← Back to Home

One Less Missing Report at Cancer Meetings: BEAGLE-MDT Earns High Clinician Ratings in Preliminary Colorectal Cancer Evaluation

From organizing medical records to finding clinical trials, this AI platform aims to fill information gaps before cancer teams make decisions. Preliminary results from 15 synthetic cases are promising, but clinicians finding it useful is still a critical step away from proving it can improve care.

By SURL BioNews

Waiting for cancer treatment sometimes begins with a missing pathology or imaging report at the meeting table. When specialists from different disciplines must jointly decide the next step for a patient with colorectal cancer, whether the information is complete and suitable clinical trials have been identified can affect whether the discussion moves forward. A preliminary study of BEAGLE-MDT presented at the 2026 European Society for Medical Oncology (ESMO) Congress is exploring whether artificial intelligence can take on these preparatory tasks.

Developer Histofy said its clinical collaborators had audited more than 14,000 records of colorectal cancer multidisciplinary team discussions spanning more than a decade. Approximately one in every 12 cases was delayed by at least one week, while clinical trials were considered in only about one in every 25 cases. These figures illustrate gaps in existing workflows; they are not evidence that the platform can already reduce delays.

BEAGLE-MDT divides the work among interconnected AI agents: checking data completeness, synthesizing guidelines and research evidence, matching clinical trials, and then producing standardized management recommendations. According to reporting by *ASCO AI in Oncology*, the system is designed to require that statements in its outputs be traceable to their sources, and all results must undergo multidisciplinary team review before action can be taken on them. The platform also includes alerts for missing information, allowing teams to identify cases that are not yet ready for decision-making.

The clearest preliminary validation currently available comes from 15 synthetic cases covering different tumor locations and stages, with outputs reviewed by five experts in a blinded evaluation. Apex AI Institute's summary of conference abstract LBA18 reported that, on a five-point scale, average ratings ranged from 4.50 to 5.00 for case summaries, 4.07 to 4.87 for management recommendations, and 4.33 to 4.93 for trial matching. This supports the feasibility of further testing the platform, but the sample did not consist of real patients receiving care.

What these high scores actually mean remains central to interpreting the results. Conference commentators cautioned that clinicians' subjective assessments of recommendation quality cannot be directly translated into clinical accuracy rates. A reviewer's disagreement with a recommendation may also reflect reasonable differences in treatment choices rather than a clear error. Similarly, identifying a potentially relevant trial does not mean that a patient meets all eligibility criteria, much less that the patient will ultimately enroll or benefit as a result.

The research team plans to first run the platform alongside real meetings in “shadow mode,” without intervening in decisions, before conducting a prospective, multicenter comparison. Such studies are needed to test whether it reduces delays, missed matches, and errors, and whether the burden of human review offsets any efficiency gains. Histofy is both the developer and a listed study funder; related reporting also disclosed that some authors work for the company and that the presenter's partner holds shares in it. For this platform, the next phase must deliver independently testable evidence of care outcomes and safety.

References

  1. ASCO AI in Oncology
  2. Histofy
  3. Apex AI Institute