How the Medical Information Data Uses For AI Semantic Analysis tool is saving pharma valuable time and resources in identifying data points.
With the recent boom in artificial intelligence (AI) across all industries, the healthcare sector, in particular, has been working diligently on how it can best be utilized. However, there is still untapped potential in this space. Healthcare sits in a unique position at the forefront of innovation and clinical science with great opportunity for streamlining tasks.
One AI tool that has been taking great strides in advancing medical affairs intelligence is Medical Information (MI) Data Uses For AI Semantic Analysis (MUFASA). An article recently published in the Journal of Pharmaceutical Policy and Practice explored how the python-coded tool is being used to advance targeted content delivery in this space.
According to the article, MUFASA, in short: “…utilizes state-of-the-art Sentence Transformer library, clustering, and visualization techniques. MUFASA harnesses unsolicited MI data with AI technology, improving efficiency and providing actionable medical affairs intelligence for targeted content delivery to HCPs.”1
The authors outline the process flow of MUFASA which consists of four major components: data preparation (step 1), text to vectors conversion (step 2), semantic search (steps 3–4), and clustering (steps 5–6).1
In the first step of data preparation, which involves a database that covers all current LEO Pharma marketed products in the year 2022, data are exported into a single CSV file. The data are then processed on Jupyter Notebook, which is a platform to execute a Python program for MUFASA. The CSV file is then loaded and converted into a Pandas DataFrame, which is a library that handles major table data.1
Following, sentences from the inquiry database are converted into vectors, which are then positioned within the embedding space based on its semantic meaning. “When a user submits a search query, the Sentence Transformer model converts the query sentence into a vector and employs the k-nearest neighbors algorithm to identify other vectors situated nearby,” the authors explain.1
In terms of presenting the data, MUFASA has the capability to visualize the sentences identified in two different formats: a list view or a chronological line plot. The AI system then reduces the dimensions of the visualization to identify more specific trends. When a sufficient number of vector points are plotted in the newly reduced space, a dense group of points is referred to as a cluster. The final step involves color-coding the data points based on their respective clusters.1 According to the article, MUFASA utilizes TensorBoard, a visualization tool provided by Google's machine learning library TensorFlow, to display these clusters in a 3D plot.2
The authors found that MUFASA has been able to save each MI team member of LEO Pharma approximately five hours per week, assuming an average consultation time of 20 minutes per case. “This, along with promoting response consistency, reducing redundancy, and mitigating compliance risks, underscores MUFASA as a compelling solution for managing routine inquiries,” they note.1
While MUFASA clearly shows great benefit, the article acknowledges limitations, such as the need for high-quality data and potential variations in semantic search results. Despite these potential hurdles, the integration of machine learning tools, such as MUFASA, presents opportunities to increase productivity within the pharmaceutical industry. It encourages medical affairs professionals, such as pharmacists, to explore the integration of machine learning technologies.
In conclusion, the authors write, “Leveraging the HDBSCAN algorithm, MUFASA's cluster analysis uncovers deep insights and discerns actionable themes from large inquiry data sets. The visualization graphs, generated from semantic searches, support evidence-based decisions by tracking the effectiveness of initiatives and monitoring trend shifts. Collectively, MUFASA enriches strategic decision-making, cultivates actionable insights, and bolsters healthcare professional engagement.”1