Data analytics hasn’t transformed healthcare in a meaningful way, yet. Yes, we have a lot of healthcare data about patients, doctors, tests and even treatment outcomes. And much of this information is moving from paper to digital format. But we are just at the beginning stages of learning how to take best advantage of it.
There are certainly pockets of innovation using data within closed care systems like Kaiser. But this is not the norm. The norm is that individuals get care from many institutions and they don’t share data with one another. Without access to large data sets, data analytics innovation stalls (as does care coordination, but that is another story). To be clear, we are talking about data in the medical records (the written record) where the actual clinical care is documented, not administrative or billing data, which is a representation of medical care for reimbursement and operations.
Obstacles to data-sharing
You would think that sharing data in this new digital age would be simple, but it isn’t — technology shortcomings and lack of incentives keep valuable data locked away. Most of the prevalent electronic health records (EHRs) were not really designed with features which would enable easy data sharing between and among systems. This is not an accident. I doubt that any EHR company lost a deal of any significance because an interoperable record or a data export feature was not present in their software.
In a fee-for-service world in which we still collectively operate, there are no real financial incentives for physicians, clinics, and hospitals to allocate time, resources, and money to share data. Payers and purchasers have contracted the rights to patient charts for administrative, payment or treatment decisions, but the burden falls on providers to provide them. What is the return to the
provider for doing so? Not much. Health plans don’t pay more for a procedure or an office visit as a result of provider data sharing. Without any financial stake in coordinated care, no two hospitals or clinics will really want to share data – it is costly, time consuming, burdensome, and many do not have the right technology or IT staff to make it happen. If there was a benefit to the institution sharing the records, there would be more of a push for EHRs to have those features.
As it stands, if a health plan or individual wants a medical record, the documents are physically printed or copied. That is not anyone’s idea of data liquidity.
Now there are ways to circumvent the lack of interoperable features in EHRs through use of queries into the database directories, to extract and securely upload and transmit documents into another database. After doing this, one still needs to make sense of the written data for insights. Some healthcare informatics and policy leaders would like for healthcare records to be structured to enable easier interoperability. But what patient wants their medical notes to be a collection of numbers? All nuance and context is lost. So a key challenge remains: how to derive meaning and value from analysis of the unstructured, textual portion of the medical record to support analytics efforts?
A Deep-dive Into Specialty Pharma
A specialty drug is a class of prescription medications used to treat complex, chronic or rare medical conditions. Although this classification was originally intended to define the treatment of rare, also termed “orphan” diseases, affecting fewer than 200,000 people in the US, more recently, specialty drugs have emerged as the cornerstone of treatment for chronic and complex diseases such as cancer, autoimmune conditions, diabetes, hepatitis C, and HIV/AIDS.
Using cognitive computing to make sense of data
Other industries have grappled with deriving meaning and value from unstructured textual data. Just think about all of the information derived from text on websites which powers search, advertising and e-commerce. Adaptive or learning algorithms have been created which learn patterns and infer knowledge within the narrative of websites. This is the essence of cognitive computing.
These algorithms are fed enormous amount of data prevalent across the web and are sent out to perform a specific activity. The statistical techniques underlying these algorithms, considered machine learning, are now used in domains to recognize voice and enable digital assistants like Apple’s Siri or Amazon’s Alexa; to integrate visual cues and patterns and enable autonomous driving;
and even beat the world champions at Chess and Go.
With 1.2 billion patient documents generated each year in the US, there is a wealth of real-world data about healthcare, which if tapped, could provide fundamental insights into the diagnosis and treatment of diseases and enable truly individualized medicine. However, these documents are written in a variety of styles by a variety of clinical specialists, in a variety of settings. Trying to decipher this information using computers is difficult and error-prone if one tries to come up with a rules-based approach to essentially tell the computer how to read and make sense of what is written. Even if the rules worked to interpret and make sense of the language and templates used by clinicians at one care site, they won’t necessarily translate at another institution or specialty practice.
Instead, the correct approach is to use adaptive or machine learning algorithms, which become more powerful and widely applicable when the machine is fed a large volume and variety of data sets. The performance gets to a one point where it can interpret clinical records from an institution for the first time without the need for customization.
How to leverage data to improve healthcare
Data mining using adaptive algorithms is a promising technology for assembling care profiles (“phenotypes or a health finger print”) with the conditions and treatment history of individual patients. Then, using accurate and timely care profiles, we can perform virtual care trials to learn what does and what does not work for an individual. This knowledge will become increasingly critical as an explosion of treatments and options makes it very difficult for a physician to recall
and apply evidence at each step along the patient disease trajectory. Moreover, with more and more expensive medical technologies in use, and patients assuming a greater percentage of the costs, a physician must consider the cost-effectiveness of each option.
Data analytics will only begin to transform healthcare in a meaningful way when we can tap into unstructured data sources and mine the information with powerful cognitive techniques. By deriving knowledge from real world clinical charts, we can learn about the best treatment options for each individual and deliver truly personalized medicine. Along the way, we will very likely overturn well-accepted norms that were based upon flawed or non-representative scientific studies. Drug therapies could be more directed and effective, medical procedures could facilitate better outcomes and it could fundamentally change the way healthcare is consumed and delivered.
Photo: Flickr user Walter Pro
