Background and Aims

The Proportional Integral Derivative controller (PID) is adopted in several applications including the artificial pancreas (AP). An advantage of PID is that only three parameters are needed and is normally subject-tailored with one personalization parameter, the Total Daily Insulin (TDI). No predictive models like in MPC are needed. The slow dynamics of the subcutaneous (SC) route limits PID effectiveness requiring the need of meal announcement. The faster dynamics of intraperitoneal (IP) delivery open the possibility of avoiding meal announcement. In this work, we propose a time-varying PID, test it in silico on the novel IP-Padova simulator (IP-T1DS) which incorporates intra- and inter-day variability of insulin sensitivity.

Methods

We identify the average IP insulin-glucose transfer function starting from data collected on the IP-T1DS, an extended version of the SC FDA accepted simulator. We consider the Insulin-to-Carbohydrate Ratio (CR) as personalization parameter as an alternative to TDI. From the CRs associated to the different portions of the day, we design a time-varying PID. We test the two PIDs simulating a 12-week protocol.

Results

Both controllers are effective without meal announcement, but the time-varying PID is less sensitive to the different day portions.

Conclusions

We propose a time-varying PID for a fully implantable IP AP. IP-PIDs are effective without meals announcement thanks to the faster dynamics of IP sensing and delivery. Personalization through CR faces insulin sensitivity variations, increasing control performance w.r.t. TDI-based personalization.

The work was supported by H2020-FETPROACT Project FORGETDIABETS, n. 951933.