Life sciences companies need to be agile in an often-changing market, and forecasting plays a crucial role in being able to do this. The increasing focus on the use of machine learning (ML) and artificial intelligence (AI) can make parts of forecasters’ jobs easier, but there is room for improvement in these and other areas. A recent IQVIA webinar discussed trends in commercial forecasting based on the company’s second annual survey, which was conducted earlier this year.
“This is a very exciting time to be in forecasting,” declared David Wolter, a vice president in IQVIA’s commercial consulting group. And while forecasting is critical to making business decisions, multiple challenges exist within the process.
One challenge is around patient dynamics. With more specialty drugs available, this leads to “more issues to deal with: longer and more complex dosing regimens; multiple lines of therapy; duration that differs, and often requires modeling persistence curves; switching and compliance topics.” All these complexities will need to be dealt with when tracking patients and subsequently forecasting how many will be treated, he claimed.
The second issue is what Wolter called “glocal,” or global and local coordination. This, he said, is “getting the local insights in a way that also enables global comparison across products and within a product across countries and then also across products.”
The third challenge is seen with tools and platforms, which are undergoing “a dynamic change,” he maintained. “The new web functionality is creating new opportunities for companies.” And with the large amounts of data that firms now have access to, they need tools that can provide “speedy calculations and quick outputs for senior management.”
Finally, what is potentially the “most interesting” challenge is in the integration of data and analytics. “We are fortunate to be in a world where there’s more and more data,” noted Wolter. “There’s not always enough data or the right data we want for forecasting, but it is improving all the time, and it’s an opportunity for all of us as forecasters to incorporate that data, including real-world data, and to integrate new analytics, including machine learning, and to get the best out of our forecasting.”
IQVIA’S web-based survey focused on three key topics: the current state of forecasting, challenges with forecasting and desired improvements in forecasting. It involved approximately 117 respondents from mostly pharmaceutical companies but also people from pharma support firms, medical device establishments and academic institutions, among others.
Current State of Forecasting Shows Room for Improvement
When asked to what extent respondents use forecasting to inform strategic decisions, 88% said that they had either moderate to high use. “So I think we can all feel good as a community that these forecasts are informing important decisions for drug companies,” Wolter remarked.
Forecasters said they spend an average of 40% of their time on data collection and reviewing results with stakeholders, making those the top two areas. This finding, he said, is “pretty positive. I think what we want as a community is spending time on the higher-level things, like insights and strategy conversations, and less time on the more manual tasks like entering data or verifying calculations,” which together represented less than 20% of forecasters’ time.
As far as the tools that respondents are using, Excel remains the dominant one, but approximately 64% are using at least one web-enabled option, such as a web tool or a web-enabled spreadsheet or dashboard. This, he maintained, reveals that “the industry is going through a shift…as new web tools become available.”
Sales data is the top type of real-world data forecasters use, found the survey, followed by primary market research and claims data.
In response to all the attention on ML and AI, IQVIA added some additional questions to the most recent survey. Respondents said that trending — via “machine learning-type trending algorithms,” surmised IQVIA — and market segmentation are the top uses for ML in forecasting. When respondents were asked about their company’s level of sophistication in the use of automation and ML, the No. 1 response was three out of five — with five being very high — followed by one out of five, demonstrating “a lot of room for improvement,” observed Wolter. The lack of forecasters ranking their company at five out of five was “not surprising. This is in some ways a new area.”
Market Access Analysis Remains Challenging
Among analysis gaps in forecasting, respondents rated market access analysis as the top challenge, followed by competitive analysis and uptake analysis. Wolter noted that market access “is one of the fundamental areas of forecasting. It’s not just about getting a drug to market, but it’s about getting access for patients,” and different strategies exist around the world to make sure people have access to medications.
Competitive analysis, he said, “is no surprise. In many of these markets, there’s an increase in competition. If you think about oncology and how the tumor types are being filled out with many competitors, even within specific lines,…it becomes more and more important and challenging for forecasters to dig into how much share we will get, where we will get it from and how we will protect share that we have.”
In examining the inputs that are most difficult to forecast, IQVIA separated those into three buckets: in-line products, or those already on the market; pipeline agents; and rare disease treatments.
The rare disease findings were “sort of what you’d expect,” with epidemiology ranked the area that forecasters were least confident in. IQVIA, he said, has efforts underway “where we’re working to find the number of patients in rare diseases using machine learning on the claims data. That can be a way to find both diagnosed patients and, importantly, potentially undiagnosed patients in rare diseases.”
