Background: Subthalamic deep brain stimulation (STN-DBS) effects may decrease with Parkinson's disease (PD) progression. There is no indication if, when, and how to consider the interruption of DBS treatment in late-stage PD. The objective of the current study was to investigate the percentage of “poor stimulation responders” among late-stage PD patients for elaborating an algorithm to decide whether and when DBS discontinuation may be considered. Methods: Late-stage PD patients (Hoehn Yahr stage ≥4 and Schwab and England Scale <50 in medication on/stimulation on condition) treated with STN-DBS for at least 5 years underwent a crossover, double-blind, randomized evaluation of acute effects of stimulation. Physicians, caregivers, and patients were blinded to stimulation conditions. Poor stimulation responders (MDS-UPDRS part III change <10% between stimulation on/medication off and stimulation off/medication off) maintained the stimulation off/medication on condition for 1 month for open-label assessment. Results: Thirty-six patients were included. The acute effect of stimulation was significant (17% MDS-UPDRS part III), with 80% of patients classified as “good responders.” Seven patients were classified as “poor stimulation responders,” and the stimulation was switched off, but in 4 cases the stimulation was switched back “on” because of worsening of parkinsonism and dysphagia with a variable time delay (up to 10 days). No serious adverse effects occurred. Conclusions: The vast majority of late-stage PD patients (92%) show a meaningful response to STN-DBS. Effects of stimulation may take days to disappear after its discontinuation. We present a safe and effective decisional algorithm that could guide physicians and caregivers in making challenging therapeutic decisions in late-stage PD.