Direct use of multivariable normal tissue complication probability models in treatment plan optimisation for individualised head and neck cancer radiotherapy produces clinically acceptable treatment plans

Abstract Background and purpose Recently, clinically validated multivariable normal tissue complication probability models (NTCP) for head and neck cancer (HNC) patients have become available. We test the feasibility of using multivariable NTCP-models directly in the optimiser for inverse treatment planning of radiotherapy to improve the dose distributions and corresponding NTCP-estimates in HNC patients. Material and methods For 10 HNC cases, intensity-modulated radiotherapy plans were optimised either using objective functions based on the ‘generalised equivalent uniform dose’ ( OFgEUD ) or based on multivariable NTCP-models ( OFNTCP ). NTCP-models for patient-rated xerostomia, physician-rated RTOG grade II-IV dysphagia, and various patient-rated aspects of swallowing dysfunction were incorporated. The NTCP-models included dose–volume parameters as well as clinical factors contributing to a personalised optimisation process. Both optimisation techniques were compared by means of ‘pseudo Pareto fronts’ (target dose conformity vs. the sum of the NTCPs). Results Both optimisation techniques resulted in clinically realistic treatment plans with only small differences. For nine patients the sum-NTCP was lower for the OFNTCP optimised plans (on average 5.7% (95%CI 1.7–9.9%, p < 0.006)). Furthermore, the OFNTCP provided the advantages of fewer unknown optimisation parameters and an intrinsic mechanism of individualisation. Conclusions Treatment plan optimisation using multivariable NTCP-models directly in the OF is feasible as has been demonstrated for HNC radiotherapy.