We seem to be missing member 24027423 in DaysInHospital_Y2.csv (he has claims that year, so should be there) -- can you give us the days for that member in that year?

S/he will be added in the next release. 

Thanks.

("He": I checked before posting.  Obviously…)

It seems that we need to predict days in hospital only for people with claims in the previous year.

I wonder what is the reason for it because I guess that practically there are also people with claims in earlier years but no claims in the previous year.

Uri Bass,

Thanks for your comment,

where did you find out that we only need to predict days in hospital for patients that made a claim in the previous year?  As you say, there are patients that may not make a claim one year but will the next, so it seems odd to excluded these patients.

I look forward to your reply - I am searching the HPN rules!

Thank you,

Sam

Hey SSRC,

Well the rule is technically that we need to predict Days In Hospital for patients in the "Target.csv" file that was in the released dataset.

It just so happens that those are precisely the same people with Year 3 claims. :-) 

A more precise answer is from the FAQ (http://www.heritagehealthprize.com/c/hhp/Details/FAQ):

Why do the DaysInHospital_Yx Tables contain only a subset of all members?

The DaysInHospital_Yx Tables only contain the members who are eligible to make a claim in Yx and the 12-month period prior to Yx. For these purposes, "eligible to make a claim" refers to individuals who were HPN members in both those years. For example, the DaysInHospital_Y4 Table contains HPN members who each were eligible to make a claim in Y4 and Y3. A member can become ineligible to make a claim if he/she ceases to be a HPN member.

Of note, there does not seem to be a prohibition against using previous years' (Y1 and Y2) data to predict the Target values if data exists -- if a person happens to be in DaysInHospital_Y2 and Y3, you can use that as an indicator in your algorithm apparently.  But the data was selected only for the eligible members as described in the FAQ, which I think answers your question.

Sam,I found it based on analyzing of the data

I did not find that I need to predict number of days even for one person who did not make claims in the previous year.