On Friday, June 6, 2014 7:29:26 AM UTC-4, vaib wrote: > Hi, > > > > I'd like to solve the problem of predicting patient bill (inpatient or outpatient) given his symptoms as he/she arrives in any hospital. > > > > Now I think I can do this without much effort but the problem I am facing is that of sample data. Can anyone point to any such data source they might be aware of? > > > > Thanks!
I work in a large academic medical center and I think that there are too many variables to get an accurate prediction. I took a recent month in which we had a bit over 4,000 admissions. The average number of admitting diagnoses per admission was 8.4 (sd 5.4) with a median of 7 diagnoses. There are hundreds of diagnoses, so the number of admissions with a specific set of diagnoses would be a rather small sample set for predicting final billings.
You could go just by the primary admitting diagnosis but there were about 1,300 different primary admitting diagnoses, the most frequently occurring only 92 times. And just using primary admitting diagnosis ignores complications; for example, someone coming in with a heart attack as the primary and only admitting diagnosis would not match well with another patient with a heart attack, pneumonia, and diabetes.
If you can come up with some good grouping mechanism, you could then look at historical billing patterns and try to do some projections.