Lecture Notes on Clinical Deterioration Prediction Model

Introduction

• Andrew, a specialist in acute internal medicine and applied data science, collaborates on a project aimed at predicting clinical deterioration in hospital wards.
• Clinical deterioration refers to situations that require escalation of care, such as transfer to ICU, rapid response activation, code blue, or mortality.

Exclusion Criteria

• Exclude patients who are pregnant, have mental illness, are under 18, or are already in monitored wards.
• These conditions make scoring for clinical deterioration unnecessary or redundant.

Evolution of the Project

• Initially focused on "failure to rescue," assessing the hospital's failure to respond to deteriorating patients.
• Transitioned into developing a nursing intuition score.
• Currently evolved into an AI model named Nadia aimed at predicting clinical deterioration.

AI Model Development

• Nadia is an AI model designed to predict clinical deterioration ahead of time to allow for early intervention.
• The model functions as a "time machine," predicting potential issues before they become critical.

Clinical Validation Approach

• No standard exists for validating such models; hence, a unique approach is needed.
• Chart Reviews: Conduct reviews for both deteriorating and non-deteriorating patients.
  • Use rapid response teams' data for patients who experienced rapid response, code blue, or mortality.
• Clinical Intuition: Use clinician assessments as a gold standard to measure the AI model's effectiveness.
• Blind Testing: Ensure that the validation is unbiased by not revealing the model's decisions to reviewers.

Discussion on Implementation

• Discuss potential integration of the model in clinical practice.
• Consideration of clinical workflow upon AI model's prediction of deterioration.
• Aim to reduce workload and alarm fatigue among nursing staff while increasing prediction accuracy.

Further Steps in Validation

• Plan involves measuring against other existing models like NEWS2, Muse, and Epic Deterioration Index.
• Focus on achieving a gold standard with at least 80% accuracy.
• Address issues of false positives and specificity in current models.

Considerations and Challenges

• Predictive nature implies reporting on interventions that prevent clinical events.
• Ensure the model does not over-predict, causing unnecessary anxiety or interventions.
• Plan includes expanding the validation beyond initial focus on sepsis, covering broader deterioration indicators.

Strategic and Operational Discussions

• Deployment in Cedars-Sinai Medical Center and potentially other locations.
• Utilization of REDCap for data collection and analysis.
• Importance of aligning with broader clinical and nursing strategic initiatives.

Academic and Educational Aspects

• Collaboration with DNP students and Ph.D. researchers to contribute to academic knowledge and practical implementation.
• Addressing educational integration with academic programs focusing on innovation, AI, and clinical care improvement.

Conclusion

• The project is a novel fusion of clinical intuition and advanced digital intelligence aimed at transforming patient care in hospitals.
• Continuous collaboration and iterative validation processes are essential to refine and implement the AI model in real-world settings.
• Emphasis on reducing nurse workload, preventing unnecessary rapid responses, and improving early intervention strategies.