شناسایی و اعتبار سنجی مدل رگرسیون لجستیک برای پیش بینی آسیب های جدی در ارتباط با تصادفات رانندگی
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی||تعداد کلمات|
|24854||2011||11 صفحه PDF||سفارش دهید||9177 کلمه|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Accident Analysis & Prevention, Volume 43, Issue 1, January 2011, Pages 112–122
A multivariate logistic regression model, based upon National Automotive Sampling System Crashworthiness Data System (NASS-CDS) data for calendar years 1999–2008, was developed to predict the probability that a crash-involved vehicle will contain one or more occupants with serious or incapacitating injuries. These vehicles were defined as containing at least one occupant coded with an Injury Severity Score (ISS) of greater than or equal to 15, in planar, non-rollover crash events involving Model Year 2000 and newer cars, light trucks, and vans. The target injury outcome measure was developed by the Centers for Disease Control and Prevention (CDC)-led National Expert Panel on Field Triage in their recent revision of the Field Triage Decision Scheme (American College of Surgeons, 2006). The parameters to be used for crash injury prediction were subsequently specified by the National Expert Panel. Model input parameters included: crash direction (front, left, right, and rear), change in velocity (delta-V), multiple vs. single impacts, belt use, presence of at least one older occupant (≥55 years old), presence of at least one female in the vehicle, and vehicle type (car, pickup truck, van, and sport utility). The model was developed using predictor variables that may be readily available, post-crash, from OnStar®-like telematics systems. Model sensitivity and specificity were 40% and 98%, respectively, using a probability cutpoint of 0.20. The area under the receiver operator characteristic (ROC) curve for the final model was 0.84. Delta-V (mph), seat belt use and crash direction were the most important predictors of serious injury. Due to the complexity of factors associated with rollover-related injuries, a separate screening algorithm is needed to model injuries associated with this crash mode.
نتیجه گیری انگلیسی
We have presented results of logistic regression analyses to predict the probability of a serious injury in a crash-involved vehicle, following the approach laid out by the CDC Expert Panel on Field Triage. These analyses are based on the information that may be obtained using an EDR in a crash, or by an operator communicating with vehicle occupants immediately following a crash (age and gender). The results of these analyses are promising for the possibility of initiating triage decisions using EDR-based crash information. The AUC of 0.84 indicates significant discrimination by the algorithm, though there is room for improvement. The key predictors in this model are delta-V (log transformed), belt use, age (presence of anyone 55 or older), and direction of impact. Additional limited predictive value comes from multiple impacts and females present in the vehicle. Vehicle type was not a significant predictor, though it was left in the model to match the predictors given by the expert panel.