How AI Enables the Use of Real World Evidence in Drug Development

Artificial intelligence (AI) has emerged as a game-changer in unlocking the potential of real-world evidence (RWE) to revolutionize the way we design, test, and deliver new treatments. RWE, which encompasses a wealth of data from electronic health records, patient-reported outcomes, and other sources, offers invaluable insights into drug safety and effectiveness in real-world settings. AI is becoming a transformative force in harnessing RWE to improve drug development and for addressing challenges that come with RWE such as data quality, patient privacy, and bias.

What is real-world evidence and its importance in drug development?

Real-world evidence (RWE) is a valuable source of information that can be used to improve the safety and effectiveness of drugs. RWE is data collected outside of traditional clinical trials, such as electronic health records (EHRs), claims data, and patient-reported outcomes. This data can provide insights into how drugs are used in real-world settings, and how they affect patients in the long term.

Historically, real-world evidence (RWE) has played a pivotal role in pharmacovigilance, the ongoing surveillance and monitoring of drug safety after regulatory approval. By leveraging real-world data from various sources such as electronic health records, claims databases, and patient registries, pharmacovigilance can identify adverse events that may have gone undetected during clinical trials. These events may occur in specific patient populations, under different usage conditions, or with long-term use. The comprehensive nature of RWE enables the detection of rare adverse events, drug interactions, and contraindications that might not have surfaced in controlled clinical trials due to their limited size and duration. Real-world data can uncover risk factors associated with adverse events, enabling the identification of susceptible patient subgroups. This knowledge empowers healthcare professionals to implement targeted risk management strategies, adjust prescribing practices, and provide personalized patient care.  

RWE can also support identifying novel biomarkers that may have been missed in traditional clinical trials; these biomarkers can then be further validated and explored to understand their relationship with disease progression, treatment response, and patient outcomes. By leveraging RWE, researchers can identify biomarkers that are specific to certain populations or diseases, leading to more targeted and personalized treatment approaches. RWE can also play an important role in patient stratification by identifying patient characteristics, such as genetic markers, disease severity, and concomitant medications, that influence treatment response.  

What are the challenges in using real-world evidence for drug development?

While real-world evidence (RWE) holds immense promise for drug development, it is not without its challenges. One significant hurdle lies in the potential for bias and confounding factors. Unlike randomized clinical trials where patient selection is tightly controlled to minimize bias, RWE is collected retrospectively based on provider treatment decisions, which can introduce selection bias. Patients in RWE datasets may differ from those in clinical trials in terms of their health status, medication adherence, and healthcare access or lifestyle factors, leading to biased results. Patient-reported outcomes can be subjective and prone to recall bias, as well.  Additionally, confounding factors such as concomitant medications, co-existing medical conditions, and variations in healthcare practices can influence the observed outcomes, making it difficult to attribute effects specifically to the drug under study. 

Another challenge in utilizing RWE is data quality. RWE is often collected from various sources, such as electronic health records (EHRs), claims databases, and patient-reported outcomes, which can vary significantly in terms of accuracy, completeness, and consistency. EHRs, for instance, may contain errors or missing data due to human input mistakes, variations in documentation practices, and differences in coding systems. It is difficult to standardize much of this data, which is unstructured by nature and sometimes even handwritten. This lack of standardization in RWE collection and reporting then poses challenges for data analysis and interpretation. Data elements, definitions, and formats can vary across different sources, making it difficult to combine and compare data effectively.  

Both these challenges result in gaps in comprehensive and representative patient understanding and severely limit detailed safety and efficacy evaluations of  different therapeutic interventions in specific patient cohorts.

How can AI address the aforementioned challenges?

To begin with, AI plays a crucial role in identifying and mitigating potential biases in real-world data. By employing sophisticated algorithms, AI can detect hidden biases that may arise from factors such as patient selection, data collection methods, or confounding variables. AI algorithms are designed to recognize these patterns and discrepancies by analyzing the data distribution and comparing it against expected norms or balanced conditions. Once biases are identified, AI can help in adjusting the data analysis processes, either by reweighting the data, excluding biased data points, or employing statistical techniques that mitigate the impact of these biases.

