Despite promising data showing that circulating tumour DNA (ctDNA) dynamics during treatment can inform real-time tumour response and recurrence risk1, how best to translate these insights into actionable clinical decision-making remains unclear. Here we report results from the EP-STAR trial-a multi-centre, ctDNA-driven, risk-adapted, non-randomized phase II study ( NCT04072107 ; ClinicalTrials.gov) testing whether a risk-adaptive treatment (RAT) strategy guided by on-treatment ctDNA dynamics can meaningfully improve survival, using nasopharyngeal carcinoma as a model. Eligible patients were enrolled and began treatment with standard-of-care gemcitabine-cisplatin neoadjuvant chemotherapy (GP-NAC; the P in this abbreviation stands for platinum)2, followed by RAT or standard-of-care chemoradiotherapy guided by ctDNA clearance trajectory during GP-NAC. Protocol-eligible patients who did not receive RAT, drawn from a prospectively registered ctDNA biomarker cohort ( NCT03855020 )3, served as a non-randomized, contemporaneous no-RAT external cohort. The primary end-point was failure-free survival (FFS) in the RAT group. After a median follow-up of 47.3 months, the 3-year FFS was 89.1% (83.2-95.0%) in the RAT group (n = 110). Patients who received RAT showed significantly improved FFS (P = 0.003, log-rank test) compared with the no-RAT external cohort (hazard ratio = 0.41 [0.23-0.75]; P = 0.004, Cox regression model). The RAT strategy was well-tolerated with no treatment-related deaths. Collectively, these data show that a ctDNA-driven RAT paradigm could be a promising strategy to improve survival, challenging the conventional fixed-course, static treatment approach.

Cancer survival analysis commonly integrates information across diverse medical modalities to make survival-time predictions. Existing methods primarily focus on extracting different decoupled features of modalities and performing fusion operations such as concatenation, attention, and Mixture-of-Experts (MoE)-based fusion. However, these methods still face two key challenges: 1) fixed fusion schemes (concatenation and attention) can lead to model over-reliance on predefined feature combinations, limiting the dynamic fusion of decoupled features; and 2) in MoE-based fusion methods, each expert network handles separate decoupled features, which limits information interaction among the decoupled features. To address these challenges, we propose a novel Decoupling-Reorganization-Fusion framework (DeReF), which devises a random feature reorganization strategy between modalities decoupling and dynamic MoE fusion modules. Its advantages are: 1) it increases the diversity of feature combinations and granularity, enhancing the generalization ability of the subsequent expert networks; and 2) it overcomes the problem of information closure and helps expert networks better capture information among decoupled features. Additionally, we incorporate a regional cross-attention network within the modality decoupling module to improve the representation quality of decoupled features. Extensive experimental results on our in-house Liver Cancer (LC) and three widely used public datasets from The Cancer Genome Atlas (TCGA) confirm the effectiveness of our proposed method. Codes are available at https://github.com/ZJUMAI/DeReF.

PURPOSES:To develop a deep learning model for automated metabolic tumor volume (MTV) delineation on routine computed tomography (CT) without positron emission tomography (PET) and to validate its prognostic value in nasopharyngeal carcinoma (NPC). METHODS AND MATERIALS:A retrospective cohort of 392 patients with NPC undergoing pre-radiotherapy 2-deoxy-2-[fluorine-18]fluoro-D-glucose PET/CT in 2021 was enrolled and randomly divided into training (n = 314, including 63 for validation) and test (n = 78) cohorts. Ground-truth MTV (GT_MTV) was generated from PET-registered CT using standardized uptake value SUV > 2.5 within the primary gross tumor volume. A 7-layer MedNext-Insight model with dual-window CT inputs and Dice-Focal loss was trained to predict MTV using CT alone. Segmentation performance was compared with no-new-U-Net version 2 (nnUNetV2), Conditional Generative Adversarial Network for Image-to-Image Translation (Pix2Pix), and three-dimensional Cycle-Consistent Generative Adversarial Network (3D-CycleGAN) p rimarily using Dice Similarity Coefficient and sensitivity. Radiomic features were extracted from predicted MTV (Pred_MTV) and GT_MTV to construct Cox proportional hazards models for event-free survival, evaluated by concordance index (C-index). Robustness was further assessed in an internal temporal validation cohort from 2022 with different scanners (n = 135) using planning CT as the sole input. RESULTS:MedNext-Insight achieved the highest MTV delineation Dice Similarity Coefficient (Mean ± SD, 0.808 ± 0.110 vs 0.740-0.782; all P < .05) with improved sensitivity. After excluding 12 patients with baseline distant metastasis from event-free survival analysis, Pred_MTV-derived radiomics showed strong reproducibility (median intraclass correlation coefficient, 0.816) and comparable prognostic performance to GT_MTV (C-index [95% CI], 0.712 [0.516-0.899] vs 0.744 [0.601-0.884]; P = .730). MTV-based radiomics outperformed primary gross tumor volume-derived features, particularly when combined with clinical variables (C-index, 0.809 [0.678-0.919]). In the internal temporal validation cohort, CT-only Pred_MTV maintained stable segmentation accuracy and prognostic discrimination (log-rank test, P < .05). CONCLUSION:MedNext-Insight enables accurate PET-free MTV delineation on routine CT with prognostic value, supporting a resource-efficient approach for risk stratification and informing potential future biology-guided adaptive radiation therapy in NPC.

