ESTRO 2021 Abstract Book

S1565

ESTRO 2021

institution. To simulate the operation of the clinic over time, we assume that, on each day, a newly diagnosed patient presents at the clinic with probability q and a varying benefit dNTCP from protons. When a new patient presents, a decision for proton or photon treatment must be taken. If no proton slot becomes available within T days, the patient must receive photons. The goal is to determine dNTCP thresholds for selecting a patient for protons in order to optimally balance the competing goals of making use of all available slots while saving slots for patients with high benefit. This problem can be formulated as a Markov decision process and the optimal thresholds can be determined via value-policy iteration. The method is demonstrated here for T =10, N =3, and q =0.4. The latter corresponds to two incoming patients per week on average. Assuming a 30-fraction treatment, 12 patients are under treatment on any day on average while only N =3 slots are available. The dNTCP values are drawn from a probability distribution with mean 10.5% and standard deviation 4.5%, motivated by a plan comparison study in combination with a grade IV xerostomia NTCP model. Results The optimal thresholds depend on the current utilization of the facility, which is shown in Fig. 1. For example, if one proton slot is available, the second frees up in 20 days, and the third is blocked for another 30 days, the optimal dNTCP threshold is 13%. If instead, the second slot becomes available in 5 days and the third in 15, it is optimal to lower the dNTCP threshold to 7.8%. The optimal patient selection policy yields a population average NTCP reduction of 3.55% compared to photons for everyone. Fig. 2 investigates the simplified strategy of using a constant dNTCP threshold independent of the current utilization of the facility. A threshold of 0% corresponds to random patient selection, i.e., giving an available proton slot to the next patient. This yields an average NTCP reduction of 2.5%, reflecting that 25% of patients are treated with protons. The optimal constant threshold of 11% yields an average NTCP reduction of 3.4%. A constant dNTCP threshold too high leads to many unused proton slots.

Conclusion In the context of limited proton slot availability at a given institution, the optimal dNTCP threshold that minimizes average NTCP in the population of all treated patients depends on the number of available proton slots, the average number of patients under treatment, and the current utilization of the facility.