Skip to content

Proposing a new prediction task: Causative Pathogen Identification before culture results #179

@netanelcyber

Description

@netanelcyber

Hi @rvandewater and team,

Excellent work on YAIB — the multi-center, modular framework is exactly what the clinical ML community needs for reproducible ICU research.

YAIB currently includes five predefined tasks: mortality, AKI, Sepsis detection, kidney function, and length of stay. I'd like to propose a sixth task that extends naturally from the existing Sepsis task:

Causative Pathogen Identification — predicting the most probable causative organism in ICU patients with suspected sepsis, using clinical admission data (vitals, labs, clinical presentation), before culture results are available (48–72 hour window).

Why this matters clinically:
The gap between sepsis onset and culture confirmation forces clinicians to prescribe empiric broad-spectrum antibiotics blindly. A pathogen prediction task could directly support earlier, targeted antibiotic decisions — a major Antimicrobial Stewardship priority.

Why YAIB is the right platform for this:

  • YAIB natively supports MIMIC III/IV, eICU, HiRID, and AUMCdb — all of which contain microbiology culture results that can serve as ground truth labels
  • YAIB's modular design allows adding new prediction tasks straightforwardly
  • Multi-center validation across datasets would be critical for this task

What I'm building:
PenuX is an in silico clinical simulation platform targeting exactly this prediction gap. I'd love to integrate it as a task within YAIB for standardized benchmarking and validation.

Would the team be open to discussing this as a new task contribution?

Repo: github.com/NetanelCyber/PenuX
Contact: Netanel Stern

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions