Indegene's Jitesh Sah, Vice President, Analytics and Vikas Mahajan, Sr. Director, Data & Analytics chat with senior editor Fran Pollaro about the challenge of data drift.
Data drift is a huge challenge in the pharma industry as companies are dealing with massive volumes of complex and diverse data that are not always generated or accessible in real-time to update machine learning (ML) models. Inaccurate, outdated, or incomplete data can have a profound impact on the accuracy of predictive models and ultimately, healthcare decisions.
Pharmaceutical Executive's Fran Pollaro sat down with Jitesh Sah and Vikas Mahajan of Indegene to discuss the current state of machine learning in pharma sales and marketing, as well as what companies are doing to ensure their data streams remain dynamic and near real-time to support high-performing models.
Jitesh Sah: Prescription data is a critical piece of information for pharma companies seeking to evaluate their marketing efforts and make informed decisions on how best to engage with healthcare professionals (HCPs). By analyzing prescription data, pharma marketers can pinpoint which HCPs are prescribing their products more frequently and which products are being prescribed more often in a particular region or demographic group – helping them accurately micro-segment HCPs and tailor marketing initiatives accordingly.
However, prescription trends are dynamic and can be influenced by a variety of factors. Changes in treatment guidelines, introduction of new drugs, shifts in patient preferences, and other factors can all have an impact on prescription patterns. For instance, when a new drug enters the market that is more effective or has fewer side effects than an existing treatment, HCPs may start prescribing the new drug more frequently, leading to a shift in prescription trends. If marketers fail to track these changes in prescription data, it can result in a significant misalignment between the information that HCPs seek and what they actually receive.
A study by DT Consulting1 found that HCPs believe new clinical data is the most relevant content for their jobs, but only 50% reported receiving it in the past three months. Furthermore, digital content related to prescription and dosage only received a customer experience score of 45 out of 100, which falls short of the excellent and more meaningful experiences that HCPs expect.
Jitesh Sah: Sales and marketing teams rely heavily on patient and HCP behavior data to analyze trends and patterns and create targeted campaigns. However, since these behaviors can change over time due to various reasons, such as evolving brand expectations, changing demographics, and economic shifts, previously reliable data may become obsolete, leading to decay in the predictive recommendation models that pharma companies are using to make personalization decisions.
One example of this is the evolution of HCP preferences over the past few years. These preferences vary at a regional, experience, and specialty level and are a critical input for pharma marketers to plan their marketing initiatives. For instance, according to Indegene's Digitally Savvy HCP report2, HCPs based out of China and India prefer using a mobile device for medical and promotional information, whereas a computer/laptop is the most preferred device for HCPs in the United States and Europe. It is crucial for pharma marketers to note these patterns and select the appropriate channel, content, format, and device to reach out to HCPs. This means creating modular content that is easily consumable on mobile devices for HCPs in India and China and more elaborate content for those in the US and Europe who prefer computers.
Granular data like this can best inform personalization moves, and weaving these data points continuously into machine learning models is crucial.
Jitesh Sah: Market saturation happens when a product or drug gains significant popularity in a particular market, resulting in an oversaturation of data.
For example, if a pharma company’s machine learning model relies mainly on sales data from a single market to make recommendations, the model output may become biased and not representative of the larger market. This could cause the company to invest more resources in that particular market and overlook potentially more profitable opportunities. Therefore, it is crucial for pharma companies to expand their data sources beyond sales data, including social media, electronic medical records, and patient forums, to gain a deeper understanding of patient preferences, treatment outcomes, and unmet needs in specific markets.
Vikas Mahajan: Predictive analytics can be particularly helpful in identifying whitespace opportunities by forecasting market trends and predicting future demand for specific products or treatments. For instance, these models can identify markets that are likely to experience growth based on demographic trends, emerging health risks, and regulatory changes.
Pharma companies can take necessary measures to ensure their data streams remain dynamic and near real-time. This can involve integrating machine learning operations (MLOps) best practices to ensure sustainable and accurate predictions:
For more details, please refer to: