Mastering Oncology Clinical Trials: A Comprehensive Guide for Statistical Programmers
Introduction
Oncology is a branch of medicine that deals with the prevention, diagnosis, and treatment of cancer. Clinical trials in oncology are essential for the development of new therapies and the improvement of existing ones. Statistical programmers play a crucial role in the design, analysis, and interpretation of data generated from oncology trials. However, there are several key differences between oncology trials and those in other therapeutic fields that statistical programmers need to understand. This article aims to provide a comprehensive overview of these differences and offer guidance to statistical programmers who are new to the field of oncology.
Endpoints
Endpoints in oncology studies differ from those in other therapeutic fields. The most commonly used endpoints in oncology trials are:
a. Objective Response Rate (ORR): The proportion of responders (complete or partial) among all eligible subjects.
b. Overall Survival (OS): Time from randomization to death from any cause.
c. Progression-Free Survival (PFS): Time from randomization to disease progression or death.
d. Quality of Life (QOL): A measure of a patient’s well-being and ability to perform daily activities during treatment.
Data Collection
Data collection in oncology trials involves gathering more information than standard safety data. This additional information includes tumor measurements, their responses, and Eastern Cooperative Oncology Group (ECOG) performance statuses. These data elements are necessary to evaluate the efficacy of the clinical trials and inform treatment decisions.
Adverse Event Reporting
The National Cancer Institute (NCI) has developed oncology-specific guidelines for adverse event reporting known as the Common Terminology Criteria for Adverse Events (CTCAE). CTCAE uses a grading system from 1 to 5 to categorize adverse events based on their severity, with grade 1 being mild and grade 5 corresponding to death. Each adverse event is coded to a preferred term and system organ class using the Medical Dictionary for Regulatory Activities (MedDRA) and is also classified by severity scales via CTCAE.
In safety analysis, treatment-emergent adverse events (TEAEs) are typically the main focus. The treatment-emergent period is defined as the time from the first dose of the study drug to a pre-specified period after the last dose date (e.g., 28 days, 30 days). In crossover studies, the derivation of the treatment-emergent period can be more complex, as it is necessary to determine which treatment is the true trigger for the adverse events. The incidence rates of TEAEs are usually summarized by system organ class (SOC) and preferred term (PT) in terms of severity, relationship to the drug, cause of dose reduction, and other factors.
Tumor Measurement and Assessment Under RECIST Guidelines
In oncology, Response Evaluation Criteria in Solid Tumors (RECIST) is the primary tool used to assess tumor progression or shrinkage for solid tumors. Tumor lesions are first categorized as measurable or non-measurable at baseline. Among the measurable lesions, target lesions are identified, and baseline measurements are recorded.
During treatment, subsequent measurements are performed for all target, non-target, and new lesions at each pre-specified time-point. The changes in tumor size determine tumor response, and the response at each time point is evaluated using specific criteria for target and non-target lesions.
Per RECIST, the best overall response at the subject level is the best response recorded from the start of the treatment until disease progression, taking confirmation requirements into account.
Oncology-Specific SDTM Domains and Time-to-Event ADaM Datasets
Oncology trials require specific Study Data Tabulation Model (SDTM) domains, including TU (Tumor Identification), TR (Tumor Results), RS (Response), and Time-to-event Analysis of Clinical Model (ADaM) datasets. These domains and datasets are crucial for capturing and analyzing tumor-related data and endpoints in oncology trials.
Censoring and Confirmation Rules
Censoring occurs when patients have not experienced any event of interest or have not had a follow-up. Most time-to-event endpoints require a careful review of the censoring rules and criteria for confirmation of progression. For example, common censoring rules for OS and PFS can be summarized in the following table:
In time-to-event ADaM datasets, the information of event or censoring is captured by the variable CNSR, with CNSR = 0 for events and CNSR > 0 for censored records. Additionally, the date of event/censoring is collected in the ADT variable, with the associated AVAL variable computing the time to event/censoring from the origin. The general formula to calculate AVAL is typically (ADT [Date of Event/Censoring] — randomization date + 1).
Confirmation criteria and pre-defined time windows for confirmatory scans should be documented in oncology studies.
Event-Oriented Efficacy Analysis
Efficacy analyses in oncology trials are typically event-triggered and will be evaluated only when the occurrence of pre-specified event counts is reached. For instance, an interim analysis and a final analysis for the OS endpoint occur when a pre-specified number of death events in the protocol have been reached.
Special Statistical Analysis for Efficacy
Oncology trials involve special statistical techniques to analyze their efficacy. Some of these techniques include:
- Log-rank test: To compare the OS/PFS between two treatment groups.
- Kaplan-Meier curve: An intuitive graphical representation of the survival distribution in different treatment groups. The 50th percentile of Kaplan-Meier estimates can be used as the estimate of the median duration (time when half of the patients are event-free) for time-to-event endpoints.
- Reverse Kaplan-Meier: Refers to reversing events such that loss-to-follow-up is treated as “events” while the outcome events are treated as “censored” to estimate the median follow-up time.
- Cox regression: Provides a way to estimate and compare the survival experiences of two treatment groups. Typically, the hazard ratio associated with Cox regression is calculated, which refers to the relative risk of experiencing an event of interest between two groups.
- Binomial test of binary proportion: Along with its confidence interval (Exact CI or normal approximation) to compare the proportion of subjects free of events or the difference in response rates at given landmark time points.
Conclusion
In conclusion, clinical trials in oncology differ from those in other therapeutic fields in several key aspects. Understanding these differences is essential for statistical programmers working in oncology trials. This article provides an overview of these key differences, including endpoints, data collection, adverse event reporting, tumor measurement and assessment, oncology-specific SDTM domains and ADaM datasets, censoring and confirmation rules, event-oriented efficacy analysis, and special statistical analysis techniques. Familiarity with these concepts will enable statistical programmers to contribute effectively to the design, analysis, and interpretation of oncology clinical trials.