Yield shocks, characterized by abrupt reductions in crop yields, represent a significant threat to food supply and global food security. Climate change and extreme weather events are critical drivers of yield shocks. Here, climate change refers specifically to changes in growing-season temperature, precipitation and related extremes. In this study, we employed data-driven random forest model to simulate the relationships between yield shocks and multiple climate- and extreme-related variables, and applied a SHAP-based attribution framework to quantify the relative contributions of different climatic stressors to yield shocks in winter wheat, spring wheat, soybean, and maize. The results show that during the historical period (1982–2016), precipitation stress was the dominant climatic driver, accounting for 42.3%%, 28.9%, 42.6%, and 59.1% of the yield shock-affected areas for winter wheat, spring wheat, soybean, and maize, respectively. The area fractions of yield shocks driven by the synergistic effects of temperature and precipitation were 37.6%, 23.4%, 26.4%, and 35.4% for the four crops. These findings are further supported by simulations from the 12 crop models. Under future climate scenarios, compared with the historical period, both the frequency and spatial extent of crop yield shocks are projected to increase markedly during 2041–2070 and 2071–2100, particularly under the high-emission pathway (SSP5–RCP8.5). In addition, the dominant climatic stressors are expected to shift from precipitation-related to temperature-related factors, and the area fractions of yield shocks driven by the synergistic effects of temperature and precipitation will further increase relative to the historical baseline. Our findings highlight the importance of addressing climate stresses when tackling the food security challenges posed by crop yield shocks.

Accurate detection of individual trees is essential for urban forest management and ecological assessment, yet remains challenging due to the heterogeneous backgrounds, variable sizes of tree crowns, and significant variations across urban scenarios. To address these issues, we propose Tree-SAM, a city-scale individual tree detection architecture built upon the visual foundation model Segment Anything Model (SAM) and equipped with three task-specific modules, i.e., Cross-Correlation Feature Backbone (CCFB), Hierarchical Instance Aggregation Neck (HIAN), and Context-Aware Adaptation Head (CAAH). These modules synergistically fuse general semantics with fine-grained structural cues, enable multi-scale feature aggregation, and adaptively refine predictions based on specific scene contexts. On the GZ-Tree Crown dataset, Tree-SAM achieves F1-scores of 0.762, 0.732, and 0.830, with corresponding AP@50 values of 0.478, 0.454, and 0.526 in the forest, mixed, and urban scenarios, respectively, consistently ranking first across all scenes and demonstrating strong adaptability to diverse intra-city landscapes. Additional evaluations on the BAMFORESTS dataset and the SZ-Dataset further confirm its robustness across varied geographic contexts. Tree-SAM provides a reliable, automated framework for large-scale urban tree mapping, providing reliable data support for urban forest management, carbon stock estimation, and ecological assessment.

