Robustness of tumor control (TCP) and normal tissue complication probabilities (NTCP) for intensity modulated particle radiotherapy in locally advanced lung cancer: a scenario-based uncertainty analysis compared to photon radiotherapy (IMRT) Authors: Xue Ming, Yinxiangzi Sheng, Jian Chen, Kailiang Wu, Xufei Wang, Jingfang Mao, Ningyi Ma Published: 2026 ABSTRACT BackgroundParticle radiotherapy offers significant dose sparing but is more susceptible to dose perturbations caused by physical uncertainties. This study aimed to assess the impact of dosimetric uncertainties on tumor control probability (TCP) and normal tissue complication probability (NTCP) for locally advanced lung cancer (LA-LC) among intensity-modulated proton radiotherapy (IMPT), intensity-modulated carbon-ion radiotherapy (IMCT), and intensity-modulated photon radiotherapy (IMRT).MethodsTen LA-LC patients were enrolled in this retrospective study with three radiation modalities of IMPT, IMCT and IMRT for treatment planning. The nominal IMPT/IMCT plans were recalculated according to the four major uncertainty factors respectively, including the type of calculation engine, range uncertainty, setup uncertainty and intra-fractional respiratory movement uncertainty. TCP values were assessed by the model for curative intent. Radiation-induced esophageal injury (RIEI) and radiation-induced lung injury (RILI) were studied as the NTCP endpoints. NTCP and the difference of NTCP between the particle radiotherapy (PT) and IMRT (∆NTCP) values were evaluated through the 6 Lyman-Kutcher-Burman models (4 models for RIEI and 2 for RILI).ResultsUnder the same prescription doses and similar dose distribution to the targets, deviations of the mean TCP values were less than 0.7% in IMPT, IMCT and IMRT plans under all the uncertainty scenarios. The deviations of the mean values of the ∆NTCP evaluation were −3.3–2.9% for the four models for RIEI and −0.8–1.9% for the two models for RILI under different uncertainty scenarios. The mean values of ∆NTCPL1 (the ∆NTCP values assessed through the first RILI-NTCP model) was −6.0%±6.1%/−6.9%±5.7% in the nominal IMPT/IMCT plans, while of ∆NTCPL2 (the ∆NTCP values assessed through the second RILI-NTCP model) was −12.0%±10.9%/−12.8%±10.1%.ConclusionsFor non-robustly optimized IMPT and IMCT plans, the stability of TCP and reduction of NTCP of RILI, especially for larger tumor volumes, maintained in IMPT and IMCT after considering the calculation engine, range uncertainty, setup uncertainty and intra-fractional respiratory motions respectively with the application of several planning strategies. Proper TCP, NTCP and ∆NTCP evaluation may support the individual model-based selection for PT to maximize the clinical benefit for LA-LC. Highlight box Key findings • First study to systematically evaluate the robustness of tumor control probability (TCP), normal tissue complication probability (NTCP), and the inter-modality NTCP difference (ΔNTCP) in locally advanced lung cancer (LA-LC) under realistic delivery uncertainties, moving beyond conventional dosimetric comparisons. • With motion-adaptive techniques and other planning strategies, target coverage remained highly stable in non-robustly optimized particle radiotherapy (PT) plans, exhibiting mean TCP deviations ≤0.7%. • PT preserved a clinically meaningful NTCP reduction for radiation-induced lung injury. In patients with larger target volume (IGTV >53.1 cm3), the sustained ΔNTCP advantage reached 10.1–13.5 percentage points (Model L2), providing quantitative support for individualized, model-based selection of PT. • The NTCP benefit for esophageal toxicity was more susceptible to uncertainties, particularly for late radiation-induced esophageal injury, which underscored the need for toxicity-specific robustness considerations in clinical decision-making. What is known and what is new? • Despite dosimetric advantages, previous evaluations overlooked inherent uncertainties that can degrade NTCP advantages. • For non-robustly optimized PT plans, the stability of TCP and reduction of NTCP of RILI, especially for larger tumor volumes, maintained by adapting several planning strategies after considering the calculation engine, range, setup and intra-fractional respiratory motions respectively. What is the implication, and what should change now? • For non-robustly optimized IMPT and IMCT plans, the stability of TCP and reduction of NTCP of RILI could still be achieved even considering the uncertainty factors. Proper TCP, NTCP and ∆NTCP evaluation may support the individual model-based selection for PT to maximize the clinical benefit for LA-LC. Introduction Intensity-modulated proton and carbon-ion radiotherapy (IMPT/IMCT) has presented better normal tissue sparing compared to the intensity-modulated photon radiotherapy (IMRT) in locally advanced lung cancer (LA-LC), even when using a fixed-beam delivery system (1). These dosimetric advantages could be translated into reduced toxicity risk as evaluated by normal tissue complication probability (NTCP) models (1). Clinical observations further support this benefit. Significantly lower radiation-induced toxicities have been reported for both proton and carbon-ion radiotherapy in LA-LC (2-5), especially when using pencil beam scanning technique. Despite these advantages, uncertainties related to beam range, tumor motion, and other factors can perturb dose distribution in thoracic particle therapy, and thereby limit their clinical implementation based on the Consensus Statement