To manage Type 1 diabetes effectively, continuous blood glucose monitoring and precise insulin administration are the most essential. While IoT-based insulin pumps have automated insulin delivery to a high level of success, security and privacy issues are still serious concerns. The study proposes an Adaptive predictive deep Reinforcement Learning (AP-DRL) integrated in the blockchain-assisted digital twin. This model of AP-DRL employs the combination of deep reinforcement learning strategies in computing glucose swings and deciding insulin doses from profiling data on a real-time and historical basis. Digital twin on the part of a healthcare professional keeps acquiring information about the physical situation of the patient to maximize the amount of insulin and glucose. Using a blockchain system, the collected data is kept secure, and a safe connection between the insulin pump, AI model, digital twin, and physicians is promoted. The suggested AP-DRL model outperforms the Long Short-Term Memory (LSTM) model, and the bidirectional LSTM (Bi-LSTM) model while validated using the ShangaiT1DM dataset regarding both precision and adaptability to an individual patient’s necessities. It exhibits a best Mean Absolute Error (MAE) of 0.02 and a best Root Mean Square Error(RMSE) value of 0.31 surpassing the performance against its counterparts in Basal insulin dosage prediction. Also, it showcases the best MAE of 0.08 and a best RMSE value of 0.63 outperforming its counterparts in Bolus insulin dosage prediction. Besides, it affirms a personalized AP-DRL model is essential to improve the quality of life of individual diabetic patients against a single unified global model for insulin dosage prediction as the glucose-insulin dynamics differ for different patients by maintaining different learning rates and gamma values for different subjects.