Spatiotemporal Framework Aims to Reshape Precision Care for Gastric and Colorectal Cancers

Gastric and colorectal cancers remain among the most lethal malignancies worldwide, and a persistent problem in their management has been the reliance on static snapshots of disease. A newly published review in LabMed Discovery argues that this single-time-point approach is fundamentally mismatched to the biology of gastrointestinal tumors, which evolve continuously in space and time. Researchers from Shanghai Jiao Tong University Journal Center have now outlined a comprehensive framework that integrates spatial multi-omics profiling, longitudinal liquid biopsy monitoring, and artificial intelligence-driven decision support into a single adaptive, closed-loop system for precision management. The work, published under the title describing full-course management based on multidimensional spatiotemporal cognition, represents one of the most detailed attempts yet to formalize how dynamic molecular information could be woven into routine clinical decision-making for these cancers.

The core critique advanced by the authors is that traditional precision management of gastric and colorectal cancers depends on static, single-point models. A patient typically undergoes a biopsy at diagnosis, receives a limited panel of biomarker tests, and then follows a largely linear treatment pathway. Modification of therapy is often triggered only by overt disease progression, meaning that the clinical response lags behind the biological reality of the tumor. This paradigm struggles to capture two defining features of gastrointestinal malignancies: spatial heterogeneity, in which genetically distinct subclones coexist in different regions of the primary tumor and its metastases, and clonal evolution, in which treatment pressure selectively expands resistant populations over time. A single biopsy from a single lesion at a single moment cannot represent this shifting landscape, and the consequences are therapeutic decisions built on incomplete and rapidly outdated information.

To address these limitations, the review proposes a new paradigm described as spatio-temporal integration, adaptive, and closed-loop. It is organized around three complementary dimensions. The spatial dimension combines multi-omics technologies, spanning genomics, transcriptomics, epigenomics, proteomics, and metabolomics, with multi-modality imaging to construct a comprehensive tumor atlas. Such an atlas would map molecular diversity across tumor regions and metastatic sites, revealing subclonal architecture that a conventional biopsy misses. The temporal dimension exploits longitudinal monitoring through liquid biopsy, most prominently circulating tumor DNA, to dynamically track clonal evolution in real time. Serial ctDNA measurements can detect the emergence of resistant clones, quantify tumor burden noninvasively, and enable functional stratification of biomarkers, meaning that biomarkers are evaluated not just for their presence but for how their behavior over time reflects active biological processes.

The third dimension is artificial intelligence-driven fusion, which the authors position as the computational engine that converts spatial and temporal data streams into actionable clinical guidance. Rather than presenting clinicians with raw molecular data, AI models would integrate multi-omics profiles, imaging features, and serial liquid biopsy trajectories to provide decision support at three precisely defined stages of patient care. These stages, designated T1, T2, and T3, form the structural backbone of the proposed framework and map onto the natural history of cancer treatment from initial therapy through adaptation to long-term surveillance.

The first decision turning point, T1, corresponds to prediction at baseline. At this stage, the framework aims to estimate treatment benefit before therapy begins, informing initial treatment selection. By combining spatial multi-omics characterization of the tumor with baseline ctDNA profiles and imaging, predictive models could identify which patients are likely to respond to a given regimen, allowing frontline therapy to be tailored from the outset rather than adjusted retrospectively. The second turning point, T2, corresponds to adaptation during treatment. Here, longitudinal ctDNA monitoring and repeated molecular assessment allow clinicians to reassess treatment response and detect emerging resistance while therapy is still underway. Instead of waiting for radiographic progression, molecular signals of failure could prompt earlier switches in regimen, sparing patients months of ineffective and toxic treatment. The third turning point, T3, corresponds to interception during surveillance. After completion of definitive therapy, the framework uses molecular markers to estimate the risk of recurrence or minimal residual disease, enabling earlier intervention at a point when disease burden is lowest and the probability of cure is highest.

Together, these three decision functions establish a closed-loop system in which clinical outcomes feed back into molecular monitoring, and molecular monitoring in turn refines subsequent decisions. The stated goal is to achieve the right treatment for the right patient at the right time, a phrase that captures the shift from reactive medicine to proactive, continuously calibrated care. Importantly, the authors emphasize that the framework is not a single technology but an architecture, one that requires standardized assays, validated algorithms, and clinical workflows capable of acting on dynamic information.

A significant contribution of the review is its careful classification of biomarkers into four functional categories: early screening, prognostic stratification, treatment response prediction, and recurrence monitoring. The authors stress that biomarker performance and clinical applicability are cancer-type and scenario-specific, and they caution against simply extrapolating findings across gastrointestinal tumors. A marker that reliably predicts response to immunotherapy in colorectal cancer, for example, cannot be assumed to behave identically in gastric cancer, and a prognostic signature validated at diagnosis may not retain utility in the surveillance setting. This scenario-specific framing is intended to prevent the overgeneralization that has undermined earlier biomarker translation efforts and to guide developers toward rigorous, context-bound validation.

The review also offers an objective assessment of pivotal clinical evidence, including the DYNAMIC-III trial, which evaluated ctDNA-guided downstaging of adjuvant therapy. According to the authors, ctDNA-guided approaches in that setting can reduce oxaliplatin exposure and its associated toxicity, a meaningful benefit for patients who might otherwise endure substantial chemotherapy-related harm. However, they note that the pre-specified non-inferiority criteria were not met in the trial, and they argue that the clinical utility of ctDNA-guided therapy de-escalation must therefore still be interpreted with caution. This balanced treatment of the evidence reflects a broader theme of the review: enthusiasm for liquid biopsy and AI must be tempered by rigorous evaluation of whether dynamic monitoring genuinely improves survival and quality of life, not merely whether it generates molecular data.

Looking toward implementation, the authors identify several translational barriers that stand between the proposed framework and routine practice. Testing standardization remains a major hurdle, as ctDNA assays and multi-omics platforms vary widely in sensitivity and reproducibility across laboratories. Cost considerations affect equitable access, particularly for serial monitoring that requires repeated sampling. Clinical validation must demonstrate that the framework improves outcomes in prospective studies, and the generalization capability of AI models must be established across diverse patient populations, institutions, and scanner and sequencing platforms. Without such validation, algorithmic recommendations risk encoding biases or failing silently in unfamiliar contexts.

The review closes by outlining future directions that could carry the field forward, including standardized regulation of dynamic biomarker testing, multimodal artificial intelligence that fuses molecular and imaging data, digital twin models that simulate individual tumor trajectories, clinical translation of spatial multi-omics, and adaptive trial designs that allow treatments to be modified based on real-time molecular feedback. If these elements mature, the authors contend, the management of gastric and colorectal cancers could move decisively beyond the static, reactive paradigm of the past toward a continuously informed, adaptive model in which every stage of care, from prediction to adaptation to interception, is guided by the evolving molecular reality of each patient’s disease.

Subject of Research: Spatiotemporal precision management of gastric and colorectal cancers using multi-omics, ctDNA monitoring, and artificial intelligence

Article Title: New spatiotemporal precision management framework proposed for gastric and colorectal cancers

Article References: New spatiotemporal precision management framework proposed for gastric and colorectal cancers. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: gastric cancer, colorectal cancer, precision medicine, spatial multi-omics, circulating tumor DNA, liquid biopsy, artificial intelligence, clonal evolution, tumor heterogeneity, minimal residual disease, DYNAMIC-III trial, digital twins