Citations

e16018 Background: Intrahepatic cholangiocarcinoma (iCCA) exhibits marked heterogeneity in outcomes that is incompletely explained by stage or treatment timing. Although treatment delay is frequently cited as a modifiable risk factor, its prognostic significance in real-world practice remains unclear. We applied machine learning (ML) to model iCCA as a dynamic disease process and to evaluate whether longitudinal disease trajectories better explain outcomes than treatment timing alone. Methods: We conducted a population-based retrospective analysis of adults diagnosed with iCCA using the Surveillance, Epidemiology, and End Results (SEER) database (2000–2022). Time from diagnosis to first cancer-directed therapy was categorized as ≤30 days, 31–60 days, or > 60 days. Associations between treatment delay, disease characteristics, and overall survival (OS) were assessed using multivariable logistic and Cox regression models. To capture real-world disease evolution, each patient’s clinical course was encoded as an ordered longitudinal sequence of diagnosis, treatment states (surgery, chemotherapy, radiation), and outcome. Unsupervised ML was used to identify latent disease-course trajectories. A supervised ML model using variables available at diagnosis was developed to predict membership in high-risk trajectories. Results: The cohort included 18,312 patients with iCCA. Treatment initiation beyond 30 days occurred in more than half of patients and was more common in localized than advanced disease. In adjusted survival analyses, shorter time to treatment was paradoxically associated with worse OS, consistent with confounding by disease severity rather than a protective effect of delay. Unsupervised ML identified four distinct and reproducible disease trajectories with markedly different outcomes. Median OS ranged from approximately 2 months in early-failure trajectories to approximately 15 months in favorable multimodality trajectories. Trajectory membership was not fully explained by stage or treatment timing. A supervised ML model predicted high-risk trajectory membership with good discrimination (AUC 0.82). Key predictors included tumor grade, age, tumor burden, and incomplete staging, whereas treatment timing had minimal influence. Conclusions: In iCCA, treatment delay reflects underlying disease biology and care prioritization rather than independently determining survival. ML-based trajectory modeling reveals clinically meaningful disease courses that better explain outcome heterogeneity than treatment timing or stage alone and enables early identification of patients at highest risk for unfavorable real-world outcomes.

e15501 Background: CP is the backbone of first-line therapy for recurrent or metastatic ASCC. In practice, Q3-week CP has been explored as a potential alternative to the standard (d1,8,15 Q28-day cycle) dosing, though its efficacy is not known. This study evaluated CP scheduling and clinical survival associations in metastatic ASCC. Methods: This multicenter retrospective cohort study included patients with metastatic ASCC treated with first-line CP, identified using the Mayo Data Explorer database. Electronic health records were reviewed to collect baseline clinical characteristics, treatment regimens, responses, progression-free survival (PFS), and overall survival (OS). Descriptive statistics were used to summarize patient demographics. Survival was estimated using Kaplan–Meier methods. Multivariable Cox proportional hazards models were used to compare survival by dosing schedule, adjusting for age, ECOG performance status, HPV (p16) status, liver metastasis, and prior chemoradiation. Results: 54 patients treated between 2020-2025 with first-line CP for metastatic ASCC were included. Baseline patient demographics were balanced between dosing groups (Table 1). Median follow-up time was 26.2 months. For Q3-week CP (n = 21), median PFS and OS were 9.8 (95% CI 8.1, NR) and 27.0 (95% CI 15.7, NR) months, respectively, while median PFS was 6.5 (95% CI 4.5,15.3) months with standard CP (n = 33) and median OS was not reached (95% CI 11.2, NR). Response rate was 76% with Q3-week and 60.6% with standard CP (p = 0.23) dosing. Survival outcomes were similar when comparing Q3-week CP relative to standard CP dosing using multivariable analysis (PFS HR 0.68, 95% CI 0.33–1.41; OS HR 1.67, 95% CI 0.67–4.14). Higher ECOG and liver metastasis were independently associated with inferior outcomes, while HPV was associated with improved survival. Subsequent therapy patterns were similar between CP schedule groups. Ten patients (16.9%) received local therapy, and 24 of 28 patients with progression received further systemic therapy, including 21 (87.5%) treated with immunotherapy. Conclusions: Survival outcomes were similar across dosing schedules, suggesting Q3-week CP may be a reasonable alternative to standard CP, with potential to reduce treatment-related burden in metastatic ASCC. Future analysis will compare toxicity profiles between CP dosing schedules, including in combination with PD1 inhibitors given the results of the POD1UM-303 trial. Patient demographics (n=54). Variable Q3-week CP (n=21) Standard CP (n=33) p-value Sex Female 16 (76.2%) 24 (72.7%) 0.77 Male 5 (23.8%) 9 (27.3%) Age Median (Q1, Q3) 62 (50.5, 69) 64 (55.5, 70) 0.57 BMI Median (Q1, Q3) 27.1 (21.9, 34.4) 27.4 (22.5, 31.7) 0.72 ECOG 0 9 (42.9%) 11 (34.4%) 0.53 ≥ 1 12 (57.1%) 21 (65.6%) Confirmed HPV (p16) Positive 16 (76.2%) 24 (72.7%) 0.77 Prior chemoradiation Yes 17 (81.0%) 21 (63.6%) 0.17

8045 Background: Surgical resection is the primary therapy for stage I-II NSCLC but not performed in a subset of patients due to comorbidities or self-election, who instead undergo radiation with or without systemic therapy (RT±ST). Little is known about the demographic, tumoral, therapeutic, and quality-of-life (QOL) characteristics of this non-surgical group. Understanding their differences from the surgical group is essential for developing precision treatment strategies and tailored survivorship guidelines. Methods: In a prospectively enrolled (1997-2016) and followed (through 2025) cohort of 20,951 primary lung cancer patients, 6,486 (31%) had stage I-II NSCLC, including 5,359 (83%) treated surgically (Surgical) and 1,113 (17%) otherwise (Non-surgical). Preexisting diseases were grouped into cancer, lung, and other comorbidities. QOL data were available for 40% of all patients. Descriptive analysis of demographics, tumor features, treatment patterns, and QOL metrics, as well as survival analysis (median years and overall survival [OS] rates) were conducted. QOL measures, on overall and 15 symptom and functional domains, were scored 0 (worst) to 10 (best), with 1-unit difference considered clinically and statistically meaningful. Results: At time of diagnosis, compared to the Surgical, the Non-surgical patients were older (median 73 v 68 years), more likely to have smoked cigarettes (91% v 83%), squamous cell and unspecified NSCLC (53% v 28%), stage IIB (19% v 12%), and preexisting lung comorbidities (71% v 58%). Non-surgical patients more frequently received RT±ST than the Surgical (70% v 21%) and had shorter median survival (2.5 v 7.8) years. Within the Non-surgical, 28% had no cancer treatment (No-Tx), 44% RT-only, 7% ST-only, and 19% both (RT+ST); their respective median diagnosis ages were 74, 76, 68, 70 years; median survival were 1.7, 2.7, 2.2, 3.2 years; 5-year OS (95%CI) at 25% (21-31), 25% (21-29), 26% (17-38), 30% (25-37); and 10-year OS at 9% (7-14), 7% (5-10), 10% (5-21), 9% (6-14), respectively. RT-only group had highest rates of comorbidities (37-84%) and the No-Tx the lowest (31-66%). Although the overall QOL scored the same for all groups, RT-only patients reported the worst scores on multiple symptoms burden, and the No-Tx reported the least pain and lung cancer symptoms but the worst on mental well-being and social activities. Conclusions: Non-surgically managed stage I-II NSCLC cases present clinically distinct patients with the shortest median survival in No-Tx group and highest 5-year survival in RT+ST. RT-only patients were oldest, had the most comorbidities, and reported the worst symptom burdens. These differences highlight the need for more in-depth analyses and studies (considering age, histology, comorbidity, QOL, etc.) to support tailored therapeutic strategies and survivorship care for patients who decline or are not suitable for surgery.

