GLEAM Multimodal Learner Use Case: HANCOCK dataset

This Zenodo record contains preprocessed, patient-level tabular datasets derived from the public HANCOCK multimodal head-and-neck cancer cohort and configured for binary recurrence prediction. The datasets follow the in-distribution split logic reported in Dörrich et al. (Nature Communications, 2025).

Files: HANCOCK_train_split: in-distribution training cohort and HANCOCK_test_split: in-distribution held-out test cohort

Each CSV row corresponds to a single patient and includes:

1. patient_id: patient identifier

2. split: in-distribution split assignment (consistent with the paper’s evaluation split) 

3. target: recurrence label (binary; endpoint used for recurrence modeling) 

4. clinical/pathological covariates, blood laboratory measurements

5. CD3_image_path, CD8_image_path: filenames/paths for the corresponding CD3/CD8 TMA core images

6. icd_codes: ICD codes in plain text

Imaging availability (CD3/CD8)

The corresponding CD3 and CD8 TMA core images referenced by CD3_image_path and CD8_image_path.

Source publication: https://www.nature.com/articles/s41467-025-62386-6

Raw HANCOCK dataset download portal: https://www.hancock.research.uni-erlangen.org/download

Preprocessing scripts used to generate these split files are available here: GLEAM Use Cases