Feasibility and potential benefits of daily online adaptive radiotherapy for head and neck cancer without margin reduction using an RTT-only workflow

Abstract

Introduction

Online adaptive radiotherapy (oART) may improve head and neck cancer (HNC) treatment, where anatomical changes during radiotherapy (RT) are common. However, oART is resource-intensive and clinical data are limited. We conducted a prospective feasibility trial implementating an RT technologist (RTT)-led workflow to evaluate dosimetric outcomes, workflow strategies, and intrafractional motion in daily oART.

Material and methods

Ten patients receiving curatively intended oART (33–34 fractions) using standard PTV margins were included (NCT: 07506902). For each fraction, a scheduled plan (original plan recalculated on daily anatomy) and an adapted plan (reoptimized to daily anatomy) were generated, and the preferred plan was delivered. Target coverage (PTV1 V95%), mean dose (D_mean) and maximum dose (D_max) to selected organs at risk (OAR), treatment time, and baseline target volume were extracted for all fractions. The effect of oART (scheduled- versus delivered plan) was analyzed using linear mixed-effects models. Intrafractional motion was assessed weekly.

Results

A total of 322 adaptive session were completed. The median treatment time was 29.9 min. The adapted plan was selected in 78% of fractions. oART improved target coverage in 76% of fractions. D_mean decreased for the oral cavity (−2.4 Gy, p < 0.001) but increased for the middle PCM (2.5 Gy, p < 0.001). D_max increased for the brainstem (2.9 Gy, p < 0.001) and spinal cord (2.2 Gy, p < 0.001). Intrafractional motion was within PTV margins, with few patient-specific exceedances.

Conclusion

Daily oART for HNC is feasible with improvements in target coverage but modest and inconsistent OAR sparing.

Highlights

• •Daily adaptive radiotherapy is feasible in a radiotherapy technologist-led workflow.
• •A decision support tool enables radiotherapy technologist-led adaptive workflows.
• •Target coverage improves with daily adaptive radiotherapy in head and neck cancer.
• •Dose reductions to organs at risk are modest and inconsistent.
• •Intrafraction motion largely remains within a 3 mm PTV margins.

1. Introduction

Head and neck cancer (HNC) patients frequently experience anatomical changes during radiotherapy (RT) [1]. Adaptive radiotherapy (ART) is a strategy that allows continuous treatment plan modification. It can broadly be categorized as offline ART or online ART (oART), the latter performed with the patient on the couch while adapting. Unlike a static, one-plan-fits-all-fractions strategy, ART introduces a dynamic approach that improves dose accuracy, with potential to improve target coverage and reduce dose to surrounding organs at risk (OAR).

Most published HNC data describe retrospective ART strategies and in silico analyses, while prospective clinical experiences are limited [2], [3], [4], [5]. In a recent randomized study, weekly ART without planning treatment volume (PTV) margin reduction did not reduce toxicity compared to standard treatment [4]. However, oART enables day-to-day adjustment and may increase the clinical benefit of ART through margin reductions and potential dose escalation [6]. Several smaller prospective studies have demonstrated technical feasibility of oART, with dosimetric improvements in target coverage OAR sparing [2], [3], [5]. Although promising, clinical studies supporting oART in HNC are limited and data are sparse [7]. A recently published randomized phase II study using daily oART with reduced margins found significantly decreased OAR doses and improved acute toxicity profile, but failed to meet its primary endpoint, of reduced patient-reported xerostomia [5]. Importantly, the study also highlighted the substantial resource use associated with oART, as each adaptive fraction required radiation technologists (RTT), physician and physicist presence, challenging cost-effective implementation [8]. Furthermore, reduced margins may raise concerns about the intrafractional movement, since prolonged treatment times in oART often approaching 30–40 min.

Based on initial experience with palliative oART in HNC [9], we initiated a prospective trial in a curative setting to evaluate the feasibility of daily oART for HNC in an RTT-only workflow, with analysis of dosimetric outcomes and intrafractional motion.

