Over the past decade, the variety and therapeutic potential of cancer medicines have escalated significantly1. New classes of treatment are prolonging the lives of many people with advanced disease2. Yet, for people with incurable cancer, there remains a desperate need for better remedies.

Physicians have relied on randomized controlled trials (RCTs) to evaluate new treatments. But there are good reasons — pragmatic and ethical — to question the continued relevance of RCTs in oncology. Just as science evolves by incorporating new methods and ways of analysis, so should clinical investigations.

Conventional wisdom holds that RCTs provide the most trustworthy form of medical evidence. Randomization, the thinking goes, eliminates bias in how trial participants are assigned to receive experimental or standard therapy. Indeed, physicians have relied on RCTs to answer basic questions in clinical oncology, such as how the benefits of lumpectomy compare with those of mastectomy to treat breast cancer.

But change is already under way. To the dismay of some medical ethicists, the US Food and Drug Administration (FDA) — which influences availability of medicines around the globe — has approved numerous cancer medicines without randomized trial data.

Critics suggest that rapid approval of relatively untested medicines engenders false hope, puts people at unnecessary risk from toxicity and jeopardizes the stepwise accretion of medical knowledge that RCTs would provide3. Others say that current regulatory processes are not fast enough to help people with short life expectancies, and favour speedy approval of drugs that have been safety-checked.

As a former oncologist and a cancer survivor, I’m more worried about denying people access to life-extending medicines than I am about the possibility of offering false hope. To paraphrase the late AIDS activist Larry Kramer: waiting for the results of an RCT is not an option for someone living with a terminal condition.

As things stand, RCTs are expensive, slow and yield frustratingly limited information. As people with cancer are living longer, each one experiencing a unique pattern of disease spread, molecular features and treatment history, the applicability of data from any one trial to an individual diminishes. Also, experimental medicines are typically pitted against existing, ineffective treatments or placebos — a strategy that benefits the investigational agents.

In the 1980s, first-in-human phase I trials were considered a last resort for people with cancer who had run out of options. The average overall response rate was a dismal 5%. But those numbers have changed: with genetic and molecular targeting of tumours, response rates are now in the range of 15–30%, with higher figures reported from studies that target specific biomarkers. Today, unlike 40 years ago, phase I trials might offer a significant chance of therapeutic benefits4.

There is a moral argument against RCTs. For a participant in a randomized phase III trial, which treatment they receive depends on a metaphorical coin toss. By contrast, a person entering a phase I trial does not submit to the unknowns of randomization. If they want to try a promising approach and their physician agrees, they can. Regardless of the outcome, they have acted with intention and have control over what treatment will be administered to their body.

Keep in mind, participants and physicians might have competing interests regarding what kind of trial is best. With precision oncology, many tumours can be considered rare diseases and so a greater reliance on single-arm trials — in which everyone receives the experimental therapy — makes sense. If a drug demonstrates impressive efficacy and safety in these early-phase trials, it should be made available by prescription, along the lines of the FDA’s Accelerated Approval Program that currently allows for expedited, tentative approval of drugs for serious conditions. Outcomes can then be monitored and analysed in phase IV post-marketing studies.

Innovative analytical platforms are already using real-world evidence drawn from registries and cancer centres to improve how treatments are developed. Investigators can use these data to run virtual control arms, populated by cohorts of people with tumour properties similar to those of people receiving an experimental agent, rather than deliberately randomizing a group to receive an ineffective treatment5. And artificial intelligence will assist physicians in sifting through all forms of evidence to identify the best treatment available for each person.

There are potential economic benefits to overhauling the way we test cancer drugs as well. In a world with fewer RCTs, cancer drugs would cost less to develop and prices could be lowered. Paradoxically, the expanded repertoire of cancer drugs that would emerge could save money in the long run, by sparing the expense and toxicity of ineffective treatments.

The future should involve prompt and transparent reporting of outcomes among people receiving all cancer medications, old and new. Industry, physicians and recipients of such drugs will need to cooperate in this endeavour — by sharing anonymized molecular, demographic, survival and toxicity data. That way, everyone affected could access the latest facts about available therapies — and make informed decisions in real time, as science advances.