independent of AV driving style or appearance. SEM analysis indicated that AV driving style moderated relationships between independent, mediating, and dependent variables, while drivers' perceived stress and comfort mediated effects of AVs, trust propensity, and recognizability on THW and SDRS. These findings provide insights into human responses to AV behavior, highlighting implications for traffic flow and safety to ensure smoother and safer human-AV interactions.
Modeling bounded rationality in pedestrian-vehicle interactions at non-signalized crosswalks: A game theoretic quantal response equilibrium approach.
Li T, Sun Z, Zhou M … +2 more , Sze NN, Zhou Y
Accid Anal Prev · 2026 Jun · PMID 41855672 · Publisher ↗
Modeling pedestrian-vehicle interactions (PVIs) is essential for automated driving systems in mixed traffic environments but challenging due to the stochasticity and complexity of human behaviors. Conventional PVI models...
Modeling pedestrian-vehicle interactions (PVIs) is essential for automated driving systems in mixed traffic environments but challenging due to the stochasticity and complexity of human behaviors. Conventional PVI models rely on strong assumptions like perfect information and unbounded rationality. For more realistic modeling of PVIs at unsignalized crosswalks, we propose a two-agent non-zero-sum non-cooperative dynamic game incorporating Quantal Response Equilibrium. This Game Theoretic Quantal Response Equilibrium (GT-QRE) approach accommodates human agents' limited cognition and bounded rationality, while accounting for unobserved heterogeneity in road users' rationality and sequential decision-making during crossing activities. We then develop an integrated simulation platform that links pedestrians' and vehicles' decisions with their movements to simulate their interactions, where GT-QRE is used to capture the decision-making behaviors of both pedestrians and vehicles, Social Force Model (SFM) and Model Predictive Control (MPC) are used to simulate pedestrian and vehicle motions. The proposed method is calibrated and validated using empirical data at four selected crosswalks in three cities in China. The key findings include: (i) both pedestrian and vehicle exhibit limited rationality in PVI; (ii) compared to conventional Nash Equilibrium (NE), our method improves prediction of crossing decisions by 13.0% for pedestrians and 36.1% for vehicles, highlighting PVIs decision process are inherently dynamic rather than static;(iii) the GT-QRE approach effectively captures the dynamic decision-making behavior of pedestrians and vehicles during interactions; (iv) the simulation platform, which integrates GT-QRE decision-making with SFM/MPC motion models, realistically reproduces vehicle and pedestrian movements at crosswalks, yielding low longitudinal/lateral displacement and velocity errors.
Exploring variations in the influence of various factors on fatal/non-fatal crash outcomes of novice driver involved crashes across diverse urban environments: Insights from partially topic-constrained modeling approach.
Sun Z, Jin S, Yang Y … +5 more , Xiao J, Wei S, Wang D, Wang J, Lu H
Accid Anal Prev · 2026 Jun · PMID 41850181 · Publisher ↗
Novice drivers present considerable challenges to urban traffic safety. Due to their limited driving experience, they may not be fully acquainted with the diverse urban environments, which can lead to variations in the i...
Novice drivers present considerable challenges to urban traffic safety. Due to their limited driving experience, they may not be fully acquainted with the diverse urban environments, which can lead to variations in the influence of factors on the crash severities across these urban settings. However, previous studies have primarily considered the urban environment as a potential influencing factor while overlooking these variations. To fill these gaps, this research employed a six-year dataset of novice driver involved crashes in Shenyang (China) and characterized the urban environment by examining the number of various types of public service facilities situated within the buffer zones surrounding these crashes. Additionally, it extracted the topics of the urban environment based on Latent Dirichlet Allocation (LDA) model, with each identified topic representing a distinct category of novice driver involved crashes. Furthermore, the research examined variations in the influence of factors on crash severities across these categories through a partially topic-constrained modeling approach. Results show that the entire dataset was subsequently divided into three topics, revealing significant differences in how factors influenced crash severities within these topics. Specifically, improper operation is observed to increase the likelihood of fatal crashes in areas with low-intensity and service-scarce environment (Topic 1), weekends and speeding are likely to increase the probability of fatal crashes in areas with high-intensity and service-concentrated environment (Topic 2); summer and possessing two years of driving experience are also linked to an increased likelihood of fatal crashes in areas with moderate-intensity and service-dispersed environment (Topic 3).
