A retrospective, deformable registration analysis of the impact of PET-CT planning on patterns of failure in stereotactic body radiation therapy for recurrent head and neck cancer

A retrospective cohort study was performed drawing upon the largest published series from the UPCI experience in treating rSCCHN with SBRT. All patients signed a written informed consent for treatment and the study was approved by the institutional review board. Data were de-identified to meet the Health Insurance Portability and Accountability guidelines. All patients had previously-irradiated, recurrent SCCHN and were originally treated with surgery, chemotherapy, and/or external beam radiation for their primary malignancies. Patients were 18 years of age or older, had a Karnofsky performance status (KPS) of 50 or greater, and received no chemotherapy or radiation for at least one month before entry into the study. Patients were excluded if SBRT was part of treatment for a primary tumor or if they had not received prior radiation therapy.

Between June 2005 and November 2009, 111 patients with recurrent, previously-irradiated SCCHN meeting these criteria received either Cyberknife^TM or Trilogy^TM SBRT, described in our prior reports [15, 16]. Fifteen patients had no post-treatment PET-CT scans and were excluded from the analysis. Of 96 final patients, 40 (42%) had distant metastases or untreated locoregional disease and were treated palliatively. Fifty-six patients (58%) had no other tumors and were treated with curative salvage intent. Seven patients (7%) were treated with adjuvant SBRT after surgical resection of rSCCHN and 89 patients (93%) were treated definitively with SBRT. All primary tumors were confirmed to be squamous cell carcinoma on pathology reports. Biopsies of recurrent lesions were not routinely performed if there was sufficient clinical suspicion of recurrence of the original tumor based on imaging, history, or physical examination.

Treatment plans for patients in this study were created using an integrated PET-CT scan, a PET-aided CT (PET-CT from a different date fused to a CT with rigid registration), or a CT-only scan. The planning CT was acquired with 1.25–mm thick slices in the target area. Physicians contoured the spinal cord and other critical structures on the planning CT. The GTV was contoured by visual inspection and treated without a margin (GTV = PTV).

Treatment plans were developed based on tumor geometry and nearby critical structures. All patients received five fractions and most (89%) received 40–50 Gy. Dose limits for normal structures were selected based on prior irradiation and the radiation oncologist’s clinical judgment. Dose limits were: 8 Gy (1 fraction) for spinal cord, 9 Gy for brainstem, 20 Gy for brain, 10 Gy for retina, optic nerves, and optic chiasm, 6 Gy for lens, 20 Gy for carotid artery, and 20 Gy for esophagus and larynx. Dose-volume histograms for at-risk structures were used to verify these limits. Phantom dose measurements were used for quality assurance. Skull or cervical spine tracking was used to localize lesions with a 1–mm spatial accuracy [17].

Before treatment, patients were placed on the treatment table with an immobilization device then fitted with a personalized thermoplastic facemask secured to the headrest. Near-real-time digital x-rays or cone-beam CT images were obtained during treatment to establish and verify target locations. Treatments were administered every other day. All patients were seen one to three months after treatment and received a PET-CT, CT, or MRI as part of their follow-up.

Patterns of failure were analyzed using VelocityAI^TM, which uses a modified B-spline deformable registration algorithm. Testing of this algorithm demonstrated a mean error of 1–2 mm for noise-free images [18, 19]. Treatment response was assessed using the RECIST (Response Evaluation Criteria for Solid Tumors) criteria [20]. Failure was defined as initial disease progression or complete response/partial response/stable disease followed later by progressive disease. Development of regional/distant disease (defined below) was also considered failure if the patient had no preexisting metastases/untreated locoregional disease.

Pre-treatment planning CT scans were deformably registered to each subsequent post-treatment scan. The post-treatment scan acted as the template image and the planning CT with its associated PTV and dose map was deformed to this image. Recurrent tumors were contoured on post-treatment scans. Mean dose received by recurrent tumors was recorded using the deformed dose map. The contoured recurrence was compared to the deformed PTV and a category of failure was assigned based on fraction of recurrent tumor falling within the PTV and distance from the closest edge of recurrence to the PTV.

