Data Mining, Algorithm Medicine and Skeptics

February 21, 2013

My recent Perspective on iHealthbeat focused on the uses of data mining of EMR data which are yet to be fully exploited. My thoughts were provoked by a New York Times article titled, Mining Electronic Records for Revealing Health Data.  Although data mining in healthcare has gotten a bad reputation, an approach which respects privacy and a focus on research discovery can yield important results. The potential uses of EMRs in research is another opportunity yet to be realized.

A new article in The Atlantic,  The Robot Will See You Now, discusses IBM Watson and other initiatives moving medicine toward what I call Algorithm Medicine and Artificial Intelligence. The potential of mining EMRs to generate real-time clinical decision support has exciting possibilities.  However, there are skeptics, especially when the predictions expand to entertain the idea of replacing physicians. Realizing the limitations of technology must be acknowledge. For instance, the concerning problem of copy-and-paste in EMRs would have a negative affect on data mining those records. Also, data mining has presents real challenges both in defining research questions and finding the correct data to answer those questions.

So data mining shows promise but a realistic approach without wild predictions can lead to real discovery and impact on practice.


Share this:
  • Facebook
  • Twitter
  • Delicious
  • Digg
  • Tumblr
  • StumbleUpon
  • Pinterest
  • Add to favorites
  • Email
  • RSS

One Response to “Data Mining, Algorithm Medicine and Skeptics”

  1. Michelle Drabik says:

    This morning I read a blog post which used the term “data informed” rather than “data driven,” which really speaks to the points you note above. Those of us who have mined and analyzed trend data with considerably less automated methods for years (my background is tech library/information management/competitive intelligence) recognize the enormous value of informing decisions with solid data analysis. Uncertainty is reduced as the relevant data sampled increases, but current methods limit how much data can be analyzed in a timely manner; so, at a certain point, the value realized with the amount of additional data that can be included using current methods becomes incremental. As we reach the incremental improvement point, the only way to boost impact is to include a huge amount of additional data — thus the promise of big data mining. The higher the sample population, he more refined the data model delivering the trend analysis can become, as well. In the end, a human needs to make the decisions, based on situational requirements. Physicians will continue to be the interpreters of data trends and resolve best bet solutions with the particular needs of the patient. Personally, the better informed my doc’s decisions are, the better I’ll feel about his/her suggested treatment plans. I can’t think of a time in history when better information resulted in overall negative outcomes. Great post, John.

Leave a Reply