on Proton Therapy for Lung Cancer from the International Particle Therapy Co-operative Group (PTCOG) thoracic subcommittee (6). Additionally, accurate dose measurement in particle radiotherapy (PT) for lung cancer is challenging due to the heterogeneous composition of alveoli along the beam path. For example, tissue characteristics greatly affect the proton range estimation for in-vivo monitoring when using an anthropomorphic lung phantom (7). Similarly, an increase in beam spot size was also reported for therapeutic carbon-ion beam passing through lung-equivalent phantoms with heterogeneous density (8). To maximize the benefits of PT, a threshold of >10‑percentagepoint NTCP advantage is recommended for model-based selection of proton radiotherapy over photon radiotherapy to optimize clinical and economic benefits (9-11) in the Netherlands. However, previous evaluations often overlooked the impact of uncertainties inherent to PT delivery, which can degrade its NTCP advantages. Four primary factors contribute to this discrepancy: (I) algorithmic variations between analytical and Monte Carlo dose engines; (II) range uncertainty caused by conversion inaccuracies from computed tomography (CT) Hounsfield units (HUs) to relative linear stopping power (RLSP) (12); (III) setup uncertainties causing geometric displacements and altered beam ranges in heterogeneous tissues (13), and (IV) respiratory motion interfering with the pencil beam scanning (PBS) dynamics (14,15). While photons are also affected by setup and motion uncertainties, particle beams exhibit heightened sensitivity to tissue density variations. In pulmonary malignancies, alveolar interfaces and respiratory motion critically amplify such range perturbations. Although Hamming-Vrieze et al. (16) reported average variations of 0–2% in tumor control probability (TCP) and 0–3% in NTCP under such uncertainties in oropharyngeal cancer, while comparable analyses for LA-LC are limited. This study aims to compare TCP and NTCP for esophageal and pulmonary toxicities between IMPT/IMCT and IMRT for LA-LC. Analyses incorporated the aforementioned uncertainty scenarios by utilizing the established TCP and Lyman-Kutcher-Burman (LKB) (17,18)-based NTCP models (19-26) (shown in Table 1) in photon and proton RT. Furthermore, although treatment planning with robust optimization is well-established standard in PT, this study employed non-robustly optimized particle plans, reflecting the continued and long-term clinical practice of our institution due to the lack of integrated robust optimization of the treatment planning system (TPS), which may occur at many particle therapy centers worldwide. This approach assesses particle therapy advantages under realistic and widely applicable delivery conditions. We present this article in accordance with the STROBE reporting checklist (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-1-0266/rc). Table 1Published LKB model parameters for esophageal and lung NTCP in lung cancer radiotherapyThoracic organLiteraturesToxicity/endpointToxicity gradeRadiation modalityChemotherapyParametersModel designation in current studynmTD50EsophagusChapet et al. (19)Acute esophagitis2 or 3aCRT± (neoadjuvant)0.440.3251–Zhu et al. (20)Acute esophagitis2 or 3a3DCRT± (sequential)0.290.1546E1+ (concurrent)0.090.4236E2Wang et al. (22)Esophagitisb2 or 3cPSPT± (concurrent)0.240.5144.83E3Chen et al. (21)Late esophageal toxicityj≥3dIMRT+ (concurrent)0.030.0376.1E4LungTucker et al. (23)Radiation pneumonitise≥3d3DCRT/IMRT± (not specified)0.570.4354.1–1f0.5652.4–Tucker et al. (24)Radiation pneumonitisg≥1a3DCRT/IMRT± (not specified)1f0.7121.6–≥2a0.4726.8–≥3a0.4328.4–≥4a0.3050.2–≥1d2.47.8–≥2d0.7121.6–≥3d0.4631.5–≥4d0.1437.0–Tucker et al. (25)Radiation pneumonitish≥3d3DCRT+ (induction or concurrent)0.410.3143.2–Tucker et al. (26)Radiation pneumonitisi≥2dIMRT/PSPT+ (concurrent)0.480.2135.7L11f0.3623.4L2a, RTOG grading; b, within 6 months from the first fraction of RT; c, Common Toxicity Criteria for Adverse Events version 3.0; d, RTOG/EORTC criteria; e, for smokers; f, set n value to be 1; g, within 6 months of the treatment; h, 12 months after the start of radiotherapy; i, at 12 months after starting chemoradiation; j, severe late esophageal toxicity was recognized as the patients with stenosis, perforation or fistula (grade ≥3). +, all patients receiving; ±, in part of patients receiving; 3DCRT, 3-dimensional conformal radiotherapy; CRT, conformal radiation therapy; IMRT, intensity-modulated photon radiotherapy; LKB model, Lyman-Kutcher-Burman model; m, defining normal tissue dose-response gradient; n, characterizing dose-volume effect steepness; NTCP, normal tissue complication probability; PSPT, passive-scattering proton therapy; RTOG, the Radiation Therapy Oncology Group; RTOG/EORTC, the Radiation Therapy Oncology Group/European Organization for Research on Treatment of Cancer; TD50, the effective dose with 50% incidence of the endpoint. Published LKB model parameters for esophageal and lung NTCP in lung cancer radiotherapy a, RTOG grading; b, within 6 months from the first fraction of RT; c, Common Toxicity Criteria for Adverse Events version 3.0; d, RTOG/EORTC criteria; e, for smokers; f, set n value to be 1; g, within 6 months of the treatment; h, 12 months after the start of radiotherapy; i, at 12 months after starting chemoradiation; j, severe late esophageal toxicity was recognized as the patients with stenosis, perforation or fistula (grade ≥3). +, all patients receiving; ±, in part of patients receiving; 3DCRT, 3-dimensional conformal radiotherapy; CRT, conformal radiation therapy; IMRT, intensity-modulated photon radiotherapy; LKB model, Lyman-Kutcher-Burman model; m, defining normal tissue dose-response gradient; n, characterizing dose-volume effect steepness; NTCP, normal tissue complication probability; PSPT, passive-scattering proton therapy; RTOG, the Radiation Therapy Oncology Group; RTOG/EORTC, the Radiation Therapy Oncology Group/European Organization for Research on Treatment of Cancer; TD50, the effective dose with 50% incidence of the endpoint. Methods Patient selection and treatment planning This retrospective study used the same cohort of 10 patients with locally advanced small-cell lung cancer previously analyzed (1) for dosimetric and NTCP comparisons between IMPT/IMCT with IMRT. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments, and approved by institutional ethics committee of Shanghai Proton and Heavy Ion Center [approve number: SPHIC-TR-THLC-2026-04(RS)], with a waiver of informed consent due to the nature of the retrospective study. The same definitions of target delineation were used of internal gross target volume [IGTV, the whole respiratory trajectory of the gross tumor volume (GTV)], clinical target volume (CTV, IGTV plus a uniform 5mm-margin), and planning target volumes (PTVs, including PTVG and PTVC, which expanded IGTV and CTV uniformly by 5 mm respectively) between this study and previous publication (1,2,5). For the adequate dosimetric coverage to the targets, the region of respiratory-induced GTV motion within lung parenchyma (IGTV excluding GTV) was overridden to an adipose-equivalent density in the planning CT. The same total doses weighted by relative biological effect (RBE) of 63 Gy (RBE) to PTVC and 69.3 Gy (RBE) to PTVG were prescribed in both IMPT/IMCT and IMRT, in which the RBE of IMCT was assessed by local effect model (LEM). To mitigate target coverage uncertainties, respiratory gating was employed to limit intra-fraction motion to less than 5 mm. The gating-on window was respiratory amplitude-based, centered at the end of exhalation and occupying with 40% duty cycle. Planning was performed on an average CT reconstructed from three respiratory phases: end-exhalation (0%Ex) and 20% amplitude on both sides of 0%Ex (20%Ex and 20%In), which was shown as Figure S1 (27). Particle plans employed 2–3 oblique beams using multiple-beam optimization and without robust optimization. Plan reproducibility was supported by daily image-guided position verification, weekly CT-based dosimetric review, and adaptive planning when indicated. Detailed planning protocols and dose constraints to organs at risk (OARs) followed the established institutional standards as previously described (2,5). Uncertainty factors and scenario-specific dose reconstruction The original IMPT, IMCT, and IMRT plans generated on the planning average CT served as the nominal plans. Four clinically relevant uncertainty factors were analyzed through scenario-specific dose recalculation: Dose calculation engine Nominal IMPT/IMCT plans were optimized using the pencil beam algorithm (PBA) in Syngo® (Siemens Healthineers, Erlangen, Germany). To evaluate computational variability, Digital Imaging and Communications in Medicine (DICOM)-RT datasets were imported into Raystation® v10B (RaySearch Laboratories, Stockholm, Sweden). Then proton plans were recalculated using a Monte Carlo algorithm (MCA) with a statistical uncertainty of 0.5%, while carbon-ion plans retained PBA due to the unavailability of MCA for carbon ions in both TPS. Range uncertainty (±3.5% RLSP) Two scenarios of range uncertainty were applied. A +3.5% shift reduced the RLSP to simulate beam overshoot (103.5% of nominal range), while a −3.5% shift increased the RLSP to mimick range underestimation (96.5% of nominal range). This analysis was restricted to particle plans for range uncertainty affected minor in RT. Three-dimensional setup uncertainty (±3 mm) Rigid isocenter shifts of ±3 mm were applied along the right-left (RL, X axis), anterior-posterior (AP, Y axis), and inferior-superior (SI, Z axis) directions, and generated 6 distinct setup-error scenarios. The dose distribution was then recalculated for each scenario. Intra-fractional respiratory motion Dosimetric perturbations due to intra-fractional respiratory motion were evaluated by recalculating nominal plans on five respiratory phases of 4-dimensional CT scans: 0%Ex, 20%Ex, 20%In, 100%Ex, and 100%In (Figure S1). Extreme inhalation (100%In) and exhalation (100%Ex) phases were prioritized to evaluate target underdosing and normal tissue overdose risks (28). TCP evaluation The TCP values were calculated using the following formula (29) based on the effective uniform dose (EUD): where vi represents the percentage target volume receiving dose Di and the tissue-specific parameter a described the volume effect of radiation response. The model parameters applied in this study were adopted from a TCP study for curative X-ray RT in non-small cell lung cancer (NSCLC) (30). The volume parameter a (reflecting volume effect) was set to −10 (31), the controlling parameter TCD50 (representing the homogeneous dose required to achieve a 50% probability of tumor control) was set to 51.24 Gy, and the response parameter γ50 (describing the tumor’s dose-response relationship) was set to 0.83. NTCP evaluation Radiation-induced esophageal injury (RIEI) and lung injury (RILI) were selected as clinically relevant endpoints for NTCP