More explicit is not always better: Boundary conditions for action guidance in hazard notifications across traffic complexity.
Ryu G, Ji YG
Accid Anal Prev · 2026 Jun · PMID 41850180 · Publisher ↗
With the expansion of vehicle automation, hazard notification systems are increasingly important for safety-critical driver-automation coordination. This study examined the effects of traffic complexity, message type, an...
With the expansion of vehicle automation, hazard notification systems are increasingly important for safety-critical driver-automation coordination. This study examined the effects of traffic complexity, message type, and delivery modality on user perceptions in a Level 3 conditional automation scenario, where participants supervised automated driving and responded to a pedestrian hazard. Forty participants completed 18 simulator trials in a within-subject design. Traffic complexity significantly affected subjective workload, and interaction effects indicated that the influence of modality and message type varied across complexity levels. Compared with a no-alert baseline, notification conditions improved usability and user experience and increased perceived safety in visual-based conditions; however, workload reductions were not consistent across modalities, and explicit action guidance was not uniformly beneficial, particularly under high traffic complexity. As a supplementary exploratory analysis, a Random Forest model with TreeSHAP was used to summarize model-based predictive sensitivity across the tested conditions (with the no-alert condition used as reference coding). TreeSHAP salience highlighted visual-advisory condition indicators as prominent contributors for usability and user experience, while the impact of modality varied across traffic-complexity levels. These findings provide empirical evidence for designing context-adaptive Human-Machine Interfaces (HMI) that calibrate information transparency to environmental demands, mitigating cognitive overload and supporting safety-relevant acceptance outcomes in conditional automation within the tested factor space.
Effectiveness of speed cameras in reducing speed: a systematic review.
Amancio EC, Cecy Gadda TM, Inocente Domingos MD … +5 more , Bastos JT, da Costa Bonetti G, Pinho Ferreira SM, Schmitz A, Oviedo-Trespalacios O
Accid Anal Prev · 2026 Jun · PMID 41846101 · Publisher ↗
Speeding has been identified as one of the most common risk factors for the occurrence and severity of traffic accidents. One of the most economical and widespread strategies for speed management is the installation of S...
Speeding has been identified as one of the most common risk factors for the occurrence and severity of traffic accidents. One of the most economical and widespread strategies for speed management is the installation of Speed Cameras (SC). In light of the growing body of evidence in this field and the need for a coherent synthesis of research findings, challenges and gaps, this paper provides a systematic review and an integrated overview of the current state of knowledge on the topic. Five electronic databases (SCOPUS, Web of Science, PubMED, TRID and PROQuest) were used to identify relevant studies. Records were identified, screened, and assessed using a structured multi-stage review process consistent with established systematic review procedures. The included studies reported investigations related to the SC impact on driving speed. A systematic classification scheme was adapted to summarize the study's characteristics. Ninety-four studies were identified. As a result, issues in the study objects, methods and procedures of SC evaluation and impact on vehicle speed assessment were discussed. In particular, due to the complex road environment, other factors also impact driving speed patterns. Additionally, it is demonstrated that the impact of SC on speeds has been assessed by four methods: self-reported questionnaires and location, time, and cross-sectional speed analysis. Complementary research on the following themes would provide interesting insights on SC related studies: understanding how other urban environment factors can influence SC effectiveness; settlement of compliance distance, continuous speed measuring methods associated with time and location speed analysis, and evaluation of the impact of road type, speed limit, and geometry.
Polysubstance impairment detected in fatally injured drivers, United States, 2018-2022.
