Did you know that the global increase in health care costs has increased at triple the rate of inflation, by more than 8% last year?
At first, you’d probably blame the pharmaceutical sector given the recent public debate on (excessive) drug pricing. But that doesn’t seem to be the real cause.
If we look at the Netherlands as an example, we see that the overall costs of medicine hasn’t really increased in the past 10 years. During that same time, including the 2007 financial crisis, employment in health care surged with 15%, while the overall Dutch labor market stayed flat.
Of course, we all want to grow old in good health and we need doctors and nurses for that. The problem is that most of that added employment was wasted on administrative, mundane tasks. One study even showed that for each hour spent with patients, physicians spend 2 hours on electronic health records (EHR) and desk work.
You might think that all that desk work in the end will help us to gather clinical (big) data and find new treatments for diseases we can’t cure today. Unfortunately this is not the case. Let’s try to peel the onion and understand why.
EHRs and unstructured data
Most of the mentioned desk work is spent entering data in electronic health record systems (EHRs). EHRs are being used for two purposes: (1) to support billing and (2) to provide a case management overview to physicians for a given patient.
While EHRs offer the ability to store structured data on various clinical aspects, most data needed for patient case management is entered in an unstructured, free text format, for instance in appointment notes and reports. Billing data, medicine data and lab data are the only types of data we’ve seen to be consistently entered as structured data in a database-like format.
The question is why physicians prefer the unstructured way? In this way, she or he can see all recent and relevant notes for a given patient at the same time, without having to look up or enter data in fields on different tabs in the EHR. Basically, it helps to ‘only’ spend those 2 hours instead of 3. And although efforts are made to get data in a more structured way in EHRs, the estimation is that at the best reachable level this will be 30-35% of the data. Let alone the fact that that structured data from different EHRs are not comparable nor interoperable. This is important if we would want to combine data from multiple health care providers.
So what happens when there is a need to gather data for research purposes on a population across multiple providers, for instance to improve personalized medicine or value-based healthcare? The free text format is normally not suitable for profiling and unsuitable for data analytics.
The impact and effectiveness of using structured clinical data for this purpose is clear. One study showed a data analytics model using clinical and social data was able to accurately predict and identify heart failure patients at risk for 30-day readmission or even death. Would you accept that your hospital isn’t able to identify you as being at risk, only because your data wasn’t entered in a structured way?
A common approach for clinical and pharmaceutical research and data analytics is setting up a Disease Registry. Such a population-based registry keeps track of a small sub population of patients (preferably national or international) with a specific condition and combines data from various sources, such as EHRs, patient data on “Quality of Life” and even medical sensor data. Now, filling the registry with relevant data from the EHR can be a challenge, since the most interesting clinical data is stored in appointment notes and other unstructured reports. We have even seen examples where the data was entered manually in a registry – a labor intensive approach prone to errors.
Intelligent automation can help us here. While mainly being used in the financial sector and shared services up to now, Robotic Process Automation, Machine Learning and Natural Language Processing are now slowly being introduced in healthcare.
Robotic Process Automation, or RPA can help to eliminate mundane, repetitive work – exactly what is being done to enter data into a Registry. Although you might associate a robot with some humanoid made out of metal and gear in a Hollywood blockbuster, the robots we describe with RPA are more friendly. They can be seen as a virtual office assistant, mimicking the repetitive data collection and data entry actions that a subject matter expert has trained them to do. By doing so, RPA delivers outcomes such as a faster turnaround time, process compliance, 100% quality in data entry and most important a motivated staff: they now can focus more time on their patients.
Machine learning has been around for several decades, but the use of these prediction and classification algorithms for decision support has surged in the past two years due to the accessibility of (big) data and (cloud) computer power. This renewed attention has also led into improvements in algorithms understanding natural language: Natural Language Processing. In the same way that RPA helps with data collection and data entry, NLP can help to extract structured data out of unstructured texts, such as the appointment notes and medical reports mentioned above.
Without the need for staff spending hours of their time each day in reviewing these notes, NLP can create valuable and life-saving insights. A 2016 study showed that NLP on free-text reports, in combination with the data analytics model mentioned above, was able to correctly predict heart failure for a given patient with a 97.5% accuracy, while correctly finding heart failures in 95.3% of the entire population in the study. In another example, NLP was even used to predict a diagnosis based on patients’ self-narratives.
Given these outcomes, we expect that any investment in RPA and NLP will dramatically improve healthcare effectiveness in the coming 10 years. The total annual cost for patient registries in the Netherlands is estimated between €160 million and €200 million and an additional €80 million is spent on other registries, focused on quality assurance, compliancy and risk.
It is without doubt that we need a sustainable solution without wasting more time on mundane, repetitive work. Given the current shortage of doctors and nurses as well as increasing complaints on patient data entry work, we cannot afford to spend another 15% more of their time in the next 10 years.
About the authors
Mathieu Jonker is the founder of Artilience. He is passionate on delivering growth for both employees and organizations and is a thought leader on intelligent automation.
Ruud Simons is CEO of Patiëntregisters.nl. He is passionate about patient wellbeing and determined to create a generic infrastructure in healthcare where patients and doctors benefit from modern technology.