Blog
Understanding AI, Data and Human Interaction in Pharmaceutical Development
Mike King, Senior Director, Product & Strategy, IQVIA
Feb 06, 2024

Mike King, Senior Director of Product and Strategy at IQVIA recently spoke alongside industry experts with Lori Ellis, Head of Insights at BioSpace and host of the Denatured Podcast. The conversation tackled concerns of data bias, data accuracy, data access and the role of human involvement in artificial intelligence (AI) use in pharmaceutical development.

One of the primary concerns of artificial intelligence (AI) use in the pharmaceutical industry revolves around ensuring data accuracy and mitigating data bias. As the volume of information increases, protecting data integrity and ensuring the right actions and decisions are taken are key to ensuring a continued focus on patient safety.

Explainability and trustability

One way that organizations are addressing these challenges is by prioritizing trustability and explainability throughout the integration of AI into industry processes. AI solutions must not only be accurate and limited in bias, but they must be financially viable, relevant to healthcare and successfully navigate validation and regulatory requirements that are mandated by government regulations and global standards. In this environment, the ability to explain and interpret decisions to patients, consumers, regulators and other stakeholders is critical to the safety and success of AI solutions.

Trustability relates to data bias and patterns of precedents that may be affecting algorithm outputs. When attempting to detect new patterns or safety signals, working with models that have been trained with select databases or prior structured datasets may influence or pose the risk of inadvertently missing critical signals. This emphasizes the need for manual touchpoints working alongside these systems to train algorithms to arrive at conclusions that we, as stewards of this information, can trust.

Value of failure mode and human in the loop

Finding common ground between checks and balances and the purpose and intended results of the AI algorithm is a critical aspect of promoting data accuracy and safeguarding against data bias. By recognizing and reconfiguring algorithms to pursue failure mode, organizations can illicit valuable information that would have previously been overlooked. As organizations increase their volume of AI mined data, there could be a tendency to overweigh a certain known failure mode (as the AI algorithm “knows what to look for”) and it is equally important to consider other signals or configurations that could point out unknown problem areas that may need further investigation. Identifying weaknesses in system performance allows stakeholders to implement proper processes to correct and optimize data analysis and subsequent outcomes. The value of AI does not rest in the system itself, but in how the user is filtering and guiding the algorithm to identify critical information that, when analyzed, yields information that increases a company’s ability to focus on patient safety and patient outcomes.

Eliminating data bias and ensuring AI algorithm accuracy are also dependent on human involvement. Though AI is highly valuable in automating processes such as signal detection and adverse event reporting in pharmacovigilance, there is still a necessary element of human interaction to make assessments, provide confidence to regulators and understand trending analysis. Improving algorithms and data analysis depends on human operators to improve models and create a virtuous feedback loop.

Use of AI and data to influence human behavior

Utilizing AI to perform statistical analysis effectively influences and modifies human behavior to improve patient and clinical trial outcomes. For example, take the instance of a nonadherent patient who may struggle with medication compliance. How can data help us communicate with the patient and encourage awareness of the treatment to navigate barriers to taking medication? In this situation, the absence of data is just as informative, as it can pinpoint potential influences on nonadherence. The interpolation of this data and the use of statistical analysis can then be used to engage in human-to-human interaction to change a patient’s behavior and to provide a company additional insight on how they can improve engagements with a broader population. This will then reinforce the data-gathering process to substantiate the effectiveness of the intervention.

The value of data and AI is particularly prominent in decentralized clinical trials, in which reporting takes place electronically through medical devices or monitoring machines. The integration of these devices with AI algorithms serves to enhance data measurements, improving data quality and compliance. This direct feedback loop on human behavior elevates the quality of real-time data throughout the clinical trial, shedding light on the potential benefits of interactions between AI, data and humans to improve healthcare outcomes.

To listen to the full conversation featuring Mike King, Senior Director of Product and Strategy at IQVIA, on the Denatured Podcast please click in chronological order: Artificial Intelligence, Part 1: Bias, Access, ROI, Potential Success and Failure, Artificial Intelligence, Part 2: Human Interaction, Liability, and Patient Safety and Artificial Intelligence, Part 3: Improving Communication to Impact Patient Behavior.

Related solutions

Contact Us