There is a saying that “In God we trust… but all others must bring data”. We agree. Without data, we trust no one. And we want analytics too. But even that is not always enough because those who crunch data in healthcare sometimes cannot deliver contextual precision. Here are a few examples:
Population health management systems are focusing on the wrong population – Data analytics companies strive to identify that segment of the population that consumes the most $$. The concept is logical – crafting tailor made intervention programs for those patients should help reduce costs. Unfortunately, this strategy misses the point that many in that population cohort will not respond to further intervention. Luckily, we are seeing some companies that are using a conditional approach to risk stratification; identifying patients that can best respond to intervention programs. And not surprisingly this methodology does not always lead to patients consuming the largest resources, but it can save the biggest $$ overtime.
Genetics interpretation industry may be working with an outdated calculator – The advances in the field of genomics are some of the most important scientific accomplishments of our time. Genomics hold enormous promise for personalized medicine which in turn can single handedly reduce our cost of care by billions of dollars. But the data analytics behind the field are still largely around correlative engines. Because we have the ability to crunch a sea of data and find correlations between them, does not necessarily guarantee that we have discovered causation. This is one of the biggest challenges of genomic interpretation. That said, we believe the dark road between association and function will gradually be illuminated as we find new approaches beyond DNA analysis such as the study of RNA, epigenetic regulations, etc.
Claims and structured EMR data are not enough for modeling predictive analytics – Using these data sets has led to some success in treatment and intervention. But the content residing in the narratives of EMRs and the inclusion of genomics information (mostly presented as PDFs for now) needs to be parsed, computed, and analyzed before we can achieve full-fidelity analytics.
There is no doubt that the continuing convergence of several technologies will accelerate progress in the field of data science. But let us acknowledge that not all data is created equal!