Chemotherapy Treatment Timelines Extraction
from the Clinical Narrative:
Task Description
Chemotherapy Treatment Timelines Extraction
from the Clinical Narrative:
Task Description
Please go to https://bit.ly/ChemoTimelines2025 if you are interested in our 2025 version of the shared task!
The goal of this task is to build a timeline of chemotherapy treatments for each patient given all available Electronic Health Records (EHR) notes of that patient. The notes range the entire spectrum -- primary care provider, oncology, discharge summaries, emergency department, pathology, radiology, etc. The success in this task will provide previously unavailable timelines contextualizing cancer treatment.
We define two subtasks:
Subtask 1: Gold chemotherapy events (EVENTs) and time expressions (TIMEX3s) are provided. The goal is to predict the temporal relations between them, then produce the final patient-level timelines which involves deduplicating and resolving conflicts in the pairwise temporal relations if pairwise temporal relations are chosen to use as the intermediate representation. A system might not rely on pairwise temporal relations to derive the timeline. Gold event attributes, including modality, the relation of the event to the document creation time (DocTimeRel), and time expression normalizations generated by TimeNorm (https://github.com/clulab/timenorm) are also provided.
THYME annotation guidelines for pairwise temporal relations (2014) with definitions, also in Styler et al, 2014. The addition of NOTED-ON temporal relation and the refinement of 2014 annotation guidelines are described in Lin, Wright-Bettner et al, 2020. We used these definitions to build our system for extracting the pairwise temporal relations from which we then derive the patient-level timeline. See Resources -> Organizers' System.
Subtask 2: In this subtask, ONLY patient's EHR notes are given, participants are expected to build an end-to-end system for chemotherapy timeline extraction.
In summary, the input to Subtask 1 is EHR notes along with the EVENTs and TIMEX3s gold annotations, output is the patient-level timelines. The input to Subtask 2 is EHR notes only, output is the patient-level timelines. In both subtasks, evaluation is against the gold patient-level timelines.
An example of a patient-level timeline is shown in Figure1.
Figure 1: An example of the patient-level timeline.
Organizers' System
We also provide the organizers' system as a Docker container. Check the Resources page for details.