[Gorodenkoff/Adobe Stock]
D’Angelo emphasized that the role of biostatistics is key and begins long before data analysis starts. “I like to say that biostatistics is at the core of any clinical trial,” she said. “Typically, your biostatisticians, and sometimes your data managers, the people overseeing the data entry, need to be involved right up front at the protocol development stage,” she said. Their early involvement can set the stage for a more efficient and potentially improve the odds of having a successful trial by optimizing sample size calculations, endpoint determination, and overall study design.
Pina D’Angelo
This early involvement of biostatisticians represents a significant shift in the industry. D’Angelo reflected on this change: “When I started off even 30 years ago, you were lucky if you got a call as a biostatistician to say, ‘Hey, can you tell me if it makes sense to have this number of subjects in my study?’ Now it’s quite regimented. Every time you suggest a certain sample size for a clinical trial, you have to defend to the agency why you think that you need this many subjects.”
This involvement is key in adaptive trial designs, where flexibility and real-time decision-making are priorities. Unlike traditional trials with a fixed protocol, adaptive designs allow for pre-planned adjustments based on accumulating data. “You’ll do the first part of your study just to decide on dose ranging and what is the best dose to carry forward with,” D’Angelo asked. “And then you do your second part with the dose that you selected from the first part along with placebo, and you adjust your sample size based on what you saw in your first part.”
Tweaks along the way could involve modifying the sample size, altering the treatment regimen, or even stopping the trial early if strong evidence emerges. D’Angelo points to the rising popularity of “seamless” Phase 2b/3 studies as a notable example of this approach. “Instead of requiring two completely separate studies, you have everything in one protocol,” she said. This streamlined approach, designed with biostatistical input, allows for a continuous flow from Phase 2b to Phase 3, eliminating the time and resource-intensive process of stopping and restarting trials.
Understanding endpoints and statistical models in adaptive trials
Another important element of designing effective adaptive trials is selecting appropriate endpoints and statistical models that make careful note of data type—whether continuous variables or dichotomous. “For continuous variables, you might use analysis of variance or t-tests,” she said, referring to statistical methods for comparing means between groups. “For dichotomous variables, you might use a chi-squared test where you’re looking at whether there’s a significant difference between people who took placebo versus active treatment in a responder rate.”
D’Angelo also underscored the importance of not solely focusing on p-values, which are used in hypothesis testing to determine the statistical significance. “In phase 2b or phase 3 studies, which are the last steps before we put a product on the market, these tend to be very regulated studies,” she said. “We control them with alpha [a predetermined significance level], which relates to the p-value.” In the industry, a p-value of 0.05 is standard. A number smaller than 0.05 means your clinical trial is a success. Something bigger than 0.05 is not. “But we always encourage people, especially in the earlier phase studies, to not focus solely on p-value,” she said. “Try to understand the data, understand if there are trends, and if there is value in moving forward.”
The role of teams in adaptive trials
While adaptive trials hold promise, D’Angelo acknowledges that they are not without their challenges. Navigating them requires close collaboration between biostatisticians, data managers, and SAS programmers. “You have to be better prepared for this type of study,” she cautions. “When you’re constructing your database, for instance, you need to think not only of the first phase but of the second phase as well. So you need to have a team that’s experienced in doing these kinds of things so that they can foresee some of the issues that may come up.”
The complexity of these trials requires a skilled team of biostatisticians and SAS programmers. D’Angelo explains, “Generally, you’d have about two SAS programmers for every biostatistician working on the study.” She notes that while other software options like R or S-Plus exist, SAS remains the industry standard.
SAS programmers are an integral part of the work. “They receive an extract from the EDC (Electronic Data Capture) system containing all needed data,” D’Angelo said. The programmers oversee data processing, transform the raw EDC data into standardized formats. They also ensure compliance with Clinical Data Interchange Standards Consortium (CDISC) standards, specifically the Study Data Tabulation Model (SDTM) and Analysis Data Model (ADaM). “These are formats that the FDA has essentially required for data submission,” D’Angelo notes. Finally, they program the Tables, Listings, and Graphs (TLGs) that form the backbone of the final study report.
Looking to the future
While skilled professionals are the backbone of current adaptive trial designs, the future could see a growing role for AI/ML, a field that offers both promise and hype. “I think AI has a wealth of possibilities,” D’Angelo said. In clinical trials though, the technology is in its relative infancy.
“The important part is that when you amalgamate this data, the data that you are putting together has to be similar,” D’Angelo said. “You have to be careful because if you’re going to put a whole bunch of data together and then you realize that one of them was a study in only males and another one was a study only in females … when you put all of this data together and if you don’t take that into consideration, you may be making very inaccurate statements overall when you’re summarizing things.”
Despite these challenges, D’Angelo sees a place for AI in clinical trials, particularly in machine learning applications. “There are advantages,” she noted, for instance in searching and extracting studies on adaptive designs in dermatology over the past decade. Yet the technology demands a careful approach. “There’s a lot of potential for AI, but you just have to be careful to ensure your statistical models are appropriate,” she said.
Filed Under: clinical trials, Drug Discovery