Multimodal survival prediction, a crucial yet challenging task, demands the integration of multimodal medical data (Whole Slide Images (WSIs) and Genomic Profiles) to achieve accurate prognostic modeling. Given the inherent heterogeneity across modalities, the feature decoupling-fusion paradigm has emerged as a dominant approach. However, these methods have the following shortcomings: (1) fail to reduce the redundant information of modality features before decoupling, which negatively affects the feature decoupling and fusion effect;(2) lack the ability to model the fine-grained relationships of the features and capture the local information interactions between intra- and inter-modality features. To address these issues, we propose a Hierarchical Decoupling-Fusion Mixture-of-Experts (HDMoE) framework with two levels of MoE and Random Feature Reorganization (RFR) modules.In the first-level MoE, shared experts and routed experts are employed to remove redundant information and extract fine-grained specific features within each modality, while the second-level MoE facilitates fine-grained inter-modality feature decoupling. Besides, we design two RFR modules following each level of MoE to finely fuse intra- and inter-modality features, which can help the model capture more fine-grained relationships between modalities. Extensive experimental results on our private Liver Cancer (LC) and three TCGA public datasets confirm the effectiveness of our proposed method. Codes are available at https://github.com/ZJUMAI/HDMoE.

Genetic mutations are clinically significant biomarkers that guide cancer diagnosis and treatment. Predicting genetic mutations from whole slide images (WSIs) provides a cost-effective alternative to traditional genetic testing, but existing methods relying on multiple binary classifiers are inefficient in modeling intrinsic biological relationships between genes and inevitably suffer from class imbalance. We present the Biological-knowledge-enhanced PathGenomic multi-label Transformer (BPGT), the first multi-label framework for genetic mutation prediction from WSIs that explicitly incorporates intrinsic biological knowledge to guide feature learning. BPGT jointly models inter-gene dependencies and spatial pathology features via: (1) The gene encoder constructs biologically informed gene priors through two carefully designed sub-modules: (a) A gene graph whose node features combine the genes’ linguistic descriptions and cancer phenotypes, and whose edges are defined by pathway associations and mutation consistencies. (b) A knowledge association module fuses linguistic and biomedical knowledge into gene priors via transformer-based graph representation learning, capturing the intrinsic relationships among mutations of different genes. (2) The label decoder integrates these knowledge-driven gene priors with spatially relevant WSI regions via a modality fusion mechanism, and employs a comparative multi-label loss to improve discrimination between mutation profiles. These designs enable BPGT to address label imbalance, capture co-mutation patterns, and leverage non-visual domain knowledge in an end-to-end learning paradigm. We validate BPGT across nine cancers from TCGA and two from CPTAC, comprising over and 48 million image patches. Across diverse cancers and genes, BPGT consistently outperforms state-of-the-art methods in genetic mutation prediction accuracy and generalization.

Amid the rapid expansion of artificial intelligence (AI) and growing concerns over ecological constraints, understanding the environmental consequences of emerging technologies has become increasingly important. This study investigates the nonlinear relationship between AI and environmental sustainability, which is proxied by the Load Capacity Factor (LCF), while considering the moderating effect of trade openness. Based on panel data from OECD countries between 1993 and 2023, the empirical results indicate a significant U-shaped relationship between AI and LCF. Specifically, AI initially reduces LCF but enhances environmental sustainability after exceeding a certain threshold. Moreover, trade openness is shown to flatten the U-shaped curve and shift its turning point. These findings provide meaningful policy insights for integrating digital transformation with sustainable development objectives.

Groundwater arsenic contamination poses a widespread yet insufficiently quantified threat to global drinking water safety and public health. In this study, machine learning-based groundwater arsenic predictions were integrated with a human health risk assessment (HHRA) framework to assess global non-carcinogenic risks from chronic ingestion of arsenic-contaminated groundwater. The hazard quotient (HQ), a measure of non-carcinogenic risk, was estimated and populations potentially exposed to HQ > 1 were identified. High-risk regions were concentrated in Southern Asia, Eastern Asia, and Central America. Bangladesh, Pakistan and Argentina are among the most severely affected countries. Approximately 350.6 million individuals (95% CI: 347.1-354.1 million) face elevated risks (HQ > 1), with 81.4% concentrated in Southern Asia. This health burden is unevenly distributed across demographic population groups, with higher estimated risks for women under sex-specific HHRA parameterization, rural communities reliant on untreated groundwater, and populations in low- to medium- human development index (HDI) countries. Using SHapley Additive exPlanations (SHAP) analysis, chronic arsenic exposure was identified as a potential environmental risk indicator for cardiovascular disease burden. These findings reveal stark global inequities in health risks associated with drinking arsenic-contaminated groundwater and underscore the urgent need for targeted risk mitigation, safe water provision, and equity-sensitive policies to achieve sustainable development.

