Monday, December 12, 2011

Sending the Questions to the Data

As hospitals and practices form accountable care organizations, they will accelerate their efforts to build healthcare information exchanges and novel analytics that support community-wide lifetime care rather than siloed episodic care,   This requires "freeing the data" from the EHRs, hospital information systems, and laboratories in which it resides.

There are two basic ways to analyze data for a panel or population.

1.  Send the data from multiple sources to a central repository for analysis.

BIDMC has partnered with the Massachusetts eHealth Collaborative on such an approach to build a quality data center supporting its ACO strategy.

2.  Send the question to the data.

The new federal Query Health initiative is a standards-based approach that enables standardized questions to be sent to multiple federated databases without moving the data itself.

In Massachusetts, we've implemented such an architecture in two ways.

I2B2/Shrine which links together the Harvard hospitals (and many other sites nationwide) with query tools supporting clinical trials and clinical research.

MDPHNet, an ONC funded Challenge grant which sends questions to data sources,  answering public health questions.

MDPHnet is being developed under contract with the Massachusetts eHealth Institute to implement a secure web-based query tool which enables predefined and ad hoc queries to be sent to participating sites, including selected practices within the Mass League of Community Health Centers and potentially, Atrius Health.

Queries are executed locally, securely returned after optional review, and then presented to the requester and displayed in a variety of ways - heat map, histogram, table etc.  Results contain no patient-identifiable data.  Data holders control authorization of requesters and their specific query capabilities.

The current focus for predefined reports is syndromic surveillance (Influenza-like illness) and chronic disease surveillance (diabetes).  It can also support other uses, such as pharmacovigilance and quality measurement.

MDPHnet uses PopMedNet open source software developed by the Harvard Medical School Department of Population Medicine at the Harvard Pilgrim Health Care Institute, with support from AHRQ and FDA. Lincoln Peak is co-developer.

There is great synergy among i2b2, PopMedNet and MDPHnet, since they use a common architectural approach.  Query Health incorporates PopMedNet in its design.

MDPHnet uses the Electronic Health Record Support of Public Health (ESP) common data model.  ESP  was developed by the HMS/HPHCI Department of Population Medicine with support from a CDC Center for Excellence in Public Health Informatics 

The Massachusetts League of Community Health Centers transforms data from their clinical data warehouse into the ESP format. Commonwealth Informatics supports the process as needed.  Additional participants will extract data from their EHR and put it into the same schema (ESP) with help from Commonwealth Informatics.

MDPHnet can be readily expanded to cover other datasources such as the I2B2 nodes which are hosted at over 60 sites nationwide.

Over the next few years I believe that for many use cases we will be sending questions to the data instead of sending the data to centralized registries.    I2B2,  MDPHnet, and Query Health will show us how.

1 comment:

Kimo said...

In thinking about the Query Health Initiative. I can see why; utilizing solutions based on traditional technologies, it would make sense to standardize the types of queries sent to a federated system of medical data repositories, but have to question the usefulness of such a solution.

Maybe I'm misunderstanding the big picture, but my thought is that they're thinking that by limiting or restricting the types of queries being run, they'll be able to more effectively handle the solution from a computational perspective, and actually have queries returned in a reasonable time frame, as opposed to waiting hours and in some cases days for results to come back.

The problem with that aspect of what's being proposed is that it leads one to question the ultimate utility of such a solution. The ultimate solution needs to be flexible enough to be useful to a broad group of users and organizations without sacrificing performance, and adding expensive hardware.

Webmedx Quality Analytics (now Nuance) is a good example of the type of solution you're talking about here, BayScribe's Semantic Search is another. They have the ability to search against many multiple data repositories, work with both structured and unstructured documents, provide sub second query times across the board (even as data sets scale to over a billion records), and operate on only a fraction of the hardware that other like solutions require.

I realize that the standardization of queries is likely just one aspect to the Query Health Initiative, but as a user, I'm looking for a solution that doesn't limit me in terms of the data I need to access.