4584 Background: EVP is the preferred frontline regimen in pts with aUC. Despite initial response to treatment, many pts treated with EVP ultimately have progression. We hypothesized that recurrence patterns would differ between pts with and without baseline liver metastases. Methods: UNITE is a multi-institutional retrospective database capturing clinical and treatment characteristics of pts with aUC receiving systemic therapies. We queried this database to identify pts who achieved an observed response (OR) to EVP. Pts were grouped by initial metastatic sites using a non-mutually exclusive approach. Descriptive statistics were used to summarize observed response rates (ORR) and recurrence patterns at time of progression. Progression-free survival (PFS) from start of EVP was estimated using the Kaplan-Meier method. The effect of liver metastases on PFS was assessed via multivariable Cox regression. Median-follow up time was estimated using the reverse Kaplan-Meier method. Statistical analyses were conducted in RStudio. Results: We identified 464 pts treated with EVP (median age 72; 74% male, 80% White). Best ORR in pts with baseline liver metastases (n=77) were complete response 9%, partial response 35%, stable disease 18%, and progressive disease 21%, compared with 18%, 37%, 27%, and 18%, respectively, in pts without baseline liver metastases (n=325). The estimated median follow-up time was 52 weeks (95% CI 48-62). Among pts with initial OR to EVP, those with baseline liver metastases showed a non-significant trend toward shorter PFS (HR 1.35; 95% CI 0.67–2.75; p=0.40) versus those without baseline liver metastases. Among pts with initial OR to EVP and subsequent progression the distribution of sites of recurrences at progression for those with (n = 10) and without (n = 40) baseline liver metastases were: liver (50% vs. 10%), lymph nodes (60% vs. 55%), lungs (50% vs. 38%), bone (20% vs 20%), and primary tumor (20% vs 25%). EVP responders with baseline liver metastases had a numerically higher rate of site-specific metastasis resolution at progression versus those with lymph node or lung metastases (Table). Conclusions: Our analysis suggests that liver metastases may be associated with shorter PFS among EVP responders, though these pts frequently do not retain liver metastases at time of progression. Study limitations include retrospective analysis, investigator-assessed response, small sample size, selection and confounding biases. Patterns of recurrence following EVP merits further evaluation. Metastatic Site Pts with Site-Specific Baseline Metastases (n) Pts with Resolution of Site-Specific Metastases at Progression, n (%) Pts without Baseline Metastases at Site (n) Pts who Remained Free of Site-Specific Metastases at Progression, n (%) Liver 10 5 (50%) 40 36 (90%) Lymph Nodes 37 12 (32%) 13 10 (77%) Lungs 26 8 (31%) 24 22 (92%)

4565 Background: EVP is the preferred 1 st -line regimen for patients (pts) with aUC. EV-related toxicities are often managed with dose adjustments, such as reductions and holds. Analysis of EV monotherapy trials showed consistent survival outcomes despite dose adjustments. We hypothesized that on-treatment EV dose adjustments would not impact survival in these pts. Methods: Pts with aUC treated with EVP in the multisite UNITE study were analyzed. Early dose reductions (DR) and holds were defined as occurring < 9 weeks from EVP start, with the remainder classified as late. Pts with DR at treatment start were excluded. Survival endpoints were progression-free survival (PFS) and overall survival (OS) from EVP start, assessed by Kaplan-Meier analysis and Cox models. Multivariable Cox analysis (MVA) included age (≥ 75 vs < 75), ECOG performance status (0-1 vs 2-3), Hgb level (< 10 vs ≥ 10) and liver metastases (present vs not). A time-dependent covariate was used to reduce immortal time bias. Complementary MVA landmark analyses were performed, aligned to the total number of treatment cycles received, with six analyses from cycles 2 to 7. Results: 360 pts from 17 sites were included (median age 70; 78% Caucasian; 76% male; 27% upper tract primary; 35% histology subtype component; 17% with liver metastasis; 78% frontline); median follow-up was 11.7 months (95% CI [10.79, 13.08]). Median number of cycles was 6 in pts with DR and 3 with no DR. Early DRs (100 pts [28%]) and holds (18 pts [5%]) were most commonly due to rash (28% of all early DRs and holds). Late DR (75 pts [21%]) and holds (50 pts [14%]) were most commonly due to neuropathy (52% of all late DRs and holds). Pts with early or late DR had longer PFS and OS vs those who did not (Table); however, there was no significant PFS or OS difference with dose hold vs no hold (PFS HR 0.65, 95%CI (0.40, 1.05), p = 0.08; OS HR 0.70, 95%CI (0.39, 1.27), p=0.24). In landmark analyses, about 50% of pts had DR; no significant difference in PFS or OS was observed at most landmarks, except for longer OS in the DR cohort at cycle 4 (HR 0.53, 95% CI (0.29, 0.98), p = 0.042) and cycle 5 (HR 0.43, 95% CI (0.19, 0.99), p = 0.048). Conclusions: In this real-world retrospective analysis, EVP efficacy was maintained despite DR and dose holds. Survival difference noted in the time-dependent analyses were not consistently reproduced in landmark analyses, suggesting exposure-related selection bias rather than causal effect. Prospective studies are warranted to validate these hypothesis-generating findings. Association of EV DR with PFS/OS (time-dependent analyses). Early DR Late DR DR vs no DR outcomes UVAHR (95% CI), p MVAHR (95% CI), p UVAHR (95% CI), p MVAHR (95% CI), p PFS 0.50 (0.35, 0.72), <.001 0.56 (0.38, 0.84), 0.005 0.32 (0.21, 0.49), <.001 0.36 (0.23, 0.57), <.001 OS 0.51 (0.32, 0.79), 0.003 0.57 (0.35, 0.92), 0.022 0.26 (0.15, 0.45), <.001 0.31 (0.17, 0.55), <.001

4173 Background: [ 212 Pb]VMT-α-NET is an alpha-particle radiation-delivering agent targeted to somatostatin receptor type 2 (SSTR2)-expressing tumors. Here, we present safety and efficacy results from the dose-finding phase 1/2a clinical trial (NCT05636618). Methods: This phase 1/2a dose-escalation study uses a Bayesian algorithm to identify an optimal dose of [ 212 Pb]VMT-α-NET for participants with advanced, well-differentiated NETs expressing SSTR2. Eligible participants must have unresectable or metastatic disease, documented radiologic progression following at least one prior systemic treatment, and evidence of SSTR2 expression (Krenning Score ³2) in ≥1 lesion confirmed on an FDA-approved somatostatin receptor PET scan. Patients previously treated with systemic peptide receptor radionuclide therapy (PRRT) are excluded from enrollment. During the dose-finding phase, participants may receive up to four administrations of [ 212 Pb]VMT-α-NET at their assigned dose level, and they are monitored for dose-limiting toxicities (DLTs) for 42 days and safety throughout the study. Efficacy is evaluated by investigators according to RECIST criteria v1.1. Dosimetry is included as a supportive analysis. Results: As of the 10-Dec-2025 data cut-off (DCO), 56 participants had been enrolled across Cohorts 1, 2, and 3 and received at least one dose of [ 212 Pb]VMT-α-NET: 2 participants in Cohort 1 (2.5 mCi), 46 in Cohort 2 (5 mCi), and 8 in Cohort 3 (6 mCi). No dose-limiting toxicities, grade 5 adverse events, treatment-related discontinuations, serious renal events, dysphagia, or clinically meaningful treatment-related myelosuppression were observed. The median follow-up for all treated participants was 40 weeks (range 6–97). Efficacy has been summarized in this abstract for 25 participants (2 in Cohort 1 and 23 in Cohort 2). Nineteen of these 25 participants remained progression-free at the time of the DCO, with a median efficacy follow-up of 49 weeks (range 6–97). In Cohort 2, investigators observed RECIST v1.1 objective responses in 9 of 23 participants (39%), including 8 confirmed responses. Both participants in Cohort 1 continued to exhibit stable disease after two years of follow-up. Updated safety for all participants and updated efficacy for participants with sufficient maturity beyond the 25 summarized here will be presented at the congress. Conclusions: Treatment with [ 212 Pb]VMT-α-NET continues to demonstrate a favorable safety profile across all treated participants (n = 56) and shows signs of sustained efficacy at the 2.5 mCi and 5 mCi dose levels. The therapy has been well-tolerated with no observed dose-limiting toxicities or serious treatment-related adverse outcomes. Based on these encouraging safety and efficacy findings, the study is ongoing, and enrollment into Cohort 3 (6 mCi) remains active. Clinical trial information: NCT05636618 .

