Merck Group’s Vice President Global Market Access & Pricing Strategic Planning, Marco Rauland, takes a look at recent advancements in Machine Learning and Artificial Intelligence and how these technologies can offer a sophisticated and flexible approach not only to predicting drug pricing, but also to estimating market access with different pricing scenarios.

 

Over the past few years, there has been a drastic increase in data digitalization in the pharmaceutical sector. This digitalization comes with the challenge of acquiring, scrutinizing, and applying that knowledge to solve complex business problems. This motivates the use of Machine Learning (ML) and Artificial Intelligence (AI), because they can handle large volumes of data with enhanced automation. The COVID-19 pandemic acted as an accelerator for AI and digital initiatives across industries and these technologies are being increasingly used to help informed decisions. But where does ML and AI stand in terms of price prediction?

 

Traditional way of estimating drug price

In pharmaceuticals, drug discovery and clinical developments are not the only factors to guarantee good return on investment. Estimation of optimum drug price is another essential part of the drug launch process on which product success depends. In the last couple of years, there has been a lot of debate on the level of incentives provided to support  innovation in the pharmaceutical field, especially with the high cost of some of life saving innovations like gene therapies and oncology drugs. The pharmaceutical industry is under a lot of pressure to cut prices or at least keep them low. Traditionally, drug prices are estimated by considering multiple factors like average cost of sustainable research and clinical development, manufacturing costs, inventory and supply costs, the cost of available similar products or therapies, the number of potential patients, etc. Detailed heath technology assessment (HTA) is done to ensure and prove cost effectiveness to ensure reimbursement status for the drug, which in turn helps the drug to get decent coverage. However, because of so many variable parameters, it is challenging to maintain a balance between serving patients while earning a decent profit for investors and further developments. Recent advancements in ML and AI can offer a sophisticated and flexible approach to predicting drug prices and also estimating market access with different pricing scenarios.

 

ML and AI in pricing solutions

More than half of all industrial product companies still create their major pricing tools in Microsoft Excel, according to a pricing maturity evaluation by BCG and the Professional Pricing Society (2020) while 25 percent of business to business (B2B) companies employ static, one-size-fits-all pricing with few inputs and few adjustments. Another survey by MIT Institute and the BCG Henderson Institute reflected that although only 12 percent of companies (from those covered in the survey) used AI to improve their pricing, their initiatives succeeded twice as often as the efforts of companies that applied AI to other functional areas. Consumer goods companies have already started leveraging dynamic pricing solutions developed with ML that focus on competitors’ pricing, consumer behaviour, real-time demand, location, time of the day, seasonality, and willingness to pay.

 

AI adoption and usage in drug price predictions

The key factors fuelling the need of artificial intelligence in drug pricing are challenges with the traditional approach, an influx of large and complex healthcare datasets (health records, diagnosis images, prevalence data, payer plans, formulary access, claims data and clinical trials data etc.), the rising number of partnerships and collaborations among different domains in the healthcare sector, and multiple stakeholders involved in drug pricing and reimbursement. Manufacturers are consistently facing pressure due to pricing regulations across the globe. Provider’s prescription decisions are being influenced by payers by motivating and shifting towards value-based care and reimbursement.

With the help of ML and AI, most of the factors impacting drug prices can be co-related in the process and a higher accuracy in the forecast can be achieved. Also, the application of AI to pricing and reimbursement work can dramatically free up the time spent on crunching datasets, modelling scenarios, and building price predictions, and allow more energy to be directed to submissions and the creation of agreements. Overall, key benefits of using AI for price prediction can be summarized as:

  1. Accuracy: AI algorithms run based on a real-time feed of market data and drug characteristics, so these provide the most accurate and unbiased price of the product with minimal human intervention.
  2. Market Volatility and Price Change: Deep learning-based AI solutions with real-time data feed of market dynamics can quickly analyse and give optimum prices by analysing drug exclusivity, patent expiry, changes in competition, changes in market prices of similar/competitive drugs, regulatory guidelines etc.
  3. Multiple Data sources: AI can analyse multiple data sources at once which can impact the drug price directly or indirectly, and instantly apply them to change the drug price. It can almost instantly reflect the change in cost of raw materials, cost of production, regulatory guidelines, the cost of inventory, and supply of drugs across globe.
  4. Profit Margin: As AI can assess the impact of price changes on market access, it provides the most optimum drug prices. Automated processes also helps in quick decisions and implementation with lower operational cost, which helps to maintain decent profit margins.

Challenges in adopting AI and ML for drug price prediction

While these technologies offer efficiency and automation in process, algorithms are written to train the models on historical data and have certain challenges associated with them:

  • As drug price depends on a lot of parameters (e.g., drug efficacy, competition, number of potential patients, drug’s ability to improve life duration and quality of life, etc.), data for these parameters is required for already launched drugs. ML models are trained on these data points to predict future drug prices.
  • Prices further need to be approved for reimbursement which requires an efficient systematic literature review (SLR), where the data required for ML will be difficult to collate.
  • Europe majorly follows reference pricing for reimbursement approval. A different level of pricing than expected can impact the market access of a drug, and therefore the sales performance of a drug.
  • As mentioned, price prediction is not the final step. For for approval of this predicted price, drugs will still have to go through an HTA process.

A lot of historical data is required for each factor to develop algorithms and ML models for efficient price prediction.

  • Available data is quite fragmented and needs lot of manual manipulations to fill data gaps.
  • There is also a chance of unintentional omission of some parameters in ML models, which can influence drug price.

Recent practical use cases in drug price prediction:

Looking at the potential benefits of leveraging ML and AI in drug price estimations, some technology experts have come up with automated solutions.

UK artificial intelligence startup Okra Technologies claims that its AI-system-based software platform ValueScope, which is powered by the OKRA Explainability Engine, can predict the price and likely outcome of an HTA assessment with more than 90 percent accuracy. This AI tool was built using historical data from more than 1,700 drugs that have been launched in Europe. It was made to give life sciences professionals power, assurance, and comfort so they may gain a competitive edge.

Another AI technology company, US-based  Konplik, claims that its AI based technology, deploys tailor-made machine learning algorithms (leveraging gradient boosting, random forest, and Bayesian networks) with chunks of game theory to predict tender price for drugs. Automated processes runs multiple model simulations to come up with the best fitted model on historical data of tender prices, and this model is used to predict future tender prices. Such tools can be leveraged for efficient price prediction in markets where competitors have frequent and up-to-date information on bidding results (e.g., the Scandinavian market).

Leading pharma information provider, Elsevier, is promoting a new medication pricing standard called Predictive Acquisition Cost (PAC) developed by Glass Box Analytics that asserts to track actual drug acquisition prices more precisely. According to the company, PAC uses predictive analytics to take into account a variety of factors when estimating the acquisition cost of a drug, including maximum allowable cost benchmarks, published price lists, existing price benchmarks, drug dispensation metrics, supply-demand measures, and survey-based acquisition costs.

Conclusion

There is lot of development and adoption of AI for price prediction in the consumer goods market. Sophisticated AI-based tools have been used for the past several years to provide real-time prices as per demand and situational factors. This technology is still new for drug price prediction. Adoption of this technology may improve in times to come with increasing availability of structured data, integration with the complexities of the healthcare market, and proven reliability. And as this technology improves and becomes more accessible , it will integrate naturally into the pharmaceutical industry’s processes. The future of pharma pricing will be AI-enabled.

 

RESOURCES: