Predicting Heterogeneous Response to Neoadjuvant and Immunotherapy

18 concepts17 papers1 proposal

The central challenge of modern oncology is that cancer is not a uniform disease, nor do patients possess uniform biology. Two individuals diagnosed with the exact same stage of lung, gastric, or rectal cancer can receive identical, state-of-the-art treatments—such as immune checkpoint inhibitors (ICIs) or neoadjuvant chemoradiotherapy (nCRT)—and experience completely divergent outcomes. One patient may achieve a pathological complete response (pCR), where the tumor is entirely eradicated before surgery, while another may derive zero therapeutic benefit while suffering severe, systemic toxicities.

To move away from a "one-size-fits-all" paradigm, a major research direction has emerged focused on predicting this treatment response heterogeneity before therapy begins. By leveraging machine learning to fuse macro-scale clinical imaging, micro-scale cellular pathology, and systemic host biomarkers, researchers are building models that explain why certain tumors resist therapy and how clinicians can accurately map the optimal therapeutic path for individual patients.

Systemic Biomarkers: The Host Macro-Environment

Therapeutic response does not depend solely on the genetic mutations within a tumor; it is heavily mediated by the patient’s systemic physiological state. The host's metabolic, inflammatory, and oxidative baselines dictate how drugs circulate and how effectively the immune system can be mobilized.

Metabolism and Body Composition

In a striking example of host-level influence, Yue An et al. conducted a large-scale meta-analysis in Association between body mass index and survival outcomes for cancer patients treated with immune checkpoint inhibitors: a systematic review and meta-analysis. They validated an "obesity paradox," demonstrating that a high Body Mass Index (BMI) is significantly associated with improved survival and delayed disease progression in patients treated with ICIs, notably without increasing immune-related adverse events. This suggests that systemic metabolic reserves play a critical role in sustaining prolonged immune responses against tumors.

Inflammatory and Oxidative Stress Indices

Routine blood tests can be transformed into powerful predictive tools by calculating composite scores that reflect systemic equilibrium. For instance, Zi‐Ning Liu et al. demonstrated in Predictive effect of the systemic inflammation response index (SIRI) on the efficacy and prognosis of neoadjuvant chemoradiotherapy in patients with locally advanced rectal cancer that the Systemic Inflammation Response Index (SIRI) serves as a robust pre-treatment predictor of long-term survival and nCRT efficacy.

Similarly, the host's oxidative state can indicate whether the microenvironment is receptive to immunochemotherapy. Ji‐Feng Feng et al. developed the Systemic Oxidative Stress Index (SOSI) in The systemic oxidative stress index predicts clinical outcomes of esophageal squamous cell carcinoma receiving neoadjuvant immunochemotherapy, combining multiple routine blood markers to successfully identify high-risk individuals who fail standard pre-surgical protocols.

Baseline Organ Function

Even standard physiological metrics can reveal a patient's capacity to withstand or respond to aggressive regimens. In Pulmonary function test-related prognostic models in non-small cell lung cancer patients receiving neoadjuvant chemoimmunotherapy, Min Zhang et al. combined pulmonary function tests (PFTs) with clinical data via a ridge regression model. Their framework successfully identified patients at high risk for early disease recurrence following neoadjuvant chemoimmunotherapy, proving that baseline organ function provides critical context for post-surgical outcomes.

Molecular and Cellular Drivers of Resistance

While systemic markers outline the host environment, the intrinsic characteristics of the tumor cells and their immediate microenvironment dictate the cellular mechanisms of resistance.

Genomic Mutations and Stemness

At the genetic level, specific mutations can fundamentally disrupt cell death pathways, rendering standard localized therapies ineffective. Miao Xue et al. demonstrated in Prognostic Value of TP53 Mutation for Transcatheter Arterial Chemoembolization Failure/Refractoriness in HBV-Related Advanced Hepatocellular Carcinoma that TP53 mutations serve as an independent predictor of transcatheter arterial chemoembolization (TACE) failure, likely by short-circuiting apoptosis and growth signaling pathways.

Furthermore, the presence of undifferentiated cell populations can drive persistent tumor regeneration. Kazuaki Hiroishi et al. established in Cancer stem cell-related factors are associated with the efficacy of pre-operative chemoradiotherapy for locally advanced rectal cancer that the cancer stem cell markers CD133 and CD24 are highly correlated with resistance to pre-operative chemoradiotherapy, explaining why universal treatment approaches fail when stem-like cells evade destruction.

Immune Microenvironment and Liquid Biopsies

The composition of the immune cells infiltrating the tumor heavily influences immunotherapy outcomes. As noted by Mengyue Liu et al. in Antitumor Activity of Ruditapes philippinarum Polysaccharides Through Mitochondrial Apoptosis in Cellular and Zebrafish Models, variations in individual immune profiles fundamentally complicate the development of universally effective agents. To map this variability, Qun Cai et al. constructed an 8-gene immune-related prognostic risk score in Prognostic model of immune-related genes for patients with hepatocellular carcinoma, allowing clinicians to characterize the tumor microenvironment and stratify patient survival before initiating immunotherapy.

To capture these cellular dynamics without invasive tissue biopsies, researchers are turning to liquid biopsies. Haiyan Chen et al. isolated plasma-derived small extracellular vesicles (sEVs) in Characterization and proteomic analysis of plasma-derived small extracellular vesicles in locally advanced rectal cancer patients. By analyzing these circulating cellular messengers, the authors developed a five-gene proteomic signature that accurately forecasts chemoradiotherapy efficacy and long-term survival.