evaluation. The LKB models (17), as implemented by the previous study (1), was applied to calculate NTCP values for both nominal plans and dose distributions in the specific uncertainty scenarios across all modalities (IMPT/IMCT/IMRT). The mathematics expressions to assess the NTCP values follows as: with: The LKB formalism incorporates: (I) n, characterizing dose-volume effect steepness; (II) m, defining normal tissue dose-response gradient; (III) TD50, the dose threshold with 50% complication probability. Four published NTCP models, designated Model E1 to E4, were utilized to assess different endpoints of RIEI (21,22). Two published NTCP models, designated Model L1 to L2, were used to evaluate the risk of grade ≥2 RILI (26). All models shared consistent radiobiological parameters (n, m, and TD50) across different radiation modalities, as detailed in Table 1. Quantification of TCP and NTCP deviations Two complementary metrics were employed to evaluate the impact of uncertainties on TCP and NTCP outcomes (summarized in Figure 1). Study workflow. IMCT, intensity modulated carbon-ion radiotherapy; IMPT, intensity modulated proton radiotherapy; IMRT, intensity modulated photon radiotherapy; NTCP, normal tissue complication probability; TCP, tumor control probability; δ(ΔNTCP), deviations in ΔNTCP; δNTCP, deviations in NTCP; ΔNTCP, the NTCP advantage of particle therapy (PT: IMPT or IMCT) over photon IMRT; δTCP, deviations in TCP. Scenario‑specific deviations (δTCP and δNTCP) For a given radiation modality (e.g., IMRT, IMPT, IMCT), δTCP and δNTCP quantify the deviations in TCP or NTCP, respectively, when a specific uncertainty is introduced relative to the nominal plan: Baseline of inter‑modality NTCP difference (∆NTCP): particle versus photon The ΔNTCP was defined as the NTCP advantage of particle therapy (PT: IMPT or IMCT) over photon IMRT under the same conditions of nominal conditions (before accounting for delivery uncertainties) or specific uncertainty scenario: A negative ΔNTCP value reflects further reduced toxicity risk by using particle therapy. Impact of uncertainties on ΔNTCP: scenario versus nominal [δ(∆NTCP)] To directly compare how a given uncertainty scenario affects NTCP difference between PT and photon IMRT, the metric δ(∆NTCP) was introduced. It quantifies the deviation of the NTCP difference (∆NTCP) between the detailed uncertainty scenario and nominal plan. Notably, of the scenarios for calculation algorithm engine or range uncertainty, the respective NTCP values in IMRT were matched to the values in the nominal IMRT plans since both uncertainty factors were considered of minor influence on the dosimetric deviations of IMRT. To quantify the contribution of each uncertainty factor, δTCP values were employed to compare the impact of different uncertainty factors as all three provided similar target coverage and TCP estimates. For NTCP, δ(ΔNTCP) values were used to quantify the contribution of each uncertainty factor, given that IMPT/IMCT achieved significant dosimetric reduction and thus potentially lower NTCP risk. To reduce bias of model selection, δ(ΔNTCP) values were calculated as relative differences. Statistical analysis For each TCP/NTCP model, results are reported as the mean ± standard deviation across the ten‑patient cohort. The deviation metrics [δTCP, δNTCP, δ(ΔNTCP)] represent the mean deviation across all patients, as illustrated in Figure 1. In interpretation, an increase in δTCP suggests improved tumor control, an increase in δNTCP indicates higher toxicity risk, and an increase in δ(ΔNTCP) represents a weaken NTCP advantage of particle therapy over IMRT, under the evaluated uncertainty scenario. Results TCP stability across RT modalities under clinical uncertainties Under equivalent prescription doses and comparable target coverage, the predicted TCP values were similar across the three radiation modalities: 75.3% (IMPT), 75.2% (IMCT), and 74.3% (IMRT) upon their respective nominal plans. All modalities maintained clinically acceptable robustness, with TCP deviations (δTCP) constrained within narrow ranges (Figure 2): −0.4% to +0.4% for IMPT, −0.2% to +0.7% for IMCT, and +0.1% to +0.4% for IMRT. Tumor control probability deviations (δTCP) under clinical uncertainties. Mean values of δTCP were shown for IMPT, IMCT, and IMRT across 4 uncertainty factors: (A) calculation engine, (B) range (±3.5%), (C) setup (±3 mm; X, Y, Z axes), and (D) intra-fractional respiratory motion (5 breathing phases). Symbols indicate maximum (△), mean (○), and minimum (◇) values of δTCP across ten patients with locally advanced lung cancer. Ex, exhalation; IMCT, intensity modulated carbon-ion radiotherapy; IMPT, intensity modulated proton radiotherapy; IMRT, intensity modulated photon radiotherapy; In, inhalation; TCP, tumor control probability; δTCP, deviations in TCP. Among total 280 uncertainty scenarios (included 1 algorithm, 2 range, 6 setup, and 5 intra-fractional respiratory uncertainty scenarios) for PT, only 1 scenario resulted in a TCP reduction >1.0%. This outlier (TCP reduction of 1.4%) occurred under a 3-mm posterior-anterior shift in IMPT plans with the patient in prone position, likely due to the shortened beam range caused by the high-density thoracic vertebra at the beam entrance. The overall TCP stability is attributable to the proposed treatment strategies, including 5-mm dosimetric margins and multiple beam directions, which smear the dosimetric uncertainty. Esophageal NTCP and its uncertainty‑induced deviations across