Trial design, endpoints and cost

After surgery for bile duct or gallbladder cancer, a blood test for leftover tumour DNA picks out the people whose cancer will return with hazard ratios of 16 to 26 in two recent cohorts. Nobody has yet tested giving those people more than the standard capecitabine.

CHALLENGE proved exercise works in colon cancer but not how much is needed. A trial comparing doses, as we would for a drug, would tell health systems what to fund.

Running an early trial properly requires monitors, data managers, safety reporting and regulatory filings that universities cannot afford from commercial providers. A public not-for-profit would do this work at cost.

Blood tests can now find leftover cancer months before scans, but most patients who test positive have nothing to enrol in. One standing trial per country would fix that.

Platform trials that test several companies' drugs against one shared control arm, such as I-SPY 2, Lung-MAP, GBM AGILE and STAMPEDE, are each built from scratch by determined individuals. A permanent non-profit sponsor holding the protocol, control arm, statistics and data, with a standard entry contract for companies, would cut the launch of a new platform from years to months.

Instead of one-off small studies, run a standing trial that continuously tests new treatments for side-effects, adding and dropping arms as evidence arrives.

Instead of starting a new trial for every drug pair, keep one always-open trial per cancer that new arms can join and leave, sharing the same control group.

Frequent tiny doses of cheap old chemotherapy pills have shown surprising benefit in some cancers. A single large trial network in India and Africa could find out where this works and where it does not.

When researchers use hospital records to ask 'would drug A have beaten drug B in a trial', they should follow a published recipe and register their plan first, so the answer can be trusted.

No company will pay to find out whether six months of its drug works as well as twelve. A dedicated public fund would pay for those trials, which save patients side effects and health systems money.

Universities often refuse to be the legal sponsor of a first-in-human trial because they cannot afford the insurance and liability. A shared public insurance pool would remove that block.

Sometimes a trial cannot randomise, so the new drug is compared with past patients' records. Clear published rules on when that is allowed, and how it must be done, would replace case-by-case guesswork.

Three antibody-drug conjugates now used in triple-negative breast cancer carry the same kind of chemotherapy warhead, a topoisomerase inhibitor. Nobody has randomised which to give first or whether the second works after the first; small series suggest it often does not. With two now approved first line, the question decides what a patient gets for the rest of her life.

The UK's RECOVERY trial answered questions about dexamethasone and other cheap COVID drugs within months by randomising tens of thousands of ordinary hospital patients with a two-page consent and routine-data follow-up. Cancer needs the same permanent platform to test generics such as statins, metformin, aspirin and propranolol added to standard care.

Today a cancer drug is approved for shrinking tumours. A drug that stopped cancer spreading would fail that test, so almost nobody develops one. A new endpoint would fix that.

Thousands of patients have received standard treatment in the control arms of past trials. Pooling their anonymised data would let new trials borrow from them and randomise fewer patients to old treatments.

Between diagnosis and surgery there are usually a few weeks. Giving a new drug in that window and comparing the tumour before and after surgery shows whether it hits its target in real people, quickly and cheaply.

Some drugs are approved for any cancer with a particular mutation. Clear rules on how many cancer types must be tested, and how to combine results across them, would make these approvals more consistent and faster.

Rather than building a new trial from scratch for every drug, keep one permanent trial open per cancer where new treatments can be slotted in and dropped out, sharing the same patients, control group and infrastructure.

Instead of a new trial for each resistance mechanism, one continuous trial could sort patients into arms based on the reason their last treatment failed.

Muscle-invading gallbladder tumours on the free side of the organ recur in the liver far less often than those against it. Two meta-analyses found the liver resection helped only the liver-side group, so a randomised trial could test dropping it for the free side.

Large adjuvant trials already follow thousands of patients for years. Adding a second randomisation to a cheap old drug would answer repurposing questions almost for free.

When a resistance mutation first appears in the blood, the current drug is often still controlling most of the tumour. Adding a second drug rather than swapping may keep both under control.

Drugs are approved on trials of fit, younger patients and then given to everyone. A required follow-on trial in older, sicker and more diverse patients would show whether the benefit holds in real life.

For symptoms such as nausea, fatigue or neuropathy, each patient can alternate the drug and a placebo over several periods and learn what works for them. Pooling these single-patient crossover trials with Bayesian models also gives a population answer, in areas where conventional trials are rare.

Statistical methods can combine information across similar rare cancers to reach an answer with fewer patients. Regulators need to say in advance when that is acceptable.

