Health outcomes analysis

Reliable answers to life's most critical questions

In the United States, we collect better quality information and provide more advanced analytics on baseball players than we do on our personal health. Every time a batter steps up to the plate or a kicker lines up for a field-goal attempt, the announcers have access to every possible statistic needed to tell us, in an instant, the chances that player will hit a home run or split the uprights. So when it comes to our health, why wouldn't we want our physicians to have the same tools?

Odds are, you or someone in your life will someday be diagnosed with cancer or another unexpected life-threatening disease. In an instant, your thoughts will race with uncertainty, worry and fear.  To help you put your risk in perspective, physicians will attempt to talk in terms of probabilities - but most risk estimates are not based on data and most people, including physicians, do not really have a complete grasp of how probabilities work.   

To complicate matters, you have to balance multiple probabilities:

• Chances of treatment success (disease cure, disease recurrence, disease control)
• Chances of side effects (short-term or long-term, temporary or permanent, mild or severe)
• Chances the disease will not pose a threat to you within your lifetime without treatment

What's even more frustrating is that most clinical probabilities are derived from clinical trials based on subjects who don't seem at all like you. In order to derive statistically valid conclusions, clinical trials must be designed to control for unknown factors by selecting patients who are clinically comparable.

Unfortunately, the world does not have enough time or money to run clinical trials for all combinations of patient characteristics, diseases, disease states and interventions (drugs, surgery, radiation therapy, or other procedures). Given this reality, is there any way to tell how someone like you will respond to the available treatment options? Do the negative side effects outweigh the benefits of the treatment?

For cancer, your oncologist provides the most reliable treatment recommendations, but both you and your oncologist would benefit from seeing real data on treatment outcomes for patients who more closely match your profile.

SAS has launched an ambitious project on two fronts to help clinical specialists, such as oncologists, and patients get answers to these types of questions. First, SAS has partnered with the National Cancer Institute (NCI) to build a cancer outcomes database that consolidates clinical data from participating NCI centers. On the second front, SAS has built and continues to expand a new solution called SAS Health Outcomes Analysis. 

Delivered from a secure hosted facility located at SAS world headquarters in Cary, NC, SAS Health Outcomes Analysis provides authorized physicians with a historical view of their patients with the option to analyze treatments and outcomes for other, similar patients. Although statistical cause-and-effect conclusions may not be possible in most queries, SAS Health Outcomes Analysis helps physicians make decisions based on published clinical guidelines as well as field results that are broader than their own personal experiences.

Single cohort analysis
In the first release of SAS Health Outcomes Analysis, we focused on a single cohort analysis. Based on a selected patient and disease, the system searches the database for patients who have the same disease and match a disease-specific set of risk factors. Once this collection of patients has been identified and saved into a temporary analysis data mart, the physician can refine the cohort analysis by adjusting the risk factors (using filters).

The system provides prebuilt cohort analyses that compare historical treatment sequences, which are built to automatically capture the interventions that were administered to the patients of the cohort. The treatment sequence can include drugs taken, chemotherapy, surgery and radiation therapy, as well as other procedures.

Analyses include basic summary statistics, distribution plots, survival analysis (time to event), and comparison statistics, such as odds ratio and least squares means. Using more advanced techniques, the system utilizes pattern and sequence recognition methods to identify opportunities for better outcomes as well as unexpected side effects (adverse events). 

Using these results, the health care provider can confirm or reject assumptions based on personal experience and avoid what Brafman and Brafman (2008) refer to as “value attribution” – basing treatment-plan decision on perceived values rather than objective data. 

Expandable design
SAS Health Outcomes Analysis has been designed to easily add new analysis and reporting beyond the single cohort analysis. The solution will support cohort comparison and selected surveillance tasks.

Although it provides a built-in Web interface, the solution was designed to support integration with other systems, such as electronic health records systems. This would enable health care providers to access our analytic services from within their usual hospital interfaces.

Although the analytics used for observational health care data can be extremely challenging, the reason we describe our project as "ambitious" is because the data to support these types of analyses is not readily available, structured, standardized or complete. While current initiatives to implement electronic health records will improve the availability of the data, they will most likely not solve the other challenges.

Structured data
Most electronic health record systems still capture information in methods that simulate paper. In other words, the information is stored as text rather than discrete fields that can be easily analyzed. Most vendors are increasing the amount of structure they provide, but more work is needed to enforce structure while still easing the workload of the physician. Perhaps touch pad and smart phone technologies will enable more efficient and accurate collection of physician notes, observations and decisions.

As we wait for more structure to be built into the system, SAS utilizes text mining and text analytics to improve the value of unstructured historical data. Through content categorization and pattern recognition, SAS can help structure the data as well as analyze the data in its current form.

Standardization
Critical information is often collected and structured but encoded with proprietary coding schemes that vary from vendor to vendor. For example, to analyze lab results over time, the system must be able to identify similar tests and use the same units of measure. Depending on where the lab tests were performed, the lab test codes and possibly the units of measure may be different. Analysis of data without standards or semantic interoperability produces meaningless results. 

Working with third-party terminology experts, we apply our DataFlux® data quality software to encode new information, retaining the original coding of the data while mapping each to a selected standard.

Completeness
In the US, the personal medical histories of most patients are spread over multiple providers and health insurance systems. None of them talk to each other. Analyzing data from one hospital only provides a partial picture of a patient's past medical history. Other countries may be in a better situation related to completeness of medical data.

On the leading edge
While the lack of quality data poses a significant challenge to routine outcomes analysis, SAS is leading the way by developing analytical methods using high performance computing and service-oriented interfaces to provide health care providers and researchers with critical information as it becomes available. Today, organizations can take advantage of our work if they have a large consistent data warehouse with common standards within the organization.   Let’s take our enthusiasm for sports statistics into the world of healthcare. 

Bio: Eric Brinsfield is an R&D Director for SAS health and life sciences.  In addition to extensive experience in clinical trials, he has also focused on drug safety signal detection, health outcomes analysis and comparative effectiveness.

References:
Brafman, Ori, and Rom Brafman. Sway: the Irresistible Pull of Irrational Behavior. New York: Doubleday, 2008.