AI in Ideal Population Finding – Finding Customers in High-Noise Environments, Intelligence by IPM.ai
Swoop works hard to ensure that biopharma marketers have the ability to find their ideal populations, groups of potential customers who may be difficult to pinpoint but who carry extremely high potential value to the marketers. By working with IPM.ai, our data and AI partner, we are able to create advertising campaigns that solve for your marketing challenge, be it increasing the addressable market, increasing market share, driving greater adherence or increasing auto dealer traffic.
Using AI, pharmaceutical marketing efforts can accomplish a lot, such as:
- Build and message undiagnosed rare disease patient models
- Build patient disease progression groups
- Improve adherence among potentially nonadherent modeled groups
However, the challenges hindering widespread AI adoption, particularly with regards to health data, are multiple:
- Notoriously dirty data (duplicate entries, missing data, incomplete coding, etc.)
- Complex interactions among multiple conditions over time
- Identification and ranking of contributing factors
- Translation of results into real-world business objectives
- Integrating multiple data sources in a HIPAA-compliant manner
Swoop and IPM.ai together can overcome these challenges and offer cutting-edge machine learning/AI solutions that can truly impact a marketing effort. Here’s how:
Natural Language Processing
NLP helps solve the problem of incomplete, missing or unspecified coding. For example, NLP lets us see ICD-10 150.8 (Unspecified Heart Failure), and using NLP with an emphasis on relationship extraction between co-morbid or concomitant codes, this unspecified code can be correctly re-mapped and indexed for level of severity, greatly enhancing the specificity of a model.
Neural Networks/Deep Learning
These cutting-edge techniques uncover the deep underlying connections between any number of medical conditions, arrayed across any time span. It is these previously-unknown interactions that are most predictive in applications such as finding undiagnosed rare disease patients. However, on their own, these act as black boxes, creating a need for:
Traditional Machine Learning
This is where features contributing to the positive class can be highlighted. These traditional regression techniques and classifiers score and rank the relative importance of all the features contributing to the final model.
Ensemble of Algorithms
Models that are carefully engineered to be as independent as possible provide an average result, more precise than any individual result. This “wisdom of the crowd” approach provides insights into the most predictive triggers that can inform other business decisions around the therapy. Swoop and IPM.ai use 3 Deep Learning and 3 Regression models to produce a result that is as actionable as possible.
Non-Parametric Models Optimizing Towards Business Goals
Traditional models measure the propensity of a population having the desired characteristic, with populations scored on a 0 to 1 scale. However, saying a population is 0.9 vs 0.8 provides no insight into the tradeoffs a marketer is faced with when trying to pick the correct population to target. IPM.ai’s models are always tuned to answer a business question, such as “What is the correct population to target that results in a number of patients where a 20% conversion rate would result in a program ROI of 100%?”