Provider-Based Population Health

It’s an enticing notion: focus resources on the sickest patients if you want to drive savings in health care, the 20% of the patients that drive 80% of the costs (or the 5% that drive 43% of Medicare‘s costs).

As I write, smart people are getting caught in the appeal of the notion. We can solve a big chunk of our health care crisis if we just take advantage of the 80-20 rule! There are health system administrators busy instructing data analysts to mine their data for high-cost patients. Those same analysts will be instructed to pass the information along to case managers who will then be told to bring order to the spiraling costs. When the administrators review their efforts, it may even look like they worked: the patients who were “case managed” showed lower costs in the period after they started getting managed. The problem is that there really wasn’t any other outcome possible.

I’ve been involved in health care analytics in one form or another for the past 15 years. The most common analytic error by far involves concluding that the inevitable was instead caused by intervention. One example is showing savings on a disease management program for heart failure patients in which the patients were enrolled right after they’ve had a hospital stay for a heart attack. The statistical problem with such an analysis is called “regression to the mean.” In practical terms, it means that most patients don’t stay acutely ill all the time. If you measure future outcomes versus the time a population was sickest, odds are that at least some of the population will improve and the outcomes will look better.

Back when I ran the analytics department at a disease management company, we did research in which we found that even for the rare diseases we worked with (diseases such as multiple sclerosis, cystic fibrosis and systemic lupus) the highest cost patients in one year were not always the highest cost patients the next. Generally, there was more than a 50% turnover of the top 10% of patients from year to year. That means that if you took the 1,000 highest costs patients in 2009, 500 or less would still be high cost patients in 2010. Usually, it was because they had some kind of an acute crisis from which they recovered. Whether we did nothing or provided intensive case management to the costliest patients, about 50% were going to get better and their costs were going down (which is why we didn’t just focus on the sickest of the sick).

One of the standard quality metrics followed by groups that track health care quality is the percentage of diabetic patients with an HbA1c value of greater than 9%. (The Wisconsin Collaborative for Health Care Quality does a good job of creating visibility around this metric.) Programs are currently being designed to target the 9%ers, to bring their scores back in line. The challenge with focusing only on the 9%ers is the same as with the high cost patients. Here at Phytel, we’ve done research on the turnover rate of the 9% patients. We’re finding that like any other outlier, they change from year to year – at a rate of over 60%. That is, for every 1,000 patients with an HbA1c score of 9% or more in 2008, only 400 of them continued to have a score of over 9% in 2009. Despite the turnover of the 9%ers, the overall percentage of patients with 9%+ HbA1c scores has remained relatively stable from year to year (at least in the data we’ve analyzed). In other words, as soon as one 9%er drops to a lower level, another is waiting to take his or her place.

I grew up in the North East where we had frozen ponds most of the winter. That meant a lot of pond hockey. The best players were always the ones that skated to where the puck was going to be rather than to where it was at any point in time. The same approach is critical when it comes to affecting the small pool of patients that drive the bulk of health care costs. You need to find them early, before they’re a high cost patient.

Why is this issue important? I wrote in the second paragraph that some health system administrators are busy creating case management programs designed to target their sickest patients. One of the reasons is to prepare their organization to take on global payments, with the goal of becoming an accountable care organization. The reasoning is that if they can just remove unnecessary costs for the sickest patients, they’ll save enough money to make those global payments profitable. But their efforts are akin to Sisyphus rolling his rock up the hill only to have it fall once he gets to the top. No sooner will a high cost patient start improving in this scenario than another will be waiting to take his or her place. If the “manage sickest 5, 10 or 20%” approach is employed on a grand scale, we won’t move any closer to bending the health care cost trend curve.

The trick is finding sick patients before a crisis event, before they’re really sick. The availability of lab results data from electronic medical records is critical in that regard. And once you find them, affecting those patients requires a skill set that our health system (which includes more than only the providers themselves) is just beginning to develop – that of reaching out to patients and motivating them to actually modify their behavior. One just has to reference the massive efforts (both public and private) that have gone into reducing the number of smokers to understand how significant this challenge truly is.

Ted Courtemanche is the Vice President of Analytics and Outcomes at Phytel.