$3 Million Prize Offered to Solve Hospital Admissions Puzzle

The head of an accountable care organization-like physician network is offering a hefty prize for the design of a predictive model that can be used to prevent unnecessary hospitalizations and readmissions – and potentially realize savings of up to $30 billion nationwide.

9 comments on "$3 Million Prize Offered to Solve Hospital Admissions Puzzle"
Rebecca Mitchell (2/6/2011 at 12:53 PM)

@Roy: I couldn't agree more, as a doctor I don't know how I could possibly make admission decisions without a patient's age and weight, they lead to entirely different differentials for what could be happening. (ie, skinny child likely has gastric reflux, whereas obese sixty year old is having a heart attack). Also this problem is already being addressed but not funded elsewhere, as discussed by Atul Gawande recently: http://bit.ly/g3B4Wr Also 3 mill is a paltry sum compared to what this is worth-how about instead providing start up capitol to a few of the best and the brightest to tackle the question? (ie Y Combinator)
Heidi Kirsch (2/5/2011 at 2:26 PM)

One of the most important variables that COULD be utilized to account for the dynamic "human component" is missing from consideration. It's so simplistically obvious, I'm surprised it has been overlooked in any of the efforts I can reseach for reference. I plan to pursue exploration of this through other channels.
Quinn (2/2/2011 at 1:21 PM)

The real question is whether "human doctors" are more accurate? Apparently, they are terribly inaccurate. So if doctors are only 50% correct in evaluating a patient for admission, wouldn't 80% be better? Overworked physicians are no good at accuracy.
IV (2/2/2011 at 12:04 AM)

"The human element" cannot be removed completely. However, to be effective, it doesn't have to be. If this is 80% predictive, it will more than pay for itself fairly quickly. You don't have to get it perfect. There are a myriad of machine learning/data mining algorithms that can be applied to the data to see if it is feasible to use for heuristics, although it will have to be adjusted on a case by case basis. The trick is always using the right size data set to allow the learning to be thorough and the test set to be telling. The data set in the description seems to fit this, which means if the problem is solved, it might be distributed to a large enough set of test sites to allow it a fair trial.
roy (2/1/2011 at 10:29 AM)

I think by not providing the patient's race, ethnicity, weight, age, socioeconomic level or geographic area of residence you are witholding meaningful information that impacts hospitalization rates. Like it or not, someone with a condition that requires hospitalization and who has good health insurance and money is probably more likely to be hospitalized than someone who doesn't and has the same condition. Similarly I'm sure older or heavier patients are more likely to require hospitalization than younger or lighter ones with similar conditions. Finally, I bet the closer you live to a hospital the more likely you are to be hospitalized.
Andy (2/1/2011 at 2:35 AM)

Heidi, I totally agree and was going to say something similar. I think somethings will be taken out of the equation, particularly some "human" components, such as hypochondriac conditions, so these will likely deal with legitimate health relapses. To be able to accurately predict the second data set you will need all pertinent factors that go into the second set, which isn't just unlikely, it's probably near impossible. Also, you will need not one person to just be part of that miracle who displayed the same initial circumstances, but miraculously doesn't have the same end case (safer to assume that perhaps it was omitted from the data set than the other improbability). This seems like a good way to lower costs, but it is absolutely ridiculous to accept nothing less than 100% when 95% would save billions of dollars. This seems like a bad waste of up to 400 dollars and lots of time.
Noah F (1/31/2011 at 3:59 PM)

There is a trivial solution, of course: kill all your patients the first time. Or perhaps you don't want game theorists working on this problem for you.
Suezanne C. Baskerville (1/27/2011 at 2:05 AM)

The story says "The $3 prize is approximately double that of the Nobel Prize in Medicine." Proofreading matters.
Heidi Kirsch (1/6/2011 at 11:04 AM)

This is not a REAL mathematical or statistical challenge. This is working with and correlating pre-determined subsets for a defined group. The formula would be "IF combinations of A,B,C exist....then X,Y,Z is the result." What this does NOT take into consideration is the unpredictable HUMAN component. You CANNOT begin to predict individual human lifestyle choices or variable environmental factors, even when patients are well aware they have specific health-altering diagnoses. Even with a formula which accurately "predicts" the hospitalization of the first set of 100,000 patients, the formula will only be able to continue to "predict" for the second patient set based on the current constant factors identified and [INVALID]ed for in the first patient set. This isn't science. It's basic analytical logic. To create this requested "sort" feature is child's play. I admire Dr. Merkin for trying to predict readmission rates and lower the unnecessary wasted expenditure of health care dollars. I believe there are steps which WOULD accomplish the means he seeks, but not via this particular route. When the independent variable of human nature can be factored into this dependent equation, he may have something." [INVALID] Heidi J. Kirsch, LPN, BA, MLT(ASCPCM), Process Improvement Analyst, HIM Covenant Medical Center, Waterloo Wheaton Franciscan Healthcare – Iowa


FREE e-Newsletters Join the Council Subscribe to HL magazine


100 Winners Circle Suite 300
Brentwood, TN 37027


About | Advertise | Terms of Use | Privacy Policy | Reprints/Permissions | Contact
© HealthLeaders Media 2015 a division of BLR All rights reserved.