RT modalities Acute grade 2 or 3 RIEI was evaluated using established NTCP models (E1–3). Model E1 In nominal plans, the mean values of ∆NTCPE1 for grade 2 or 3 acute esophagitis (AE) were −6.9%±11.0% for IMPT and −7.2%±10.8% for IMCT, corresponding to nominal NTCP values of 26.2%±29.3% (IMPT), 24.9%±29.0% (IMCT), and 33.1%±32.5% (IMRT). Mean scenario-specific deviations in ∆NTCP [δ(∆NTCPE1)] ranged from −3.3% to +1.4% for IMPT and from −1.9% to +2.2% for IMCT across all scenarios (Figure 3). Impact of clinical uncertainties on the inter-modality NTCP difference (∆NTCP) for radiation-induced esophageal injury. Mean deviations of ΔNTCP are shown for three esophageal NTCP models (E1–E3) across four major uncertainty factors: calculation engine (A,E,I), range (±3.5%; B,F,J), setup (±3 mm; X, Y, Z axes; C,G,K), and intra-fractional respiratory motion (5 breathing phases; D,H,L). Panels in (A-D), (E-H), and (I-L) correspond to models E1, E2, and E3. Symbols indicated maximum (△), mean (○), and minimum (◇) values of ∆NTCP across ten patients with locally advanced lung cancer. ∆NTCP, the NTCP advantage of particle therapy (PT: IMPT or IMCT) over photon IMRT; Ex, exhalation; IMCT, intensity modulated carbon-ion radiotherapy; IMPT, intensity modulated proton radiotherapy; IMRT, intensity modulated photon radiotherapy; In, inhalation; NTCP, normal tissue complication probability. Model E2 The nominal values of ∆NTCPE2 were −1.9%±3.7% for IMPT and −2.0%±2.8% for IMCT, corresponding to nominal NTCP values of 79.2%±23.1% (IMPT), 79.1%±22.9% (IMCT), and 81.1%±21.4% (IMRT). The δ(∆NTCPE2) values varied from −2.3% to +2.6% for IMPT and −1.7% to +2.9% for IMCT (Figure 3) on average. Model E3 For grade 2 or 3 AE, the nominal values of ∆NTCPE3 were −2.8%±5.3% for IMPT and −2.8%±5.3% for IMCT, corresponding to nominal NTCP values of 40.9%±19.9% (IMPT), 40.9%±19.8% (IMCT), and 43.7%±20.2% (IMRT). The δ(∆NTCPE3) ranges of −2.3% to +1.2% for IMPT and −2.0% to +1.3% for IMCT were observed under different uncertainty scenarios (Figure 3). Late RIEI (Model E4) Both nominal and uncertainty-disturbed ∆NTCPE4 values remained less than 0.2% for all RT modalities. This reflected the stringent esophageal dose constraints applied in planning under the prescribed dose of 69.3 Gy (RBE) (Figure S2). In summary, for nominal IMPT plans, the number of patients (out of 10) achieving a ∆NTCP advantage exceeding the 10‑percentage‑point threshold for esophageal toxicity was 3, 0, and 1 for NTCP models E1, E2, and E3, respectively. For nominal IMCT plans, the corresponding number were 2, 0, and 1, respectively. Therefore, the dosimetric advantages of IMPT or IMCT for LA-LC could be translated into a reduction in acute (but not late) RIEI, particularly in patients who partly receive sequential chemotherapy. This may be attributed to the difference in TD50 values between models E1–E3 [36–46 Gy (RBE)] and E4 [76.1 Gy (RBE)], indicating that the volume receiving doses around TD50 is more critical for the respective injury endpoints. Pulmonary NTCP and its uncertainty‑induced deviations across RT modalities Two NTCP models (L1 and L2) were used to evaluate the risk of grade ≥ 2 radiation-induced lung injury (RILI). In nominal plans, particle therapy (IMPT/IMCT) consistently showed lower pulmonary NTCP than IMRT. The uncertainty‑induced variations in ΔNTCP remained within limited ranges, supporting the robustness of this dosimetric advantage. Overall cohort analysis Model L1 yielded nominal ∆NTCPL1 of −6.0%±6.1% for IMPT and −6.9%±5.7% for IMCT compared to photon IMRT, with nominal NTCP values of 7.3%±8.9% (IMPT), 6.3%±8.8% (IMCT), and 13.2%±14.0% (IMRT). The corresponding mean δ(∆NTCPL1) values ranged from −0.8% to +1.5% for IMPT and −0.8% to +1.4% for IMCT across different uncertainty scenarios (Figure 4). Impact of clinical uncertainties on the inter-modality NTCP difference (∆NTCP) for radiation-induced lung injury. Mean deviations of ΔNTCP are shown for two lung NTCP models (L1–L2) across four major uncertainty factors: calculation engine (A,E), range (±3.5%; B,F), setup (±3 mm; X, Y, Z axes; C,G), and intra-fractional respiratory motion (5 breathing phases; D,H). Panels in (A-D) and (E-H) correspond to models L1 and L2. Symbols indicated maximum (△), mean (○), and minimum (◇) values of ∆NTCP across ten patients with locally advanced lung cancer. ∆NTCP, the NTCP advantage of particle therapy (PT: IMPT or IMCT) over photon IMRT; Ex, exhalation; IMCT, intensity modulated carbon-ion radiotherapy; IMPT, intensity modulated proton radiotherapy; IMRT, intensity modulated photon radiotherapy; In, inhalation; NTCP, normal tissue complication probability. Model L2 showed similar trend in NTCP differences: nominal ∆NTCPL2 values were −12.0%±10.9% for IMPT and −12.8%±10.1% for IMCT, with nominal NTCP values of 9.4%±9.6% (IMPT), 8.7%±10.4% (IMCT), and 21.4%±20.1% (IMRT). The mean uncertainty-driven shift δ(∆NTCPL2) varied between −0.8% to +1.9% for IMPT and between −0.7% to +1.9% for IMCT under different scenarios (Figure 4). Subgroup analysis by IGTV volume For patients with IGTV >53.1 cm3 (the median volume), the nominal NTCP advantage was more pronounced: −9.5%±7.2% (IMPT) and −11.2%±5.3% (IMCT) in ∆NTCPL1; −17.7%±13.3% (IMPT) and −18.9%±11.1% (IMCT) in ∆NTCPL2. Uncertainty‑driven deviations of ΔNTCPL1 and ΔNTCPL2 in this subgroup ranged between −1.2% and +3.0% for IMPT and between −1.6% and +3.1% for IMCT (Figures S3,S4). For patients with IGTV ≤53.1 cm3, the absolute values of ∆NTCP were smaller