When attempting to predict commercial demand for a drug, forecasters generally use one of two methods to design the forecast structure:

Patient-based: Forecasters often start with an epidemiology-based approach, using data and assumptions around prevalence, persistence, compliance, and market share to determine how many patients are taking a drug, and use this to forecast future revenue. This model is most commonly used when a product is new to the market, or where patterns of usage are complex (e.g. oncology).

Demand-based: When commercial sales data or real-world evidence are available, forecasters often use a demand-based model fueled by historical sales data (volume or revenue) to predict short- and long-term future sales. This approach ‘trends’ past performance into the future and is particularly valuable when a drug’s sales have reached ‘steady state’, where the past is a good predictor of future performance.

Both models bring value to the forecasting process, but they also have their shortcomings.

A patient-based model aims to build up a clear picture of disease progression and treatment by focusing on the patient journey and decision. It looks at who has the disease, how it’s typically detected, how patients are treated and by whom. These models are uniquely built for pharma, aiming to understand the patient journey and how patients arrive to – and branch away from – certain therapies. These insights can be useful for building a commercial strategy since they establish a deep and more causal relationship between patients and resulting commercial sales. However, they require a lot of research, and are often based on data that is difficult to collect and infrequently updated, making the forecast less accurate or responsive to current market trends.

A demand-based model is based on real-world sales numbers. With the explosion of detailed market trends and more granular views of sales, these models have become increasingly accurate forecasting tools for in-line brands. These models are frequently updated and so provide timely insights that alert of changing market dynamics. However, trend-based models offer a very narrow view. They focus how the product is currently performing in the market and basic correlations with patient behavior, but not the underlying cause of these trends. If sales are dropping in a specific region, they typically can’t tell you whether it’s due to physician preference, patient preferences for competing products, or other causes. The lack of context makes it difficult to use the forecast to drive change or react effectively to market disruptions.

Traditionally, forecasting has taken a binary approach. Forecasters pick a model – patient or demand – and use it for predictions. It’s considered too hard to build both, and even if they did the two projections rarely line up. That creates an entirely new set of work for forecasters to try and reconcile the two.

However, pharma-forecasting thought leaders are increasingly challenging this binary “patient or demand” choice. Their increased access to more granular real-world data, along with growing understanding of longitudinal patient journeys, are poised to usher in a new approach to forecasting.

Blended Models Using Real World Data

Today’s forecasters are increasingly looking at ways to ‘bridge the divide’ between these two forecasting world views by using real-world data and machine learning to integrate components of both approaches. Here’s how it works:

• Both patient and demand models can be improved using longitudinal data of patients’ sequential visits and treatments to their healthcare provider. This allows us to understand the likely number of patients suffering from a disease - even if they have not yet been diagnosed or treated.
• Machine learning tools can quickly analyze gaps in patient-based models and actual demand and determine the likely cause of variations and patterns of bias. The resulting reconciliation offers a valuable source of forecasting insight.
• Demand models can be enriched by increasing the granularity of patient data to understand more about the types of patients being prescribed, including socio-economic and sub-national geographic data, new vs. continuing users, and how long patients are persisting on therapy. This allows for more causal relationships to be defined with greater certainty.
• Patient models can benefit from increasingly sophisticated analogue analysis to identify the ‘perfect’ synthetic analogue that merges data from tens of thousands of real-world products. The real-world demand data can be used to estimate key assumptions in patient-models around market share and other factors.

The Forecast Horizon platform can now analyze both patient-based and demand-based views of an asset. Central to the tool’s approach is a set of user-friendly tools that bridge the gaps between data from both models. It allows users to input persistence, patient flow over time, and epidemiology information as patient-based metrics, as well as to use trending and integrated real-world data in an environment specifically designed to incorporate both forecasting models.

This has allowed best-in-class pharma forecasters to move away from a ‘binary’ forecast approach towards blended solutions that incorporate the best of both worlds. These users have observed more accurate revenue predictions complemented by deeper insights into the reasons behind changes to forecast performance.  This gives forecasters the data they need to analyze the specific effect of an event or epidemiological fact on future sales outcomes. It also allows them to break out the data by market segment, disease severity, competitor sales, physician behavior or other dynamics to hone their commercial strategies.

All of this knowledge transforms the forecast from a static prediction into a living document that can help commercial teams be more competitive.