Patients’ treatment journeys often are complex as they move through lines of therapies and switch treatments. This can make it a challenge for pharmaceutical companies to accurately predict how their products, as well as those of their competitors, will do on the market. However, certain steps can be taken to improve this essential task.
“Forecasting drugs is challenging,” stated David Wolter, M.B.A., vice president of consulting services at IQVIA, during a recent webinar. While there aren’t many studies on forecasting accuracy within the pharma industry, he referenced one from 2013 that found “the majority of consensus analyst forecasts — so bank analysts for new drugs — are off by more than 40%.” He maintained that part of the reason this is so difficult is tied to “getting the patient part of the forecast correct.…How many patients are being treated, when they’re being treated and when we get the revenue and the volume associated with those patients.”
Across the patient universe, multiple differences may exist for different people. For one thing, he said, “if you just think across drugs, the treatment time is very different, from chronic drugs that patients must take for their whole lifetime, semi-chronic drugs that patients may take for a course of therapy and then one-time or acute treatments that may even be curative with some of the new therapies. So there’s a whole range across drugs that need to be thought of.”
Complexities within a drug regimen also exist “because patients are moving through what we call a treatment journey. So they’re moving between lines of therapy in some cases, between drug holidays, between changes in their disease, reactions, maybe tolerance of a drug and switching situations. So they’re in this treatment journey, and where your drug falls in that treatment journey varies, and it’s important to think of that as a forecaster.”
And finally, “each individual patient doesn’t follow the same path,” explained Wolter. “Even if you’re in the same box within the treatment journey, and you’re on the same drugs,” some people may have more or fewer doses than others, while some patients may discontinue therapy sooner or remain on it longer.
In considering treatment flow, patients will go through different stages of therapy and varying stages of their disease. “Increasingly, drugs have labels where they are in only certain parts of this journey, so they’re only for certain lines of therapy, and maybe their labels require treatment with prior therapies,” he noted. “So forecasters have to deal with this complexity of” properly segmenting the market. “When patients are in a segment and start on therapy, they don’t all stay on therapy the same length of time. So some patients might drop off after a month or two, and some patients may stay on treatment several months. And this might vary across your drug and your competitor drugs. That adds to the complexity.”
A range of forecasting models exists, from simple approaches to complex ones. According to Wolter, “simple models feel so logical, but they can miss the dynamics going on in the situation.…In simple models, you can imagine having an annual model or a quarterly model where you drop in the number of patients who start, and you say, ‘OK, in this year, we’re going to have 100 patients, and we’re going to assume all those patients stay on our therapy for the average duration’ — maybe it’s 10 months of therapy, maybe it’s 12 months of therapy. But that’s not actually what happens in the real world, of course. Patients don’t all enter at the same time of the period being modeled; they enter over time,” and they discontinue over time, which these models do not consider.
When these two issues are not accounted for, it can result in overestimation of a company’s revenue and drug volume. “So you can get fairly far off in your forecast, and not only that, but you could argue that the forecast in the launch year is crucial because that’s when companies are figuring out the size of their sales force and probably most closely monitoring the uptake of that drug,” he pointed out.
The situation can be more complex when multiple lines of therapy are involved, Wolter explained. “What happens here are companies modeling patients coming in at first line, then moving to second line, then moving to third line. Each of these lines of therapy may have different durations of therapy, different curves on how fast patients drop off, etc.…To get a more accurate volume and revenue forecast, it’s necessary to forecast those dynamics.”
One-time treatments such as chimeric antigen receptor T-cell (CAR-T) therapies that essentially are a cure as opposed to maintenance medication add an additional forecasting wrinkle. “It’s very important for companies to think about removing patients who have been cured from the system because they no longer can be candidates for treatment in future years,” he asserted. “Hopefully, as there are more cures with gene therapy and other products…that we will see more and more drugs where this is an issue, but it’s important for forecasters to take treated patients out of the system for future years.”
However, Wolter added that a “simple forecast is not wrong in all cases. It may be that a business development analysis is being done very quickly, or a back-of-the-envelope forecast is needed, or for process reasons in your company, you don’t want to go to patient flow. There are many process elements and situation elements to consider here.”
Static models, which look at a pool of patients that can be treated, “are sometimes appropriate for a single dose or short duration of therapy,” said Wolter. Dynamic models — also known as patient flow models — keep track of patients entering and leaving the pool. These models start with epidemiological data and then narrow down the patient funnel using patient flow data, he explained. “Those funnel cuts can be topics like what percent of the incident patients will receive any drug treatment, what percent will have insurance, what percent will be eligible for this particular class of therapy, taking out contraindications, etc., and then finally, what percent will be selected as patients by physicians for this particular drug of interest vs. competitive drugs. And that all leads to an available set of patients for the forecast model. All of this happens before patient flow, before we get to thinking about, ‘OK, when are they going to start and how long will they stay on?’”
Wolter explained that there are three approaches to patient flow models:
(1) Straight line, which is essentially “the absence of patient flow. It says all patients are going to come in, we’re not going to worry about what month they’re going to come in, and they’re all going to stay on therapy for the same amount of time.”
(2) Continuous flow, which models people starting treatment at different times. This approach, however, assumes all patients will be on a drug for the same length of time.
(3) Persistence flow. This is the most accurate model, he maintained, but companies must have the necessary data for this approach. This method as well starts patients on therapy at varying times but also drops off patients at different times. Moreover, it considers other factors such as the number of doses a patient takes over a period of time, the volume of dosing or number of vials administered, which can allow for an accurate forecasting for a certain period, such as a week or a month. Armed with this data, companies “can calculate volume and revenue,” he explained.
Because a drug may be indicated for only a particular line of therapy, a company “might be coming in at different parts of the patient journey, and so keeping track of this is another complexity that needs to be added,” noted Rick Johnston, Ph.D., senior software solutions lead at IQVIA, during the webinar.
For medications that have been available for a while, with consistent sales over time, “the overestimation and underestimation from patient flow won’t be that big of an issue,” Wolter stated. “It’s really when the drug is ramping up because that’s when you’re getting more patients each year and when the drug is ramping down. But of course, those are the times when the forecast might be the most important to the company.”
In addition to the launch year, Johnston said that the information also is important in the year that a drug loses exclusivity. “Those are the periods where we’re seeing a lot of dynamics, a lot of change, and when we see a lot of change, modeling these types of issues like flow and patient dynamics is really, really important.”
Traditionally, forecasting has been done using spreadsheet-based persistence models, but these “can be complex to build and slow to run, [while] forecast engines on the web are fast and user-friendly,” contended Johnston. “Basically Excel is very poorly designed to deal with this concept of flow over time, where patients are continually entering or exiting treatment. So things that you might want to track” — such as new patient starts, number of patients exiting therapy, average number being treated and the number of patients transitioning from first line to second line — “those kinds of things are poorly dealt with.”
However, he said, “there are ways to do this in Excel,” and many companies use spreadsheet models with persistence calculations in them. “And what we’re noticing is that they’re really slow.” Johnston said that IQVIA has such a tool, as well as a web-based tool that perform similar calculations. The Excel tool, though, is “not as quick. It’s probably 1/100th as quick” as the web tool.
“We see the web as being really kind of a game-changer in terms of people’s ability to be able to model this level of detail…about individual patients and their treatment progression…and how important that is…to get those…numbers right to avoid over- and underestimation,” stated Johnson.