however still favored particle therapy: −2.5%±1.0% (IMPT) and −2.7%±1.0% (IMCT) when using ∆NTCPL1; −6.3%±3.3% (IMPT) and −6.7%±3.9% using ∆NTCPL2. Uncertainty‑induced deviations of ΔNTCPL1 and ΔNTCPL2 in this subgroup ranged from −0.4% to +0.8% for IMPT and from −0.3% to +0.8% for IMCT (Figures S5,S6). For pulmonary toxicity, ∆NTCP advantage exceeded the 10‑percentage‑point threshold in 3 (IMPT) and 2 (IMCT) patients according to model L1, and 4 (IMPT) and 6 (IMCT) patients according to model L2. Notably, all patients with IGTV volume >53.1 cm3 met this criterion under IMCT when assessed with model L2. Preserved NTCP gain of PT after uncertainty analysis The mean δNTCP values for all 6 NTCP models evaluating the risk of esophageal and pulmonary toxicities were summarized in Table S1. An overview of mean δNTCP, δ(∆NTCP) and ∆NTCP values after considering the disturbance across the 14 evaluated uncertainty scenarios for both IMPT and IMCT was presented in Figure 5. Heatmap of NTCP deviations and inter-modality NTCP differences under clinical uncertainty. Mean values of (A) δNTCP (deviation from nominal NTCP), (B) δ(∆NTCP) (deviation from nominal ΔNTCP), and (C) ∆NTCP (difference between particle therapy and IMRT are shown for 6 NTCP models under 14 uncertainty scenarios. Colors represent cohort-average values of each metric across ten patients with locally advance lung cancer: red indicates smeared advantage, blue indicates strengthen advantage. ∆NTCP, the NTCP advantage of particle therapy (PT: IMPT or IMCT) over photon IMRT; Ex, exhalation; IMCT, intensity modulated carbon-ion radiotherapy; IMPT, intensity modulated proton radiotherapy; IMRT, intensity modulated photon radiotherapy; In, inhalation; NTCP, normal tissue complication probability; PT, particle therapy; δ(∆NTCP), mean deviations of the ∆NTCP values; δNTCP, mean deviations of NTCP values. After incorporating all these uncertainty factors, PT showed sustained reductions in NTCP compared to IMRT. For RIEI, the preserved NTCP advantage ranged from 5.0–10.2% (Model E1), 0.0–4.2% (Model E2), and 1.5–5.1% (Model E3). For RILI, the maintained NTCP benefit of PT varied from 4.5–7.7% (Model L1) and 10.1–13.5% (Model L2). This analysis further identified specific uncertainty factors which might have more significant impact on these comparative outcomes. For RIEI, a 3-mm posterior to anterior (y-3) setup shift in IMCT plans, a −3.5% range uncertainty in IMPT, and the use of PBA calculations for IMCT planning were the predominant reasons that caused deviations >2% in mean ∆NTCP values. For RILI, deviations of ∆NTCP >1% were mainly attribute to respiratory motion uncertainty, particularly at the extreme inhalation and exhalation phases of the whole breathing cycle (100%Ex and 100%In). To distinguish the core influence factors without introducing bias from NTCP model selection, the δ(∆NTCP) values were calculated as relative differences (listed in Table S2). Compared with the lungs, the ∆NTCP values for esophagus tended to present more obvious deviations in response to uncertainty factors. For esophageal toxicity, range and setup uncertainties emerged as the core influencing factors, with the greatest impact on smearing the NTCP advantage for RIEI. For pulmonary toxicity, respiratory motion presented as the core influencing factor, potentially compromising the accurate assessment of RILI. Discussion Proton and carbon-ion radiotherapy offer the potential to reduce radiation-induced toxicities in LA-LC through dosimetric advantage, especially in the low-dose areas (1). This study systematically evaluated the impact of four major uncertainty factors (dose calculation engine, range uncertainty, setup uncertainty, and intra-fractional respiratory motion) on the prediction of TCP and NTCP in IMPT/IMCT/IMRT plans. Our analysis revealed three key findings, which are detailed in the subsequent sections. Key findings in this study First, TCP estimates remained highly stable across all RT modalities and uncertainty scenarios. The mean δTCP values in this study confined to a narrow range of −0.2% to +0.7%. Second, although PT maintained an NTCP advantage over IMRT in most scenarios, some uncertainties may attenuate this benefit. The mean δ(∆NTCP) values were −3.3% to +2.9% for RIEI and −0.8% to +1.9% for RILI. This reflected certain scenarios (e.g. setup shifts, range uncertainties, or respiratory motion) partially diminished the dosimetric advantage of particle beams. Third, after full incorporation of uncertainty-induced dosimetric deviations, PT retained an NTCP advantage over photon IMRT. For RIEI, the preserved NTCP advantage ranged from 5.0–10.2% (Model E1), 0.0–4.2% (E2), and 1.5–5.1% (E3). For RILI, the maintained NTCP benefit of PT varied from 4.5–7.7% (Model L1) and 10.1–13.5% (L2). These ranges confirmed that the dosimetric superiority of PT translated into a sustained NTCP advantage under realistic delivery conditions. Robustness of TCP Our study found that TCP deviations across different treatment modalities were relatively minor. This stability is attributable to the series of treatment strategies adopted for IMPT and IMCT in LA-LC, including defining IGTV of the full respiratory trajectory of GTV, using an average CT reconstructed from the gating window as the planning CT, density overriding in high‑motion regions and etc. The observed stability of TCP could be then attributed, in which target dose coverage remained robust under the investigated uncertainties. Therefore, the clinical significance of this study lies in that when without robust optimization, the proposed planning strategies could ensure acceptable TCP stability if only intrafractional uncertainty is considered. Additional planning strategies should be employed to compensate the situation where commercial robust optimization is unavailable so that PT for lung cancer could be implemented with confidence in target coverage and TCP expectation. Regarding the analytic algorithms, the minor deviation between PBA and MCA in IMPT (δTCP =0.1%) contrasted with reported overestimations in earlystage lung cancer (32). This was likely due to the larger volumes and more solid densities of the tumor targets in LA-LC. For IMCT, cross-platform PBA comparisons revealed a slightly higher mean δTCP of 0.7% when commercial MCA was unavailable. Range (±3.5%) and setup (±3 mm) uncertainties induced negligible TCP degradation: −0.4% to +0.1% for IMPT and −0.2% to +0.0% for IMCT. These findings were different from oropharyngeal cancer data (16), where IMPT showed the TCP losses up to 2.0% under similar uncertainties. Notably, among 280 uncertainty scenarios in our study, only 1 IMPT case presented with a TCP reduction >1.0%, which was a 1.4% decrease caused by a 3-mm posterior-anterior setup shift, emphasizing the necessity of high-precision positioning in PT. Esophageal NTCP and dominant uncertainty sources For RIEI, the impact of uncertainty factors was relatively major, driven primarily by calculation algorithm, setup errors, and range uncertainty. Our findings indicated that the NTCP reduction might be mitigated when these four uncertainty factors are taken into consideration. Range and setup uncertainty emerged as the core influencing factors disturbing the accurate evaluation of ∆NTCP. This aligns with the conclusion of Park et al. (33), who reported that the esophagus, as a small-volume OAR adjacent to the target, is more sensitive to range and setup uncertainties in passive-scattering proton radiotherapy for lung cancer. In the present study, setup shifts, range errors, and algorithm differences led to ΔNTCP deviations exceeding 2% across esophageal models, while respiratory motion contributed relatively minor deviations [δ(∆NTCP) ≤1.1%]. The largest δ(∆NTCP) of 2.9% occurred with a 3-mm posterior-anterior shift in IMCT plans. Similar deviation (less than 3.0%) on a weighted-average level was reported in the NTCP robustness evaluation of IMPT and VMAT plans when considering range and setup uncertainties in head and neck cancers (16). Range uncertainty represents a critical, systematic, particle-specific factor which is seldom considered in the photon therapy (34). More precise conversion from CT HU values to the relative stopping power, e.g., via proton CT (35), could improve range accuracy. Although no ΔNTCP difference >10 percentage points was observed between the particle and photon plans for esophageal toxicity, clinical studies in limited-stage small-cell lung cancer (5) and locally advanced NSCLC (2) have reported significant reductions in esophageal toxicity with particle therapy. This discrepancy addresses the need for careful consideration of RBE estimation accuracy in LEM model (36) and linear energy transfer (LET) impacts (37) in particle-specific NTCP estimation. Pulmonary NTCP and the impact of respiratory motion For RILI, the impact was relatively minor, and the observation from this study indicated that intra-fractional respiratory motion accounted for the most dominant uncertainty factor influencing pulmonary NTCP. This is likely because, as a parallel organ, RILI risk correlates with the volumetric doses to the lungs (e.g., mean dose, low-dose region) rather than only high-dose areas. Respiratory motion most readily alters lung’s volume and position, leading to marked changes in the volumetric dose distribution. With respiratory gating (20% amplitude window around end-exhalation), mean ΔNTCP deviations (compared to IMRT) within the gating window were <0.5%, and increased to 1.3–1.9% at the extreme phases (100%In and 100%Ex) due to PBS interplay effects. These findings reinforce the application of respiratory motion management, especially for solid peripheral lesions. Gorgisyan et al. (38) also emphasized the necessity of respiratory motion management to balance dosimetric reliability and treatment efficiency. When inter- and intra-fractional 4D dose reconstruction is performed in the pencil beam delivery system, overall NTCP variations remained less than 2.0% in proton therapy for lung cancer (28). Importantly, after considering all 4 uncertainty factors in our study, the pulmonary dose advantages of IMPT/IMCT (1) were preserved and translated into NTCP benefit, particularly for patients with larger target volumes (IGTV >53.1 cm3). In such subgroups, ΔNTCP advantage for pneumonitis (Model L2) consistently exceeded 10 percentage points in both nominal plans and uncertainty scenarios, providing a quantitative basis for selecting PT for LA-LC. This aligns with model‑based selection criteria proposed for other tumor sites, where a ΔNTCP threshold of 10 percentage points is used to justify particletherapy use (39). ∆NTCP threshold-driven patient screening A ∆NTCP advantage exceeding 10 percentage points is recommended for model-based selection of proton radiotherapy (9-11) in the Netherlands. Considering the maximum deviation of 3% observed for intra-fractional uncertainty in this study, together with an additional 2% inter-fractional variation (28) reported for proton therapy