This paper presents an artificial pancreas algorithm implemented with a discrete-time sliding-mode method combined with a nonlinear block controllable form using a unidirectional control law and a discrete adaptive observer. The algorithm is both unidirectional, using insulin as unique input, and hybrid with a percentage of the pre-meal bolus administered in advance to complement the control action. Personalisation combines patient clustering and individual gain calculation after analysing glucose time-in-range. The stability of the complete closed-loop system involving the observer bounds is tested using the Lyapunov methodology. The performance of the algorithm is evaluated with 256 patients, simulated using Hovorka’s model. The evaluation defines two scenarios (closed-loop and open loop with continuous subcutaneous insulin infusion, CSII), divided into three 1- week intervals, to respectively compare the impact of glucose control under the expected meal plan as a result of changes in carbohydrates quantities and meal schedule, and in response to more extreme events. The performance of time-in-range in each period obtained with CSII is improved by hybrid closed-loop in all cases, i.e., with the expected meals (weekdays 74.9 vs 78.4; weekends 71.5 vs 73.4), with meal variability (weekdays 74.3 vs 77.6; weekend 71.6 vs 72.7), and in the presence of transgressions (forgetting meal announcement: 50.1 vs 63.4; meal omission: 78.2 vs 82.03; copious meal: 60.1 vs 63.4). Compared to CSII, the personalised nonlinear block controller was able to increase the time-in-range without glucose presence in time below range level 2 for 98 % of the cohort for expected meals, meal variability, and transgressions.