Personalising treatment plan quality review with knowledge-based planning in the TROG 15.03 trial for stereotactic ablative body radiotherapy in primary kidney cancer

In the FASTRACK II trial, participants receive a single fraction of 26 Gy for tumours smaller than 4 cm, or 42 Gy in three fractions for larger tumours. A sequential cohort of 53 previously treated primary kidney cancer patients were used as the training set for the KBP model. These had been treated at a single institution either as part of a prospective Phase I trial or off-trial following the FASTRACK II protocol (23 patients treated with single fraction, 30 treated with three fractions). These patients were selected as they represented the expected cohort that would be recruited in the FASTRACK II trial. This technique was previously demonstrated to be safe with a signal for efficacy [9, 25]. All patients were planned and treated with 3DCRT, using 7–9 beams typically including 1–2 non-coplanar beams. The plans were added into an initial KBP model (RapidPlan, v13.6, Varian Medical Systems), and a series of objectives were created using this KBP model (Additional file 1: Supplementary Table 1). It was expected the majority of plans in the present trial would be treated with a Volumetric Modulated Arc Therapy (VMAT) technique, therefore all patients in this cohort were optimised with a class solution VMAT plan created with two ipsilateral arcs spanning 210° from the posterior-anterior direction around to 30° past midline. The plans were optimised, and refinements were made to objectives to meet the FASTRACK II protocol requirements. In particular, the priority of maximum dose objectives to organs at risk were adjusted to ensure the plans met constraints, and further reduction of mid-range OAR doses was achieved through use of the maximum generalised Equivalent Uniform Dose (gEUD) objective. A new KBP model was created with the final plans. Due insufficient numbers to create two separate models, we created one model for both fractionations. For most OAR constraints, the two fractionations have equivalent OAR constraints as a percentage of the prescription dose. Where this was not the case, we set the objectives to meet the lower constraint of the two fractionations. Validation is described in the Additional file 1: supplementary material.

Real-time review was performed for all patients treated on the trial, and involved review of the CT imaging used for treatment planning, including relevant 4DCT phases, diagnostic imaging used for contouring, structure sets, treatment plans and dose distributions. The plan was submitted ideally seven days prior to treatment start date. The contouring and dose distributions were reviewed by an independent radiation oncologist, and the treatment plan, dose distributions were reviewed by a medical physicist. The review ensured adherence to protocol for contouring, treatment planning parameters, motion management and target and OAR dosimetry. Feedback was provided to the site via a PDF case report with itemised violation criteria, and the site had the opportunity to correct any violations prior to treatment delivery. Subsequent revisions underwent the same review as an initial treatment plan.

KBP review was performed centrally by TROG Cancer Research RTQA Services. In the Eclipse treatment planning system (v13.6, Varian Medical Systems), a generic linac model was created using representative beam data and machine characteristics for each linac type expected to be used in the trial. Only C-arm linear accelerators were used in this trial, and included Varian TrueBeam, Varian TrueBeam STx, Elekta VersaHD and Elekta Synergy linacs. For each patient, a plan was created in Eclipse using the same beam angles and a generic linac model matching that submitted by the centre. The KBP model was applied and the plan was run through one full optimisation without user interaction. An application programing interface (API) script then generated a PDF report which was sent to the site within 24 h of receiving the plan.

For each of the OAR constraints, the KBP generated plan was compared with the submitted plan; the submitted plan value and the KBP plan values were presented, along with the minor and major constraints. Possible improvements were flagged in the report if the submitted plan violated the trial constraint and the KBP plan met the constraint, or if the KBP plan OAR dose was less than 90% of the submitted plan dose. The individual centre then had the opportunity to revise the plan, or continue with the submitted plan, based on their clinical judgement. No plan alterations were mandated. The centre was then invited to provide feedback on the process, using the form presented as Additional file 2: Supplementary Table 2.

For all patients treated in this phase of the FASTRACK II trial, the output of the API script was collected which included the target and OAR metrics for the submitted and KBP plans. Comparison was performed per patient and per metric based on classifications in the report as described above.

The PTV near-minimum and near-maximum doses were similar between the submitted and KBP plans. In the 26 Gy cohort, the PTV_2600 D99% was at or above prescription dose for 13/15 submitted plans, and 11/15 KBP plans, but the D95% was at or above prescription dose for all submitted and KBP plans. In the 42 Gy cohort, the PTV_4200 D99% was at or above prescription dose for 10/16 submitted plans, and 3/16 KBP plans, but the D95% was at or above prescription dose for 15/16 submitted and KBP plans. In one patient the large bowel was adjacent to the target and the PTV coverage was compromised to ensure large bowel constraints were met.

Figure 2 classifies the KBP result for each OAR constraint. Figure 3 shows the near-maximum doses for the small bowel, large bowel and spinal cord for both fractionation regimes, and Fig. 4 shows the mid-range DVH metrics for the stomach and small bowel. The near-maximum doses and the DVH constraints were however well below constraints for all plans. There was a relatively even split between KBP being within 10%, or either KBP or the submitted being superior by more than 10%. Improvement was suggested most often for the large bowel D1.5 cc, and Kidney_I V50%. In the three fraction schedule, the near-maximum and mid-range DVH metrics were close to constraint for many patients (Figs. 3, 4). As a result, there were five instances where the submitted plan met the constraint but the KBP plan did not; three were near-maximum doses (small bowel, large bowel and spinal cord) and two were small bowel D30cc. For the small bowel D30cc, there were three submitted plans that did not meet constraints, but the KBP plan did (Fig. 4).

The small bowel D30cc violation was detected by the reviewer for Patient #51 (see Additional files 4, 5: supplementary reports), with the KBP suggesting this constraint could be met. Figure 5a shows the initial plan, with intermediate isodose lines extending across the bowel structures in the submitted plan. The replan (Fig. 5b) shows significant improvement in the intermediate dose spillage, and all constraints met. Patient #54 showed improvement was possible such that the small bowel D30cc constraint was met, but the centre elected to not replan. In this case, objective functions for the small bowel were only used for the near-maximum doses, not the intermediate doses.

There were three plans that were resubmitted based on dosimetric review. For #46, the reviewers recommended reducing target coverage to keep the prescription isodose line off the large bowel. The KBP showed minor improvement, but due to its proximity to the PTV, minimal gains could be made. For patient #60 the reviewers indicated there was a high risk of bowel constraints being exceeded if bowel loops were in a different position at time of treatment. The KBP plan had a similar dose distribution to the submitted plan. In #71, small and large bowel doses were within constraints, but the KBP was more than 10% less than the submitted plan. The site elected to replan this case, in the process reducing small and large bowel dose (Fig. 5c, d), with the aim of improving robustness to bowel position variation at time of treatment.

Feedback was provided by the treating centre for five of 30 participants, and is provided in Additional file 2: Supplementary Table 2. One centre (two patients) indicated the plans could be improved based on the KBP report, however due to time constraints they did not replan. Another centre (two patients) indicated they preferred to retain their chosen clinical trade-offs prioritising one OAR over another. Despite no changes being made in the subset of plans for which feedback was provided, in each case the centre indicated that KBP based QA was a valuable tool for plan quality improvement in clinical trials. Specifically, the participating centres who provided feedback thought analysis of plan quality was useful for their own practice, and for the overall quality of the plans in the trial.