in lung cancer. The clinical significance of a 10‑percentage‑point threshold may indicate a clinically meaningful benefit in reducing radiation‑induced injuries, thereby supporting the legitimate use of particle therapy even when accounting for delivery uncertainties. For esophageal NTCP models (E1–E3), less than 30% of 10 patients exceeded the 10‑percentage‑point ∆NTCP threshold in both nominal IMPT and IMCT plans. For pulmonary NTCP models (L1–L2), this proportion ranged from 30% to 40% for IMPT and 20% to 60% for IMCT. Notably, all patients with and IGTV >53.1 cm3 met this ΔNTCP criterion under IMCT when assessed using model L2. As a preliminary recommendation for patient screening, those with large target volume, who may be unable to tolerant photon radiotherapy, could particularly benefit from the dosimetric sparing of lung and reduced risk of RILI when receiving IMPT or IMCT for LA-LC. An IGTV of 53.1 cm3 corresponds approximately to a sphere of 4.65 cm in diameter. Therefore, the clinical implications of this ΔNTCP analysis can be summarized as follows: ❖ For patients with a tumor diameter >4.65 cm, proton and carbon-ion radiotherapy may be considered the optimal radiation technique to achieve a reduced risk of RILI. ❖ For patients with a tumor diameter >4.65 cm, proton and carbon-ion radiotherapy may be considered the optimal radiation technique to achieve a reduced risk of RILI. ❖ For patients with tumor diameter ≤4.65 cm, the optimal technique should be determined by individualized dosimetric comparisons between particle and photon radiotherapy to quantify the ∆NTCP difference. ❖ For patients with tumor diameter ≤4.65 cm, the optimal technique should be determined by individualized dosimetric comparisons between particle and photon radiotherapy to quantify the ∆NTCP difference. Although ΔNTCP advantages failed to reach the 10-percentage-point threshold in esophageal models (E1–E3), the ΔNTCP values of approximately −7.2% to −6.9% for model E1 still indicate that patients who partly receive sequential chemotherapy may experience a reduced risk of grade ≥2 acute radiation-induced esophagitis when treated with IMPT or IMCT for LA-LC. Limitations of this study This study is subject to four primary limitations. First, this analysis assumed that the LEM I-calculated RBE-weighted doses for particle therapy were equivalent to the physical doses delivered in IMRT. For protons, a constant RBE value of 1.1 was applied per LEM model, which does not account for the known variation of RBE, especially of the increased RBE at the distal edge of the Bragg peak (40). For carbon ions, emerging dosimetric studies and clinical observations suggested that within the fractionation range employed in this study, the LEM I model may overestimate RBE-weighted doses for carbon ions with the constant application of α/β =2 Gy (41). Consequently, the actual clinical benefit of carbon-ion therapy is likely greater than estimated here. Future comparisons between particle and photon therapies would benefit from biologically refined models that provide more accurate RBE estimations through more basic physics variables, such as LET gradients. Second, the robustness evaluation was confined to intra‑fractional uncertainties in this study and was performed by discussing the uncertainty factors independently. This is because through the independent analysis on a single scenario which disturbed the accurate evaluation of TCP and NTCP, detailed technical guidance could be then attempted to smear the dosimetric compact specifically for these uncertainty scenarios. Still, incorporation of a sub-set of uncertainty scenarios might provide comprehensive evaluation on the dosimetric robustness of PT. Also the inter-fractional robustness of the dosimetry in particle plans (including anatomic changes, setup errors, and respiratory pattern variability) for LA-LC is also warranted to be assessed in the future studies. Thirdly, the role of fractionation was in absence of its dosimetric and followed biological compact when assessing the TCP and NTCP values especially in IMCT. This was still originated from the lack of accurate translation of the equivalent effective doses between the particle and photon radiation (41). Therefore, PT-specific LKB-based NTCP models for severe radiation-induced toxicities of thoracic OARs are expected to predict such toxicities accurately and support clinical decision-making. Finally, the relatively small sample size of ten patients might result in insufficient statistical power of this study. Therefore, larger patient cohorts with extended follow-up and detailed clinical observation are warranted to validate the TCP and NTCP findings reported here for PT in LA-LC. Conclusions After incorporating the impact of four key uncertainty factors (algorithm engine, range, setup, and respiratory motion), non-robustly optimized particle therapy (PT: IMPT and IMCT) plans remained robust target coverage (mean TCP deviation <0.7%) by applying our institutional planning strategies. Compared to IMRT, PT preserved clinically meaningful NTCP advantage for RILI (particularly in patients with larger IGTV), while esophageal benefits were more susceptible to degradation (especially for late toxicities). Proper TCP, NTCP and ∆NTCP evaluation under realistic uncertainties supports individualized model-based selection of PT to optimize clinical outcomes in LA-LC. Supplementary The article’s supplementary files as Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2026-1-0266/prf