A cancer patient runs out of standard options. Their tumour carries a genetic change that is the target of an approved drug, but that drug is approved for a different cancer. Should the oncologist prescribe it anyway? For years the honest answer was a shrug. The drug might help. It might do nothing but cause harm. And whatever happened to that one patient usually vanished into a chart, teaching the next doctor nothing.
A study published in Nature on April 15 tries to replace that shrug with numbers. Researchers in the Netherlands report results from 1,610 patients enrolled in the Drug Rediscovery Protocol, or DRUP, a trial built specifically to track what happens when cancer drugs are used off-label based on a tumour's genetics. Between July 2016 and May 2024, these patients started treatment with 37 different drugs, each matched to a genomic alteration in their tumour rather than to the organ where the cancer began.
Matching drugs to mutations, not organs
The logic behind DRUP is that cancers sharing a genetic feature might respond to the same drug even if they arise in different tissues. A drug approved for breast cancer with a particular mutation might, in principle, work against a lung or bladder tumour carrying that same mutation. Oncologists already prescribe this way, off-label, but the results are almost never collected in a systematic way. DRUP is the collection system. Every patient had advanced solid cancer, no remaining standard treatment, and an actionable alteration that pointed to a specific drug.
Of the 1,610 who started treatment, 1,363 could be evaluated for response. A striking 39.1 percent of those had rare cancers, the kind that almost never get their own dedicated trials because there are too few patients to run one. That alone is part of the point. A basket approach lets people with uncommon tumours access precision drugs that would otherwise be out of reach.
Modest overall, dramatic for a few
The headline results are sobering and encouraging at the same time. The clinical benefit rate, meaning a confirmed response or stable disease lasting at least 16 weeks, was 34.9 percent. The objective response rate, where tumours actually shrank, was 15.7 percent. Median progression-free survival came to 3.4 months and median overall survival to 8.2 months. These are not miracle numbers. Across all the drug and tumour combinations lumped together, the activity was modest.
But the average hides the interesting cases. Seven percent of patients were exceptional responders whose tumours retreated far beyond what anyone would expect. Certain molecular subgroups did much better than the pooled figures suggest. The value of a trial like this is partly in finding those pockets of benefit, so that a drug-mutation pairing that helps a defined slice of patients can be pulled out of the noise and studied properly.
Safety was in line with the drugs' known profiles. Grade 3 or higher treatment-related adverse events, the more serious kind, occurred in 28.4 percent of patients. For a population with advanced cancer and no other choices, that is a familiar tradeoff.
From data to reimbursement
What sets DRUP apart from a one-off study is where its findings went. The Dutch authorities used evidence generated inside the trial to make reimbursement decisions, meaning the data helped determine which off-label uses the health system would pay for. That is a rare loop to close. Careful outcome tracking fed directly back into policy.
The authors are candid about the limits of what they built. This was not a randomized trial with a control group, so a 34.9 percent benefit rate cannot be read as proof that any single drug beat doing nothing. The patients were heavily pretreated and diverse, which makes comparisons across subgroups slippery. And "clinical benefit" counts stable disease, a softer endpoint than tumour shrinkage or longer life. The team is careful to frame the modest overall numbers as a reason for discipline, not enthusiasm.
Their recommendation follows from that caution. Off-label precision drugs, they argue, should be used only inside frameworks that systematically measure whether they work and whom they harm, that refine the biomarkers over time, and that feed toward possible formal approval. They call for prioritizing high-confidence targets, treating patients earlier rather than as a last resort, using endpoints regulators recognize, and pooling data across countries. The through-line is that genomics-guided treatment is worth doing, but only if someone is keeping score.
Comments