By: Daniel McLaughlin, M.H.A.
The expansion of Electronic Health Records is presenting an unprecedented opportunity to make significant improvements in the American health care system. However, for this opportunity to be realized, new methods of data management and analysis that are uncommon in health care will need to be deployed.
Organizations that have mature electronic health records have conquered the challenge of moving data from operating systems into data warehouses and are using them for substantial improvements. For example, a question that had challenged researchers for many years was whether traditional low-priced blood pressure control was as effective as newer, more expensive drugs. To answer this question, NIH conducted an extensive trial that took eight years and cost $120 million. The results indicated that: the oldest and cheapest of the drugs, known as thiazide-type diuretics, were more effective at reducing hypertension than the newer, more expensive ones.
However, some patients did not respond to these drugs and needed to use the newer drugs – but which ones? Unfortunately, NIH did not have the funds to conduct a follow up study. By the time the NIH study was complete, however, Kaiser Permanente had an extensive electronic health record and data warehouse. By using real patient data in their warehouse and traditional statistical methods, the researchers had the answer in 18 months for $200,000.
Although traditional scientific methods and statistical tools work well for some health care questions, they cannot be easily applied to many interesting questions such as:
- Which doctors have the most cost effective risk adjusted care patterns based on actual cost of care – not charges?
- What are the characteristics of patients that can predict the level of non-compliance with discharge orders and the probability of re-admissions?
The challenge of answering these questions is best illustrated by the complexity of the data bases. A standard electronic health record for a patient will have over 2,700 fields. A charge master for a hospital can easily contain 20,000 separate services and prices. Traditional statistical methods flounder in this environment.
Fortunately, data mining professionals (particularly in retail) have developed new tools such as market basket analysis, classification algorithms, association rules, cluster analysis and neural networks to understand these massive data bases. Hopefully, these techniques will soon migrate to health care to support substantial improvements in care delivery.
To learn more about how the new tools of data mining and other technologies are changing the business of health care, attend the UST Executive Conference on the Future of Health Care on Friday, November 9, 2012 at the University of St. Thomas Minneapolis campus.