patient record systems; patient billing systems; data marts with PHI and HR.
IN all cases we find that our test systems have to mimic the loaded production systems as closely as possible. We have tried various strategies to de-indentify the data. Sometimes it works, mostly it fails to be maintained and that causes testing to drift back to production clones.
My experience is based in a very large patient traffic setting. Even the presence of a small percentage of 'failures' such as billing rejects costs large sums of money. Because no set of 'model data' has ever really shown these small scale production failures, testing is driven to use production clones. We take all due care that only authorized users have access to PHI and this has greatly complicated testing as we must do it with production staff.
An estimate of the cost of implementing an effective (for testing purposes) de-identified data set has run into the multi-million dollars and sits waiting cost justification and identification of funding sources.
We mainly use 'model data sets' for early stage functional verification, not for final testing.
-- Wayne Wilson An attachment containing my pgp-signature is included. My public key fingerprint is: 9325 05AD 866B BCCB 45BF E86A 63E1 C6ED 4130 5461 My public key can be downloaded from wwwkeys.us.pgp.net
pgp00000.pgp
Description: PGP signature
