Trajectory Health AI emphasizes transparency and accountability in evaluating healthcare AI and population health initiatives. Trajectory, as an independent third party, requires program leaders to submit defined metrics, population data, and objectives to verify and validate both the initial and ongoing effectiveness of client models/programs.
Through a rigorous review process based on standards adopted by the Consumer Technology Association and the Population Health Impact Institute, Trajectory promotes evidence-based medicine in today’s rapidly changing environment. The Trajectory certification review process allows program sponsors and their clients to update and adjust their programs and devices in a timely fashion to support generally accepted methodological standards of evaluation to ensure quality, accuracy, and effectiveness.
Principles
Trajectory Health AI emphasizes transparency and accountability in evaluating healthcare AI and population health initiatives. Trajectory, as an independent third party, requires program leaders to submit defined metrics, population data, and objectives to verify and validate both the initial and ongoing accuracy of client models.
Through a rigorous review process based on standards adopted by the Consumer Technology Association and the Population Health Impact Institute, Trajectory promotes evidence-based medicine in today’s rapidly changing environment. The Trajectory certification review process allows program sponsors and their clients to update and adjust their programs and devices in a timely fashion to support generally accepted standards of care and to ensure quality and effectiveness.
Trajectory Health AI supports the following three principles for Predictive Health AI models.
Transparency
Transparency is “making clear” the objectives of the product, clarity of metric results and the definitions, the characteristics of populations in development, and other important revelations, all without violating important proprietary methods.
Verification
Verification involves the overview of relevant documentation—including population characteristics and completeness of underlying data and the design of the evaluation—that occurred during model development.
Validation
Validation involves evaluating performance of models or programs in environments using real world data sets over time and across multiple settings and populations, conducting usability testing, and setting quality controls to monitor and address future changes in product use.
Products
Certification Products
Basic Certification
Based on CTA “shall” standards
i.e., one accuracy metrics, duration for one year
Advanced Certification
Based on CTA “should” standards
i.e., multiple accuracy metrics, duration for 2-3 years
Program Model Verification
The TRC framework assesses how AI and healthcare models are developed and tested. This includes highlighting the methodology associated with the predictions and purported outcomes in model populations.
Real World Validation
The TRC framework helps implementation experts evaluate their model in real-world situations, including a comparison to original model results and a plan for testing over time in different populations. A primary goal is to assess and validate the ongoing impact of the AI including the accuracy and clinical usefulness.
Return on Investment
An optional product offering is to deploy the Trajectory system to help assess cost and benefit of the model in specific real world or intended use environments.
AI Drift
The TRC assessment process can also be used to address algorithm drift and program degradation in original model or in varied intended use populations.