Virtually all of the risk adjustment models with which I’m familiar are based on retrospective data. This data clearly has some predictive benefits – maybe it’s too cliché to say the past is prologue – and is already in our hands.
To look at just one example of what existing data archives can do, we need go no further than the pages of this blog. Late last year, I shared the story of a group of French hospitals which are working to predict admission rates as much as 15 days in advance by mining a store of historical data. Not surprisingly, the group’s key data includes 10 years’ worth of admission records.
The thing is, using historical data may not be as helpful when you’re trying to develop risk-adjustment models. After all, among other problems, the metrics by which evaluate care shift over time, and our understanding of disease states changes as well, so using such models to improve care and outcomes has its limitations.
I’ve been thinking about these issues since John shared some information on a risk-adjustment tool which leverages relevant patient care data collected almost in real time.
The Midas Hospital Risk Adjustment Model, which is created specifically for single organizations, samples anywhere from 20 to 600 metrics, which can include data on mortality, hospital-acquired complications, unplanned readmission, lengths of stay and charges. It’s built using the Midas Health Analytics Platform, which comes from a group within healthcare services company Conduent. The platform captures data across hospital functional areas and aggregates it for use in care management
The Midas team chooses what metrics to include using its in-house tools, which include a data warehouse populated with records on more than 100 million claims as well as data from more than 800 hospitals.
What makes the Midas model special, Conduent says, is that it incorporates a near-time feed of health data from hospital information systems. One of the key advantages to doing so is that rather than basing its analysis on ICD-9 data, which was in use until relatively recently, it can leverage clinically-detailed ICD-10 data, the company says.
The result of this process is a model which is far more capable of isolating small but meaningful differences between individual patients, Conduent says. Then, using this model, hospitals risk-adjust clinical and financial outcomes data by provider for hospitalized patients, and hopefully, have a better basis for making future decisions.
This approach sounds desirable (though I don’t know if it’s actually new). We probably need to move in the direction of using fresh data when analyzing care trends. I suspect few hospitals or health system would have the resources to take this on today, but it’s something to consider.
Still, I’d want to know two things before digging into Midas further. First, while the idea sounds good, is there evidence to suggest that collecting recent data offers superior clinical results? And in that vein, how much of an improvement does it offer relative to analysis of historical data? Until we know these things, it’s hard to tell what we’ve got here.