Can Data Analytics Help Solve our Healthcare Crisis?
The US spends far more than any other nation on healthcare. In 2009, health care costs reached $2.5 trillion dollars - - more than $8000/person and 17% of GDP, up from 13.4% in 2000. Yet, as is widely cited in healthcare circles, the US fails to achieve acceptable results from all this spending, lagging some other industrial countries that spend far less.
The healthcare industry is now turning to information and analytics to improve quality, while simultaneously reducing costs. This is similar to what many other industries have done - from retail to manufacturing - - taking advantage of the internet, access to data, and data analytics to transform and succeed. Some say it can’t be done with healthcare, but at a recent University of Maryland/Smith School of Business Workshop, cosponsored by IBM’s Analytics Solution Center, speakers provided many examples of how analytics are helping in the healthcare field.
Dr. Thomas Cantillina, Lt Col, USAF, Deputy Chief Medical Information Officer and keynote speaker, discussed how the Air Force is now collecting and digitizing not just the data needed for billing but medical data about each patient. This “big data”collection of 95 TB can now be used to learn from their mistakes and develop better medical practices. Dr. Cantilliana talked about a virtuous cycle that the Air Force was trying to create where the data will be analyzed by researchers and then new best practices instilled in the workflow and automation systems, which then collect more data to further improve the healthcare provided. He cautioned the audience that important medical data is still in the “White Space,” because either the patient was not interacting with the USAF’s medical providers, or the information does not follow the airman or dependents when the person moves.
Professor Ritu Agarwal, Director, Center for Health Information and Decision Systems, Smith School, next spoke about the promise of big data. She gave 5 areas where she thought data analytics could help: Gaining new insights into disease progression, Understanding the effectiveness of interventions, “Discovering” new knowledge, Speeding the pace of science, and Moving towards personalized medicine. Professor Agarwal then describe how the University of Maryland is helping the Military Health System to develop a health services data warehouse which will be used for research into condition/disease specific research, healthcare delivery research, and data mining of new relationships that can be discovered from the data. She hopes to better understand such problems as obesity, PTSD, and TBI (Traumatic Brain Injury) to provide better diagnosis and treatment. UMD is just starting to provide data warehouse access to the broader federal research community, so much more to come from this work.
Dr. Basit Chaudhry, IBM Research Scientist, talked about “Putting Watson to Work in Healthcare.” He described how IBM is working with the medical community to make Watson into a decision support system for doctors. Doctors today have problems dealing with the huge mass of data, including patient history, current symptoms and test results, and all the medical literature that might pertain. A decision support system, such as several healthcare providers are developing with IBM, could continuously ingest the latest medical literature and help doctors decide on next steps in patient care.
Jianying Hu, IBM Manager of Healthcare Analytics Research, described a physician’s care delivery analytics tool which could compare a new patient with other patients in the tool to determine if the patient was on the right treatment plan and had a similar pattern of utilizing healthcare services, and then could predict the patient’s risk of congestive heart failure(CHF). A longitudinal study of the predictive CHF analysis showed a 1 year accuracy of 78%. The work on patient similarity analytics could also be used to help with comparative effectiveness research. While patient similarity can be valuable, one concern in a clinical setting is the issue of sharing confidential patient data. Jianying showed an approach around this by sharing model data rather than specific individual data.
Margret Bjarnadottir, Smith School of Business, started off by showing a graph indicating that the use of big data for health care research has dramatically grown in just the last 5 years. She then described some of her work in predicting future patient health care costs. She showed that she could provide good predictions using only previous cost data, ignoring medical information, except for existing high cost members. Professor Bjarnadottiir then described an active drug surveillance system using claims data. She found that claims data (money) can be a key indicator in drug surveillance.
Next up was Rashmi Mathur, an IBM Business Analytics consultant. She described work being done on behalf of Healthcare Insurance companies. In the first one, a major Blue Cross health care plan wanted to optimize its customer communications between voice and digital modes to reduce service costs and improve customer satisfaction. Through a regression analysis, her team determined that patients requiring more doctor visits were more likely to need voice communications. In one unexplained anomaly, students were also more likely to want voice communications. The second study used text analysis to develop customer specific education requirements based on life and medical events.
Professor Bruce Golden, Francis-Merrick Chair in Management Science at Smith School of Business, started by noting the emergence of healthcare analytics talks at INFORMS meetings starting in approximately 2005 and currently running over 250 per annual meeting. He then presented 3 examples of the use of analytics to improve hospitals. In the first example, the team used simulations to maximize cardiac surgery throughput at University of Maryland Medical Center resulting in a 20% increase. In the second example, the team looked at how quality varies within hospitals between shifts for trauma cases. Using regression analysis, the team concluded that outcomes for night/early am admissions were worse than those admitted during the daytime in spite of the fact that waits for surgery were shorter at night. Comparing data between hospitals, they found that the decrease in quality care is largest at small (presumably resource constrained) hospitals. The final study looked at patient discharge and readmission rates. Not surprisingly, the discharge of patients increases as bed utilization increases. This results in an increase of readmissions, implying that some of these patients were discharged before medical prudence would dictate. Dr Golden noted that there are more opportunities to apply analytics to healthcare than the OR community can handle – a growth industry.
A Panel of subject matter experts – Guodong Gao, Smith School; Karoline Mortensen, UMD School of Public Health; William Rollow, IBM Global Services; Abdul Shaikh, NCI; Terry Sharrer, retired-INOVA – then discussed the question of the day: Can Analytics and Big Data have a Major Impact on Healthcare Costs? A lively discussion then ensued. My conclusion listening to the discussion is that so many factors are contributing to driving UP the costs of healthcare (including the aging of the population to which I’m a part) that much greater research, development, and application of data and analytics is needed in order to have a chance at bending the cost curve.
The presentations from The Second Annual Robert H. Smith School of Business and IBM Business Analytics Workshop are located here. Other presentations from the Analytics Solution Center can be found here.
What are your thoughts for applying analytics to health care?