2. Material and methods

2.1. Patient cohort

The Daily Adaptive RadioTherapy of Head-Neck cancer (DART-HN) (NCT: 07506902) trial is a prospective feasibility study exploring daily oART. The study is formally ongoing but currently paused for enrollment. Patients eligible for inclusion were adults (> 18 years) with histological proven stage I-IV squamous cell carcinoma of the pharynx (excluding nasopharynx) or larynx (excluding: glottic larynx stage I/II), planned for primary curative intended RT and with ECOG/WHO performance status 0–2. Patient were excluded if they had distant metastases, previous curatively intended surgery for the primary cancer, previous RT of the head and neck or were enrolled in conflicting protocols.

2.2. Radiotherapy planning

Patients treated with curative intent underwent RT-specific PET-CT and MRI acquired in treatment position immobilized with a five-point mask covering head and shoulders. All included patients were planned for 66–68 Gy in 33–34 fractions, six fractions/week. Target volumes were defined according to current DAHANCA guidelines [10]. The gross tumor volume (GTV) comprised all macroscopic disease identified on imaging and clinical examination. The high-risk clinical target volume (CTV1) was a 5 mm expansion of the GTV to account for microscopic spread. CTV2 defined a 10 mm expansion of the GTV representing regions at intermediate risk of spread. The elective clinical target volume (CTV3) included CTV2 and nodal regions at low risk of involvement. All patients were treated using intensity-modulated radiotherapy (IMRT) with 7–16 fields and standard PTV margins of 3 mm (transverse) and 4 mm (cranio-caudal). The reference plan was always based on the pCT and was not updated using subsequent imaging during the treatment course.

All fractions were adaptive sessions (Fig. 1). The oART workflow was delivered on the Ethos platform (Varian Medical System) with Hypersight imager. The first five patients were treated on version 1.0 and the subsequent five patients on version 2.0 which use a different optimization algorithm and image handling [11]. The difference in workflows is described below.

Fig. 1.

Workflow of the online adaptive session on Ethos: After patient positioning, a CBCT1 was acquired. Target at OARs were deformed or segmented (automatic in Ethos 2.0 or via synCT on Ethos 1.0) and contours were edited if needed. Two plans were compared: Scheduled (recalculation of original plan) and adapted (re-optimization) and preferred plan was selected for treatment. A final CBCT2 verified patient positioning before treatment delivery. Abbreviations: CBCT, cone beam computed tomography; OAR, organs at risk; synCT, synthetic CT.

Each adaptive session consisted of 5 steps.

• 1)A setup CBCT (CBCT1) was acquired. To minimize intrafractional motion, patients were instructed not to swallow during CBCT acquisition. In Ethos 1.0 the planning computed tomography scan (pCT) was deformably registered to the CBCT1 generating a synthetic CT (synCT) used for dose calculation. The deformation was guided by “influencer structures” (brainstem, spinal cord, mandible, parotid glands), which were reviewed and edited if needed prior to the deformation. In Ethos 2.0, dose calculation was performed directly on the CBCT1 and generation of a synCT was no longer required [12].
• 2)Target structures were deformably registered from the pCT to CBCT1, and OARs were auto-segmented with the built-in AI model. Target volumes and critical OARs were reviewed and edited if necessary. Target editing was restricted to predefined minor corrections of the deformed target volumes (removal of air, Fig. 2 step 3).
• 3)Two plans were compared: A scheduled plan representing a recalculation of the original plan on the daily CBCT1 showing what would have been delivered in standard non-adaptive RT. An adapted plan generated by reoptimizing the original plan to the daily CBCT1 using the same beam arrangements and objectives. After review of both plans, the most favorable was selected for treatment delivery (adaptive or scheduled), prioritizing target coverage followed by doses to OARs.
• 4)A second CBCT (CBCT2) was acquired for patient positioning verification. If substantial intrafractional motion was detected, couch shifts were applied, or the adaptive session was restarted.
• 5)If acceptable positioning, the selected plan was delivered for treatment.

Fig. 2.

Guide for an RTT-only adaptive workflow including a decision support for plan choice (scheduled versus adaptive).

