Haywire Algorithms

Retraining

Retraining machine learning models is crucial for maintaining their effectiveness, especially in clinical settings. Ravi Parikh emphasizes the need for systematic retraining and recalibration, suggesting a cadence of every six to twelve months to adapt to changes in feature distribution and input utilization 1. He argues that flexibility in model recalibration can lead to better patient care, as models often need adjustments beyond their initial published versions.

Oftentimes the best model is going to require some recalibration and be a little bit different than the model that appeared in your peer-reviewed publication.

--- Ravi Parikh

Parikh also highlights the importance of collaboration between clinicians and data scientists to develop impactful machine learning models for clinical care delivery 2.

   

Human Oversight

Human oversight plays a pivotal role in the deployment and effectiveness of machine learning models in healthcare. Ravi Parikh shares an example from his clinical experience where a machine learning-based risk score was not effectively utilized due to a lack of clinician input in its deployment 3. He stresses that involving clinicians in the initial deployment can maximize the clinical impact of these models.

Having a clinician in the loop beforehand to maximize impact could have been really helpful.

--- Ravi Parikh

Furthermore, Parikh discusses the importance of end-user feedback in developing predictive tools that influence clinical care, emphasizing the need for human factors research to optimize machine learning applications in healthcare 4.