 |
By Doug Willson, Ph.D.,
Senior Vice President,
GfK Strategic Marketing
While pharmaceutical markets have experienced unprecedented growth over the past few decades, many observers are now concerned about an innovation gap. Fewer and fewer new products are coming to market, and drug development costs are rising rapidly. Moreover, it doesn’t appear to be getting any easier after launch. Competitive pressures continue to increase as blockbusters lose patent protection and lower-priced generics become readily available. Regulatory authorities have also become more cautious on safety issues. From a reimbursement perspective, payors are more regularly applying restrictions – prior authorization, step therapy, indication restrictions and quantity limits – before reimbursement can occur. Flat research productivity, rapidly escalating development costs and heightened competitive pressures after launch have clear and troubling implications for profitability across the industry and have increased scrutiny and pressure on development pipelines for many manufacturers.
In this challenging environment, pharmaceutical manufacturers have increasingly turned to early-stage market research to help identify the most promising development options from a commercial perspective, and to provide information about the impact of alternative clinical trial designs and results on future market opportunities. Early-stage market research helps manufacturers align commercial and clinical development interests before major development costs have been incurred, increasing the odds of commercial success for drugs that ultimately make it to market. Choice modeling is often used as the basis for these early-stage forecasts; since clinical characteristics of the new drug are still largely unknown in early-stage research, the choice modeling approach is attractive because it can be used to develop forecasts for many different possible clinical measurements and outcomes (e.g., efficacy and safety attributes), product characteristics (e.g., routes of administration, formulations), and competitive scenarios.
One of the most challenging aspects of early-stage forecasting is that it must be done when some of the most important clinical properties of the drug are not yet known. A common starting point for developing a forecast is a drug's most likely clinical profile. Unfortunately, forecasts based on the most likely scenario suffer from the "flaw of averages” – market shares for the average or expected clinical profile can differ significantly from the average or expected market share that takes into account all possible clinical outcomes. The “flaw of averages” effect is well known in capital budgeting and investment fields. To avoid it, some forecasters will examine best and worst case scenarios in addition to the most likely case. However, best and worst cases don’t really represent reality; they provide information on the range of possible outcomes but provide no information on how likely each will occur. As a result, examining best and worst cases provides a very limited view of the overall possible opportunities for a particular drug.
A more thorough examination of the risk and return characteristics associated with developing a particular drug can be conducted using quantitative risk analysis tools in combination with a market simulator developed from a standard choice model. The risk analysis approach generates random values for the unknown attributes, while conditioning on those attributes that are known or under the control of the manufacturer (e.g., price, formulation). Randomly generating values for unknown variables over and over is often called Monte Carlo simulation, named after the random outcomes in games of chance such as roulette and dice that are played in Monte Carlo, Monaco. In comparison with standard scenario-based forecasting, the risk analysis approach provides:
- More accurate estimation of expected market shares, to assess market opportunities for products in development, or as a precursor to in and outlicensing valuation discussions.
- More thorough sensitivity analysis surrounding the drivers of market potential, because the complete range of possible outcomes for each unknown attribute is included in the analysis.
- Better identification of crucial valuation thresholds. What is the probability that a certain market share will be achieved?
Early-stage forecasts are typically based on surveys of physicians who are asked about their prescribing practices for a sequence of hypothetical future market scenarios. The scenarios are developed using an experimental design and reflect different assumptions about the characteristics of the new drug in development and perhaps competitor products. Responses to the survey questions are used to develop a model of physician prescribing for combinations of new product characteristics and competitive situations that were incorporated in the design. The primary output of the market research is a choice model simulator that allows the user to generate a market forecast for any future market scenario.
A screen shot from a hypothetical Alzheimer's disease choice model simulator is shown below. The input template allows the user to select specific values for clinical attributes concerning efficacy and side effects and additional product attributes such as FDA indication, price and route of administration.
The model estimates the share of patients who are prescribed the new product (product X) as mono or combination therapy, for the specific combination of inputs (efficacy, safety, administration, indication, price) selected by the user. The model also provides a forecast for gradual uptake for the new product in the future. In a typical forecast, the share of patients estimate for product X is translated to prescriptions and revenues using assumptions about the size of the patient population, average duration of therapy and perhaps other factors.
In the Alzheimer's example above, the clinical attributes concerning efficacy (MMSE, ADAS-cog, ADL) and side effects (nausea and vomiting) are not under the control of the manufacturer and not known for certain at the time of the research; the price and route of administration are under the control of the manufacturer. Although the manufacturer's NDA is submitted with an indication in mind, the indication is also uncertain until the application and labeling are approved by the FDA.
To perform a risk analysis using the current example, we would:
- Set price and route of administration on the simulator input template at attractive values, since they are directly under the control of the manufacturer.
- Click on the Risk Analysis button. A random value for each of the remaining uncertain attributes is then generated, and the resulting market share is calculated. This process is repeated many times using different random values for the uncertain clinical attributes (you only have to click the Risk Analysis button once).
- The results are then tabulated and can be summarized graphically as shown in the following histogram.
As shown in the chart above, E(X*) represents the share associated with the most likely scenario, and E(X) is the average share calculated from the sequence of shares from the risk analysis. This is an example of the “flaw of averages” – the average share from the risk analysis can differ significantly from the forecast based on the most likely scenario. In this situation, basing a forecast on the most likely scenario would understate the opportunity associated with the drug. The sequence of forecasts generated in the risk analysis can be used to estimate other quantities as well. For example, the probability that market shares will exceed a certain threshold value (X) is simply estimated using the proportion of shares above X in the sequence of risk analysis simulations. The variability of market shares and confidence intervals can be also calculated using the results from the risk analysis simulations.
Monte Carlo simulation and risk analysis are actually widely used in pharmaceutical forecasting; however, the connection between Monte Carlo simulation and choice modeling for early-stage forecasting is less widely known. In our experience, the market share simulators that are typically developed rarely include risk analysis capabilities; and when risk analysis is performed, the explicit link between uncertain clinical attributes and the choice model forecasts is often ignored or underutilized. Risk analysis using the market simulator demonstrated here can be accomplished using traditional Excel-based risk analysis tools (like @RISK or CrystalBall) or through special purpose programming. In either case, these tools are easy-to-use and allow for a much more thorough exploration of the risk and return characteristics of early-stage pharmaceutical products.
Critics of risk analysis sometimes argue that the extra level of effort required to specify distributions rather than point estimates for unknown attributes introduces an extra degree of arbitrariness in the forecast. However, the counterargument – that simpler analyses that condition only on specific values for these unknown attributes can be seriously misleading – is even more powerful. In practice, risk analysis professionals employ simple and robust choices for probability distributions. To be successful, the views of internal clinical and business experts regarding these distributions need to be reflected in these choices. The risk analysis approach focuses attention on what is really known with a fair amount of certainty (perhaps represented by a normal distribution with probability concentrated tightly around the peak) and what is really unknown (perhaps represented by a uniform distribution spread over a wide range).
One final advantage of the risk analysis approach is that it can be used to link the choice model to more sophisticated capital budgeting tools such as discounted cash flow (DCF) and net present value (NPV) analyses, and real options valuation. These tools all employ Monte Carlo simulation and are being used increasingly by pharmaceutical forecasting professionals. Overall, adding risk analysis tools to our choice models can provide better estimates of both the market risks and opportunities associated with a particular drug.
|
|
Risk Analysis, Choice Models and Early-Stage Forecasting in Pharmaceutical Markets
