What you are allowed to do will depend both on the participation contracts you have signed with the individual participants, your ethics approvals and the individual legislation of the participants countries (which should really be reflected in the contract, but might not be). An ideal model for registry participation is the STS Society for thoracic Surgeons (USA). This is one of the largest, longest running and most comprehensive clinical registries in the world with over 5 million records! It is only recently become international, however. The Duke Clinical Research Institute hosts the database and uses best practice management processes. There are many resources for the data managers also online. You might find some information re practice for data quality here as there are some technical refs there. Try exploring this site starting here. You may also find an expert contact here from Duke who can provide some advice.

http://www.sts.org/sts-national-database/database-participants

this question is legislative issue rather than technical (database) issue .Specially it deals with directly with profiling and personal issues

Mehdi, I guess this question is in relation to your KAPPA database, it looks like great project to assist with improvement of haemophilia A patients treatment. I am the data manager for the Australian idiopathic pulmonary fibrosis registry, so there are many similarities. I think you might need to consider what is the purpose of exporting data into your 'research' database. If you are managing the data, this is probably not considered 'research' per se and you may have more scope in the use of identifiers for this process. You also need to consider what your regulations (ie ethics, legislation, contract) allows with regard to de-identification and re identification of records. The definition of these terms vary and can be very confusing. In Australia we have many differences across our states regarding privacy legislation, similar to the US and I am sure for the countries of Europe also. Trans border data flows are always more difficult and restrictive, but I think The OECD has worked to provide some more unified legislation. It is probably useful to become familiar with the health data privacy issues in the registry research field. Here is an interesting document that might assist - note that many useful documents may not be published in the scientific peer reviewed literature. http://www.euroreach.net/sites/default/files/EuroREACH-WP3%20Final%20Report.pdf

This document also highlights a couple of data matching processes from Australia -the Western Australian model WASM Is considered a global leader, however this is for linkage between disparate systems. You should be able to create a simpler solution with your single database. Perhaps what you are really looking for is a pseudonimisation process that allows you to work on the anonymous data, but re identify it for the participants

Medhi and Susan, The NCDR is a US registry sponsored by the American College of Cardiology (ACC) for cardio procedure and patient outcomes, mainly. It was started in 1997, and contacts there may be useful for addressing some of your administrative and research safeguarding issues. They have annual conferences with over 2,000 hospitals participating in the registries, and much research and quality improvement work spins out of its use. https://www.ncdr.com/webncdr/home/trainingeducation

Another point of contact is a group that have contact information with key stakeholders that additionally could provide assistance around any of the issues you are grappling with, which can be found here: . http://www.linkedin.com/groups?gid=4360721&trk=myg_ugrp_ovr

I recently submitted a letter about "under use" of the NCDR PCI registry concerning a study about timely treatment for AMI, where an available date element in that registry (the NCDR has six) representing time of pain onset was ignored in the study I was commenting about. You may see my brief comments here https://www.researchgate.net/publication/257944896_Letter_to_NEJM_on_Timely_treatment_for_AMI_with_PCI?ev=prf_pub which references the published data dictionary for this particular NCDR registry.

Data Letter to NEJM on Timely treatment for AMI with PCI

Thank you very much Dear Susan and Edwin. I appreciate your time and effort on answering my query. You are quite experienced and I hope I could get your help in case again in the future.

Mehdi, maybe that your question "What are best practice to link our data to the source data in anonymous way?" can't have an answer. It strongly depends on your local legislation. In Italy data can be reckoned as anonymous only if they cannot be referred, in no way, with the original personal data.

This is what our privacy law says:

"dato anonimo", il dato che in origine, o a seguito di trattamento, non può essere associato ad un interessato identificato o identificabile"

("anonymous data", the data that originally, or following treatment, cannot be associated to a person identified or identifiable ").

In the past, the Italian Privacy Authority said:

L'informazione originariamente non associabile ad uno specifico interessato (c.d. dato anonimo) può divenire "dato personale" ex art. 1 della legge n. 675/1996 allorché, attraverso una successiva operazione di collegamento ad informazioni di diversa natura, risulti comunque idonea a rendere identificabile un soggetto. Ne consegue che, ai sensi dell’art. 27, comma 3 della legge n. 675/1996, in mancanza di specifiche norme di legge o di regolamento, non può ritenersi consentita la comunicazione a privati, da parte di un soggetto pubblico, di dati statistici apparentemente anonimi, qualora il campione dei dati da analizzare, benché richiesto per scopi scientifici e di ricerca, per genere e consistenza numerica, consenta di risalire ai diretti interessati.

(The information originally not associated to a specific person (so-called anonymous data) can become "personal data" [...], through a subsequent operation of connection to information of different nature, become valid to identify a subject. It follows that, [...] in the absence of specific laws or regulations, can not be allowed the transferring, by public organisations to private organisations, statistical data, apparently anonymous, if the data sample to be analyzed, although required for scientific or research purposes, by type and numerical consistency, allow to refer to the specific individuals.)

Sorry for the translation if it is not perfect but I hope it's clear.

So, if you find a practice to link your data to the source data you obtain only pseudonimisation, as Susan said. Is this allowed in your country?

I Suggest you design some SQL Code to extract the data are needed. It is possible to remove Data is not allowed to be in your research data and exchange the data is necessary with sequential number in same SQL code. But, it is depend on source Database and research structures. in fact, you must delete any information is define personal information.

Something like global unique identifiers would work ? Using one way hash from combination of determined variables that uniquely identify subjects. Every time you want to link data; regenerate hash to match up with the existing hash. NDAR uses similar approach developed by simons foundation for their autism database. http://jamia.bmj.com/content/17/6/689.abstract

Hello, since the previous answers didn't mention it, I think you should look at anonymization techniques. The point is that even if you remove all obviously identifying fields, like names, unique IDs and so forth, it may still be possible to uniquely identify records by the association of particular fields. For example, Sweeney demonstrated about 10 years ago that 87% of the american population could be uniquely identified by an aggregation of Gender, Age and Zipcode where they lived.

Narayanan and Shmatikov also could deanonymize data published by netflix for research purposes, data that had been previously anonymized by removal of identifying fields.

If you are interested in this, I would suggest you look at these specific terms: k-anonymization, l-diversity, t-closeness. These all apply to manipulations of database records. You lose in accuracy of the data, but you gain in privacy. If you want to keep the data as exact as possible but only for aggregate queries, then you might be better served by other techniques like data perturbation and permutation of sensitive data.