By Simon MacGibbon
Addressing Alert Fatigue and Clinician Burnout
Implementing Five Elements for Scalability
- Recognize individualized baselines. Using a sound statistical approach to computing individual physiological and functional baselines, by monitoring entropy in patient data such as vital signs, will allow computational assessments to recognize – and flag – their significance. For instance, one heart failure patient might experience frequent changes in weight that revert without clinical consequence, making weight gain less concerning than a heart failure patient who suddenly experiences an abnormally high rate of weight gain. Systems should be built to account for these individual differences.
- Establish clear safety thresholds. This will effectively guarantee a fail-safe design together with the individualized baselines. It will also be used to incorporate established medical guidelines to prevent any one person from dismissing the system. Fail-safe designs are an important feature of automated systems and virtual care should not be an exception.
- Build a learning system that has the facility to be overridden by clinicians. Already, clinicians use non-medical sticky notes and spreadsheets to take note of potential problems which aren’t often caught by medical monitoring (ex. patient looks more pale than normal; worried patient may not tolerate this new medication). The right virtual care system takes this into account, making it easy for clinicians to apply their own judgement.
- Use machine learning to make sense of a wide range of data and its aggregate meaning. While medical professionals have been trained to provide expert analysis on a patient’s condition based on a combination of factors, the human brain is not designed to make sense of high dimensional datasets. This is the ideal application for Machine Learning, to augment clinicians to see concerning changes in patient health status that might otherwise go unnoticed.
- Take a data driven individualized approach to establishing scheduled visits. Just like the in-patient setting, unplanned alerts ought not be the only way patients are surfaced. This is critical for patients with chronic disease. While patient data should be used to monitor a patient’s condition, periodic visits (virtual, if possible) are essential for those living with chronic conditions. AI doesn’t detract from that, but rather supports the ongoing patient-clinician relationship.