Protocol Writing

[amadeosalvador]

Protocol for Developing a Federated Learning Model for Mortality Prediction in Spinal Cord Injury Patients

Overview

This protocol outlines the development of a federated learning (FL) model with the aim of predicting mortality in spinal cord injury patients. The protocol aims to compare the performance of federated learning models against single-database models, compare feature sets derived using different methods, and develop a clinically useful mortality prediction model.

1. Objectives

Aim 1: Create a federated learning model and compare its external validation against single-database models.

Aim 2: Compare feature sets created using:

A data-driven approach

Clinical expertise

Gen-AI

Aim 3: Develop a clinically useful mortality prediction model for patients with spinal cord injury.

2. Cohorts Description

TPL Cohort

Definition: Patients with tetraplegia.

Inclusion Criteria: Specific medical diagnosis code related to tetraplegia.

Exclusion Criteria: Non-relevant medical conditions that could skew the data.

Death Cohort

Definition: Patients whose records are included in the death table.

Inclusion Criteria: Documented date of death.

Exclusion Criteria: Missing or unreliable death records.

3. Target Variables

Outcome: Mortality (Death)

Predictors: Membership in TPL or Death cohorts, and binary procedure variables.

[schillil]

1. Table of contents
2. List of abbreviations
3. Abstract
4. Amendments and Updates
5. Milestones
6. Rationale and Background
7. Study Objectives
   • Primary Hypotheses: Predict the mortality post-procedure within <6 months> (PEG, tracheostomy, colostomy) for persons hospitalized with acute tetraplegia
   • Secondary Hypotheses
   • Primary Objectives
   • Secondary Objectives
8. Research methods
   • Study Design: From a cohort of patients hospitalized with a new diagnosis of tetraplegia due to acute spinal trauma we designed a patient prediction model to estimate the risk of death with <6 months post-procedure.
   • Data Source(s)
   • Study population
   • Exposures
   • Outcomes
   • Covariates
9. Data Analysis Plan
   • Calculation of time-at risk
   • Model Specification
   • Pooling effect estimates across databases
   • Analyses to perform
   • Output
   • Evidence Evaluation
10. Study Diagnostics
    • Sample Size and Study Power
    • Cohort Comparability
    • Systematic Error Assessment
11. Strengths and Limitations of the Research Methods
12. Protection of Human Subjects
13. Management and Reporting of Adverse Events and Adverse Reactions
14. Plans for Disseminating and Communicating Study Results
15. Appendix: Negative controls
16. References