Objective To apply cluster analysis to classify road traffic crash (RTC), identify crash occurrence patterns, and provide a basis for formulating targeted intervention programs to specific RTC patterns.
Methods The entries of traffic police conventional accident handling records in Shanghai Pudong New Area from January 1, 2010 to December 31, 2016 were retrieved, and 3 135 primary responsible persons (PRP) were identified. Nine variables including PRP's age and gender, as well as RTC occurring time, season, weather condition, road category, road section, transport means, and recorded RTC cause were selected for latent class analysis and system clustering respectively to generate cluster results and injury outcomes.
Results For the selected dataset, latent class analysis revealed more hidden patterns than traditional system clustering method The results of latent class analysis classified the accidents into six categories of RTC occurrence patterns, namely "young and middle-aged PRP+motor vehicle+highway group" "young and middle-aged PRP+passenger vehicle+general road group" "young and middle-aged PRP+evening+motorcycle and passenger vehicle+drunk driving without license group" "middle-and old-aged PRP+electric vehicle and bicycle group" "middle-and old-aged PRP+morning and evening+pedestrian group" and "young and middle-aged PRP+late night+motor vehicle group". The differences in injury outcomes between the six categories were statistically significant (χ2=1 492.492, P < 0.05), and a correlation between injury outcomes and the categories was also identified (r=0.568, P < 0.05). The largest health impact was contributed by the "middle-and old-aged PRP+morning and evening+pedestrian group", followed by the "middle-and old-aged PRP+electric vehicle and bicycle group", and the "young and middle-aged PRP+passenger vehicle+general road group" did the least. After comparing the results of logistic regress models using original data with the results using the generated categories, new information on injury risk factors was added, and the contribution of the same independent variable to injury outcomes varied in different RTC models.
Conclusion Latent class analysis on the specific dataset shows better performance than conventional system clustering in terms of RTC pattern recognition in current study. Middle-aged and senior pedestrians, motorbicycle riders illegally crossing roads, and young or middle-aged drivers driving motor vehicles in late night are the high-risk variable combinations for RTC-related injury in this area.
DownLoad: