We are at a crossroads in our collective quest to improve healthcare delivery. For decades, thousands of academic studies have been conducted to tell us what might or might not work in patient care. Like clockwork, what was truth yesterday is overturned with the latest study. Women should take estrogen supplements to help prevent chronic conditions – but wait, we now find out that side effects are potentially more harmful than the conditions they were prescribed to address, such as breast cancer, coronary heart disease and stroke, reversing the previous guidelines.
Commonly used study designs have their limits, so their conclusions are met with understandable skepticism. Random clinical trials are treated as the “gold standard” given their efforts to isolate the effect of the intervention being studied. But they are often funded by big pharma or device makers and are therefore conducted to answer narrow questions. Additionally, the patient cohorts are small by design and not representative of the larger population. The conclusions from such studies are the basis of “evidence-based medicine,” or recommendations which physicians are encouraged to follow in their everyday practice. Houston – we have a problem.
With the greater adoption of electronic health records (EHR), we now have a growing amount of digital data about real-world clinical practice. Where was this data before? In unusable handwritten records, filed away in manila folders, stored in dusty mahogany-lined shelves. This was the dark ages of medicine: the “practice” of medicine, not the “science” of medicine.
Now that information about clinical care can be machine-read, we can assemble and read this clinical data to create a living laboratory. We can learn from the everyday clinical care of millions of patients. Rather than depending upon narrowly designed studies which do not relate to you and me, we can start to learn about healthcare delivery from everyday real-world care data. Given that this new medical model is based upon real historical practice and not test situations, it has been termed “practice-based evidence,” or PBE.
Under PBE, improving healthcare will depend upon assembling individual patient profiles and aggregating these profiles across a population to determine what works, when, for whom, and by whom. Since more than eighty percent of all healthcare data for an individual is unstructured (text), it is critical that both text and coded data are processed for analytics. Newer internet-based technologies, like what Apixio has built, can read clinical notes to augment coded data from billing and lab data to create a comprehensive patient profile. Non-clinical data could be added as well, if available.
With these comprehensive patient profiles, we can identify similar people and study what works and what does not for them. While each of us is unique, we can be grouped together with respect to our invidual profiles: for example, working thirty-year-old healthy men living in San Francisco suffering from occasional muscle injuries. Within each cohort, there are natural studies which can be performed to figure out the best treatment for a torn tendon injury – not only whether surgery is best but who is best to perform the operation. With a greater number of profiles, the signal on what does and doesn’t work gets stronger.
For example, researchers at Stanford data mined eighteen years of hospital records and were able to establish that the use of a specific type of pain medication (rofecoxib) for rheumatoid arthritis patients was related to a higher incidence of heart attacks in the patient population.
The major obstacle to PBE is the ability to assemble a large collection of patient medical data from diverse sites across the country and the world. This is not a technology problem, it is a business problem. The reticence of healthcare systems and health insurance plans to share their data is the primary obstacle. They regularly hide behind patient privacy regulations while freely using the data for their own operations.
It is only with a large, regularly updated pool of patient data that we can use PBE to improve healthcare delivery so we are free of the limitations of EBM. We need real-world data to solve real-world problems, not a sliver of data here and there.