2.3. RTT-only workflow

The experiences from the first five patients treated formed the development of an RTT-only workflow. Through iterative feedback from the RTTs, an adaptive guide with a decision-support tool for plan selection was developed (Fig. 2). The RTT-only workflow was fully implemented after treatment of the fifth patient and applied on the subsequent patients. The workflow involved physician presence during the first six fractions and physicist presence during the first 12 fractions. Afterwards, both physician and physicist transitioned to weekly offline reviews of the deformed CTVs and RTT-edited contours, allowing for continuous feedback to the RTTs and, if necessary, withdrawal of patients from the protocol in cases of unexpected findings raising safety concerns. To evaluate RTT performance in plan selection, the RTT-only fractions for each patient were divided into six consecutive treatment periods and last fraction in each period was selected for evaluation. Each fraction was independently reviewed offline by three physicians, blinded to the plan selected for treatment. For each fraction, the physicians indicated preferred plan (scheduled or adapted). Fractions were then classified as consensus (full agreement in plan choice) or as non-consensus (disagreement in plan choice). In consensus cases, the unanimously selected plan served as reference. In non-consensus cases, the majority choice served as reference. Agreement between RTT-selected plan and the reference was quantified as the percentage concordance.

2.4. Data analysis

For each fraction, PTV1, PTV2 and PTV3 target coverage, (as percentage of PTV1 receiving ≥95% of planned dose (V_95%)), hotspots within the PTV1 (D_1.8cm³), PTV2 (PTV1 excluded) and PTV3 (PTV2 excluded) (as percentage receiving ≥107% of planned dose (V_107%)), absolute mean doses (D_mean) and maximum doses (D_max) to selected OARs were extracted for both the scheduled plan and the adapted plan. The plan selected for treatment delivery (scheduled or adapted), session duration and baseline target volume were also extracted. The dosimetric effect of oART was analyzed as scheduled plan versus the delivered plan, delivered plan being either the scheduled or adapted.

Dose was analyzed per fraction; to account for repeated measures among the same patient, a linear mixed effects model was applied. The model included plan type (scheduled versus delivered), reference dose, Ethos edition (Ethos 1.0 versus Ethos 2.0), and baseline target volume as fixed effects, as well as interaction terms between plan type and Ethos edition and between plan type and baseline target volume. The model was adjusted for the random effect of patient number.

Intrafractional motion between CBCT1 and CBCT2 in all planes was assessed weekly (fraction 6, 12, 18, 24, 30) throughout the treatment course using predefined anatomical landmarks: 1) The midpoint of the superior border of the hyoid body, 2) The posterior aspect of the tongue base at the level of the epiglottic tip, 3) The anterior surface of the vertebral body at the inferior border of C2.

Data were structured in PSPP (ver. 2.0.0) and analyses performed in R (ver. 4.5.0). P-values below 0.05 were considered significant.

3. Results

3.1. Study cohort characteristics

From November 1, 2023, to October 1, 2025, ten patients were included. Patient accrual was slower than anticipated due to technical challenges, as described in the supplementary material. For patient characteristics see Table 1. The cohort was predominantly male (90%), with a median age of 59 years (range 49–80 years). All were diagnosed with oropharyngeal carcinoma (tonsil 80%, tongue base 20%), and 90% were p16-positive. Most patients had N1 disease (90%) and presented in stage I (60%) (UICC 8th). Seven out of ten patients received concurrent weekly chemotherapy.

Table 1.

Patients characteristics of the study cohort.

	Median [min, max]
Age	59 [49, 80]
	No. patient (%)
Sex
Male Female	9 (90%) 1 (10%)
Site
Oropharynx Tonsil Base of tongue	10 (100%) 8 (80%) 2 (20%)
T-Classification
T1 T2 T3 T4	1 (10%) 6 (60%) 0 (0%) 3 (30%)
N-Classification
N0 N1 N2 N3	1 (10%) 7 (70%) 1 (10%) 1 (10%)
Stage
I II III IVa	6 (60%) 0 (0%) 3 (30%) 1 (10%)
p16
Negative Positive	1 (10%) 9 (90%)
Smoking
Never Former Current	3 (30%) 5 (50%) 2 (20%)
Use of chemotherapy
No chemotherapy Chemotherapy	3 (30%) 7 (70%)

All patients were alive at time of analysis with a median follow up time of 8.5 months (IQR 5.8;13.8). One patient developed local recurrence with subsequently regional spread at 8.3 and 11 months after end of treatment, respectively. Both recurrence sites were located within the high-dose CTV1 volume. Date of last follow up was January 2026.

3.2. Plan adaptations

A total of 322 adaptive sessions were completed. One patient withdrew from protocol halfway through treatment due to difficulties coping with prolonged time on the couch. The adaptive plan was chosen over the scheduled plan in 253/322 (79%) fractions. Among the 69 scheduled plans delivered, 61 was from the Ethos 1.0 cohort due to synCT-related uncertainties (see supplementary materials), resulting in a conservative strategy favouring the scheduled plan unless clear superiority of the adaptive plan. Technical challenges with the synCTs of Ethos 1.0 resulted in acquisition of new pCT in three out of the five patients, with two, three and four replannings respectively (excessive weight loss was the predominant cause). No patients treated with Ethos 2.0 needed replanning.

Median time per fraction (CBCT1 to end of treatment delivery) was 29.9 min (IQR: 23.3;34.7) with a median adaption time (CBCT1 to CBCT2) of 23.7 min (IQR: 17.6;28.8) (Fig. 3). A trend toward longer adaptation times was observed from patient 5 onward, coinciding with the implementation of the RTT-only workflow. Adaption time was 24.6 min (IQR: 21.7;29.8) in the physician-led workflow, compared with 33.2 min (IQR: 29.9;37.4) in the RTT-only workflow. This increase was attributable to prolonged contour review and consensus-based decision-making between the two treating RTTs, while plan selection times were unchanged or slightly reduced.

Fig. 3.

Median adaptive session times, shown overall and per patient, including a breakdown of workflow components. Time intervals comprise CBCT1 acquisition with contour review and editing if required (blue), plan review and selection with CBCT2 acquisition (orange), and treatment delivery (purple). Abbreviations: CBCT, cone beam computed tomography. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

3.3. RTT workflow

Across all RTT-only fractions, none were rejected during offline review, and no retrospective corrections or interventions were required.

In RTT performance evaluation, 30 RTT-only fractions were assessed. Fifteen fractions were consensus cases, and 15 fractions were non-consensus cases. In non-consensus cases plans were clinically comparable and both suitable for treatment. An example of a non-consensus case is provided in the supplementary material. The overall agreement in plan choice between RTT-selected plan and the physician reference plan was 90% (27/30). In consensus cases concordance was 100% (15/15), while an 80% agreement (12/15) was observed in non-consensus cases. The three discordant cases reflected physicians individual weighting of OAR sparing between otherwise clinically comparable plans, whereas RTTs selected plan according to the predefined decision-support tool filled out at fraction 6 (Fig. 2 step 4).

3.4. Dosimetric outcomes

The adaptive plan improved PTV1 coverage in 76% (244/322) of fractions and reduced PTV1 target coverage 2/322 (<1%) fractions. For the latter, the adaptive plan was delivered despite inferior PTV1 coverage compared to the scheduled. Both fractions originated from the same patient: in one fraction, the delivered plan did not match the RTT-documented plan choice suggesting a human error. For the other, the adapted plan was deliberately chosen to spare OARs. PTV1 coverage on delivered versus scheduled plan is illustrated in Fig. 4.

Fig. 4.

Comparison of coverage, as percentage of the PTV1 receiving ≥95% dose, between the delivered and scheduled plan for all fractions. Each point represents a single fraction colored by patient number. Coverage on the scheduled plan is plotted on the x-axis and coverage on the delivered plan (adapted or scheduled) on the y-axis. The gray diagonal line indicates equal coverage between the two plans meaning that dots on the line represents fractions where the scheduled plan was chosen for delivery. Points above the line indicate fractions where adaption improved coverage, whereas points below the line indicate fractions where adaption reduced coverage.

