Predicting Product Demand: Leveraging An Integrated Approach
The analytics team at Snowfish has worked with multiple life science companies helping identify patterns in large quantities of data. As part of our offerings we have assisted companies in demand planning. Predicting and planning for demand is a common and critical challenge pharmaceutical companies confront. Multiple factors can affect demand including the product category and exogenous events. Demand for products such as flu vaccine varies tremendously from year to year depending on the severity of the flu that year and the amount of media coverage. Demand for allergy therapies is highly seasonal. Furthermore, the seasonality may differ by geographic region and may in fact vary in the same region from year to year depending upon weather patterns. Variation in demand may even be evident at the individual physician level for infrequently prescribed therapies such as oncology where treatment regimens can vary significantly. Therefore, predicting demand for the overall market, individual products, and individual physicians can be particularly challenging.
The Right Approach is Critical
In all cases, it is important to first pick an approach that is most appropriate for the challenge. It is all-too-tempting to throw all of your data into an algorithm without first analyzing the nuances of the situation. If seasonality is a major concern, then it will be necessary to first discover seasonality trends in past years. Keep in mind that those trends may shift in time and magnitude from one year to the next and may even vary depending on geographic region within a single year. Once seasonal trends are identified, the next step is to see how they fit the current year in both timing and magnitude. In some years the flu may peak early, and in other years it may peak late. How this year fits past trends can be used to predict the next several weeks.
For infrequently prescribed medicines, it may be necessary to simply predict the probability that a particular physician is going to prescribe at all in the next three, six, or twelve months. Such predictors include whether the physician has ever prescribed a particular medicine in the past, how recently, and their overall level of prescribing of all related drugs. For new products, it is helpful to determine if certain physicians have been early adopters of other therapies when they were new. A physician that lagged his peers in the past in prescribing new medicines is likely to lag on future new products even if he eventually ends up being a high-volume customer. Likewise, if a physician has been an early adopter in the past, this increases his chances of being a leader in prescribing new products.
Data Is Often Incomplete: Do Not Rely on a Single Data Source
Incomplete and/or incorrect data are limitations that are going to negatively impact any technique. Certain sales channels may not be reported to data vendors. This tends to occur when a certain prescription is processed through an unreported channel. Demographic data about the geographic area where a physician practices while very helpful when available may be incomplete for certain regions. Therefore, it is important to include data from multiple sources such as internal sales data, prescriptions of competitors’ products, diagnostic codes, etc. With multiple data sources, problems with any one source are less likely to have a major impact.
Use the Right Tools and Integrate Them
Approaches that integrate analytical skills, clinical knowledge, and business acumen yield the best results. It is helpful to use the latest analytical tools such as Gradient Boosting, Random Forests, and Support Vector Machines as well as more traditional tools such as linear regression. Additionally, clinical knowledge is crucial for understanding the disease state the medicine treats. How long does typical treatment last? Is the therapy used to treat multiple different diseases? How is it administered? Clinical knowledge affords the necessary background for data selection and creation of features for predictive statistical analysis. Finally, business knowledge of outside factors that drive demand and purchasing decisions also should also be incorporated into data selection and feature creation.
Using an integrated approach to predicting product uptake and demand not only increases prediction accuracy, it also yields better insights into your data. Trends that might have just been lost in the shuffle of statistical analysis might jump out at a team member with clinical knowledge of the disease state or business knowledge of the market. When pharmaceutical companies face the challenges of predicting product uptake and demand, integrating analytical, clinical, and business skills into the team is the key to addressing those challenges.
Please feel free to reach out to Snowfish to learn more about our integrated approach to predicting demand. Companies are already benefiting from our insights.