Normally I would centralize both the data *and* the application in most cases of ERP systems, but you're dealing with some unusual factors when it comes to healthcare. I am familiar with an implementation where standalone applications with their own datastores, i.e. in-memory were used. While you would want to aggregate data whenever possible, these terminals needed to be taking into operating theaters where cables were considered obstructive to personnel pacing around and wifi connections were disabled for fear of affecting sensitive equipment.

However another consideration to make here is the fact that you're dealing with data that enjoys a much greater deal of protection than corporate data, dealing with privacy in the public sector is an incredibly sensitive issue. One problem the above project ran into was the risk of data leaks through theft of the laptops they were using. While you can mitigate this risk by caching the exact datasets you need when disconnecting from the network and keeping the rest stored server-side, I believe it was still found unacceptable and in the case of operating theaters would have needed wallmounted terminals with a fixed connection, which would lead us back to a data warehouse.

I'm unsure where the project went from there, but in any case I think the answer to your question is going to depend on the functionality you need, the constraints you're dealing with and the hardware infrastructure you come up with to support this.

Assuming; the amount of data is over few millions of subjects, each of such has few thousand clinical variables from demographics, procedures, medications, vital signs and the list goes for ever. The task can be expensive in both financially and effort. My opinion, I would look for existing platforms serving the clinical domain to organize data into warehouse structure . Ex: I2B2(https://www.i2b2.org/), OMOP (http://omop.fnih.org/ETLProcess). Once we have the structured data, can configure BI tools like http://www.pentaho.com/ to do analytics on top.

Thanks for the suggestions, Rick. Actually I am only currently thinking of 'small data' and mostly it would be in some sort of structured or tabular form, but needing the capacity to scale up and evolve. I think a hybrid approach would be ideal, but there is the cost factor to consider also. And thanks Venkateswara for the suggestions of the existing platforms. I have heard if i2b2, but I thought that was specifically meant for research data - I shall investigate!

Hi Susan

You might consider NoSQL if you want something that scale up with performance and economy and evolve with heteregenous data sources,  I have a PhD Student who is working in the Healthcare sector as a practitioner in this space and his company has done some amazing things with NoSQL for a distributed patient electronic healthcare records system that has been implemented nationally. He is now looking at a whole range of opportunities for decision support systems sitting on top on this system

Best regards Michael