Abstract
Background: Cachexia, a systemic wasting syndrome, affects up to 50% of cancer patients and is a known driver of resistance to immunotherapy. Early identification of patients at risk for cachexia is critical for improving clinical outcomes. This study investigates whether pre-treatment $^{18}$F-FDG-PET/CT-based radiomics can predict cachexia and subsequent clinical responses in patients with advanced non-small cell lung cancer (NSCLC) undergoing immunotherapy.
Methods: This retrospective multi-institution study analyzed 210 patients with stage IIIB-IV NSCLC treated with immune checkpoint blockade. A radiomics signature was trained and tested using 175 patients from Moffitt Cancer Center and further validated using an external cohort of 35 patients from the James A. Haley Veterans’ Hospital (VA). The signature's ability to predict cachexia, durable clinical benefit (DCB), progression-free survival (PFS), and overall survival (OS) was evaluated.
Results: The radiomics signature predicted cachexia with an Area Under the Curve (AUC) of 0.77 (training), 0.75 (test), and 0.73 (external validation). It also predicted DCB (AUC 0.66–0.72) and was significantly associated with PFS and OS across all cohorts ($p < 0.05$).
Conclusion: PET/CT radiomics can identify cachexia risk before immunotherapy begins, offering a non-invasive tool to stratify patients and guide early intervention.
Introduction
Cancer cachexia is a complex metabolic syndrome characterized by the progressive loss of skeletal muscle and adipose tissue. Unlike simple starvation, cachexia cannot be fully reversed by conventional nutritional support and accounts for approximately 20% of cancer-related deaths. In the context of immunotherapy, specifically Immune Checkpoint Inhibitors (ICIs), cachexia poses a significant challenge. The metabolic wasting associated with the syndrome is thought to accelerate the clearance of therapeutic antibodies, leading to suboptimal drug exposure and primary resistance.
Currently, cachexia is often diagnosed only after significant weight loss has occurred. There is an urgent clinical need for pre-treatment biomarkers that can identify "cachexia-prone" patients before they start therapy, allowing for earlier nutritional or pharmacological intervention.
Methodology
Radiomics and Habitat Imaging
The researchers utilized Radiomics, a method that extracts high-dimensional quantitative features from medical imaging that are often imperceptible to the human eye. To capture the heterogeneity of the primary tumor, the study employed "habitat imaging." This involved dividing the tumor into distinct metabolic regions (habitats) based on PET and CT intensity levels (e.g., PEThigh, PETlow) using Otsu thresholding.
Feature Selection and Model Training
From these habitats, 1,053 quantitative features—including shape, texture, and intensity—were extracted. To ensure the model's robustness and prevent overfitting, the team used a rigorous four-step selection process:
- ICC Filtering: Retaining features consistent across different segmentations.
- Statistical Testing: Identifying features with significant differences between cachectic and non-cachectic patients.
- Correlation Analysis: Removing redundant, highly correlated features.
- LASSO Regression: Selecting the final nine most predictive features to form the "Radiomics Signature" (RS).
Finally, a nomogram was constructed by combining the RS with clinical variables such as Body Mass Index (BMI), ECOG performance status, and metastasis stage.
Key Insights
The Tumor as a Systemic Mirror
A central finding of this research is that the primary tumor’s metabolic and structural characteristics reflect the systemic metabolic state of the patient. The tumor's radiomic profile can signal the catabolic (breakdown) processes it drives throughout the body, effectively acting as a mirror for the patient's overall metabolic health.
Independence from Tumor Size
The study confirmed that cachexia risk is not merely a function of tumor volume. A small tumor with a specific aggressive radiomic signature can be more metabolically destructive than a larger tumor lacking those features. Furthermore, while the Skeletal Muscle Index (SMI) is a common research metric for cachexia, it failed to significantly predict cachexia in this study’s general cohort, highlighting the superior predictive power of tumor-focused radiomics.
Results and Clinical Implications
Predictive Performance
The Radiomics Signature demonstrated consistent predictive accuracy across all datasets:
- Internal Training Cohort: AUC 0.77
- Internal Test Cohort: AUC 0.75
- External VA Cohort: AUC 0.73
Prognostic Value
The RS proved to be a powerful indicator of how patients would fare on immunotherapy. Patients classified as "High RS" (predicted cachexia) experienced significantly worse outcomes:
- Progression-Free Survival (PFS): Median PFS was 4–6 months for the High RS group compared to 11–17 months for the Low RS group.
- Overall Survival (OS): Median OS was approximately 19 months for the High RS group, while the median was not even reached in the Low RS group during the study period.
- Durable Clinical Benefit (DCB): The signature identified patients likely to benefit from therapy for longer than six months with an AUC of up to 0.72.
Limitations and Future Directions
While the results are promising, the study's retrospective nature means the findings should be validated in prospective clinical trials. Additionally, the external validation cohort was relatively small ($N=35$), and some historical data regarding weight loss prior to the study was unavailable.
Moving forward, this radiomics signature could be used to stratify patients in clinical trials. Patients identified as high-risk for cachexia could be prioritized for early nutritional support or anti-inflammatory therapies alongside their immunotherapy, potentially improving both treatment response and quality of life.
Summary updated