The dosimetric effect of oART on target coverage and OARs, as dose difference between the scheduled and delivery plan is summarized in Table 2. oART was associated with a small but significant increase in D_max to the brainstem (0.085 Gy/fraction, p < 0.001) and spinal cord (0.063 Gy/fraction, p < 0.001), corresponding to cumulative increases of 2.9 Gy and 2.2 Gy, respectively. The cumulative D_max estimates represent a worst-case scenario, as the voxel receiving the highest dose unlikely remains spatially consistent throughout treatment.

Table 2.

Dosimetric difference between the scheduled and delivered (scheduled or adapted) plan. Differences are presented on a per fraction basis and as cumulative estimates over a full treatment course (34 fractions). *Cumulative maximum dose estimated represents worst case scenarios.

Organ at risk	Fractional dose difference scheduled versus delivered (95% CI)	Cumulative dose difference (34 fractions) scheduled versus delivered (95% CI)	p-value
Maximum dose (Gy)
Brainstem	0.085 (0.062;0.11)	2.9 (2.1;3.7)*	<0.001
Spinalcord	0.063 (0.045;0.082)	2.2 (1.5;2.8)*	<0.001
Mean dose (Gy)
Ipsilateral PG	−0.0038 (−0.029;0.022)	−0.1 (−1;0.7)	0.8
Contralateral PG	−0.011 (−0.032;0.0097)	−0.4 (−1.1;0.3)	0.3
Oral Cavity (−excl. PTV)	−0.070 (−0.088;-0.052)	−2.4 (−3;-1.8)	<0.001
Upper PCM	−0.024 (−0.048;-0.00005)	−0.8 (−1.6;-0.02)	0.05
Middle PCM	0.074 (0.043;0.11)	2.5 (1.5;3.6)	<0.001
Lower PCM	0.0060 (−0.016;0.028)	0.2 (−0.5;1)	0.06
Target coverage (pp)
PTV1 (V95%)	0.8 (0.5;1)	–	<0.001

Abbreviations: CI: confidence intervals, PG: Parotid gland, PTV: Planning Target Volume, PCM: pharyngeal constrictor muscle, pp.: percentage points

oART significantly reduced D_mean for the oral cavity (−0.070 Gy/fraction; −2.4 Gy cumulative, p < 0.001). In contrast, a significant increase in D_mean was found to the middle PCM (0.074 Gy/fraction; 2.5 Gy cumulative, p < 0.001), while no significant differences were observed for the remaining OARs.

Larger baseline target volumes were significantly associated with greater dose reductions to five of eight OARs (brainstem, spinal cord, upper PCM, middle PCM, lower PCM). The upgrade from Ethos 1.0 to Ethos 2.0 significantly reduced the effect of oART on D_max to the brainstem while it increased D_mean to the PCMs. Full mixed-model results including all coverage- and hotspot metrics are provided in the supplementary material.

3.5. Intrafractional motion

Intrafractional motion between CBCT1 and CBCT2 was assessed on a total of 47 fractions. Results are presented in Fig. 5. Largest motion was observed at the tongue base, followed by the hyoid bone. Overall, intrafractional motion remained within the standard PTV margins, although exceptions were observed. Motion exceeding the PTV margins was mainly in the anterior-posterior direction for the tongue base and lateral direction for the hyoid bone and spinal column. Exceedances were largely confined to two patients, who accounted for 14/16 transverse-motion violations (motion >3 mm) and 6/8 cranio-caudal-motion violation (motion >4 mm). Motion estimates were not corrected for compensatory couch shifts which were applied after CBCT2 in 36 out of 47 fractions. Maximum-, median and mean motion of all anatomical landmarks are provided in supplementary. Applied couch shift corrections were small, with maximum displacements of 2.1 mm left/right (median = 0 mm), 1.2 mm superior/inferior (median = 0.1 mm) and 1.0 mm cranio/caudal (median = 0).

Fig. 5.

