Despite the continuous efforts from federal and state agencies to improve safety at traffic work zones, incidents continue to occur, resulting in 675 fatalities in the U.S. during the last decade. Current safety measures at work zones, such as visual and audible alarms are ineffective due to the lack of calibration studies to evaluate the attention span of workers towards alarms, given variances in the alarm modality, frequency and duration. This study proposes an integrated approach combining virtual reality (VR) and wearable sensors to capture data regarding workers' behaviors towards safety alarms when workers are exposed to simulated dangerous situations in VR. This dataset is needed to understand the relationships between workers' responses (i.e., react or dismiss) and the characteristics of the received alarms (i.e., modality, frequency, and duration). The proposed approach was implemented on an urban intersection from a real-world work zone in New York City for user studies.