AI and Multimodal Fusion: Bridging Macro and Micro Scales

The true breakthrough in predicting response heterogeneity lies in multimodal data fusion. Single-modality approaches (e.g., using only radiology or only genomics) often fail because they capture only one facet of the disease. Macro-scale imaging (radiomics) captures the spatial architecture and gross heterogeneity of the entire tumor mass but lacks cellular resolution. Micro-scale pathology (pathomics) captures cellular morphology and spatial immune infiltration but lacks the three-dimensional context of the whole tumor. Machine learning models act as the mathematical bridge between these scales.

[Macro-Scale Radiology] (3D MRI, Spatial Structure) 
                       \
                        +---> [Deep Learning Multimodal Fusion] ---> Accurate Response Prediction
                       /
[Micro-Scale Pathology] (Whole Slide Images, Proteomics)

Radiopathomic Synergy

Lizhi Shao et al. pioneered this multi-scale integration in Multiparametric MRI and Whole Slide Image-Based Pretreatment Prediction of Pathological Response to Neoadjuvant Chemoradiotherapy in Rectal Cancer: A Multicenter Radiopathomic Study. By combining pre-treatment MRI data with AI-extracted features from digital biopsy slides, their "radiopathomic" signature achieved an accuracy of up to 87.66% in predicting response to chemoradiotherapy, demonstrating that the combination of macro and micro features is significantly more predictive than either modality alone.

Advanced Deep Learning Architectures

To optimize how different imaging sequences interact, a deep learning framework detailed in Attention mechanism based multi-sequence MRI fusion improves prediction of response to neoadjuvant chemoradiotherapy in locally advanced rectal cancer introduced a "channel attention" mechanism. This architecture automatically weights the most informative MRI sequences, outperforming traditional radiomics by focusing the model's attention on the specific imaging channels that capture tumor regression patterns.

On the pathology side, Dan Han et al. utilized weakly supervised deep learning on standard pre-treatment pathology slides in Whole slide image-based weakly supervised deep learning for predicting major pathological response in non-small cell lung cancer following neoadjuvant chemoimmunotherapy: a multicenter, retrospective, cohort study. Their model automatically extracts predictive pathomics signatures across entire whole-slide images without requiring tedious, manual cell-by-cell annotations by pathologists, making it highly scalable for clinical workflows.

Integrating Proteomics with Imaging

Taking fusion to the molecular level, Yan Li et al. developed a multimodal framework in Deep Learning-Based Multimodal Prediction of NAC Response in LARC by Integrating MRI and Proteomics that integrates 3D MRI with mass-spectrometry-based proteomic data. Crucially, they engineered a "zero-imputation" design. In real-world clinical settings, advanced proteomic data is frequently missing due to cost or tissue constraints; this architecture allows the deep learning model to remain functional and highly accurate even when the molecular data stream is incomplete, solving a major barrier to clinical adoption.

Algorithmic Optimization and Real-World Validation

For these predictive models to guide clinical decisions—such as opting for a non-operative "watch and wait" strategy versus immediate radical surgery—they must be rigorously optimized, interpretable, and validated across diverse populations.

Model Selection and Interpretability

The mathematical combination of algorithms matters immensely when dealing with high-dimensional survival data. The authors of Performance of AI-based machine learning models for overall survival prediction in advanced hepatocellular carcinoma patients receiving immunoradiotherapy screened 101 different machine learning combinations, ultimately identifying a specific "StepCox + Ridge" model that maximized predictive accuracy for patients undergoing complex combined immunoradiotherapy.

To make these complex mathematical outputs actionable for clinicians at the bedside, researchers often translate machine learning models into visual nomograms. Hao Cui et al. successfully implemented this in Development and validation of nomogram for predicting pathological complete response to neoadjuvant chemotherapy and immunotherapy for locally advanced gastric cancer: a multicenter real-world study in China. Their multicenter study integrated clinical features into a simplified nomogram, allowing doctors to visually calculate the precise probability of a gastric cancer patient achieving complete remission before initiating therapy.

Cross-Continental Validation

The ultimate test of any predictive model is its generalizability across different healthcare systems and demographics. Zoltán Kiss et al. addressed this in 1294 External validation of machine learning models to predict efficacy and toxicity of immune checkpoint inhibitors using real-world pan cancer cohorts by training machine learning models on clinical data from the United States and externally validating them on a large, independent cohort from Germany. Their success in predicting both patient survival and treatment-induced toxicity proved that real-world, heterogeneous clinical data can indeed generate robust, globally generalizable tools.

By combining host physiology, tumor genomics, and multi-scale AI fusion, this research direction provides the predictive framework necessary to transform cancer care from a reactive protocol into a proactive, personalized strategy.

History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_treatment_response_heterogeneity_20bffca6,
  title = {treatment response heterogeneity},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/treatment-response-heterogeneity/rdr_20bffca67443440184baafb939701ced},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 1, frozen text):

@misc{lacuna_treatment_response_heterogeneity_20bffca6_v1,
  title = {treatment response heterogeneity},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/treatment-response-heterogeneity/rdr_20bffca67443440184baafb939701ced/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-19},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake