Did you know that you may be introducing risk unnecessarily to your forecasts without even knowing it?

Forecasts are used in a variety of ways in business for a wide range of decisions – mostly relating to revenue and cash-flow. What’s something worth or going to be worth? How much cash is going to come in and when?

In healthcare, we see a wide range of applications of forecasts across many different functions. Here are a few different groups within companies and a very high-level view of what they use forecasts for:

Department / FunctionUse of Forecast
ManufacturingPlanning for drug supply & production
FinanceSetting targets and budgets
Sales & MarketingSetting targets and budgets
Strategic groups / portfolioLong term planning for investment
Business developmentDetermining acquisition costs for assets
Investors & analystsLong term planning for investment

Each forecast that is developed tends to have a specific purpose or business question / decision to answer. There is often a substantial amount of money riding on each forecast, so they are important to get ‘right’, or at least invest as much time and effort as makes sense in order to minimise any likelihood that the forecast could potentially be wildly inaccurate.

While the art of predicting the future is inherently rife with unknowns (it is the future!), how can we be more confident in what we are predicting?

Forecasts are often the result of some sort of forward projection of a series of values based on a combination of assumptions and historical data. Variability, and thus risk and uncertainty, can reside in either of those two areas when we look at a forecast projection – how have we reached our forecast, based on what assumptions and data? What did we take as fact and what did we have to estimate?


The data on which the forecast is based has a significant impact on the future projection. We have to be sure that the historical data we’re using is reflective of what actually happened (i.e. did we really sell 55 million units of the product last year? Were there really 70 million people in Japan with osteoarthritis of the hand?). For data and especially history, we should know this. There is no reason why we should be basing a forecast on incorrect data. However, this is not always seen as the case. It is often taken for granted or assumed that our base data is accurate, otherwise, why use it?

These seems to be especially the case when it comes to epidemiology data and its use in healthcare forecasting.

The following case studies for healthcare help to illustrate the importance of good data when it comes to forecasting and making sound investment decisions.

Case study #1: inaccurate epidemiology data - musculoskeletal

A client came to us to commission a forecast model for development of a systemic treatment for osteoarthritis in Japan & the USA. They had previously bought some epidemiology data for the prevalence of the condition in both the USA and Japan that they had been using to date in their company. To maintain consistency across the business, they were keen for us to use this data as the basis for the forecast model.

The first problem we were faced with when trying to use the data was the actual “structure” of the data – the asset in development was a systemic treatment for ‘osteoarthritis’ and the forecast model needed to reflect this, whereas the data was structured by site / joint affected rather than giving an estimate of the total population size. Patients receiving a systemic treatment would have the same treatment for any joint, not specific treatments for each affected joint (as would be the case with a joint-replacement therapy or injection by affected site)

This is a problem since patients with OA may have multiple joints affected by the condition. Building a model based on adding up patients by location would leave to inclusion of overlaps and therefore double-counting patients, overinflating the patient forecast.

Raised hand

The second issue we picked up on with the data was the estimation of the overall number of patients affected with the condition. In both countries, the estimates included a very high percentage of the population affected with OA which was post likely not possible. In one instance over 50% of the population were assumed to have at least one joint affected by OA (of the TOTAL country population – 70 million people in a total population of 120 million for Japan), while in the USA they estimated at least 1/3rd of the population were affected by the condition (80-90 million people). Slightly more likely, but still not entirely probable especially as the arthritis foundation state ~27 million in the USA with OA.

It was not possible to define exactly what the estimates had been based on, but one estimate may have been based on the number of joints affected rather than the number of people affected (which have a significant impact on a forecast).

In reality it is more likely that there are ~20 million people in Japan with OA, not 70+ million (of a 120 million population). Thus, the foundation forecast data had a catastrophic impact on the resulting market forecast for the Japanese market not to mention the US market.

The shame of the situation was that there is plenty of data available for estimating the prevalence of a condition as wide spread as OA, but care and attention had not been brought to bear on either the collation of the data or the interpretation of the information and sanity checking before publication.

Case study #2: inaccurate epidemiology data - ophthalmology

One client that we were asked to assist with, on behalf of one of our business partner companies, had been ‘sold’ a vision of a future franchise in ophthalmology. However, a fundamental mistake had been made in the disease definition vs. the geographic markets in which they were planning on developing and launching their asset.

The issue was with a relatively complex cluster of diseases known as “uveitis”. Uveitis is a more general term for a type of inflammation of the eye. It can affect three distinct locations within the eye (anterior – front of the eye or the anterior chamber, posterior – back of the eye, intermediate – middle of the eye, or it can be termed as pan-uveitis, affecting the entire eye). The causes of uveitis in different locations are also dramatically different (anterior are often caused of presence of a foreign body or parasite), frequency of the eye segment affected varies by country and cause.

Uveitis – split by location and aetiology:

Site Affected% of total uveitis% of site that caused by autoimmune disease
Total % uveitis eligible for asset9.4%

The client’s asset was more to treat uveitis as caused by autoimmune conditions, which meant more pan-uveitis than other locations.

Case study #3: incomplete patient segmentation data – heart failure

In this case, a client came to us with a very specific question regarding product performance. Their issue was that they had successfully developed and launched a product into a market they felt they knew quite well – heart failure. Despite a strong sales force and good clinical data for the product, they were hitting a ceiling in their achieved sales to date and wanted to understand why.

When we dug a little deeper, we discovered that they had fundamentally missed a key segmentation of their market. Their forecasts were all based on a premise of treating all patients with chronic heart failure whereas their label restricted their use to patients with reduced ejection fraction within specific NYHA classes. This more than halved their eligible population, leaving them with ~30% of their previously assumed eligible treatment population.

Patient segmentation to align with licenced drug use is now a core part of forecasting which MUST be included within any forecast model. Understanding those that are eligible or excluded from being prescribed a product on the basis of clinical profile, comorbid condition or severity is essential and should be included within the core population data on which a forecast is based.

Case study #4: impact of hidden dynamics

A long-standing client asked us to help them with an opportunity to bring a product back in-house that had previously been out-licensed in a specific geographic region. In this instance the product was related to treatment of chronic hepatitis B in Asian markets – a region where HBV was considered endemic by the WHO.

They had previously received estimates from local sources regarding the state of the market, growth and treatment of the populations along with other estimates regarding epidemiology and prevalence of the condition in the territories in question.

One of the main issues that our client flagged was that they had little confidence in any of the estimates of the market size and population growth and really wanted an independent review of all of the available information and a view on what the key dynamics and drivers of the market were.

Starting with the fundamental question of how many patients have chronic HBV and how many will there be in the next 5-10 years, we had to consider several factors:

  1. The natural history of HBV infection – not all patients that are infected with the virus become chronic sufferers with liver dysfunction
  2. Incubation period of the disease – length between first infection to emergence of chronic HBV
  3. Age groups or population groups at particular risk of both initial infection and eventual transition to cHBV (hint, children infected with the condition are 90% likely to become chronic HBV patients versus adults exposed to the first time to infection are only ~ 10% likely to transition to being chronic).
  4. Impact of a universal vaccination programme introduced in the early 1990’s.

Vaccinations tend to be introduced in order to address a significant public health issue – in the case of HBV, the vaccination was developed and introduced to Asian markets in the early part of the 1990’s to combat the endemic prevalence of the HBV virus.

The main route of transmission in Asian was via vertical transmission (mother to child), meaning that a high proportion of children were exposed to the virus and would go on to be chronic sufferers in later life. The universal vaccination programme focused on vaccinating infants prior to exposure, preventing them from initial infection and subsequent progression to chronic disease.

However, the incubation period for cHBV in its natural course is ~ 20-30 years (depending on the individual). In 2010 when we were asked to quantify the number of patients with cHBV, the vaccines had been around and used for about the same length of time as the incubation period.

The expected consequence on the available population with cHBV was that it would start to decline over the coming decade, leading to a significant reduction in patient numbers requiring treatment.

Every other estimate of this market had failed to take into consideration the HBV vaccination and the long-term impacts on the market.

Consequences of inaccurate data:

The consequences of having an inaccurate dataset on which to base a forecast are obvious, but it is shocking how many people / companies do not invest sufficient attention to selecting and using the best quality data they can find for use in their investment decisions.

If you are going to invest millions in an asset, why not be sure that you have the most up to date and robust data possible on which to make that decision? Why accept the introduction of unnecessary risk where it does not have to be? Forecasts already contain risk, why not reduce as much of that risk as you can by opting to use better data?

If you would like to talk to us about how we can help you to reduce risk in your forecasts through better patient data or would like to commission specific patient quantification, please do not hesitate to get in touch with a member of our epidemiology services or forecasting teams. Here we also have a selection of checklists and infographics to help you review the data you have at hand.  

Written by Nic-Talbot-Watt