Balancing technology services for our customers is a tremendous challenge. Figuring out where to invest scarce IT funds has always been difficult. In Life Sciences, the confluence of protected data (PHI) and ‘big data’ creates an even higher hurdle than those we’ve faced in the past. At the UMass Medical School, we’ve tailored an IT strategy that leverages the strengths of our great institution, and works to avoid the costly IT pitfalls of over-engineering, and relying on what’s worked in the past.
Fifteen or twenty years ago, IT departments throughout the Life Sciences world were forced to rethink their compute infrastructures when high performance computing (HPC) became the norm. This wasn’t easy, and many IT departments tried to fit this new square peg of HPC into their existing round hole of data center expertise. That didn’t work out well for many institutions, and led to labs building their own mini-clusters. For IT to truly support research at that point, specific HPC skills and a nimble, well thought out strategy were required. It also required shifting IT investments from data center services to HPC. This shift in thinking at UMass resulted in a centralized, high performance cluster that rivals premier institutions around the country.
"Changing strategic course is what CIO’s are required to do"
About five or ten years ago, the advent of Next Gen Sequencing, high throughput screening, modern microscopy and other instrumentation forced the need for more powerful data aggregation. Once again, IT’s strengths at the time didn’t match well with this new challenge. Relational databases, the bread and butter of IT, are perfect for transaction systems, and lousy for data analytics. Since IT departments are rarely given a new source of funding to meet these challenges, the answer was to retool. UMass Medical School shifted from Oracle to NoSQL solutions, like MongoDB. Once again, the key was shifting resources and investments, taking risks, and communicating closely with the labs.
Today, maturing electronic medical record systems provide an opportunity to link structured patient data with unstructured research data such as sequenced patient DNA, and PACS imaging output. Precision medicine that leverage these combined data sources can tailor therapies for individual patients and speed research. Herein lies the latest challenge for IT. The security and compliance needs of patient data require IT to invest heavily in powerful information security frameworks. The marrying of structured and unstructured data tests the abilities of our data engineers. And the hardware infrastructure needed to support all of this grows by the day. What is an IT department to do with this latest challenge?
The temptation is to invest heavily in the perfect, scalable, secure data warehouse. If IT can provide perfectly aggregated data, researchers and clinicians should be able to analyze data more rapidly, bringing novel treatments to bear more quickly, right? The complexity of this task is nearly overwhelming, however, and the aggregation goals of today (e.g. genomic scores by phenotype) will most certainly change faster than an IT group can keep up. There is no google map that IT can create to traverse biological pathways.
In this new scenario, building a “perfect” data warehouse is not simply the enemy of good data access. It is the death knell of cost effective, rapid analytics. A successful research ecosystem simplifies access to data, and provides powerful consultative services and tools for accessing and analyzing that data. Academic Medical Centers have robust Institutional Review Boards (IRB) and data governance structures in place. IT does it best work when it leverages those structures instead of designing a new ‘data governance mouse trap.’
Here is the roadmap UMass Medical School is following to enhance our research ecosystem:
• Start with the analytics—the end product. Hire talented data scientists who have gravitated to the IT world. Standardize on cheap, powerful analytical tools that your researchers probably already use (don’t force IT mandated analytics on them—you will have a very quiet phone).
• Digitalize your IRB’s data access mechanism. Don’t reinvent the wheel, but instead leverage and enhance your institutions data governance model. One great way is to streamline identified data access requests through online workflows. Work closely with your clinical partners (UMass Memorial Health Care, in our case) to ensure that Quality governance is included in the same data access mechanism.
• Build the cheapest, fastest infrastructure you can, and plan on refreshing it often. Don’t overspend. Don’t over design. Work fast. Utilize the Cloud where possible.
• Don’t worry about structuring your data lake too carefully. Whatever structure feels right today will be upended when a new data source comes along that we never predicted.
• Work on data curation. This is really hard, and really essential. IT cannot be in the life sciences data cleanup business like we are for administrative systems. The magnitude is just too great, and the life sciences field is too specialized. Empower those who create data to curate. Technology frameworks like TAMR enable curation to belong to those closest to the data, instead of the poor IT folks trying their best to keep up in the background.
We are beginning to see the benefits of this approach already. Changing strategic course is what CIO’s are required to do, and this turn of IT’s rudder may help improve healthcare at a critical juncture.