In a former role, I was often called upon to assess the merit of health analytics proposals. The well-intentioned pitches often promised all the right things; to leverage information assets to give analysts the right data – anywhere, any time, and to target known opportunities to reduce health care costs and improve health outcomes. Frequently, an initial proposed step involved gathering data assets into a consistent format. Unfortunately, more often than not, when funded, even the seemingly best opportunities failed to deliver on their initial promise.
How can we best gauge which offers should be pursued? Obviously, there is no perfect answer, but here are four factors to consider that reduce the risk of getting less than we bargain for:
1) What business problem are we trying to solve?
To achieve success in our endeavors, we need to be clear on what we are trying to deliver. Vague notions often result in vague solutions and vague success. For example, “Build a data warehouse” may result in a pile of code and a pile of data that might be used if analysts can find what they want in it. But if we have a specific business problem we are trying to solve, like “Reduce patient risk for complications following outpatient surgeries,” then we have a guiding focus on a tangible target that tells us what data to go after, how to process it, and what to do with the insights we glean.
2) How do we know it is a problem
I once reviewed a proposal to identify new targets for treatment pre-approvals that promised to reduce the costs associated with unnecessary treatments by a wide margin. But when digging into the details, I found that the initiative had forecast savings for treatment denials, but had not considered the cost of alternate treatments; after all, simply denying a particular treatment will not make a patient better. After adjusting for this oversight, the net result was that overall, the proposal would not result in significantly lower costs (compared to the cost of implementation), nor did it project significantly better health outcomes. The takeaway was that the existing process balanced patient needs, doctor recommendations, and care delivery as well as, or better than, the proposal.
We must take care to make sure we are pursuing a solution to something that is a solvable problem. Quite often we gravitate toward issues in our own sphere and work to solve them, only to find that sometimes, in optimizing locally, we have made the end-to-end outcome no better, or actually worse. Often what we see as a problem is actually a balancing of opposing concerns that, over time, have reached equilibrium. That isn’t to say that the current balance is optimal for all time – sometimes the greatest opportunities arise from new forces coming into play (like technology innovation) and driving new equilibriums.
So how can we recognize solvable problems? One key indicator is that variation exists in outcomes across locations, regions, providers, treatments, and other differentiators. Where these variations exist, our analyses can identify the drivers for the differences, and we can work to uniformly optimize all outcomes. Another indicator is when others’ outcomes are better than ours – it signals that potential improvements are possible; we have the opportunity to leverage our own data and analytics to follow (and potentially improve upon) their lead.
3) What does our success look like?
It isn’t enough to just deliver data and information. Gathering data across our systems, analyzing that data, and then trumpeting our insights doesn’t do much toward affecting real-world outcomes. Our solution should envision pushing those insights back into our processes (human and technical) and monitoring outcomes for the impact of our efforts. With this end-to-end view of our problem, we will more often deliver end-to-end solutions that have real-world impact.
We must also accept that our data and tools limit what we can achieve. Health analytics insights are limited by three factors: (1) the amount of data we have; (2) the quality of that data; and (3) available analytics tools. The more data we have, the more subtle the effects we can detect. The better the quality of our data, the less that the more subtle effects will be lost in noise. The more sensitive our analytics tools (including technical platform and effect-specific analysis methods), the more subtle the effects we can uncover.
Regarding the amount of data, we only have as much as we have, which is a function of how many patients we have had across the span of time under consideration. This is often a function of how long we have been operating, but also a function of the lifespan of the particular situation we ultimately will affect. If we haven’t enough data of our own to support the analyses, we might consider acquiring additional relevant data from other external sources. However, given that the data would have been produced under a process different than our own, there may be some impediment in pushing insights gleaned from that data back into our own processes effectively.
4)How much should we be willing to invest?
It may seem obvious that if the proposal would result in a net improvement in costs or health outcomes relative to its implementation cost, then we should support the proposal. However, oftentimes the bottleneck for health analytics efforts is less about funding and more about organizational bandwidth. Knowledgeable experts spanning operations, care delivery, and data analytics are typically a scarce resource and should be applied to efforts that give the greatest return. Given this, the value of a proposal should be weighed against the opportunity cost of being able to pursue other proposals based on organizational bandwidth – or in other words, an investors’ view.
Organizations that adopt this investor’s view of weighing health analytics proposal values against one another will generate a history of return-on-investment for funded proposals. They often will find that they initially realize a factor of four or more times return on their investments. Then, as time goes on and the backlog of high-value proposals is depleted, the realized value will settle to an average of two times return on the investment. With this understanding, data analysts putting together a proposal only need a good estimate of the value of gross improvements (cost savings and health outcomes), but do not need a detailed estimate of implementation costs. Instead, decision-makers can make funding decisions based initially on delivery team estimates, whether the proposal can be fulfilled for one-quarter to one-half (or whatever current fraction is indicated by their funding history) of the value of gross improvements. This can speed along promising health analytics initiatives while avoiding development of detailed estimates to deliver less promising ones.
Time is also an important consideration when deciding which health analytics efforts to pursue. It is important not only for the obvious concern for timeliness of realized benefits, but also from a risk perspective. Often the longer that a health analytics initiative takes, the less likely it is to produce promised outcomes. This is because modern data analytics technology largely removes the uncertainty of technical performance, leaving issues associated with health data and analytics protocols to drive the timeline. To that end, any health analytics initiative should be able to produce demonstrable results within a short time-frame of 4-6 weeks; otherwise it should be re-evaluated.
Finally, despite considerations of focus, value, outcomes, organizational bandwidth, return, and timeliness, another important factor is strategy. Looking back, sometimes we see that the proposals we funded were not exactly aligned with our stated strategy – it can be said that strategy is how you spent your time and money! In some cases, proposals that according to all measures are less attractive can have longer-term benefits when executed as part of an overall strategy, since such efforts are often reinforcing. Special consideration should be given to proposals that align with current organizational strategy.
In summary, health analytics efforts require scarce organizational resources that must be allocated to initiatives based on both cost and health outcomes. When we have a specific problem that is known to be solvable, the right approach to affecting real-world outcomes, streamlined decision processes, and robust, competitive funding practices, then our health analytics investments will more often yield the positive results for our patients and our communities that they promise.