BACKGROUND:Nasopharyngeal carcinoma is an aggressive malignancy originating from the nasopharyngeal mucosa and associated with genetic factors. Many nasopharyngeal carcinoma susceptibility loci have been identified by genome-wide association studies (GWASs), but their underlying functional insights are largely unexplained. RESULTS:A meta-GWAS including 5073 nasopharyngeal carcinoma patients and 5860 controls from nasopharyngeal carcinoma endemic areas identifies a total of 863 significant SNPs, including SNPs at a novel locus 3p24.1 (rs56365817; nearby genes: CMC1/EOMES). By integrating the GWAS signals with single-cell and bulk profiles, we find nasopharyngeal carcinoma susceptibility robustly associated with T cells in different methods and datasets. In nasopharyngeal carcinoma-associated cell type, we identify 234 putative susceptibility genes (81.62% of them novel), mainly enriched in immune-related biological processes. Five putative causal genes are prioritized. We perform in-depth bioinformatic analysis and functional experiments for EOMES, finding that the nasopharyngeal carcinoma-risk alleles of four functional SNPs upregulate EOMES expression by promoting the activity of regulatory elements in T cells, and EOMES participates in nasopharyngeal carcinoma tumorigenesis via regulation of CD8+ T cell exhaustion in the tumor microenvironment. CONCLUSIONS:This study uncovers novel nasopharyngeal carcinoma susceptibility genes and their functional cell types, which improves the understanding of nasopharyngeal carcinoma genetic etiology.

Hepatocellular carcinoma (HCC) early identification is crucial for improving patient outcomes. Current screening methods are often complex and costly. This study developed a simplified, cost-effective HCC screening model using serum marker data. A diverse study population from two Chinese hospitals was recruited, including cancer patients, hospital patients, and healthy individuals. A two-stage screening model was created: LASSO logistic regression for preliminary screening, followed by logistic regression incorporating alpha-fetoprotein (AFP). The model's performance was evaluated in multiple cohorts. Across five populations, the model showed strong performance with AUC-ROC ranging from 0.868 to 0.907, accuracy between 87.43% and 96.96%, and sensitivity over 75% with specificity above 90%. Compared with solely AFP models, the second-stage model improved HCC risk estimates in healthy populations, with significantly higher AUC (0.930 vs. 0.827) and net reclassification improvement (NRI) up to 56.2%. This two-stage model offers a practical, cost-efficient tool for early HCC detection, addressing a significant public health need.

To quantify morphological and dosimetric variations in nasopharyngeal carcinoma (NPC) radiotherapy via autosegmented fan-beam computed tomography (FBCT) and to inform decision-making regarding appropriate objectives and optimal timing for adaptive radiotherapy (ART). This retrospective study analyzed 23 NPC patients (681 FBCT scans) treated at Sun Yat-sen Cancer Center from August 2022 to May 2024. The inclusion criterion was as follows: ≥1 weekly FBCT via a CT-linac with ≤ 2 fractions between scans. Four deep learning-based autosegmentation models were developed to assess weekly volume, Dice similarity coefficient (DSC), and dose variations in organs at risk (OARs) and target volumes. A systematic review of autosegmentation on FBCT scans demonstrated satisfactory accuracy overall, and missegmentation was manually modified. Linear decreases in volume and/or DSC were observed in the parotid glands, submandibular glands, thyroid, spinal cord, and target volumes (R² > 0.7). The linear dose variation included coverage of the low risk planning target volume (-3.01

Under climate warming, effective regulation of urban thermal environment has become a major challenge, which depends on fine-grained identification of spatial heterogeneity. Although land surface temperature offers good spatial continuity, its limited spatial resolution and inability to accurately reflect human thermal perception constrain its applicability in fine-grained thermal environment assessments. This study introduces an innovative analytical framework to address this gap. We first employed the SOLWEIG microclimate model, integrating meteorological data with high-resolution 3D urban morphology, to simulate the mean radiant temperature (Tmrt) at a 1-m resolution in Mumbai. Using a local spatial autocorrelation analysis, we identified statistically significant hot spots (HSs) and cold spots (CSs), then establishing a quadrantal classification framework to explore spatial differentiation of urban thermal environment types. Furthermore, we combined a Structural Equation Model (SEM) to test hypothesized linear pathways with an interpretable machine learning model to explore complex, non-liner interactions among driving factors. Our results show that 30.63 % of urban areas is identified as significant HSs, with 35.4 % classified as "overall high-temperature" areas, primarily driven by high building density and a lack of shading. The analysis reveals that while tree shading is the dominant cooling factor (total effect in SEM = 0.94), its efficacy is non-linear, diminishing in dense urban canyons. Moreover, built environment exerts a compound effect through both heat accumulation and shading. This study recommends implementation of vertically layered greening networks and patch-based cooling strategies to enhance fine-grained thermal regulation and heat risk management in urban areas.