Intrafractional motion assessed between CBCT1 and CBCT2 of the tongue base, hyoid bone, and spinal column in the transverse and cranio-caudal directions, evaluated on one weekly fraction per patient. Each dot represents a single fraction, colored by patient. PTV margins are marked in gray. Motion estimates do not include couch shifts corrections.

4. Discussion

In the present study, we demonstrated that daily oART for HNC using an RTT-only workflow is feasible. Intrafractional motion largely remained within the applied PTV margin, but its magnitude underscores the need for careful monitoring of any potential margin reductions. This challenge is amplified by already small safety margins and the anatomical complexity in HNC. Nevertheless, by implementing an RTT-only workflow resource demands can be reduced compared with current multidisciplinary presence, enabling scalable clinical implementation.

In the RTT-only workflow, physician and physicist transitioned to structured offline reviews. This transition changed their role from active participation to a condensed, quality-focused function with clearly defined decision criteria and high demands on sustained attention. No clinically relevant deviations were identified in the offline reviews, suggesting that RTT-edited contours were generally acceptable, although more systematic evaluation of contour quality would be valuable. This will be particularly important in future studies with reduced PTV margins, where contour accuracy becomes increasingly critical. However, even with the contour-editing task restricted to limited predefined edits, the process remained time-consuming. This highlights that future RTT-led workflows with more extensive contour editing will require further training and validation, as well as system optimization, as the current workflow is limited by the need to manually repeat identical corrections across all dose levels. The RTT-only approach has also been shown feasible in other sites, with reduced resource demands and maintained treatment quality [13], [14].

With resource constraints addressed, the benefits of daily oART in a real-world clinical setting were evaluated. To our knowledge, only two studies have investigated daily CT-based photon oART in HNC, both demonstrating improved target coverage and reductions in target D_max [2], [5]. In contrast we found a D_max increase to the spinal cord and brainstem, but smaller interfractional D_max variation in oART. This indicates that reoptimization alone improves dose homogeneity. However, the full potential of oART is suggested to lie in second-order effect of margin reductions [15], [16], [17], [18]. This was investigated by Sher et al. [5] in a cohort of 24 patients treated with daily oART and PTV margin reduced to 1 mm. This margin reduction amplified the dosimetric benefits, resulting in more extensive OAR sparing compared with standard-margin oART. This was particularly evident for the salivary glands, with dose reductions to the ipsilateral parotid and both submandibular glands (−4.5 to −14.1 Gy) [5], whereas significant reductions in standard-margin oART was confined to one submandibular gland (−1.4 Gy) [2]. Without margin reduction, salivary gland dose sparing is limited, and significant dose differences were not expected given the small sample size of our study.

Sher et al. [5] also found further dose reductions with smaller PTV margins to the superior and middle PCMs (−6.1 Gy). In contrast, our results suggest that oART without margin reduction redistributes rather than reduces PCM doses. However, PCMs are challenging structures to evaluate, as their small volumes render dosimetric results highly sensitive to variability in autocontouring. Our study also found reduced dose to the oral cavity with oART, partly consistent with Blumenfeld et al. [2], who reported statistically significant reductions emerging over the treatment course. In contrast, Sher et al. [5] observed no dose reduction to the oral cavity.

While sparing of OARs with daily oART appears limited in the absence of margin reduction, improvements in target coverage may still justify its clinical relevance in select cases. Observational studies have shown that actual delivered dose to target volumes are prone to dose inhomogeneities, resulting in unintended “cold spots” and worse local control [19], [20], [21]. This is highly important in HNC, as analyses from our center have demonstrated that most recurrences occur within the target volume [22], [23]. By eliminating “cold spots”, daily oART may improve tumor control, suggesting that even modest improvements in coverage could translate into meaningful clinical benefit.

Despite these potential clinical benefits, sustainable delivery with modest resource demands is a challenging factor [24], [25]. Consequently, offline ART strategies have been used as a pragmatic alternative. In the ARTIX study, weekly offline ART, even without margin reduction, was associated with significant improved salivary function (8%) relative to IMRT although not reflected in patient-reported outcomes [4]. However, this is maybe not surprising in a study powered to detect a 25% difference, despite previous estimates suggesting a likely difference of only 8–11% [26].