Leveraging AI can also result in higher levels of confidence in the data quality of RWE. AI enables the automation of data collection and processing, streamlining the handling of vast amounts of unstructured data from sources such as electronic health records (EHRs) and patient-reported outcomes. This automation significantly reduces the manual effort and potential errors associated with traditional data processing methods, ensuring the integrity and consistency of the collected data. AI's ability to extract relevant information from unstructured data sources is another critical advantage. Advanced algorithms, especially those that leverage Natural Language Processing (NLP) techniques and/or Large Language Models (LLM), can sift through complex medical records, identify pertinent information, and structure it for analysis. This capability enables researchers to uncover hidden patterns and insights that may not be readily apparent from a cursory review of the data.

How does VeriSIM Life build upon and improve standard AI approaches for using RWE?

While advanced AI methods help reduce bias and improve data quality, several key challenges are still left unanswered by conventional AI approaches that limit its benefit to drug development. These challenges relate to data availability on a limited set of patients, underlying patient heterogeneity, and difficulty to assess the  relative efficacy and safety of different treatments on the same patient/patient cohorts. To address this shortcoming, BIOiSIM leverages, in addition to the conventional AI approaches highlighted above to reduce bias and improve data quality, AI-driven virtual patient generation and quantifying therapeutic outcomes in patient cohorts through a Translational Index score.

Our virtual patient approach utilizes AI-driven methods to compensate for a small patient population by creating virtual patients that accurately reflect the diversity and complexities of real-world populations. By doing so, we can enhance the representation of underrepresented groups, enabling more accurate predictions of treatment outcomes and better identification of high-risk subgroups, thus supporting both personalized medicine and effective risk management strategies.  In order to generate virtual patients, VeriSIM Life utilizes advanced algorithms such as Generative Adversarial Networks (GANs) and Synthetic Minority Oversampling Technique (SMOTE) to augment relevant RWE datasets with synthetic data and  enhance the representation of  patient groups. Further, such virtual patients reduce patient population heterogeneity. By enriching the dataset in this manner, AI/ML models  predict key outcomes across a more diverse and comprehensive patient population.

BIOiSIM, further, connects treatment outcomes of different interventions in different patient cohorts through VeriSIM Life’s proprietary Translational Index scoring algorithm. The Translational Index is driven by knowledge AI hybrid (hybrid AI) models that connect efficacy, safety, and dose-exposure of different therapeutics. Upon combination of the Translational Index with virtual patients, one can assess the likelihood of translational success of different therapies in specific (virtual) patient populations - providing comprehensive drug development insights.

In conclusion, there are several key benefits of the VeriSIM Life approach that transcend conventional uses of RWE and enable the use of RWE to supercharge drug development: a virtual patients approach that helps quantify therapeutic response in different patient cohorts and reduce patient heterogeneity; the integration with Translational Index provides assessment of success likelihood of different therapies in specific patient cohorts and help select the best therapy for each cohort; and the ability to accurately stratify patient populations. Such benefits of the platform are exemplified in a BIOiSIM-driven comparison of two diabetes drugs. In this study conducted using BIOiSIM, 10,000 virtual patients were generated from longitudinal data on approximately 100 patients. BIOiSIM demonstrated several key comparative drug response outcomes. For example,  patients with low (BMI, HbA1c, proinsulin) and high (BMI, proinsulin/IRI ratio, Glucagon and LDL) showed a factor of 2 higher reduction in HbA1C when using drug A vs when using drug B. In this specific patient cohort, inclusion of dose-exposure and toxicity profiles enabled estimation of Translational Index as 8.2 and 6.8 for drugs A and B, respectively.  The Translational Index scores show that drug A is more likely to be translationally successful in that specific cohort.