There are two major matching challenges in EHM:

1.   matching intervention to control groups in order to make evaluation of results scientifically valid and reliable; and
2.   matching the risk/reward potential and preferences of individual EHM participants to interventions that are most likely to deliver the desired results and return on investment (ROI)

To some extent, the two challenges can interfere with each other.  When matched control vs. intervention groups are needed, pure science would argue that the population of interest should be split in half in order to achieve comparable samples.  This automatically reduces the potential gain through a successful intervention by half, since it would reduce the number participating in that intervention by that much.  This would likely be enough to discourage employers from doing any matching, since their success and financial gains would be so markedly reduced.

One employer, at least, got around this dilemma by choosing as its control group employees who were like participants in the EHM intervention in all important respects, save that they worked for another employer.  This enabled rigorous matching of controls to the members of the intervention group, while still enabling the vast majority of its employees to participate, thereby maximizing the gain that was found to have resulted. [B. NAydeck, et al. “The Impact of the Highmark Employee Wellness Program on 4-Year Healthcare Costs” JOEM 52:2 Feb 2008  146-156

Matching the intervention to what has the best chance of succeeding with each individual participant, while adding greatly to the probability and amount of gains likely, will also add to the costs of the intervention, hence threaten ROI ratios in particular, and amounts as well, though not as dramatically.  To make the best matches, a lot has first to be learned about individual participants.  Then, this learning must be applied to customizing the interventions, making it more difficult to achieve economies of scale, as well as threatening the scientific rigor of the evaluation.

The science of learning which prospective participants have the greatest risk of future costs, in both the immediate and long terms, has been dramatically improved in recent years.  The science of learning how best to predict the chances of success for individuals, and for tailoring the intervention to a mix of what can be justified considering that chance, and what will most likely realize that chance, is still in the dark or barely light ages.

Fortunately, the examples of the technology being used to predict costs can guide the development of technologies used to optimize chances of success.  A host of insurance firms, for example, have decades of history in using technologies to gauge the risk of individuals.  This same technology should prove equally useful in predicting the likelihood of EHM success.

Instead of using predictive modeling (PM) to “underwrite” populations, select who should be rejected or charged more because of their health risks, the growing number of insurance firms that are in the EHM business could use it to select the best prospects.  PM technologies would have to be further developed to determine the best method for intervening with individuals, predicted costs vs. success, but the technology is certainly capable of doing that.

Tailored interventions have proven themselves, at least in the few cases where they have been tried.  A customized asthma intervention program, for example, was able to increase the number of symptom-free nights among participants, reduce ER use by 37%, and enable three times as many patients to have their asthma under control as an undifferentiated educational effort. [“Tailored Asthma Intervention Shows Promise” Yahoo! News,  Apr 10, 2008 ]

Tailoring has already been partially applied in EHM, with populations typically divided into low, medium, and high-risk segments, and differentiated interventions based on the costs vs. potential of such segments used in EHM, even with just healthcare cost reductions in mind.  Once EHM includes the total economic benefit of its success — counting healthcare, disability and workers compensation claims along with absenteeism and presenteeism reduction, plus productivity and performance improvements – a far more accurate and optimistic expectation of its value should enable better matching.

While matching interventions to predicted gains and participant characteristics that affect their success chances based on a few segments is better than nothing, it is nowhere near as promising as is matching by individuals.  There are likely to be far more than three different cohorts of participants, since the factors that determine probability, extent, and best method to achieve success are likely to measure in the dozens, if not more.  And differentiating for dozens of segments, or even tailoring to individuals will cost more, it is likely to yield more, as well.

If there are three or even four of five segments, the intervention must be geared to a probable gain that reflects only the average of all.  Since half of the members of the segment will normally promise above average returns, while half promise below, the use of the same intervention for everyone within that segment is simply the same as a one-size fits all approach, multiplied by three, four or five.

Using interventions designed for cohorts of one each will have a far greater chance of succeeding, plus should ensure that success for no member of the population never costs more than it is worth.  And if the gains for every individual participant exceed the costs for that participant, the overall results must be success, where with segments, it depends entirely on how well the intervention works with the roughly half of the members of the segment who have better than average potential.

The interventions and costs for individuals can be set based on whichever intervention yields the best combination of probability of success times economic gain potential, not simply some standard amount below potential, in order to ensure acceptable results.  Planning everyone’s intervention to cost 10%, 30% or even 50% of the potential, while ensuring, in theory, a positive ROI, would simply ensure the precise gain as the ROI ratio chosen delivers.  Picking the best combination of gains times probability could do far better, by not ensuring costs must be any set amount.

The technology of PM is already close to being where it needs to be in order to achieve individual participant matching.  All that is needed is to learn enough about which interventions work best in EHM based on the characteristics of participants, not just of the interventions.  This may take some years to perfect, but it is something that we ought to be working on already.