The key question remains whether dosimetric improvements translate into meaningful patient benefit. To that end, Sher et al. [5] found no improvement in patient reported outcomes despite daily oART with reduced margins. It has therefore been suggested that current adaptive strategies may be approaching a plateau in the dose-response relationship, where further optimization alone has limited clinical impact [27]. Consequently, future studies should focus on efficient workflows while maintaining an emphasis on margin reduction.

In HNC, margins are associated with concerns regarding intrafractional motion, which may offset potential gains if unaccounted for. Largely, intrafractional motion remained within the 3 mm PTV. Only a small subset of patient-specific fractions exceeded the PTV, most pronounced at the tongue base. However, the tongue is highly mobile with intrafractional deviations approaching ±5 mm [28]. Additionally, motion increases near the swallowing structures, greatest in caudal tumors (hypopharynx and larynx), and minimal to none in nasopharyngeal tumors [29]. These findings should be considered when reducing margins. In a subsequent analysis, Sher et al. [30] retrospectively evaluated intrafractional motion between CBCT1 and a near-end-of-session CBCT acquired weekly. They reported that intrafraction target motion remained within the 1 mm PTV margin, supporting the safety of their approach. While this target-based strategy is clinically relevant, it relies on CBCT-based contouring with limited soft-tissue contrast. In contrast, we relied on clearly identifiable anatomical landmarks, providing a more robust motion estimates, although not target specific. Thus, the two approaches reflect a trade-off between anatomical visibility and direct target relevance [5]. It should be noted that our motion assessment reflects movement during the adaptive workflow (CBCT1 to CBCT2) rather than during beam delivery. Therefore it may represent a conservative estimate of motion affecting dose delivery, given the shorter treatment delivery interval after CBCT2 than the interval between CBCT1 and CBCT2. However, individualized margins may represent a logical next step. Anatomical position could be considered, with for example, larger margins around mobile structures such as the tongue, either upfront or after early fraction assessment. In addition, margins may also be adaptor-specific, reflecting variability in experience with daily recontouring.

An alternative future strategy for oART in HNC may be ad hoc or on-demand adaptation, with adaptive sessions performed at predefined intervals or when clinically relevant anatomical changes are observed. Such approach could potentially reduce resource requirements while retaining the benefits of oART. However, on-demand adaptation is currently not supported within the system, and the resource demands associated with available workarounds are substantial. Furthermore, uncertainties remain regarding robustness of the adaptive plan for consecutive use. Consequently, patient accrual in the DART-HN trial is paused, pending a strategy for future continuation.

In conclusion, oART is feasible, and an RTT-only workflow can reduce resource requirements. Daily oART provides more conformal and homogeneous daily dose distributions, with significant reductions in doses to selected OARs. Further reductions may be achieved through margin reductions, although intrafractional motion must be carefully considered.

Declaration of generative AI and AI-assisted technologies in the manuscript preparation process

During the preparation of this work the author(s) used ChatGPT (OpenAI) in order to improve grammar and perform minor text editing. After using this tool/service, the author(s) reviewed and edited the content as needed and take(s) full responsibility for the content of the published article.

Ethics declaration

Written informed consent to take part in the study and to publish the article has been obtained from all participants or their legal representatives. The privacy rights of participants have been observed.

This study was performed in compliance with relevant laws, regulatory frameworks and guidelines where the research took place.

This study was approved by the Danish Research Ethics Committees (De Videnskabsetiske Medicinske Komitéer, VMK).

(Approval No. #2202672)

The results of this clinical trial and any associated work have been posted in a registry. This clinical trial was registered with number NCT07506902 (ClinicalTrials.gov).

Declaration of competing interest

The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: The study has received financial support from Siemens Healthineers, through its Business Area Varian. The funding source had no role in the study design, data collection, data analysis, interpretation of results, or manuscript preparation.

Appendix A. Supplementary data

Supplementary material 1

Supplementary material 2

Supplementary material 3

Supplementary material 4