According to a survey from PPD, pharma companies and clinical research organizations (CRO’s) frequently grapple with a host of challenges when executing clinical trials. Topping the list is patient enrollment and recruitment while meeting the right inclusion/exclusion criteria, a concern faced by 55% of respondents. It often takes around 19 months to enroll the requisite number of patients in trials which in turn increases the cost. This is mostly owing to a lack of precise cohort selection strategy and tool, leading to ineffective and time consuming patient enrollment.
Closely following this is the issue of trial diversity and complexity, which 51% of respondents cite as a major obstacle, given the increasing intricacies of modern medical research. Regulatory hurdles, cited by 46%, present another significant barrier, with ever-evolving compliance standards and guidelines. Lastly, 32% respondents find it challenging to keep up with the rapid advancements in technology, integral to streamlining and enhancing the trial process.
Top Challenges in executing clinical trials as cited by respondents from a PPD survey
Finarb’s upcoming virtual session: How a data-driven approach can address such challenges.
Reserve your seat here. September 28th | 9:30 AM - 11:00 AM Pacific Standard Time.
On September 28th join IQVIA and Finarb Analytics Consulting as we delve into industry use cases and explore how enterprises can adopt a data-driven approach and custom AI to maneuver such challenges and advance clinical trials at various stages. Two key Finarb’s use cases which will be showcased during the session, are discussed below -
AI ENABLED PATIENT MEDICATION ADHERENCE MONITORING IN CLINICAL TRIALS FOR RHEUMATOID ARTHRITIS
Finarb has worked with a number of clients on improving medication adherence. Below is the use case specific to improving adherence in clinical trials thereby improving efficacy of trials.
Development Strategy
Data Requirements:
- Patient adherence records from trial databases, pharmacy medication refill data, wearables data, and patient self-reported medication adherence data through primary data collection surveys/interviews.
Exploratory Data Analysis (EDA) & AI Modelling:
- We determined medication adherence in terms of Proportion of Days Covered (PDC) for the specific medication for the disease mentioned. PDC is the proportion of days in a period that a patient could have taken their medication, given their supplied medicine amounts.
A patient PDC score ≥0.9 is conventionally considered adherent, and below is considered non-adherent.
Prior to building the PDC model, the EDA stage is critical to understand the data structure, trends, and patterns, leading to informed decision-making in dependent variable creation, feature selection and subsequent model building. During EDA we detected outliers, which were treated appropriately to ensure the model's accuracy. We also employed data imputation methods for missing data.
Post EDA, key features influencing medication adherence were identified and analyzed. New potential predictive variables were also selected. We used 35 different features including patient demographics, prior adherence performance, SDOH data, medication features etc. For more accuracy, real world data (RWD) or patient sentiment analysis data was also fed in, we used NLP techniques with BERT (Transformer Models) for analyzing unstructured patient feedback data from follow-up assessments, to gauge improvement of patients on various parameters such as pain, mental health, treatment effectiveness, side effects etc,.
We proceeded to assess the importance of these features in predicting medication non-adherence using ML classifier 'Random Forest,' a powerful ensemble based non-parametric, tree-based learning algorithm that excels at learning both linear and non-linear relationships between explanatory variables and dependent variables. The 'Random Forest' model was trained using the dataset with the engineered features and learned the relationships between the features and medication adherence status.
The model conducted risk stratification, to separate high risk individuals from low risk, based on predicted non-adherence. For the subsegment of high risk patients, the model also recommended personalized interventions and follow ups based on top drivers of non- adherence
Key Results
- 35-40% reduction in risk of non-adherence
- Marked improvement in clinical trial effectiveness
AI ENABLED PATIENT COHORT IDENTIFICATION IN TRIALS FOR DIABETES MELLITUS
Development Strategy
Data Requirements:
- Patient Records - predominantly EHR, Lab information systems, historical trial data, also public healthcare datasets for data augmentation, unstructured data such as clinical notes, digital forums for patient exchange information to detect if there are certain locations or regions where this medical condition might be more prevalent to narrow down and to speed up cohort identification
- Clinical Trial Inclusion Exclusion Criteria: Detailed criteria for the clinical trial, conditions for eligibility (e.g., age range, HbA1c levels, prior treatment history)
Data processing and AI modeling
- Data Preprocessing included data cleaning to ensure data consistency, data Integration into a unified dataset
- Identified and selected features from the dataset that are critical for cohort identification, such as age, gender, specific biomarker levels in this case HbA1c levels, comorbidities, and medication history.
- Trained the ML with the chosen features and predicted the eligibility of patients based on the predefined inclusion and exclusion criteria.
- Scored and ranked patients based on who will benefit the most
- Assessed the model's performance using metrics like accuracy, precision, recall, and F1-score.
- Refined the model by iteratively adjusting feature selection, hyperparameters, and validation criteria to ensure the identified cohort is both eligible and representative of the target population.
Key Results
Predictive and accurate Patient cohort Identification who meet the trial’s inclusion/exclusion criteria
- 30% reduction in patient enrollment time
- 95% Enrollment rate – exceeding initial target
Early identification of potential drop-outs
- 85-87% accuracy in identifying patients at risk of discontinuing study
A data-driven approach in clinical trials is not without its challenges. Some of the common bottlenecks faced with mitigation measures are discussed below:
Scarcity of Data:
The availability of diverse and comprehensive data remains a challenge, especially in a global context. Clinical trials often require data from various regions and demographics to ensure broad applicability, but accessing such data can be hard due to regulatory and privacy constraints. Establish partnerships with healthcare institutions and research organizations worldwide to access diverse datasets.
Prospects for Utilizing and Acquiring Real-World Data (RWD):
Integrating real-world data, collected from sources like EHR, patient surveys and wearables, into clinical trial planning is promising but complex. Ensuring the accuracy and reliability of this data while navigating ethical considerations is a hurdle that requires careful navigation. Engage patients in data collection by incorporating patient-reported outcomes through user-friendly apps and surveys. Collaborate closely with ethics committees and institutional review boards (IRBs) to ensure ethical data acquisition and usage. Implement data quality control measures to verify the accuracy and reliability of RWD sources
Increasing Demand for Specialized Data:
As trials become more targeted and personalized, the need for specialized data, such as genetic profiles or patient-reported outcome, grows. Procuring such data necessitates innovative approaches to data collection and management. Collaborate with genetics labs, healthcare providers, or patient advocacy groups to access specialized data.
Data-driven clinical trial is leading the way into a new era of efficiency, precision, and collaboration. Join Finarb’s session on September 28th 2023, as we delve deeper into these use cases and discuss common roadblocks and mitigation measures.
Top comments (0)