Previous studies on patterns of failure after Intensity-Modulated Radiation Therapy (IMRT) for primary head and neck cancer have defined in-field recurrences as 95% involvement within clinical target volume (CTV), border-of-field recurrences as 20–95% involvement within CTV, and out-of-field recurrences as below 20% involvement within CTV [21, 22]. The current study differs in that we treated recurrent rather than primary head and neck cancer. There are no standard definitions for categories of failure after re-irradiation of recurrent head and neck cancer. Because re-irradiation involves smaller target volumes and minimal margins (no CTV, GTV = PTV), we used a less stringent definition of in-field recurrence. Few local recurrences would otherwise meet a >95% in-field involvement requirement by the time they were detected. We divided the recurrent tumors into four categories of failure: In-field–75% or more of recurrence within PTV, Overlap–20–75% of recurrence within PTV, Marginal–20% or less of recurrence within PTV but closest edge within 1 cm of PTV, Regional/Distant–1 cm or more from PTV. Examples of different patterns of failure are shown in Figure 1.

Endpoints were time to SBRT failure and overall survival. Failure analysis endpoints included time to overall (any) SBRT failure, time to combined In-field/Overlap/Marginal failure (local control), and time to combined Overlap/Marginal failure (near misses). Endpoints were evaluated in both the overall and definitive SBRT cohorts. Multivariate and univariate Cox regressions were used to model predictors of failure and survival. Variables considered in the failure analysis were use of PET-CT planning, tumor volume, use of cetuximab, and time from prior radiation to failure. Variables in the survival analysis included these as well as age, In-field/Overlap/Marginal failure, prior surgery, KPS, and distant metastases/untreated locoregional disease. Variables were entered into the final multivariate model if they reached a significance of p = 0.10. Kaplan-Meier curves were used to estimate time to failure and log-rank tests were used to compare differences between PET-CT planned and non-PET-CT planned patients. SPSS software package was used for statistical analyses [IBM®SPSS® Statistics software Version 19.0].

Table 1 summarizes patient, tumor, and treatment characteristics. Ninety-six patients with rSCCHN meeting criteria were treated with Cyberknife™ or Trilogy™ SBRT between June 2005 and November 2009. In this cohort, 83% of patients originally presented with stage III–IV primary tumors, with 60% receiving prior chemotherapy, 79% receiving prior surgery, and 48% receiving both. All patients received prior definitive radiation therapy and many received subsequent re-irradiation for pre-SBRT recurrences. Median prior radiation dose was 70 Gy (range, 36–139 Gy). Median time between prior radiation and recurrence was 10 months (range, 1.4–188 months). At the time of SBRT, 40 patients (42%) had other untreated locoregional or distant tumors and were treated palliatively while 56 patients (58%) had no other sites of disease and were treated with curative intent. SBRT was used for positive margins after surgical resection of rSCCHN in seven patients (7%) whereas 89 patients (93%) were treated definitively with SBRT.

Forty-five patients (47%) were planned using PET-CT whereas 51 patients (53%) were planned using non-PET-CT (CT-only: 33 patients, PET-aided CT: 19 patients). Median tumor volume was 26.3 cc (range, 1–205 cc). Median radiation dose was 44 Gy (range, 25–50 Gy) in five fractions and did not differ significantly between the palliatively and curatively treated patients. Forty-six patients (48%) received concurrent biologic agents or chemotherapy with SBRT, with cetuximab being the most common agent (41 patients). All patients completed treatment without toxicity-related breaks and median follow-up was 7.4 months (range, 2.6–52 months)

SBRT failure outcomes are graphically represented in Figure 2. Of 96 patients, 47 (49%) developed post-SBRT failures. The distribution of failure events was In-field–seven (12.3%), Overlap–14 (24.6%), Marginal–21 (36.8%), Regional/Distant–15 (26.3%). Eleven patients developed two types of failure. Estimated median SBRT failure-free survival was 10.2 months. Among patients that developed post-SBRT failure, median time to failure was 3.8 months. There was no significant difference in prescribed dose for different categories of failure but there was a significance difference in mean dose received within In-field, Overlap, and Marginal failures (44.6 Gy, 37.8 Gy, and 21.2 Gy, respectively, p < 0.001).

