Prev
The Heritage Health Prize , a $3 million competition to predict who will go to hospital and for how long, was launched on Apr 4.
The data will be released in three waves. The Apr 4 launch allows people to register and download the first instalment, which includes enough data for people to start trying out models. It includes claims data from Y1, information on members and the details of hospitalizations recorded in Y2.
The next instalment, released on May 4, will include a more comprehensive dataset, with claims for later years and the test dataset against which entries will be judged. At this point the competition will be open to entries, reveal the performance threshold and begin posting the leaderboard.
Finally, the last release happens on June 4 and will include some ancillary data of prescriptions and lab tests.
The official site for the competition is
www.heritagehealthprize.com/c/hhp
People can also sign in to the competition with the Kaggle account.
From the
HHP Evaluation rules
Entries will be judged based on the degree of accuracy of their predictions of DaysInHospital for Y4 (or if applicable Y5), carried to six (6) decimal places. An entry's predictive accuracy will be judged by comparing (i) the predicted number of days a member will spend in the hospital reflected in the entry and (ii) the actual number of days a member spent in the hospital as reflected in the Scoring Data Set. Prediction accuracy will be evaluated based on the following metric:
Where:
- i is a member;
- n is the total number of members;
- p is the predicted number of days spent in hospital for member i in the test period;
- a is the actual number of days spent in hospital for member i in the test period.