Therapeutic Assistance and Decision-making Algorithms in Hepatobiliary Tumor Boards (ADBoard)

Evaluation of the Trustworthiness of the Application of Artificial Intelligence and Decision Support Systems for the Creation of Tumor Conference Protocols

The goal of this observational study is to compare the recommendations of the artificial intelligence clinical decision support system 'ADBoard', with the recommendations of physicians by tumor conferences in patients with hepatobiliary tumors. The main questions it aims to answer are: Can ADBoard achieve a high level of similar recommendations as physicians' tumor conferences? Can ADBoard consider a more complete set of patient-related data than in physicians' tumor conferences? Can ADBoard reduce the time between the first time the patient is discussed at the tumor conference and the start of the recommended treatment plan? Participants will have their hepatobiliary tumor treatments determined by either tumor conference with ADBoard, or tumor conference without ADBoard.

Hepatocellular Carcinoma, Fibrolamellar Hepatocellular Carcinoma, Cholangiocarcinoma, Perihilar, Intrahepatic Cholangiocarcinoma, Metastasis to Liver, Gallbladder Carcinoma

decision support systems, clinical, tumor conference, tumor board, multidisciplinary team meeting, artificial intelligence, machine learning, natural language processing

The ADBoard study is a monocentric, prospective, parallel randomized controlled trial with a non-inferiority framework. Participants will be randomized 1:1 into one of two groups: either a) tumor conference with ADBoard, or b) tumor conference without ADBoard.

Participants' cases are discussed in tumor conference without the use of ADBoard, according to conventional practice.

Participants' cases will be evaluated by ADBoard directly before they are discussed in the tumor conference.

Utilizing the clinical decision support system 'ADBoard' for therapy selection in participants with hepatobiliary tumors

Concordance (yes/no) of interdisciplinary tumor conference treatment recommendation with ADBoard-supported recommendation

Examples of possible treatment recommendations are liver resection (curative), regimen of chemotherapy, immunotherapy, radiotherapy, transarterial chemoembolisation, diagnostics, best-supportive care, follow-up by re-presentation. The interrater reliability of the recommendations of ADBoard and the tumor conferences with regard to their agreement will be measured.

The intrarater reliability will be measured by testing all participant cases several times by the ADBoard according to the required sample size with sufficient statistical power (test-retest). Interrater and intrarater reliability will be evaluated descriptively (percentage of agreement, contingency tables), and finally the Cohen-Kappa value will be assessed.

For each type of tumor examined, a series of relevant parameters is defined by the medical staff. The objective is achieved if complete documentation, defined as documentation status 'present' or 'detected as missing', is present in more than or equal to 75% of all ADBoard-assisted decisions.

The System Causability Scale will be reviewed by 3 specialist physicians. The score ranges from 0 to 1. A higher score means greater suitability of the user interface, explanation, or explanation process itself for the intended purpose of tumor conference treatment recommendations.

The time between the initial presentation of the case to the tumor conference and the time of recommendation implementation will be recorded.

Inclusion Criteria: Age ≥ 18 years Valid informed consent Patient information available in the hospital information system or Health Data Platform (HDP) Enrollment in the hepatobiliary tumor conference Diagnosis of any of the following entities: Colorectal liver metastasis, Gallbladder carcinoma, Hepatocellular carcinoma (HCC), mixed cell carcinoma, or fibrolamellar carcinoma, Perihilar cholangiocarcinoma (Klatskin tumors), Intrahepatic cholangiocarcinoma (iCC) Exclusion Criteria: Patient does not consent / incapable of giving consent Missing findings in the hospital information system Patient is seeking for a second opinion and is not being treated at the study institution

Surgical Clinic, Campus Virchow-Klinikum / Campus Charité Mitte, Charité - Universitätsmedizin Berlin