Precision Treatment of Unresectable HCC Guided by Multi-omics Deep Learning Models

Precision Treatment of Unresectable Liver Cancer Based on Multi-omics Deep Learning Model: a Multi-center Prospective Single-arm Study

Surgery is the main curative treatment for hepatocellular carcinoma(HCC) patients, but 70%-80% of HCC patients are in the middle and advanced stages at the time of diagnosis and cannot be surgically resected. Local and systemic therapy are the main treatments for unresectable HCC. Two recent trials of HAIC combined with PD-1 monoclonal antibody and targeted therapy reported objective response rates (ORR) as high as 43.3% to 77.1%.

Surgery is the main curative treatment for hepatocellular carcinoma(HCC) patients, but 70%-80% of HCC patients are in the middle and advanced stages at the time of diagnosis and cannot be surgically resected. Local and systemic therapy are the main treatments for unresectable HCC. Two recent trials of HAIC combined with PD-1 antibody and targeted therapy reported objective response rates (ORR) as high as 43.3% to 77.1%. However, the selection of patients who will benefit from the therapy remains a major challenge for the individualized treatment of HCC, which requires more accurate prediction of combination therapy.

With the advancement of sequencing technology, more and more fine-grained biological data can be obtained, including radiomics, pathology, genomics and immunomics. In recent years, the development of new methods such as graph neural network and multi-scale PHATE makes it possible to integrate multi-omics data. The use of artificial intelligence models to integrate multimodal data is an effective means to predict treatment response more accurately, which is helpful for more accurate and detailed classification of patients with different treatment outcomes, and to explore the internal mechanism of treatment response or not.

We constructed a multi-omics deep learning prediction model based on the retrospective cohort data from multiple medical centers (who received HAIC combined with target therapy and immunotherapy). The model could better distinguish the patients who would benefit from combination therapy, with an AUC of 0.86.

Therefore, the investigators conducted this multicenter, prospective, single-arm study to explore the response and prognosis of combination therapy in a population screened by the model and to evaluate the predictive power of the model.

Inclusion Criteria:

1. Aged 18-75.
2. No previous local or systemic treatment for hepatocellular carcinoma.
3. Child-Pugh liver function score ≤ 7.
4. ECOG PS 0-1.
5. No serious organic diseases of the heart, lungs, brain, kidneys, etc.
6. Enhanced MRI determines that the tumor is technically unresectable.
7. Pathologic type of hepatocellular carcinoma confirmed by puncture biopsy.
8. Multimodal Deep Learning Model Screening Based on Pathology, Imaging, and Genetic Data Suggests Benefit from HAIC in Combination with Lenvatinib and PD-1 inhibitors.

Exclusion Criteria:

1. Pregnant and lactating women.
2. Suffering from a condition that interferes with the absorption, distribution, metabolism, or clearance of the study drug (e.g., severe vomiting, chronic diarrhea, intestinal obstruction, impaired absorption, etc.).
3. A history of gastrointestinal bleeding within the previous 4 weeks or a definite predisposition to gastrointestinal bleeding (e.g., known locally active ulcer lesions, fecal occult blood ++ or more, or gastroscopy if persistent fecal occult blood +) that has not been targeted, or other conditions that may have caused gastrointestinal bleeding (e.g., severe fundoplication/esophageal varices), as determined by the investigator.
4. Active infection.
5. Other significant clinical and laboratory abnormalities that affect the safety evaluation.
6. Inability to follow the study protocol for treatment or follow up as scheduled.