BackgroundInfection is a leading cause of mortality in idiopathic inflammatory myopathies (IIMs). This study aimed to develop a nomogram for predicting severe infection risk in IIM patients.MethodsPatients with IIMs admitted to Zhongshan Hospital, Fudan University, from January 2015 to January 2022 were enrolled. They were randomly divided into derivation (70%) and validation (30%) sets. Univariate and multivariate Cox regression identified independent risk factors for severe infection, and the Akaike information criterion (AIC) was applied for model selection. A nomogram was constructed to predict severe infection risks at 6 months, 1 year, and 3 years. Predictive accuracy and discriminative ability were evaluated using the concordance index (C-index), calibration curves, and the area under the receiver operating characteristic curve (AUC). Decision curve analysis (DCA) assessed clinical utility. Kaplan-Meier (K-M) curves were used to analyze survival differences between high- and low-risk groups stratified by nomogram scores.ResultsAmong 263 IIM patients, 81 experienced 106 severe infection events, with lower respiratory tract infections being the most common (47.2%). Independent risk factors included age at onset (HR 1.024, 95% CI 1.002-1.046, p=0.036), lactate dehydrogenase (HR 1.002, 95% CI 0.999-1.005, p=0.078), HRCT score (HR 1.004, 95% CI 1.001-1.006, p=0.002), and lymphocyte count (HR 0.48, 95% CI 0.23-0.99, p=0.048). The nomogram demonstrated strong predictive performance, with AUCs of 0.84, 0.83, and 0.78 for 6 months, 1 year, and 3 years in the derivation set, and 0.91, 0.77, and 0.64 in the validation set. Calibration curves showed good agreement between predicted and observed risks, while DCA demonstrated significant net benefit over individual predictors. Kaplan-Meier curves revealed significant differences in the cumulative risk of severe infection between high- and low-risk groups. Further validation in DM and ASS subgroups demonstrated that the nomogram effectively predicted severe infections, with AUCs of 0.86, 0.81, and 0.73 for DM and 0.86, 0.83, and 0.74 for ASS at 6 months, 1 year, and 3 years, respectively.ConclusionWe have developed a new nomogram to predict severe infection risk in IIM patients at 6 months, 1 year, and 3 years. This model aids clinicians and patients in formulating treatment and follow-up strategies.

Emerging evidence reveals that microbiota plays a crucial role in multiple cancers. Nasopharyngeal carcinoma (NPC) tissues harbour microbiota, highlighting the need to investigate the clinical implications of tissue-resident microbiota in the development of NPC. Here, we aim to clarify the specific profile of tissue-resident microbiota and its influence on NPC outcomes. This retrospective study included 491 NPC patients from Sun Yat-sen University Cancer Center (Guangzhou, China) and the Affiliated Hospital of Guilin Medical College (Guilin, China). We profiled the microbial composition of 343 NPC and 36 normal nasopharyngeal tissues through sequencing of the genes encoding the 16S rRNA subunit of bacterial ribosomes. There were significant differences in microbial composition, alpha diversity (Shannon index, P = 0.007; Simpson index, P = 0.036), and beta diversity (Bray–Curtis distance: R2 = 0.016, F = 5.187, P = 0.001; unweighted UniFrac distance: R2 = 0.017, F = 5.373, P = 0.001) between NPC and normal nasopharyngeal tissues. A bacterial signature comprising four risk bacterial genera, including Bacteroides, Alloprevotella, Parvimonas, and Dialister, was constructed in the training cohort (n = 171). Patients in the high-risk group had shorter disease-free (HR 2.80, 95

LBA6003 Background: This study aimed to assess the efficacy and safety of toripalimab combined with induction chemotherapy and radiotherapy alone in locoregionally advanced nasopharyngeal carcinoma (LANPC). Methods: Patients with non-metastatic T4N1 or N2–3 (AJCC 8 th edition) NPC were recruited from 13 centers in China from Aug 2021 to Jul 2022 and randomly assigned (1:1) to receive either toripalimab plus gemcitabine-cisplatin induction chemotherapy and concurrent cisplatin-radiotherapy ( standard arm) or standard therapy sparing concurrent cisplatin ( cisplatin-free arm). Toripalimab was administered at a dosage of 240 mg once every 3 wks for up to 17 cycles (1.06 year), covering the induction (×3 cycles), radiotherapy (×3), and adjuvant (×11) phases. The trial would be considered positive if both coprimary endpoints, failure-free survival (FFS; non-inferiority) and the incidence of all-grade vomiting (superiority), were significantly met, maintaining a 1-sided type I error of 5% without α splitting. A total of 532 patients were needed to achieve 80% power to detect a HR of 1.74, with non-inferiority defined as the lower limit of the 1-sided 95% CI for the difference in 3-year FFS greater than -8%. Quality of life (QoL) was assessed based on EORTC and FACT systems. Tolerability was measured by PRO-CTCAE questionnaires. Results: After a median follow-up of 36 mo, intention-to-treat analysis in 532 patients (266 vs 266) showed that the estimated 3-year FFS was 88.3% in the cisplatin-free arm and 87.6% in the standard arm, with a difference of 0.7% (1-sided 95% CI, -4.8% to ∞; p non-inferiority = 0.002); the stratified HR was 0.92 (95% CI, 0.66 to 1.79; log-rank p = 0.731). The incidence of all-grade vomiting in safety dataset was 25.6% (68/260) in cisplatin-free arm and 69.0% (156/261) in standard arm (χ 2 p < 0.001); the incidence of grade 3–4 vomiting, 3.8% vs 10.3%. Acute grade 3–4 adverse events (AEs) occurred in 136 (52.3%) and 166 (63.6%) patients, including immune-related AEs in 13 (5.0%) and 22 (8.4%) patients, in the cisplatin-free and standard arms, respectively. No treatment-related death was observed. Compared to standard arm, cisplatin-free arm had significantly better QoL in global health status, physical function, role function, nausea/vomiting, constipation, swallowing, sexuality, and H&N total score, as well as higher tolerability to nausea, vomiting, constipation, and fatigue during radiotherapy. Conclusions: Removing concurrent cisplatin from toripalimab plus chemoradiotherapy provides comparable survival, lower toxicity, and better QoL and tolerability for patients with LANPC. Clinical trial information: NCT04907370 . 3-yr survival (%) Cisplatin-free arm ( n = 266) Standard arm ( n = 266) p non-inferiority OS 96.1 96.5 < 0.001 LRRFS 92.9 93.6 0.001 DMFS 93.2 91.6 < 0.001

Bodyweight loss is a common occurrence in Nasopharyngeal Carcinoma (NPC) patients during Radiotherapy (RT). Previous studies found that the prognostic value of percentage weight loss (pWL) during RT is not credible. We aimed to develop a novel progression predictor surrogated to pWL by modelling all bodyweight records measured during the treatment interval. This retrospective study included two independent hospitals of 624 patients. The Predicted Progression Probability (PPP) was obtained from deep learning-guided differential equation solution, model by the patient’s age, sex, body height, and the weekly measured bodyweight records. The performance of PPP in predicting disease progression was assessed, its association with prognosis and adjuvant chemotherapy response was evaluated. The PPP was learnt from the training cohort (N = 257) with 7 weeks of bodyweight records. The prediction performance was validated with 367 patients of the testing cohort sub-divided according to the number of bodyweight records found. The area under of curve for patients with 7 weeks (N = 155), 6 weeks (N = 176), and 5 weeks bodyweight records (N = 32) were 0.76, 0.73, and 0.95 respectively. PPP was significantly associated with progression-free and remained an independent prognostic factor adjusting for clinicopathologic variables in multivariate analysis in all study cohort (adjusted hazard ratio [HR] range: 2.50–7.04, all p < 0.001). Patients with high-PPP derived progression benefit from adjuvant chemotherapy (HR: 0.41–0.54, all p < 0.03), whereas those with low-PPP did not for both cohorts. The trajectory of bodyweight change during RT is more robust than the pWL to give a progression prediction after RT. The PPP is a reliable predictor for estimating the risk of residual diseases after RT course, which also helps to predict adjuvant chemotherapy response in locally advanced NPC patients.