Feedback systems are frequently used to encourage safer driving habits by targeting behaviors like harsh braking, harsh accelerations, and speeding. While research consistently shows that these interventions lead to short-term improvements during the feedback period, their long-term effectiveness is less examined. Many drivers revert to their old habits after the feedback ends, highlighting the need to better understand what drives this relapse. This study focuses on the relapse patterns of harsh braking behavior once feedback interventions are withdrawn. Using survival analysis methods, this paper evaluates how long drivers sustain improved behavior, identify key relapse predictors, and explore the influence of individual and contextual factors on long-term outcomes. This study examined data from 31 drivers who participated in a naturalistic driving experiment over 21 months, generating a total of 24,904 trips. Harsh braking events, calculated per 100 kilometers, were measured across three phases: baseline (before feedback), feedback, and post-feedback. During the feedback phase, drivers received regular feedback, which was withdrawn in the post-feedback phase. The Kaplan-Meier survival curves revealed that relapse rates increased steadily over time. Initially, 81.5% of drivers maintained improved harsh braking behavior during the first 50 trips of the post-feedback phase. However, this dropped to 61.4% by 100 trips and further declined to 40.3% by 150 trips, meaning nearly 60% of drivers reverted to their pre-feedback harsh braking levels by the end of the study. This pattern underscores the temporary nature of behavior improvements once feedback is withdrawn. The study highlights the critical role of sustained feedback in maintaining long-term improvements in driver behavior. Although feedback interventions significantly reduced harsh braking rates during the feedback phase, relapse was common once feedback was withdrawn. Key factors such as age and vehicle engine capacity influenced the likelihood of relapse, suggesting that tailored feedback systems could enhance the effectiveness of these interventions.