Understanding Time Series Data is Critical to Addressing ‘Notification Overload’
“Notification overload” is a term used in mobile circles where average people complained about the noise from popular social media apps and mobile games. In the world of healthcare ‘notification overload’ is a much more serious problem.
Physicians and nurses are experiencing ‘notification overload’ that is rooted in health IT systems that are ‘rule-based systems’ pre-programmed around a broad set of rules that are generic for clinicians and for the patient being monitored. The result is “notification fatigue” — which can cause deafness in clinicians who are bombarded with continuous alerts without value-justification. This may sound silly. But seriously, this is a big problem and must be resolved.
Healthcare has a notification problem that goes beyond EHR notifications. Alarms are found on most medical devices used at the bedside. These alarms sound every hour of every day. The purpose of clinical alarms is to enhance safety by alerting clinicians to deviations from a predetermined normal status, however, the definition of normal is generic for all patients.
A perfect alarm system would never miss a clinically important event. The alarm would not go off needlessly or when there is no clinically important event. By design, alarms should be highly sensitive as well as specific, so that they do not miss an important event, but do not waste time on alerts that are meaningless. Today however, their high sensitivity without specificity results in a flood of false alarms, leading to alarm fatigue.
Alarm fatigue occurs when nurses and clinicians become overwhelmed by the sheer number of alerts, which can result in alarm desensitization and, in turn, can lead to missed alarms or a delayed response to alarms. The desensitization to alarms occurs largely because the devices have “cried wolf” too often. Studies have shown that 72% to 99% of clinical alarms are false. The large number of false alarms has caused nurses to turn down the volume of audible alarm signals, adjust the alarm settings outside limits that are safe and appropriate for the patient, ignore alarm signals, or even deactivate alarms, which has resulted in sentinel adverse events and patient deaths.
Making Notifications Intelligent
The solution for this problem is to adopt the use of artificial intelligence or deep learning to understand the data and personalize alarms and notifications for each specific patient and be smart enough to notify the right provider at the right time. These are systems that learn automatically. They’re not pre-programmed, they’re not rule-based systems, and they’re not generic for all patients. They take raw information and can learn how to use data to drive action, classification or predictions to make alarm and notifications personalized, accurate and potentially, pre-emptive.
Key to making notifications and alarms intelligent is to effectively understand ‘time series data,’ I.e., a series of measurements of the same variable over time. The volume of clinical time series data is growing rapidly. Every medical device and monitor is generating time series data. In 2009, intensive care units in the United States treated nearly 55,000 patients per day, generating millions of individual measurements, most of those forming time series. As our healthcare system becomes less dependent on acute care and more on virtual care with remote monitoring, the amount of clinical time series data is set to explode and so are the false alarms. As a result, we will need sophisticated tools to unlock the true value of this data to improve the lives of patients and the workflow of clinicians.
This is why we built Flow Health, and why Flow Health is on the leading edge of this transformation — we combine artificial intelligence and automation under one operating system. This system sifts through massive amounts of data: aggregating and integrating data and applying algorithms developed through advanced deep learning neural networks to deliver proactive, automated health care. We’ve laid the foundation to solve this pressing problem. The question remains: are health systems, clinicians and the FDA ready for artificial intelligence?
Imagine the Potential
Think of deep learning neural networks as an intelligent filter that is continuously monitoring every patient, and based on every available piece of data, intelligently determines the significance of the data from real-time monitoring and then routes the appropriate notifications to the right provider. Imagine the right notifications being sent to just the covering nurse, while other notifications that are more critical are sent not only to the nurse, but also to the charge nurse and attending physician or even directly to a specific on-call specialist. The opportunities are endless when we move from rule-based (‘dumb’) systems to intelligent, self-learning algorithms that can process large amounts of data in real time. At Flow Health, we are focused on using health data to move from being reactive to proactive, and eventually, to pre-emptive healthcare.