4013 Background: Pancreatic ductal adenocarcinoma (PDAC) is driven predominantly by oncogenic KRAS signaling, yet prior attempts to target the MAPK pathway have been limited by toxicity and emergence of resistance. Atebimetinib is a next-generation, deep cyclic inhibitor (DCI) of MEK designed to achieve pulsatile pathway inhibition with improved tolerability. We report a Phase 2a cohort evaluating atebimetinib in combination with modified gemcitabine/nab-paclitaxel (mGnP) in the first-line treatment of advanced or metastatic PDAC (NCT05585320). Methods: In this multicenter, open-label, nonrandomized Phase 2a study, patients with previously untreated advanced or metastatic PDAC were eligible for participation and were treated with orally administered atebimetinib daily in combination with mGnP (modification: every other week dosing). The primary endpoint was objective response rate (ORR) per investigator assessment using RECIST 1.1 criteria in response-evaluable patients. The cohort was powered (~80%, one sided alpha 0.05) using an optimal Simon’s 2-stage design. Results: Fifty-five patients were enrolled and treated with 320 mg atebimetinib daily + mGnP. ECOG performance status was 0-1 (100%); median age was 68 years (range 42-85 years), 62% were ≥65 years, and 55% were male. No grade 5 treatment-related adverse events (AE) were observed. No AE related to atebimetinib were higher than grade 3. Grade 3 AE related to atebimetinib occurred in 29% of participants, with the most frequent being rash (5%), ALT increased (5%), and AST increased (5%). Among the 50 response-evaluable patients, best overall response (BOR) achieved were 21 PR, 21 SD, 7 PD, 1 NE. The ORR was 42%, and DCR was 84%. With a median follow-up of 10.4 months (data cutoff Jan 7, 2026), 6- and 9-month OS rates were 88% and 79%, respectively; median PFS was 8.3 months and median OS was not reached. Conclusions: Atebimetinib in combination with mGnP demonstrated favorable safety and promising efficacy compared to historic standard-of-care (SoC) GnP. 1 A phase 3 registrational trial is planned to start in mid-2026 for evaluation of atebimetinib + mGnP versus SoC GnP. 1. Von Hoff DD, Ervin T, Arena FP, et al. Increased survival in pancreatic cancer with nab-paclitaxel plus gemcitabine. N Engl J Med . 2013;369(18):1691-1703. doi:10.1056/NEJMoa1304369 (PMID: 24131140). Clinical trial information: NCT05585320 .

3146 Background: The MTA-cooperative PRMT5 inhibitor BMS-986504 selectively binds to the PRMT5-MTA complex, a synthetic lethal target in MTAP -del cancer cells, while sparing MTAP –wild-type normal cells. In the phase 1 CA240-0007 study, BMS-986504 was well tolerated and had antitumor activity in pts with MTAP -del advanced solid tumors (ORR, 23%; median DOR, 10.5 mo). BMS-986504 also demonstrated dose-dependent reductions in plasma symmetric dimethylarginine (SDMA) levels, a PD biomarker of PRMT5 inhibition. Here, we report expanded PD and genomic analyses and initial results of exploratory proteomic analyses from CA240-0007. Methods: Pts with advanced, unresectable, or metastatic solid tumors with homozygous MTAP -del received BMS-986504 (50–800 mg) in 3-wk cycles. Analyses of SDMA levels and proteomic profiles from paired plasma samples (baseline and C2D1) were performed using mass spectrometry and SomaScan 11K, respectively. Tumor genomic information was extracted from local pathology reports. ORR was assessed per RECIST v1.1. Results: Among 78 pts with paired samples at data cutoff (22 Sept 2025; median f/u, 15.9 mo), dose-dependent reductions were seen in plasma SDMA levels, with the greatest reductions at 400 mg QD (median, 60.3%; n = 31) and 600 mg QD (median, 60.3%; n = 23). Proteomic pathway enrichment analysis demonstrated downregulation of PI3K/Akt/mTOR, Myc target, cell cycle, and DNA damage repair (DDR) pathways with BMS-986504 on C2D1 (n = 87), consistent with the expected effect of PRMT5 inhibition; a trend toward deeper modulation of PRMT5-regulated pathways was observed at higher doses. No proteins at baseline (n = 94) were found to be significantly associated with ORR (false discovery rate > 0.05). Responses in clinically evaluable pts with NSCLC (n = 33) or PDAC (n = 35) were observed regardless of EGFR , KRAS , or TP53 status. In pts with NSCLC, STK11 and SMARCA4 alterations were numerically higher in nonresponders. No gene alterations were associated with response in PDAC, except GABRA6 alterations (n = 2, both responders). A trend toward higher ORRs was seen in pts with alterations in DDR genes ( CHEK2, FANCD2, XRCC4 , and/or MSH2 ), consistent with the role of PRMT5 in DDR via regulation of mRNA splicing. ctDNA analysis will be presented. Conclusions: BMS-986504 demonstrated robust dose-dependent PD effects in pts with advanced solid tumors in plasma SDMA and proteomic analyses. Downregulation of PRMT5-regulated pathways was observed with BMS-986504 treatment. Response was seen regardless of EGFR, KRAS , and TP53 alterations in NSCLC or PDAC, and DDR gene alterations were associated with improved response. These findings further support the MOA of BMS-986504 and its ongoing investigation in the phase 2/3 MountainTAP-9, -29, and -30 studies as a potential treatment option in pts with MTAP -del advanced solid tumors. Clinical trial information: NCT05245500 .

