In the ideal world of clinical research, every new treatment is tested against a concurrent control group in a randomized controlled trial (RCT). This setup allows researchers to isolate the "signal" of the drug from the "noise" of natural recovery, the placebo effect, or the Hawthorne effect—where patients improve simply because they are being observed.
However, in many modern scenarios—such as rare diseases, aggressive cancers, or post-deployment vaccine monitoring—maintaining a placebo group is either ethically impossible or practically unfeasible. This has led to a surge in single-arm clinical trials, where every participant receives the experimental intervention. While these trials are faster and more patient-centric, they face a fundamental evidentiary crisis: without a baseline for comparison, how do we know the treatment actually worked?
The Attribution Problem
The primary challenge of a single-arm design is the lack of a counterfactual. As noted by Antoine Vanier and colleagues at the French National Authority for Health in Rapid access to innovative medicinal products while ensuring relevant health technology assessment, the absence of a control group makes it nearly impossible to distinguish a drug’s therapeutic effect from the natural progression of a disease.
Consider a study on mobile health interventions for stroke survivors. Jaewoo Kang and Woo‐Keun Seo observed clinical improvements in patients using a self-monitoring app, but because there was no control group, they could not definitively attribute these gains to the technology itself rather than routine clinical care (Feasibility of Using a Mobile Application for the Monitoring and Management of Stroke-Associated Risk Factors). This pattern repeats across disciplines, from educational scaffolds for 5th graders (Morris Siu-Yung Jong) to complex surgical techniques for hiatal hernias (Zhewen Feng).
Synthetic Baselines and External Controls
To solve this, researchers are developing methods to "create" a control group where one does not exist. The most prominent strategy is the use of external control cohorts—data drawn from historical trials, electronic health records, or disease registries.
Jérôme Lambert and Sylvie Chevret propose a rigorous three-step framework to formalize this process (How to improve the quality of comparisons using external control cohorts in single-arm clinical trials?). By using statistical techniques like propensity score weighting, researchers can adjust the external data so that it matches the characteristics of the patients in the single-arm trial. This allows for a "pseudo-randomized" comparison, providing the comparative baseline necessary for regulatory drug approval.
Borrowing Strength Across Baskets
In oncology, the challenge is often compounded by genetic diversity. "Basket trials" test a single drug across multiple cancer types that share a specific mutation. Because these sub-groups are often small and treated in single arms, estimating the treatment effect for a specific tumor type is difficult.
Zhixuan Chu and Sheng Li introduced Multi-Task Adversarial Learning (MTAL) to address this (Multi-Task Adversarial Learning for Treatment Effect Estimation in Basket Trials). Their approach uses machine learning to "borrow strength" across different tumor types. By learning shared representations of how a drug interacts with a mutation, the model can mitigate selection bias and provide more accurate predictions for how an individual patient might respond, even when the data for their specific cancer type is sparse.
The Ethical and Statistical Frontier
The move toward single-arm trials is often driven by necessity. In vaccine science, once a vaccine is proven effective, it becomes unethical to give future trial participants a placebo. Peter B. Gilbert and Youyi Fong argue that we must shift from simple associations to causal frameworks that can evaluate "immune correlates of protection" without a concurrent control arm (Assessment of Immune Correlates of Protection via Controlled Vaccine Efficacy and Controlled Risk).
While single-arm trials—such as those for the CDK 4/6 inhibitor dalpiciclib (Pin Zhang) or novel gene therapies for blindness (Jinyuan Wang)—provide essential early-phase signals, the consensus across the field is that they remain "hypothesis-generating." The goal of this research direction is to provide the statistical rigor needed to turn these single-arm observations into definitive clinical evidence.