Hacker News new | past | comments | ask | show | jobs | submit login

I've done study of these exact datasets with Splunk and applied machine learning to detect anomalies in prescribed drugs comparing each doctor/provider behavior to his peer group.

Doing this on a past HHS/CMS datasets - i was able to discover anomalies that was proven to be actual fraud/crime:

Was interesting to discover anomaly in 2015 provider behavior data and then read about that same provider just being raided by DEA in 2018.

Interesting data!

https://imgur.com/a/eIySC

https://imgur.com/a/5BTo6




Do you have any papers or writeups on your methodology and dataset? Would be very interesting to see whether this could direct law enforcement (or, barring that, making it public knowledge so consumers can choose to avoid doctors where there is evidence of questionable behavior).


Yes. Here was initial write up:

https://www.splunk.com/blog/2017/09/28/building-a-60-billion...

https://www.splunk.com/blog/2017/10/03/building-a-60-billion...

Also I combined prescription dataset with provider payment dataset and it allows to discover potential conflict of interest where provider got compensated by drug manufacturer (speaking fees, stock, conferences).

Good idea to make this data publicly searchable.

It’s not that difficult.

So anyone could search for his favorite doctor name and find how much and for which drugs the doctor billed Medicare for. And how often, how much and when he was compensated by drug manufacturers.


Did you publish your findings? How do I get my hands on the data?


Data is available on data.cms.gov

I published initial research (with 2014 datasets) on splunkbase.com Search “Splunk security essentials for fraud detection” app.

(Free app with anonymized data embedded into it. You need to download/install free Splunk to use it).

Latest discovery for 2015 datasets (for which I attached snapshots) are not published yet - I plan to do it.

Also - I’ll be speaking about it next week (Splunk for fraud detection seminar)


why cant this be used by the law enforcement as probable cause for shutting down dodgy providers?


Because some statistical analysis showing a particular entity is an outlier isn't sufficient evidence to prove they are drug dealers. It is enough to start an investigation, of course.


They do!

Similar approaches are used to identify Medicare and Medicaid fraud.




Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: