End of treatment cone-beam computed tomography (CBCT) is predictive of radiation response and overall survival in oropharyngeal squamous cell carcinoma

We identified 64 patients with squamous cell carcinoma of the oropharynx diagnosed between 2016 and 2018 with available follow up and treatment outcomes who received curative-intent radiation therapy at the University of California, San Diego. Patients must have completed definitive therapy with radiation with or without chemotherapy. Patients were excluded if they received post-operative radiotherapy. Patients were considered post-operative if they had received any of the following: subtotal resection, gross total resection, or neck dissection (ipsilateral or bilateral). Additionally, patients were excluded if the course of radiotherapy was prolonged by greater than one week.

All patients received external beam radiation therapy with intensity modulated radiation therapy (IMRT) to a minimum dose of 60 Gray (Gy) (range 60–73 Gy, median 70 Gy, standard deviation (SD) 2.56 Gy) using 2.0–2.12 Gy/fraction. The median number of radiation treatments was 35 (range 30–35) over an average of 48 days. The average patient age was 61 at the time of diagnosis. T-Classification and N-classification were collected according to the American Joint Committee on Cancer (AJCC) classification 8th edition. Fifty-seven patients received concurrent systemic therapy, while 7 patients received radiation alone. Of the patients who received systemic therapy, 51 (89.5%) received cisplatin-based chemotherapy. A total of seven patients received cisplatin combined with pembrolizumab. The remaining six patients receiving either cetuximab, carboplatin/paclitaxel or pembrolizumab.

The largest nodal conglomerate (LNC) was evaluated for measurement due to distinct fat planes and tissue heterogeneity in the cervical chain. Three timepoints were selected for measurements: (1) CT Simulation Scan (CT Sim), (2) End of Treatment (EOT) CBCT and (3) Post-Treatment Follow-up Imaging (Post-RT). The largest nodal conglomerate was identified on the CT Sim. Measurements were then collected of the LNC including anteroposterior (AP), mediolateral (ML) and craniocaudal (CC) dimensions. The AP and ML measurements were obtained from multiple axial slices of the LNC. The axial slice that resulted in the largest product of AP and ML measurements was recorded as the 2D maximum (2D). These measurements were then collected from the EOT CBCT and Post-RT with matching of bony landmarks and soft tissue structures to ensure consistency among measurements. All measurements were collected by a single operator to ensure consistency. An example of the measurement technique is shown in Fig. 1.

The EOT CBCT measurements were reported as a percent change in size relative to CT Sim. The Post-RT measurements were reported as response based on RECIST 1.1 criteria.

Statistical analyses were performed using SPSS V23.0 (SPSS Inc., Chicago, IL) and R, version 3.5.1(R; Vienna, Austria). Median follow up was calculated using the reverse Kaplan–Meier method (KM) [19]. Correlation of percent change in size seen on EOT CBCT to Post-RT response was evaluated utilizing Pearson correlation. The Post-RT response was dichotomized to complete response (CR, ‘Responders’) or less than CR (‘Non-responders’), to include partial response (PR), stable disease (SD) or progressive disease (PD) thus creating a point biserial correlation coefficient. Logistic regression was utilized to derive the relationship between the percent change in size on EOT CBCT and response on Post-RT imaging. OS was examined using the KM method. Univariate survival analysis (UVA) was performed with the log-rank test and unadjusted Cox proportional hazards models to estimate hazard ratios (HR), with HR > 1 corresponding to worse OS and progression free survival (PFS).

On post-therapy imaging including PET/CT or CT Soft Tissue Neck, a total of 40 patients had experienced a CR, while 19 patients experienced PR and 5 patients had SD. There were no statistically significant differences in any patient characteristics between Complete responders and Non-Complete responders. Of the patients who experienced a CR, 75% had stage I-II disease as compared to 50% in non-responders (p = 0.19). Approximately 90% of both responders and non-responders were p16 positive. Representative examples of CR, PR and SD are shown in Fig. 1, and patient characteristics according to treatment response are shown in Table 2.

The median percent change in CC and 2D measurements of the LNC from CT Sim to EOT CBCT was 28.8% and 56.9% respectively in Complete responders compared to 7.7% and 38.2%, respectively, in Non-Complete responders (Table 3). The percent change in both the 2D and CC measurements from CT Sim to EOT correlated to response on Post-RT imaging with r = 0.313, p = 0.02 and r = 0.318, p = 0.02, respectively.

Logistics regression of the percent change in CC measurement from CT Sim to EOT CBCT based on treatment response on post-therapy imaging yielded an OR 1.05 (95% CI, 1.01–1.11) reflecting a 5% increase in CR for every 1% change from CT sim to EOT CBCT (Fig. 2). For patients who experienced \(\ge\) 40% (n = 8) and \(\ge\) 30% (n = 17) reduction in CC measurement from CT Sim to EOT CBCT, the rate of CR was 87% and 82% respectively. Conversely, among patients who experienced < 10% (n = 11) reduction CC measurement from CT Sim to EOT CBCT, the rate of CR was 45%. Additionally, a \(\ge\) 30% reduction in CC measurement from CT Sim to EOT CBCT was significantly associated with CR on post-therapy imaging (r = 0.274, p = 0.039) which supports our hypothesis that EOT CBCT may have utility as an early indicator of treatment response.

Out of 64 patients a total of five patients experienced local failure with two of these patients additionally developing distant metastases in the lung and chest wall, respectively. A single patient developed isolated liver metastases. Two-year OS was 100% in patients who had a \(\ge\) 30% reduction in the CC measurement compared to 78% for those with \(<\) 30% reduction ( Binary Chi-Square HR 4.85, p = 0.028; continuous HR 1.07, 95% CI 1.02–1.15, p = 0.045) (Fig. 3A). Two-year PFS in these cohorts with \(\ge\) 30% reduction versus \(<\) 30% reduction was 100% and 86%, respectively (HR 1.06, 95% CI 0.99–1.14; p = 0.11). Analysis of PFS and OS based on 2D measurements of the LNC from CT Sim to EOT CBCT did not reach statistical significance (HR 1.03, 95% CI 0.996–1.07 and HR 1.07, 95% CI 0.99–1.05, p = 0.28, respectively). In order to compare the prognostic capability of a ≥ 30% reduction in the EOT CBCT metric to the standard 3 month post-treatment diagnostic scan, we analyzed overall survival and progression free survival in patients that manifested a CR versus non-CR on diagnostic imaging. Two-year OS in patients that had a CR versus non-CR on post-treatment diagnostic imaging was 97% and 67%, respectively (Fig. 3B). Two-year PFS in patients that had a CR versus non-CR on post-treatment diagnostic imaging was 100% and 56%, respectively. Kaplan–Meier curves for PFS are shown in Additional file 1: Figure 1. Importantly, the 2 year overall survival in patients with a ≥ 30% reduction in the EOT CBCT CC measurement was almost identical to the 2 year overall survival estimates predicted in patients that demonstrated a CR on post-therapy imaging, namely 100% and 97% respectively (Fig. 3). These findings suggest that EOT CBCT imaging data may have utility as an early prognostic indicator of treatment response.