Clinical trials are the cornerstone for evaluating the efficacy and safety of new therapeutic agents and procedures. However, the process of recruiting a sufficient number of participants within a defined timeframe is notoriously complex and often jeopardizes the successful completion of trials. The task of determining a patient’s eligibility is knowledge and data intensive, requiring careful examination of medical records against detailed protocol documents that specify study design, entry criteria, and monitoring procedures. Criteria can range from specific diagnoses and laboratory results to demographic characteristics, prior medications, and even subjective factors like patient compliance.
This process is dynamic, time-consuming, and prone to error, especially given the sheer number and complexity of criteria, and the fact that patient conditions can change. In large academic clinics, where dozens of clinical trials may be active simultaneously, potentially eligible patients are frequently overlooked. This often leads to significant delays, and in some cases, the premature termination of trials due to insufficient recruitment.
A Computer-Assisted Solution for Enhanced Eligibility Determination
To address these critical recruitment challenges, a novel methodology for computer-assisted determination of patients’ eligibility for clinical trials has been developed. This system aims to assist clinical researchers by automating parts of the eligibility-determination process, thereby reducing the information load on clinicians and enabling a more systematic identification of potentially eligible patients. The methodology is designed to:
- Identify areas where patient information is incomplete.
- Indicate the possibility of a patient’s condition changing over time.
- Highlight characteristics that can be modified or acted upon to render a patient eligible.
Core Components of the Methodology:
- Modeling the Clinical Eligibility Process: The methodology is grounded in a model of how clinicians identify eligible patients, which involves several steps:
- Basic Entry Requirements: Initial screening for fundamental criteria (e.g., a diagnosis of AIDS).
- Routine Laboratory Tests: Checking if common lab results fall within protocol-specified ranges (e.g., platelet count greater than 75,000/mm³).
- Collecting Additional Information/Ordering Special Tests: If initial criteria are met, clinicians may order specific, unusual tests required by the protocol (e.g., lumbar-puncture studies).
- Subjective Evaluations: Assessing factors like patient compliance or estimated survival.
- Adjusting Treatment Therapies: Modifying a patient’s treatment (e.g., changing medications) to meet eligibility requirements. The computer program is built to support clinicians through this iterative and often dynamic process.
- Computer-Interpretable Language for Eligibility Criteria: A simple language has been developed to express the diverse eligibility criteria found in clinical trial protocols. This language handles:
- Time-stamped parameter values: Such as platelet counts at a specific date.
- Interval-based events: Like the duration of medication administration.
- Complex expressions: Including simple comparisons (e.g.,
platelets > 75000), arithmetic combinations (e.g.,SGOT < 2*upper_limit_of_normal), prior conditions (e.g.,no prior(medication, drug_name = rifampin, 30, day)), and logical conjunctions/disjunctions (e.g.,(HIV+ = true) OR (AIDS = true)). These expressions are translated into database queries to check patient data against protocol requirements within defined “windows of acceptability”.
- Classification of Eligibility Criteria: To enhance the program’s sensitivity to the variability and controllability of patient conditions, eligibility criteria are organized into five distinct groups:
- Stable Requisite: Immutable preconditions that persist unchanged (e.g., disease history, drug intolerance).
- Variable Routine: Criteria based on routinely collected, relatively stable data over short periods (e.g., CD4 count, WBC count, platelet count).
- Controllable: Patient circumstances that a physician can modify (e.g., medication restrictions that can be resolved by switching drugs or waiting a period).
- Subjective: Criteria requiring a physician’s judgment (e.g., patient compliance, survival estimates, Karnofsky score).
- Special: Criteria depending on results from unusual, often costly or invasive, laboratory tests not typically performed in routine care (e.g., lumbar-puncture studies). This classification allows the program to alert clinicians to patients who may become eligible later or with specific interventions, and to analyze reasons for ineligibility retrospectively.
- Approaches to Computing and Summarizing Eligibility Status: The methodology offers two main approaches, which are then combined for optimal performance:
- Qualitative Approach: This approach defines discrete, nonnumeric eligibility scores (P, PP, N, FP, F) to represent the likelihood of a patient meeting a criterion. It uses heuristic missing-value assumptions (e.g., Default, Immutable, Stable satisfied, Assume satisfied, No assumption, Complex) to assign scores when current data are absent. These assumptions are designed to be inclusive to avoid overlooking potentially eligible patients. A multivalued propositional logic with AND/OR truth tables is then used to combine individual criterion scores into a summary eligibility status. This approach allows for sensitivity-versus-specificity trade-offs, where more inclusive assumptions identify more potential candidates, albeit with more false positives.
- Probabilistic Approach: This method uses Bayesian belief networks to represent eligibility criteria, dependencies among data values, and prior beliefs about patient states. Eligibility is defined as a probability between 0 and 1. The network nodes represent individual criteria, combinations of criteria, and propositions affecting criterion satisfaction. Bayesian inference algorithms update posterior probabilities based on observed patient data, providing a direct probability that a patient satisfies all eligibility criteria simultaneously. This approach offers a stronger, probabilistic statement about overall eligibility but requires extensive modeling of dependencies and elicitation of prior/conditional distributions, and can be computationally intensive.
- Combined Approach: Recognizing the complementary strengths, a combined approach is proposed. It leverages the Bayesian belief network formalism for consistent representation of individual eligibility criteria and missing-value assumptions, allowing for a principled computation of probabilities for each criterion. These probabilities are then mapped into qualitative scores, which are finally summarized using the simple, logical AND/OR truth tables from the qualitative approach. This synthesis provides robust management of uncertainty for individual criteria while benefiting from the qualitative approach’s straightforward summary procedure, effectively avoiding the complex dependency assumptions needed for a full probabilistic summary of all criteria.
Demonstrated Impact and Broader Applicability
The qualitative approach was implemented in the THERAPY-HELPER system for HIV and AIDS protocols at Santa Clara Valley Medical Center. A retrospective analysis of 60 HIV-positive patients over a seven-month period revealed compelling results:
- 20% of provisionally eligible patients were not enrolled in trials, indicating significant missed opportunities.
- 90% of patients could not be ruled out as ineligible, suggesting a large pool of potential candidates.
- Frequent missing data (e.g., platelet counts missing for 31% of visits, CD4 counts for 43%) often obscured eligibility. These findings strongly suggest that an automated eligibility screening program can increase patient accrual by actively alerting clinic staff to necessary data collection and controllable criteria.
Beyond clinical trial recruitment, this methodology holds great promise for determining the applicability of medical practice guidelines in specific clinical situations. Practice guidelines, much like trial protocols, define specific criteria for appropriate tests and procedures, facing similar challenges with missing data and changeable patient conditions.
This computer-based methodology represents a significant advancement in leveraging information technology to streamline patient recruitment, optimize resource utilization, and ultimately enhance the success of clinical research and the adherence to best medical practices.
Reference:
Tu, S. W., Kemper, C. A., Lane, N. M., Carlson, R. W., & Musen, M. A. (1993). A Methodology for Determining Patients’ Eligibility for Clinical Trials. Methods of Information in Medicine, 32(4), 317–324. https://doi.org/10.1055/s-0038-1634933