This is the first installment of a three-part series where I will be exploring methods of estimating physician treatment behavior. I’ll be looking at three common approaches used in healthcare market research: patient allocations, patient chart pulls, and patient simulation studies – and exploring the trade-offs associated with each.
Predicting how physicians will act in the future is at the heart of almost every healthcare commercialization challenge. Having accurate information on how physicians are using currently-available medications as well as how likely they are to use future medications informs a multitude of decisions, from pricing to positioning to in-licensing, among others. For example, creating a launch forecast for a new drug is an extremely demanding task, and there is too much at stake to risk having anything but the most accurate information possible on future physician behavior.
Patient allocation studies are perhaps the most straightforward method of estimating physician behavior. An example allocation study might ask a physician simply, “For what percentage of your breast cancer patients are you using drug a, drug b, or drug c?” If you need only simple answers, this could be an adequate approach. However, there are some potential challenges associated with allocations that users should be aware of:
As a result of the way that humans think, responses to allocation questions tend to get rounded to the nearest 5% (e.g., 3% becomes 5%, 1% becomes 0%, etc.). This increases error and ultimately means that patient allocations are on a 21-point scale. For example, I looked at this question on an internal project and found that 95% of all allocation responses ended in either a 0 or a 5 (across 5,160 total allocations).
It is also a common occurrence that shares for new products get overestimated within allocation studies. However, the allocation methodology itself offers no way to correct for overestimation other than to discount share values of new products by some number. However, this approach leaves room for lengthy debate about the value of that reduction factor.
Comparison by Patient Attributes
In conducting an allocation study, it can be difficult to look at more specific cuts of physician use. Often, it is important to know how specific doctors might behave, say in an academic setting, which could be done in the above example. However, if you want to know how certain patients will be treated (e.g., brca2+ patients), the above way of asking the question isn’t adequate. You would need additional questions specifically asking physicians what drugs they use for their brca2+ patients and what medications they use for their brca2- patients – adding two additional allocation questions. If we follow this out for all of the patient variables of that might be of interest, this can quickly add up, increasing survey length and respondent fatigue. Additionally, you may find an unexpected result and decide that you want to look at a cut of the data by a patient variable that you didn’t originally think to include.
Lastly, given the increasing complexity of treatment options available to physicians, allocations make it extremely difficult to examine how physicians are using medications in combination with one another. You might be able to identify that there is overlap somewhere among treatment options, but it’s extremely difficult to figure out which treatments are being used together and in what patients.
Given the weaknesses of allocation methodology, is there a place for them and can we can do anything to make them better? I believe that the answer to both of these questions is yes. Patient allocations are the quickest way to do market research. If you need an answer to a question fast and you have a small budget to get it done, patient allocations are the way to go. These studies can often be done quickly to get a snapshot of a therapeutic area. Think of it as a quick, 10,000-foot view, which sacrifices the ability to examine extreme levels of granularity for speed. Patient allocations are appropriate when you don’t need to cut the data by more than a few variables and when you are looking at current physician behavior only. But how can we improve the current approach to patient allocations?
Patient allocations become much more accurate with some slight modification. By using the “Wisdom of Crowds” methodology, patient allocations quickly become better and more accurate estimates of physician behavior. In short, the Wisdom of Crowds methodology has physicians answer allocation questions from the perspective of their peers Instead of reporting on their own behavior (assuming that they have a solid understanding of how their peers behave). Instead of getting estimates of use from one individual, you’re getting a perspective on use covering many individuals. If for example, each respondent can think of 10 peers they are reporting on and you have a sample of 30 physicians, you’re getting the perspective on 300 physicians using the Wisdom of Crowds methodology compared to the perspective of 30 physicians using standard allocation methodology.
Patient allocations are currently widely used (perhaps overused) to provide solutions to the questions facing the healthcare industry. Caution should be used when weighing whether or not an allocation approach is appropriate for that business questions. When patient allocations are deemed the most appropriate solution, the Wisdom of Crowds methodology should be employed to maximize the accuracy of the results.
Please stay tuned for my next blog post which will examine patient chart pulls…