Health officials are constantly talking up the importance of clinical decision support, more popularly known now as evidence-based medicine. We’re owning up to the awkward little fact–which really should embarrass nobody–that most doctors lack expertise on many of the conditions they encounter and can’t read the thousands of relevant studies published each year. New heuristics are developed all the time for things such as predicting cardiac arrest or preventing readmissions after surgery. But most never make their way into the clinic.
Let’s look at what has to happen before doctors and patients can benefit from a discovery:
The researcher has to write a paper with enough detail to create a working program from the heuristic, and has to publish the paper, probably in an obscure journal.
A clinician or administrator has to find the article and line up staff to write and thoroughly test a program.
If the program is to be used outside the hospital where it was created, it has to be disseminated. The hospital is unlikely to have an organization set up to package and market the program. Even if it is simply put out for free use, other institutions have to learn about it and compile it to work on their systems, in order for it to spread widely. Neither the researcher nor the hospital is likely to be compensated for the development of the program.
The program has to be integrated into the doctor’s workflow, by being put on a menu or generating an alert.
Evidence-based medicine, therefore, is failing to tap a lot of resources that could save lives. A commonly cited observation is that research findings take 17 years to go into widespread practice. That’s 17 years of unnecessary and costly suffering.
I have often advocated for better integration of analytics into everyday medical practice, and I found a company called Apervita (originally named Pervasive Health) that jumps off in the right direction. Apervita, which announced a Series A round of funding on January 7, also has potential users outside of clinical settings. Pharma companies can use it to track adverse drug events, while payers can use it to predict fraud and risks to patients. There is not much public health data in the platform yet, but they’re working on it. For instance, Leapfrog group has published hospital safety info through their platform, and Diameter Health provides an all-cause 30-day readmissions prediction for all non-maternal, non-pediatric hospitalizations.
Here’s how the sequence of events I laid out before would go using Apervita:
The researcher implements her algorithm in Python, chosen because Python is easy for non-programmers to learn and is consequently one of the most popular programming languages, particularly in the sciences. Apervita adds functions to Python to make it easy, such as RangeCompute or tables to let you compute with coefficients, and presents these through an IDE.
The researcher creates an analytic on the Apervita platform that describes and publishes the analytic, along with payment terms. Thus, the researcher derives some income from the research and has more motivation to offer the analytic publicly. Conversely, the provider pays only for usage of the analytic, and does not have to license or implement a new software package.
Clinicians search for relevant analytics and upload data to generate reports at a patient or population level. Data in popular formats such as Excel or comma-separated value (CSV) files can be uploaded manually, while programmers can automate data exchange through a RESTful web service, which is currently the most popular way of exchanging data between cooperating programs. Rick Halton, co-founder and Chief Marketing Officer of Apervita, said they are working on support for HL7’s CCD, and are interested in Blue Button+ button, although they are not ready yet to support it.
Clinicians can also make the results easy to consume through personalized dashboards (web pages showing visualizations and current information) or by triggering alerts. A typical dashboard for a hospital administrator might show a graphical thermometer indicating safety rankings at the hospital, along with numbers indicating safety grades. Each department or user could create a dashboard showing exactly what a clinician cares about at the moment–a patient assessment during an admission, or statistics needed for surgical pre-op, for instance.
Apervita builds in version control, and can automatically update user sites with corrections or new versions.
I got a demo of Apervita and found the administration pretty complex, but this seems to be a result of its focus on security and the many options it offers large enterprises to break staff into groups or teams. The bottom line is that Apervita compresses the difficult processes required to turn research into practice and offers them as steps performed through a Web interface or easy programming. Apervita claims to have shown that one intern can create as many as 50 health analytics in one week on their platform, working just from the articles in journals and web resources.
The platform encrypts web requests and is HIPAA-compliant. It can be displayed off-platform, and has been integrated with at least one EHR (OpenMRS).
Always attuned to the technical difficulties of data use, I asked Halton how the users of Apervita analytics could make sure their data formats and types match the formats and types defined by the people who created the analytics. Halton said that the key was the recognition of different ontolgies, and the ability to translate between them using easy-to-create “codesets.”
An ontology is, in general, a way of representing data and the relationships between pieces of data. SNOMED and ICD are examples of common ontologies in health care. An even simpler ontology might simply be a statement that units of a particular data field are measured in milliliters. Whether simple or complex, standard or custom-built, the ontology is specified by the creator of an analytic. If the user has data in a different ontology, a codeset can translate between the two.
As an example of Apervita’s use, a forward prediction algorithm developed by Dr. Dana Edelson and others from the University of Chicago Medical Center can predict cardiac arrests better than the commonly used VitalPAC Early Warning Score (ViEWS) or Modified Early Warning Score (MEWS). Developed from a dataset of over 250,000 patient admissions across five hospitals, “eCART” (electronic Cardiac Arrest Triage) can identify high-risk hospital ward patients and improve ICU triage decisions, often as much as 48 hours in advance.
The new funding will allow Apervita to make their interface even easier for end-users, and to solicit algorithms from leading researchers such as the Mayo Clinic.
Halton heralds Apervita as a “community” for health care analytics for authors and providers. Not only can the creators of analytics share them, but providers can create dashboards or other tools of value to a wide range of colleagues, and share them. I believe that tools like Apervita can bridge the gap between the rare well-funded health clinic with the resources to develop tools, and the thousands of scattered institutions struggling to get the information that will provide better care.