Healthcare Personalization in the Age of Ubiquitous “Omics”

Big Data Evangelist, IBM

I’m learning a lot this week at Personalized Medicine World Conference 2013 in Mountain View, Calif. My chief takeaway from the event is that genomics is a hugely innovative field. You can see that in the range and depth of genomics presentations by practitioners, researchers and solution providers. It’s also quite evident in the sophistication of the many specialized genomic analytic and data management solutions on the expo floor.

dna.jpgActually, a larger takeaway is that “omics” is a hot buzzword (or, perhaps, buzz-suffix) in the life sciences these days. In addition to genomics (the study of gene sequences), there are discussions of “proteomics” (the study of the structure and functioning of proteins), “metabolomics” (study of chemical process in metabolism), and so forth. If you want a good overview of all of these, check out the Wikipedia entry on “omics.” The conference also includes a presentation on cutting-edge research into something called the “human gene connectome,” which is of interest both to evolutionary biologists and to those exploring the roots of hereditary diseases associated with particular regions, races and nationalities.

My interest in “omics” is more than just a personal infatuation with new scientific lingo. In the healthcare arena, “omics” refers to new sources of big data and to a wide range of cutting-edge analytics that are key enablers for fine-tuned healthcare personalization. After all, we can’t truly say we’re delivering personalized treatments if we aren’t basing them on the patient’s unique genomic, proteomic and metabolomic make-up—as well as on each patient’s unique profile of “nutri-” “pharmaco-,” “toxico-,” “psycho-” interactions at and among these various molecular levels.

Of course, healthcare personalization demands more than just a 360-degree low-level molecular portrait of each patient. For starters, a full 360-degree view must include details on an individual’s health history, medications, family status, diet, exercise, temperament and the like. Just as important, the view must also consider the extent to which an individual matches other patients on various health-relevant dimensions, such as whether they have experienced similar health issues or undergone similar treatments. We can’t fully personalize healthcare delivery if we don’t first identify whether a specific individual’s case is truly unique, or whether it has clear precedents in others who were given specific treatments that resulted in desirable outcomes.

All of which brings me to “patient similarity analytics,” the topic that I co-presented with Jianying Hu, IBM manager of Healthcare Analytics Research. It refers to an IBM-developed decision-support technology that helps healthcare providers and payers answer the following critical question: from historical data, which other types of patients were similar to a given patient—in characteristics, treatments and outcomes—for the purpose of identifying the most appropriate physician and treatment plan for a particular condition faced by this specific patient?

Incorporated into the recently released IBM Patient Care and Insights solution, patient similarity analytics works as follows:

  • Analyzes aggregated demographic, social, clinical and financial factors along with unstructured data such as physicians’ notes
  • Factors the specific health history of each individual patient into the creation of a personalized healthcare delivery plan.
  • Enables healthcare professionals to examine thousands of patient characteristics at once to generate personalized treatment plans.
  • Identifies other patients with similar clinical characteristics to see what treatments were most effective or what complications they may have encountered.
  • Supports patient-physician matching so an individual is paired with a doctor that is optimal for a specific condition.
  • Allows healthcare professionals to better tap into the collective memory of the care delivery system to uncover new levels of tailored insight or “early identifiers” from historical/long term patient data

Patient similarity analytics is a big-data application par excellence. The tsunami of electronic healthcare record data powering patient similarity analytics flows from the care delivery network of healthcare providers and payers. In most real-world usage scenarios, the volumes of data involved, the variety of data sources, and the time-criticality of the analyses will push patient similarity analytics into big-data territory in no time.

It’s also a big-data application for which cloud-based big-data platforms might prove optimal. Massively parallel computation fabrics are absolutely essential to support scalable statistical analysis, predictive modeling, patient stratification and visual cohort refinement on data sets in excess of 10,000 features. The analytics pipeline for calculating customized similarity patterns, with learned context and endpoint-specific distance metric tailored to specific purpose (outcome, diagnosis, utilization etc.), demands heavy-hitting horsepower.

Is there “omics” inside patient similarity analytics? Not yet, given the application’s initial focus on aggregating electronic medical records. But nothing is stopping any IBM customer or partner from adding, aggregating and analyzing new data sources, “omics” or otherwise.

Related Information