Using Big Data and Predictive Analytics to Improve Healthcare
Search across Healthcare Data to Create a Better Patient Treatment Model
Electronic Health Records (EHR) in conjunction with Electronic Medical Records (EMR) have been steadily increasing in use over the last 15 years. In the time from 2001 to the end of 2014 EMR usage in physician offices rose from 20% to over 82%. With the introduction of the Health Information Technology for Economic and Clinical Health (HITECH) Act of 2009, Meaningful Use incentives for higher billing and reimbursement rates from the federal government continue to drive adoption rates.
With the adoption of EMRs, the increase in EHRs has grown exponentially. EHRs are a broader view of a patient’s collective individual EMR experience and contain a historical 360-degree view of a patient’s medical history. While the exchange and sharing of this EHR data has been a primary focus in recent years (through Continuity of Care Documents (CCD) and Consolidated Clinical Document Architecture (C-CDA)), the massive collection of clinical data by large health systems and treatment centers (public, private, and academic) has moved into the realm of big data.
Coming from the healthcare space, one of the things that always fascinated me was the ability to use this wealth of data to do predictive analytics on treatment plans to improve patient outcomes. With big data, big answers and meaningful analytics can be extrapolated from the healthcare continuum.
The Problem: Weighing Patient Outcomes and Operating Costs
Health systems continually face the dichotomy of medicine: improving patient outcomes and maintaining effective operating costs. It is difficult to maintain both and to give the highest level of satisfaction to a patient at the same time. By having to meander through treatment plans, trying one treatment after another, a physician continues to prolong the condition or symptoms of a condition leading to patient dissatisfaction.
Additionally, having to repeatedly see a patient without results may cause patient drift (the concept of a patient either going elsewhere or discontinuing to seek treatment), reducing revenue opportunities, and perhaps discouraging other patients from seeking treatment.
The Solution: Selecting the Best Treatment Plans with Predictive Analytics
Ultimately, the treatment plans for a patient are determined by the physicians and their patients based on what is known about the patient and their medical history. But what if physicians could be given another tool in their arsenal of medicine to make more informed choices about a patient’s treatment based on patient population cohorts? What if physicians could quickly and accurately determine the best treatment plan for any given patient based on their medical history and demographics?
Take the example presented here:
1. A patient of a certain age bracket, with certain race characteristics, a given gender, specific medical history, problem lists, allergies, etc. presents at a clinic with a newly diagnosed condition.
2. Based on this condition, taking into account what is known about a patient’s existing conditions, medications, personal history, etc., search is used to find other patients within a population cohort that are similar.
3. Based on those specific facets (now aggregations) of data, the treatment plans given to these patients could be analyzed by the physician to determine the most appropriate option for a patient.
4. This could then be used as an additional tool for the attending physician to make determinations about what their treatment plan should be, either one that they would have suggested reinforced, or another path that may not have been seen, with a level of certainty that this initial treatment plan would be best for this particular patient based on the population cohort.
A large portion of this relies on standardized, normalized, and centralized data from disparate data sources which can be done efficiently with content processing and data lakes (read more about how the data lake is leveraged in the biotech and health industries).
With the methodology outlined above, a patient could be directed to the most accurate treatment plan for their given conditions based on their existing conditions and the observed outcomes of other patients in the cohort. With this knowledge in hand, a physician can provide a treatment plan that will have a better chance of improving a patient’s outcome.
Use Cases for Predictive Analytics in Healthcare
1. Aiding Diagnosis:
In addition to identifying treatment plans, search can also be used in aiding diagnosis. With a battery of tests performed, a host of symptoms, results, and observations, guidance could be provided for potential diagnosis. Often the approach of “not knowing” or “it could be this” contribute to patient dissatisfaction and prolonged durations of being able to diagnose or misdiagnosing a condition.
2. Applying to Elective Processes:
Similarly, predictive analytics need not be limited to diagnosed chronic conditions. Search can also be applied to elective processes like physician assisted weight loss clinics for example. In this use case, a patient’s conditions are not only known, but additional data related to activity and diet are also recorded.
Other data gathering techniques like wearable health monitoring tools can be used to automatically populate EMRs or Personal Health Records (PHR) that can be consumed by the EMR. With this information, a patient can immediately be given the best treatment plan for their age, race, gender, BMI, etc. that includes exercise plans, diet plans, and any assistive medications based not only on the population cohort, but on the highly visible and measurable results of those plans. This is very crucial for something like weight loss, as patients will want to see effective results as quickly as possible to have faith in the plan they have been given, and to maintain participation in the clinic. This drives continued participation, as well as results-driven analytics encouraging recommendations to other participants.
The Future: Integrating Machine Learning to Improve Healthcare Results over Time
While showing treatment plans for patients that most closely resemble the patient being attended to by a physician are valuable to a physician, the process can be taken a step further. What if not only showing treatment plans for patients who have similar conditions as the patient being seen, the best treatment plan could be recommended based on how other patients that are in the same cohorts have responded to any given treatment?
This approach goes beyond identification of populations and treatments, and transitions into the natural extension of predictive analytics. By identifying patients and their treatment plans, the related observations associated with that treatment plan can be utilized to make determinations about what plan is best. This is a much more difficult problem to solve, as machine learning must be integrated to understand what a “positive outcome” is in relation to a treatment plan.
For example, if a patient presents with high blood pressure, the initial search would identify what treatments have been done for patients that are similar and have the same conditions. However, this still leads the manual interpretive step of the physician deciding what is best. Search can be extended beyond identification to interpretation by understanding that the condition is high blood pressure, and that a positive outcome is not only a reduction in overall blood pressure, but what other contributing factors caused that to happen. This would include observations of:
- A patient’s related conditions that may or may not impact the presenting condition
- The vitals obtained on each visit (like their blood pressure readings, weight, etc.)
- The patient’s self-reporting of their activities (like exercise)
- The medications that they (may) have been prescribed
The recommended treatment plan would not only be the best for the patient, but explained to the physician as to why the plan is the best.
Understanding every facet of the treatment plan, the related observations, and what a positive outcome is, in conjunction with the presenting condition, is what truly makes machine learning and predictive analytics useful in improving healthcare for patients. This is the power of trending analysis on specific observations related to the condition and the results of those treatment plans.
We’re very excited to announce that we’re now part of Accenture! Read the announcement here.