Cox regressions are shown in Tables 2, 3, and 4 for time to overall (any) failure, time to In-field/Overlap/Marginal (local) failure, and time to Overlap/Marginal (near miss) failure, respectively. On multivariate analysis of the entire cohort (n = 96), PET-CT treatment planning was associated with longer overall failure-free survival (p = 0.042) and Overlap/Marginal failure-free survival (p = 0.042) than non-PET-CT planning. Analysis of the definitive SBRT subgroup (n = 89) showed a more significant association between PET-CT planning and longer overall failure-free survival (p = 0.011) and Overlap/Marginal failure-free survival (p = 0.009). PET-CT planning was also associated with borderline improvement in In-field/Overlap/Marginal failure-free survival in the definitive SBRT subgroup (p = 0.051). Tumor volume, use of cetuximab, and interval from previous radiation to recurrence were not found to be significant predictors of failure.

PET-CT treatment planning was found on Cox regression to be a significant predictor of longer failure-free survival. There were no significant differences in age, gender, time from prior radiation to recurrence, prescribed dose, use of cetuximab, tumor differentiation, and tumor volume between the PET-CT and non-PET-CT groups (p > 0.05). A greater number of patients in the PET-CT group had distant metastases/untreated locoregional disease (p = 0.013) but more patients in the non-PET-CT group eventually went on to receive chemotherapy or biologic agents post-SBRT (p = 0.001).

Table 5 shows the frequencies of SBRT failure events and chi-square comparisons. In the overall cohort (n = 96), PET-CT (vs. non-PET-CT) planning was associated with a lower frequency of both overall (38% vs. 59%, p = 0.044) and Overlap/Marginal (24% vs. 47%, p = 0.033) failure. In the definitive SBRT subgroup (n = 89), PET-CT was more strongly associated with a lower rate of both overall (38% vs. 64%, p = 0.020) and Overlap/Marginal (24% vs. 52%, p = 0.009) failure.

On Kaplan-Meier analysis of PET-CT planning vs. non-PET-CT planning for the overall cohort, there was a significant improvement in time to both overall failure (Figure 3A, p = 0.037) and Overlap/Marginal failure (Figure 3B, p = 0.037) in the PET-CT group. There was no significant difference in time to In-field/Overlap/Marginal failure (p = 0.159). In the definitive SBRT subgroup, PET-CT planning was also associated with a significant increase in time to overall failure and time to Overlap/Marginal failure (Figures 3C, 3D, p = 0.008 and 0.009, respectively). There was a borderline significant increase in time to In-field/Overlap/Marginal failure (p = 0.046).

Median overall survival for the entire cohort was 8.2 months. Patients without (vs. patients with) distant metastases/untreated locoregional disease had a significantly longer median overall survival (11.5 months vs. 7.2 months, p = 0.003). Table 6 displays Cox regressions for overall survival. Significant predictors of decreased overall survival on multivariate analysis of the entire cohort (n = 96) were higher tumor volume (p = 0.03), In-field/Overlap/Marginal failure (p = 0.025), older age (p = 0.021), and presence of distant metastases/untreated locoregional disease (p = 0.016). Significant predictors of increased survival were prior surgery (p = 0.004) and higher KPS (p = 0.007). On multivariate analysis of the definitive SBRT cohort (n = 89), tumor volume (p = 0.024), age (p = 0.012), and presence of distant metastases/untreated locoregional disease (p = 0.037) remained significant predictors of decreased overall survival. Prior surgery (p = 0.01) and KPS (p = 0.028) remained significant predictors of increased survival. PET-CT planning, use of cetuximab, and interval from prior radiation to recurrence were not significant predictors of survival.