It’s a fundamental of science that today’s failures are stepping stones to tomorrow’s successes.
But that tends to get lost when clinical trial failures hit the headlines. Stock prices take a hit, investors demand answers and patients’ hopes diminish.
Fortunately, drug developers have new tools to reduce the likelihood of clinical trial failures, thanks to the emerging field of data science. Pretty much every industry is embracing data science — especially analytics and predictive modeling. It’s helping aeronautics experts anticipate when an aircraft needs servicing or parts need to be replaced, and enables financial companies to maximize profits by identifying subtle correlations between oceans of transaction data.
Data science also is driving the evolution of quantitative clinical development, which is helping clinical trial supervisors anticipate where things can go wrong and plan accordingly. That sounds encouraging, but it’s important to understand what a quantitative approach is, and what it isn’t. Though quantitative development can produce predictive models, it’s not a magic box that reveals the future. It’s a tool that does certain things well.
You don’t use a hammer to turn a screw, and you can’t expect quantitative methods to fix all challenges in a clinical trial. But you can look to a more quant-driven approach to provide key insights, potentially reducing the risk of clinical trial errors and failures that can potentially cost millions, should a trial need to be repeated. Modeling helps us understand the relationship between the drug, target, patient and disease.
The emergence of quantitative clinical development poses three critical questions for biopharma leaders:
- When is the optimum time to use quantitative tools?
- Where does quantitative modeling prove its value?
- What are the inherent limitations of these models?
Let’s dig into the answers to these questions.
When is the optimum time to deploy quantitative tools?
Quantitative modeling uses computers, algorithms and large datasets, essentially looking to the past for an idea of how the future might shape up.
“We begin the modeling process very early, even before the first clinical trial,” says Frank Hoke, PhD, vice president for quantitative clinical development for PAREXEL, a leading biopharmaceutical services company. Some of the most critical decisions of the entire drug development process – which can last a decade or more — happen in the human trials that begin in Phase I.
“We're making go, no-go decisions” in phase I, Hoke says. In the ensuing development phases, modeling can help establish a proof-of-concept and glean accurate dosage data to help target optimal patient populations. Still, “there may be some of the greatest value in the early phase,” Hoke says.
Where does quantitative modeling prove its value?
Clinical trials can succeed or fail based on getting dosages right. In trials, scientists look for the smallest and largest concentrations of a drug that produce a desired effect. “By understanding the two ends of that spectrum as well as in between, we can determine which doses or dose levels are most appropriate to study in a clinical trial,” Hoke says.
The challenge with dosages is that too much produces too many side effects, and too little does not improve people’s conditions.
“It’s critical to understand the clinical benefit vs adverse events,” Hoke says. “Sometimes these models reveal where you need to be — the concentration range for maximum benefit and minimum risk that enables you to choose the most effective dose.”
A model is not static. It evolves through an iterative process and pulls in data from ever more sources, accounting for variables like age, ethnicity, gender and many more. And it can combine models. An exposure-response model, for instance, can be combined with a disease-progression model.
“If you understand the disease model and the drug exposure-response model, then you can combine the two to give you an even better understanding of all these previous components,” Hoke says.
What are the inherent limitations of quantitative modeling?
“Some people may shy away from a model because they say a model cannot capture the complexity of the biology and pharmacology of a human,” Hoke says. “Well, I think I would argue otherwise that we can.”
Though models are becoming more robust, it’s still crucial to understand exactly how a model was built, what assumptions it uses and what it can — and cannot — predict.
“We don't want clients to think the approaches we’re talking about in model-based drug development are going to be able to answer all of their questions,” Hoke says. “That's not realistic.”
Inevitably, quantitative modeling will always be about confronting what you don't know -- testing hypothesis that are costly, challenging or not feasible in patient populations.
“If you have a very early model with very little data, despite providing some insights there could be a large amount of uncertainty,” Hoke says. “As we refine it, add more information, it becomes more robust. Less uncertainty means it's going to be far more predictive.”