SUMMARY Intraductal oncocytic papillary neoplasms (IOPNs) are rare tumors that develop from biliary and pancreatic ductal epithelium and progress to lethal cancers. Human IOPNs and IOPN-associated carcinomas are known to harbor the DNAJB1-PRKACA gene fusion, however the role of the oncogenic fusion in carcinogenesis is poorly understood. We developed a Cre-inducible mouse model of human DNAJB1-PRKACA expression and substantiated that the DNAJB1-PRKACA fusion gene is a bona fide driver of IOPN-associated biliary and pancreatic carcinoma. By analyzing the unique histopathologic and transcriptional changes that occur at each stage of tumor development, we found that these murine tumors closely mimic human DNAJB1-PRKACA driven tumors. Furthermore, we identified that many of the salient features of invasive carcinoma are established at the pre-invasive stage, including evidence of tumor cell metabolic dysregulation, immunosuppressive tumor stroma and expression of genes strongly associated with invasive DNAJB1-PRKACA driven cancer in humans. Finally, we found that Slc16a14 , a characteristic DNAJB1-PRKACA regulated super-enhancer associated gene, serves as a robust biomarker of malignant transformation from IOPN to invasive carcinoma.

Abstract Background Accurate and efficient dose calculation is essential for online adaptive planning in proton therapy. Deep learning (DL) has shown promising dose prediction results for pencil beam scanning proton therapy (PBSPT) in recent years, but existing DL‐based dose prediction methods still suffer from limited generalizability and an inability to effectively handle outlier clinical cases. This may lead to inaccurate dose delivery to targets or excessive irradiation to organs at risk (OARs), thereby compromising the safety and efficacy of online adaptive proton therapy. Purpose To design a physics‐aware and generalizable AI‐based PBSPT dose prediction method that incorporates underlying physics to enhance generalizability, particularly in handling outlier clinical cases. Methods This study analyzed PBSPT plans of 103 prostate (93 for training and 10 for testing) and 78 lung cancer patients (68 for training and 10 for testing) from our institution, with each case comprising CT images and structure sets. Using the doses generated by our Monte Carlo‐based dose engine as the reference standard, we compared three methods: the region of interest (ROI)‐based method, the beam mask and sliding window method, and the proposed noisy probing dose method, which rapidly generates a low‐statistics dose via uniformly weighted spots on an expanded spot‐placement target volume without optimization. To evaluate the generalizability of these methods to rare treatment planning scenarios, 12 cases with uncommon beam angles or prescription doses were used to assess their performance, which was evaluated using dose‐volume histogram (DVH) indices, 3D Gamma passing rates (3%/2 mm/10%), and Dice coefficients for dose agreement, while prediction times were measured to gauge model efficiency. Results The proposed noisy probing dose method consistently outperformed the ROI‐based and beam mask baselines across all evaluated metrics, with more accurate dose agreement and superior generalizability. For DVH indices, the noisy probing dose method achieved the smallest deviation in clinical target volume (CTV) dose coverage: in prostate cancer, CTV D98 deviation was reduced by 45% (from 0.53 ± 0.22 Gy [RBE] for ROI‐based) and 29% (from 0.41 ± 0.28 Gy [RBE] for beam mask) to 0.29 ± 0.06 Gy [RBE]; in lung cancer, similar improvements were observed, with CTV D98 deviations reduced to 0.34 ± 0.12 Gy [RBE]. The 3D Gamma passing rates improved to 99.65% ± 1.15% for prostate targets and 97.04% ± 1.17% for lung targets. The dice coefficients of the 90% iso‐dose lines were also the highest with the noisy probing dose method (prostate: 0.983 ± 0.005; lung: 0.967 ± 0.01). For the 12 outlier cases, the noisy probing dose method maintained superior generalizability, yielding higher 3D Gamma passing rates (prostate targets: 96.79% ± 0.83%, OARs: 94.29% ± 1.01%; lung targets: 93.38% ± 1.34%, OARs: 93.95% ± 1.32%), demonstrating robust generalizability to rare clinical scenarios. The dose predictions for all testing cases were completed within 0.3 seconds. Conclusions A novel noisy probing dose method was proposed for PBSPT dose prediction in prostate and lung cancer patients. By embedding more proton‐specific physics, this method demonstrated an improvement in the generalizability of dose prediction.