Solid programs won’t last on shaky data foundations

There’s a great reason why many cultures and religions – around the globe and across time – have at least one allegory about the importance of building a structure upon a solid foundation: It’s nearly universally true. A house built on sand or mud is likely to slide and crumble.

This is no less true in the Land of 1s and 0s, which is where we in the healthcare IT community live. The most exciting new *thing* -- be it a software or service – may be shiny and promise to open new horizons but if its base is weak, it likely will underperform/fail over time.

There are many exciting breakthroughs we see and there clearly will be many more to come. The promise of machine learning and the even-more-heralded Artificial Intelligence (AI) is only now starting to come to fruition.

To wit: Researchers at Massachusetts General Hospital have announced that they were able, via machine learning, to use clinical data in their electronic health records (EHR) system to estimate the chances a patient would develop dementia. This clearly is an exciting development in the study of a tragic disease.

I have two related concerns about these kinds of announcements and, to be clear, they are far more about the longevity of the potential projects than the advancements they may bring:

1. Data quality – Machines do their learning with the information they are presented with. The old adage about chicken salad applies here … it applies everywhere! We cannot be sure we are getting legitimate outcomes if there are questions about data in the systems used.
2. Patient ID/matching – This is critical. We must be able to connect data to real people and yet the industry continues to whistle past the graveyard on this issue.

These incredible new things people continue to roll out really have the potential to change lives … if they are available universally. That means pulling data from the myriad EHR systems in use.

How clean is the data in those systems (that are not at one of the nation’s preeminent research institutions)? Will it be meaningful? And how will we be able to help Mrs. Jones (vs. Mrs. Smith) if we still don’t have a viable patient-matching system?

The healthcare IT community must decide these two things are absolute priorities – and then act to remedy them (we don’t need another steering committee) – lest so many great ideas end up thrown upon the scrap heap because “they didn’t work” when, in fact, we set them up to fail by not ensuring a solid foundation.