Mathews JN, Chihuri S, Li G
Accid Anal Prev · 2026 Jun · PMID 41846100 · Publisher ↗
Polysubstance-impaired driving is an emerging public safety concern. However, research on polysubstance-impaired driving is scant. We used 2018-2022 Fatality Analysis Reporting System data and logistic regression modelin...
Polysubstance-impaired driving is an emerging public safety concern. However, research on polysubstance-impaired driving is scant. We used 2018-2022 Fatality Analysis Reporting System data and logistic regression modeling to assess the prevalence of, and factors associated with, polysubstance impairment among drivers aged 14 years and older who died within 1 h of motor vehicle crashes and who had toxicological testing data available (n = 60,741). Polysubstance impairment was defined as having a blood alcohol concentration (BAC) ≥ 0.08 g/dL and testing positive for one or more nonalcohol drugs. The prevalence of polysubstance impairment increased from 14.9% in 2018 to 18.8% in 2022 (p < 0.0001). Cannabis was the most frequently detected nonalcohol drug (25.0%), followed by stimulants (19.2%), and narcotics (8.3%). With adjustment for demographic characteristics, drivers who died in nighttime (7 PM to 6 AM) crashes were almost three times as likely as those who died in daytime (7 AM to 6 PM) crashes to be polysubstance-impaired (adjusted odds ratio [aOR]: 2.94, 95% confidence interval [CI]: 2.80-3.10). Relative to drivers who died in 2018, the odds of polysubstance impairment increased 15% in 2019 (aOR: 1.15, 95% CI: 1.07-1.24), 25% in 2020 (aOR: 1.25, 95% CI: 1.16-1.34), 32% in 2021 (aOR: 1.32, 95% CI: 1.23-1.42) and 38% in 2022 (aOR: 1.38, 95% CI: 1.28-1.49). Results of this study indicate that polysubstance-impaired driving is increasingly involved in fatal motor vehicle crashes. Interventions to reduce polysubstance impairment among drivers are urgently needed for improving traffic safety.
Human-inspired emotion and attention encoding for autonomous vehicles' decision-making: a lane-change timing optimization case.
Han T, Liu T, Bao Q … +1 more , Shen Y
Accid Anal Prev · 2026 Jun · PMID 41831329 · Publisher ↗
In mixed traffic environments, autonomous vehicle (AV) lane-changing requires coordination with multiple surrounding vehicles, posing significant challenges. Improper lane-change timing can lead to traffic accidents, so...
In mixed traffic environments, autonomous vehicle (AV) lane-changing requires coordination with multiple surrounding vehicles, posing significant challenges. Improper lane-change timing can lead to traffic accidents, so current AVs tend to adopt conservative strategies, especially on busy highways. However, flexible and reliable lane-changing is often crucial and sometimes indispensable; otherwise, it not only reduces efficiency but may also cause traffic disruptions. To address this, we draw on research from cognitive psychology and neuroscience to propose an emotion and attention encoding framework that enables AVs to change lanes in a human-like manner. Specifically, based on Cognitive Energy Theory, Attenuator Theory, and Prospect Theory, we construct the neural encoding processes of drivers' physiological arousal, subjective experience, attention allocation, and emotional utility under multiple risk stimuli. We introduce an emotional utility model (EUM) and a human-like lane-changing decision (HLD) method to help AVs adaptively optimize lane-change timing. Finally, we evaluate our approach using a data-driven simulation built from 3,129 lane-change segments. Results show that under the 3-sigma rule, the HLD achieves a lane-change rate exceeding 99.8%, with lane-change timings closely matching real driver behavior and demonstrating even greater safety. This success is mainly attributed to the EUM's ability to adaptively adjust weights according to the relative urgency of risks, thereby better balancing overall utility and individual risk. Furthermore, the success of this case study will provide insights for AVs' decision-making in other complex tasks and scenarios.
Modeling heterogeneity in fault attribution of Pedestrian-Vehicle crashes using a Random parameter Binary Logit approach.
Ergin ME
Accid Anal Prev · 2026 Jun · PMID 41831328 · Publisher ↗
Crashes between pedestrians and motor vehicles remain a significant traffic safety problem, especially in urban areas. This study aims to analyze factors that systematically influence whether the driver or the pedestrian...
Crashes between pedestrians and motor vehicles remain a significant traffic safety problem, especially in urban areas. This study aims to analyze factors that systematically influence whether the driver or the pedestrian is considered at fault in crashes, affecting them in terms of road, vehicle, environmental, temporal, and behavioral factors. The study uses data from 7,213 crashes that occurred in Istanbul between 2022 and 2023, where fault was attributed solely to the pedestrian or solely to the driver. First, a Binary Logit Model was created as a benchmark model. Subsequently, a Random Parameter Binary Logit Model was constructed using the Likelihood Ratio test to identify heterogeneity not observed in fixed-parameter models. As a result, "intersection", "traffic sign", "traffic lights", "adverse weather", "public transport", and "the number of vehicles" were determined as random parameters. Crashes occurring at intersections, in areas with traffic signs, in multi-vehicle crashes, and in adverse weather conditions increase the likelihood of attributing fault to the driver, while crashes at traffic lights and involving public transport vehicles increase the likelihood of attributing fault to pedestrians. These findings suggest that interventions addressing different road user-related and behavioral risks, rather than uniform safety measures, are more effective.
DRPVLM: A generative multimodal large language model for real-time driving risk prediction.
Wang J, Zhang W, Fu T … +1 more , Shangguan Q
Accid Anal Prev · 2026 Jun · PMID 41825169 · Publisher ↗
Large language models (LLMs), known for their general knowledge comprehension capabilities, have recently been integrated into certain in-vehicle systems. However, their potential to enhance driver understanding of traff...
Large language models (LLMs), known for their general knowledge comprehension capabilities, have recently been integrated into certain in-vehicle systems. However, their potential to enhance driver understanding of traffic environments and support driving risk identification remains underexplored. This study proposes a Driving Risk Prediction Vision-Language Model (DRPVLM) to recognize real-time driving risks. The framework is fine-tuned using LoRA on several open-source multimodal LLMs of different parameter sizes, including three Qwen-2.5-VL models (32B, 7B, and 3B), Gemma-3-12B-it, and Llama-3.2-11B-Vision. DRPVLM processes video and image data from the Shanghai Naturalistic Driving Study to extract multi-dimensional features, including road environment, traffic conditions, and driver states, which complement structured trajectory data obtained from in-vehicle sensors. These features are subsequently fed into a Long Short-Term Memory (LSTM) neural network for risk prediction. In addition, we compare DRPVLM, equipped with each of these multimodal LLMs, with the model using only structured trajectory data collected from in-vehicle sensors to evaluate their predictive performance. Results indicate that Multimodal LLMs significantly enhance driving-risk prediction, with fine-tuned Qwen2.5-VL-32B achieving an accuracy of 0.89 to 0.92 and an F1 score of 0.88 to 0.91 across observation windows. LLMs with different parameter sizes also perform well and clearly outperforming the baseline model that relies solely on structured trajectory data collected from in-vehicle sensors, whose performance drops below 0.7 for longer prediction horizons. Feature importance analysis shows that all five LLM‑extracted variables make meaningful contributions, effectively supplementing structured trajectory features. These findings demonstrate the effectiveness of multimodal LLMs in enhancing risk feature extraction and improving driving risk prediction performance, highlighting the strong potential of LLMs for real-time driving risk prediction.
A deep reinforcement learning algorithm for optimizing safety and efficiency of traffic signals using traffic conflict technique and artificial intelligence-based video analytics.
Tahir HB, Haque SMM
Accid Anal Prev · 2026 Jun · PMID 41812491 · Publisher ↗
The advancements in computer vision have opened doors to estimate crash risks in real-time and revisit traffic signal systems to optimize safety and efficiency within a unified framework. While the efficiency reward can...
The advancements in computer vision have opened doors to estimate crash risks in real-time and revisit traffic signal systems to optimize safety and efficiency within a unified framework. While the efficiency reward can be based on operational characteristics, there remains a critical need for methodologies that integrate non-stationary, conflict-type-specific, and cycle-level crash probability estimates into traffic signal design. This study proposes a deep reinforcement learning technique to optimize the safety and efficiency of traffic signals using AI-based video analytics. Specifically, the proposed framework builds on a Deep Q-Network (DQN) that integrates real-time cycle-level crash risks, estimated from a non-stationary Extreme Value Theory model, with traffic delays (waiting times), extracted from a microscopic traffic simulation platform. The proposed framework is trained and tested on two isolated signalized intersections in Queensland, Australia, and compared with the observed adaptive traffic signal control system. The estimated rear-end crash risks from the developed non-stationary extreme value model were validated by using crash frequency estimates against the Poisson confidence bounds of observed crashes. The developed model was utilized in an integrated Deep Reinforcement Learning-based framework to optimize the safety and efficiency of traffic signals. Compared to the observed adaptive traffic signal control system, the proposed Deep Reinforcement Learning-based signal system has been found to reduce crash risk and delay by 69.95% and 33.14% at the Gold Coast Rd - Hope Island Rd intersection, and 87.75% and 37.91% at the Granard Rd - Beaudesert Rd intersection. The trained DQN model has also been found to consistently dissipate the queues without causing excessive delays in any direction. The safety and efficiency weights of around 0.5 have been found to form the optimal policy for traffic signal optimization. The proposed traffic signal design framework has the potential to enhance both safety and efficiency, offering improvements beyond approaches that focus solely on efficiency optimization.
Impact assessment of contributing factors in direct-impact work zone crashes using heterogeneity and constrained-based discrete choice models.
Chowdhury TI, Starewich M, Barua S … +2 more , Tusti AG, Das S
Accid Anal Prev · 2026 Jun · PMID 41806800 · Publisher ↗
Work zone intrusion crashes, where vehicles enter active work areas and strike workers or equipment, remain a critical safety concern with high potential for severe outcomes. This study analyzes five years of Texas crash...
Work zone intrusion crashes, where vehicles enter active work areas and strike workers or equipment, remain a critical safety concern with high potential for severe outcomes. This study analyzes five years of Texas crash data (2020-2024) to identify the roadway, environmental, driver, vehicle, and crash dynamics influencing injury severity in these events. Using both multinomial logit (MNL) and advanced Random Parameters Logit (RPL) models with heterogeneity in means and variances, this study accounts for unobserved heterogeneity and evaluate temporal stability across individual and pooled years. The analysis incorporates a partially constrained estimation approach, which allows selected parameters to remain fixed across time periods while permitting others to vary, balancing statistical efficiency with flexibility in capturing year-to-year differences. Results indicate that environmental conditions such as cloudy or rainy weather and nighttime lighting affect crash severity, potentially due to reduced visibility or altered driver behavior. Roadway context, including intersection type, ramp versus main lane location, and rural versus urban settings, shows distinct temporal patterns, highlighting the importance of context-specific countermeasures. Driver and vehicle attributes are also critical, with male drivers, younger drivers, and heavier vehicle types such as pickups and SUVs more frequently associated with fatal or serious injury outcomes, while some passenger car crashes are linked to reduced severity. Incorporating unobserved heterogeneity and partial constraint improved model fit and revealed risk factors that standard models overlooked. Temporal stability testing demonstrated that certain effects shift over time, underscoring the need for continuous monitoring. These findings provide targeted, evidence-based strategies for engineering, enforcement, and education to reduce injury severity in work zone intrusions, while demonstrating the value of advanced, partially constrained modeling for safety policy and decision-making.
About
- Frequency
- Sun
- Papers found
- 200
- RSS feed
- Subscribe