Until recently, healthcare, pharmaceutical and medical technology companies have focused most of their AI investments on clinical applications. AI is being employed to help diagnose breast cancer, identify potential patients in clinical trials, and accelerate the process used to determine whether a device or new drug may work or fail.
Currently, medical centers are investing in AI in backend use, including creating contracts, managing and speeding up reviews.
To do this effectively, companies need to build data for AI-powered analytics, said David Gould, chief customer officer at Encompaas, a technology company that helps organizations prepare data for the implementation of AI tools. Business leaders should also recognize that AI recruitment is an ongoing process that requires employees to invest in upskills, healthcare supply chain experts told Business Insider.
Prepare data for effective and accurate algorithm input
Gould said creating a uniform dataset is an essential step for organizations looking to build large language models or deploy effective chatbots.
He said that the data used to train a particular LLM must be accurately classified or organized into a database.
“By its nature, structured data is categorized,” Gould said. This includes information already in the healthcare system's database, such as customer ID numbers, diagnostic codes, and supplies prices.
However, unstructured data must also be properly prepared, such as PDFs trobes that outline contracts, to know what information the AI algorithm extracts and what information to do.
Contracts from different vendors are not written in a uniform structure, so if you want to extract the required data from a specific table, for example, you need to train your machine learning AI to find a specific one, Gould said.
Imagine a PDF that is a contract. AI software “can tell you it's a document. That's not what the context is, whether it's a contract, an amendment, a notice or a notification,” Gould told BI.
“An algorithm works well only if it matches what you're looking for,” he said. “OK, chatbots, looking at my entire enterprise for information and looking at what you can find is not effective because processing costs can be very expensive. And that's where you get drifting and biasing.
Preparing the data also means calculating historical information that has not been properly classified or recorded and being updated with the correct metadata for AI-based analysis. That could take time, Gould said.
First, you need to classify your data correctly and within the appropriate context. That might mean identifying and correcting previously incorrect tags. Data must also be stored in a way that maintains compliance with privacy and other safety regulations, such as HIPAA, which protects patient privacy.
Consider the impact on personnel
In a recent webinar on AI integration, Matt Parker and Jacob Thompson, AI tools for pharmacy procurement, according to Matt Parker and Jacob Thompson's Webinar for AI integration, automation of tedious manual tasks is transformative, especially due to the role of procurement and compliance.
“In the world of pharma and healthcare, those who are asked to do this job are highly educated, expensive and ambitious,” Gould said. However, they have historically been tasked with “cutting and pasting contracts into spreadsheets” for hours a day. By automating these tasks, implementation of AI can reduce working hours.
Still, some employees may need retraining or additional training as AI often changes employment capabilities, Jeremy Strong, vice president of supply chain at Rush University Medical Center, told BI. He said ensuring that there is a plan to address AI's high-class skills and recognize the importance of changing work functions for employees is strong as it helps in managing the transition.
The better employees are asking accurate questions, the more AI algorithms can improve by providing accurate answers, Gould said.
He gave an example of knowing whether the number of contracts with a certain type of clause will expire in 30 days. This process usually takes weeks, if not months. Record Managers or the entire Records department are looking into thousands of contracts. However, AI allows employees to learn how to ask powerful and accurate questions that capture this information more quickly.
