When researchers develop a new drug or treatment for a disease, their first steps include verifying it is safe and determining a safe dose for specific patients. This process can sometimes be the slowest stage of development, with clinical trials for oncology treatments taking anywhere from months to several years. Alexandra Curtis, a graduate of the Biostatistics PhD program, is seeking to expedite this process.
The lengthy timeframe of clinical trials make them expensive to run, but there are other costs that emerge from the slow process. Individuals who volunteer to participate in a clinical trial must accept the possibility of negative outcomes, including unknown side effects and the possibility that the treatment is not effective. Other people who may benefit from a promising new drug stuck in trials are also forced to wait as their potentially life changing medicine clears the gradual testing process.
Curtis has focused her dissertation research on making this process more efficient. She has developed a statistical model for determining the best doses of new drugs and treatments that balances safety and efficacy for patient subgroups. While many researchers in pharmaceutical firms and academia tend to use rule-based algorithms for clinical trial designs instead of statistical methods, the latter approach offers more innovation.
“There are a lot of simulation studies out there showing that statistical models like this are more precise and sometimes safer than using simple rule-based algorithms,” says Curtis. “They can be complicated and require more work to implement than simpler models, but they’re more efficient and can help clinical trials determine optimal doses more quickly.”
The statistical model Curtis introduces is particularly beneficial in determining optimal doses when multiple patient subgroups stand to benefit from a new treatment. Her model allows subgroup data to be compared and even borrowed between groups based on their similarities to each other, so researchers can make inferences for each group more efficiently than if they were to run separate trials.
Researchers balance safety and efficacy to identify the optimal dose of a new treatment. Curtis’s model helps researchers identify the optimal dose of a new treatment for multiple groups of patients when researchers are unsure how similar or different the groups are at the start of the trial. For example, sometimes the same gene mutation will cause cancer in breast tissue or prostate tissue in different patients. Although these patients have different types of cancer, they may both benefit from a treatment which targets the mutation. The proposed statistical model adaptively changes the amount of information borrowed between groups (e.g. cancer types) based how similarly patients from these groups respond to treatment during the trial. This information borrowing allows for more efficient selection of the optimal dose for each patient group.
Curtis’s model makes fewer assumptions about specific subgroups than previously proposed approached which identify the optimal dose(s) for patient subgroups, and simulation results indicate that it selects the optimal dose for patient subgroups more often. In certain scenarios, the model was substantially more effective in dose-finding than previously proposed approaches. While Curtis focused on oncology treatments in her dissertation research, the model’s adaptive approach could be used for making a myriad of other clinical trial designs featuring dose finding for subgroups.
“The driving force for this research is to identify doses that are safe and effective for different groups as efficiently as possible,” says Curtis. “Our focus is on speeding up the pipeline of new treatments being tested and developed. That’s something everyone can benefit from.”
A model for practice
Statistical models in clinical trials require more collaboration between clinicians and statisticians in setting them up compared to simpler, rule-based designs. Implementation thus relies on interdisciplinary work from both perspectives to ensure these statistical models are effective. While Curtis, a 2020 Ballard and Seashore Fellow, is grateful for the additional time she had to focus on her dissertation in her final semester, much of her collaboration experience comes from her graduate research assistantship.
Curtis credits her research assistant position in the College of Dentistry with providing her more practical experience managing data and collaborating with interdisciplinary teams. As someone who recognizes the importance of working with other experts, she is excited to continue her new research in that type of setting at the pharmaceutical firm Eli Lilly and Company.
“For the past couple years, I have had the pleasure of working with Alex on her dissertation topics of developing clinical trial designs to better optimize safety and efficacy of new drug therapies,” says Brian Smith, a professor in Biostatistics. “She has been a tenacious researcher and a never-ending fountain of new ideas. With her motivation and talent, I have no doubt that she will continue to succeed as a professional biostatistician at Eli Lilly and that they will enjoy working with her as much as her colleagues here at Iowa have.”