For pipeline products, forecasters were least confident in market share and price, “which also seems to make sense.” And for in-line products, forecasting market share/competition was the top area.
Responses to the question of how well forecasters think their company is doing across a range of categories also reveal that there is “room for improvement in all areas,” said Wolter. Integrating advanced analytics approaches was the area respondents felt companies were most in need of improvement, “which was sort of a theme we saw a little bit through this survey.”
That was followed by generating management reports quickly with minimal forecasting effort, integrating real-world data and incorporating uncertainty into forecasting.
The top areas in which respondents said they believed their companies were stronger were quickly updating forecasts with new data changes and global and local integration.
Data, Analytics Are Top Desired Solutions
When it comes to overall organization improvements that forecasters would like to see, IQVIA offered multiple select questions. The top response was increased integration of advanced analytics, cited by about 65% of respondents. That was followed by better data sources, more robust forecasting methodologies, and better models and tools, which were cited by 44% to 50% of respondents.
“When we looked across the whole sample, every forecaster who responded included at least one of those top four in their responses,” noted Alexandra Tataru, an associate principal in IQVIA’s commercial consulting group and the product lead for IQVIA Forecast Horizon, the company’s platform for life sciences forecasters. “So there’s pretty widespread consensus around those topics.”
“As we might expect, analytics and data are coming to the top once again,” she pointed out. “It’s worth noting, though, that these improvements are not necessarily independent. So, for example, good data and effective tools are both important to support setting up advanced analytics.”
Less of a priority — “which means they’re working pretty well” — were improved data pre-processing, global and local team coordination, and better internal processes, communications and reviews.
When IQVIA drilled down into the top tool improvements that forecasters want to see, 70% of respondents said they wanted to see automation of parts of the forecast. “We need to improve a lot,” remarked one respondent. “Currently we use traditional methodology — we need automation and streamlining of manual efforts.”
Facilitating analysis and enabling collaboration were the second and third improvements that respondents said they wanted. Those were followed by facilitating input data and plugging into data pools.
“With the increase in the amount of data available and being used by forecasters, we’re seeing that there’s a lot of value in tooling up to reduce manual burden,…hence the automating parts of the forecast,” said Tataru. “When you’re copying and pasting data from one place to another 50 times or having to sort through a really large data file, it’s just something that humans are not super good at and machines are becoming increasingly more skilled at.” When that runs smoother, it “frees the forecasters up to focus on things like scenario planning, asking the right questions: What are the big drivers of the forecast? How can we move them? Which ones do we need to focus on in getting those right insights?”
The top process improvements that respondents said they would like to see were consistent forecasting approaches, cited by 70%, followed by better management review processes, selected by 67%. Slightly more than half of respondents said improved collaboration, while almost one-third said better communication.
“It was interesting that so many forecasters are saying that they want structural consistency,” she asserted of the top desired improvement. “Sometimes it seems like the opposite, where everyone wants to set up their own special way of looking at their market and work in their own Excel. But what we’ve seen with our folks that we work with is that there’s a lot of efficiency gained after synchronizing a forecasting approach across the portfolio. Making variations from a common structure, it’s more intuitive, everybody understands what’s going on. It usually becomes more straightforward vs. trying to consolidate many incongruous models.”
Respondents said that trending was the top ML-driven improvement they hoped for, recalling the finding that forecasters said that trending is the No. 1 use for ML currently. “As expected, many folks responded that trending was an area that could see even more improvement,” she said. “Most feel that their sophistication is still fairly low, so it’s not surprising that it turned up here, too.”
Pre-processing of real-world data was the second desired ML-driven improvement, which “was not a huge problem area, but it seems to be a real opportunity area where machine learning could make this better.”
The last survey question was about desired AI-driven improvements, and trending came out on top again, followed closely by market segmentation, sales for pre-launch products and patient segmentation.
“As everyone knows, there’s been an explosion of development in the AI and ML space,” said Tataru. “We are particularly excited about those use cases beyond trending so, for example, using longitudinal patient data and characterizing things like unexpected patient grouping segments like clusters of different behavior or drug response. Conceivably every input in the forecast could be facilitated by AI or ML,…even if it’s just looking more broadly across existing data sets and just pulling out those patterns that are relevant to a specific asset.”