Intelligent Data From Large
Data Sets

AnswerMine Healthcare Services

The healthcare industry is facing ever-growing demands for information related to performance across a spectrum of areas - clinical, satisfaction, financial, to name a few. These demands come from and affect providers, pharmaceutical companies, health plans, government and other parts of the health care marketplace. Huge pools of data are available for analysis, but often sit underused.

AnswerMine and its staff have experience working with a variety of forms of health care data. We have worked with claims, patient reported outcomes, and electronic health record data. Our approach of "letting the data speak" can provide both operations and research-oriented investigators with a productive way to pull value from huge data sets. By starting with outcomes of interest, and looking for the clusters of characteristics that are associated with those outcomes, we can drive hypothesis testing much faster than traditional methods of analysis.

For example, a provider group might be interested in examining care of diabetic patients. The outcomes of interest might be HbA1c and cholesterol level. The first step would be designating results that might be classified as high, normal, and low for each test. We would then find patients whose results are low HbA1c-low cholesterol, high HbA1c-low cholesterol, etc. The next step would be to find the characteristics (or variables) from the available data associated with each pair of findings. Assuming we have clinical, demographic and administrative data, we might find that patients in the low-low group are 50-64 years of age, have no co-morbidities, have their HbA1c checked semi-annually, and live in a certain geographic area (proxy for socioeconomic conditions). The high-high group might be 20-35, overweight, be asthmatic, and on 5 medications. The stratification gives someone trying to understand the care being delivered the information necessary to plan, execute, and monitor a quality improvement effort. Targeted interventions of this sort would not be possible using average results, or examining clinical information in silos.

Our methods can also be used in a research setting when one is interested in hypothesis generation - determining what opportunities may exist and what variables might be most promising to study.