Measuring tumours on scans for trials is slow, expensive and inconsistent between readers. Software that measures lesions and flags changes, checked by a radiologist, could make trial endpoints cheaper and more reliable.

Once a class of antibody such as PD-1 blockers is proven, later copies could be approved on smaller trials showing equivalence, forcing price competition and freeing patients and money for genuinely new drugs.

Trials often measure a stand-in for survival, such as time until the cancer grows on scans. An independent body would test, for each cancer and treatment type, whether the stand-in actually predicts survival, and publish the answer.

Three genetic classifications of the commonest aggressive lymphoma exist and none of them yet decides anyone's treatment. The trial that would change that has not been run.

Trials sometimes quietly swap the outcome they promised to measure for one that looks better. Software can compare the registered plan with the published paper and flag the switch.

Trials look at dozens of patient subgroups and some will look good by chance. A statistical method that pulls extreme subgroup results toward the overall result would make these claims more honest.

When the standard treatment has been given to thousands of similar patients in earlier trials, a new trial could randomise fewer people to it and lean on that history, as long as the old data still matches.

Gut bacteria appear to influence whether immunotherapy works, and diet and antibiotics shape gut bacteria. Yet almost no drug trial records what patients ate or which antibiotics they took. Recording it would cost almost nothing.

Trials often stop following patients once the main result is in, so we never learn whether the drug extended life. Linking participants to national death and cancer registries costs almost nothing and would answer that question.

Giving chemotherapy before the operation is standard when the tumour is borderline operable, but for tumours that can be removed straight away two trials have failed to show it beats operating first. Two more, one Dutch and one American, are directly comparing chemotherapy before and after surgery with chemotherapy after alone; the proposal is to wait for them and to pool them.

Gallbladder cancers found by chance often come back within six months of the second operation, usually far from the gallbladder, which suggests the disease was already in the bloodstream. Two randomised trials are testing chemotherapy first; one closed small, the other reports in 2028.

Three companies testing three drugs against the same standard treatment each recruit their own control group. Pooling those controls in one shared study would need fewer patients and answer faster.

Radiotherapy trials need physicists to check every plan and central review of every target drawn, which nobody pays for. Fund that infrastructure permanently so trials are faster and results are trustworthy.

After pre-surgery chemotherapy leaves tumour behind, a blood test for tumour DNA triples the risk of distant relapse when positive. The one trial that acted on it found the DNA usually appeared too late. A trial that tests at surgery with a tumour-informed assay, escalates positives to an antibody-drug conjugate and observes negatives has not been run.

Trials report side effects as a table of percentages that hides how long they lasted, how bad they felt and whether people stopped treatment. Tolerability should be measured with defined endpoints and a decision rule, like efficacy.

Trial staff still retype data from the hospital record into the trial database, and monitors then check every entry by hand. Piping data directly and checking by risk would cut cost and errors.

Most people with pancreatic cancer lose muscle and weight in a way food alone cannot reverse, and that wasting is a common reason chemotherapy is cut or stopped. A 2024 trial showed an antibody against the hormone GDF-15 restored weight and activity in twelve weeks, a third of the patients having pancreatic cancer. The proposal is to test it inside the chemotherapy trials, not alongside them.

Before spending millions on a randomised trial, analyse existing patient records as if the trial had already happened. If the answer is obvious or the question is unanswerable, skip or redesign the trial.

Trials stop following patients after a few years, so late side effects and late relapses are missed. Link trial participants to national records so follow-up continues automatically for decades.

The PERSEPHONE trial showed six months of trastuzumab after surgery is as good as twelve, halving the drug cost; no company will run such trials, so public funders and charities must.

Two thirds of women treated on the KEYNOTE-522 regimen have no tumour left at surgery and about 92 percent of them are alive without relapse at five years, yet all receive nine more cycles of pembrolizumab. Trials are now testing whether the best responders can stop early, skip the anthracycline, or in lymphocyte-rich stage I tumours skip chemotherapy altogether.

Every survivorship intervention currently raises its own small trial with its own control arm and its own endpoint. A platform trial with a shared control and a common outcome set would test several at the cost of one and a half.

Rare cancers often share a broken cellular machine even when they arise in different organs. Grouping patients by that shared fault makes trials possible.

Simulate trials of drug combinations in populations of virtual patients to decide which real trials to run, and keep score of how often the simulations were right.

Before a company runs its phase 3 cancer trial, FDA, EMA, MHRA, PMDA, Health Canada and TGA would agree the endpoints, comparator, population and statistical plan in one joint written advice letter, so a single trial can support approval everywhere. Today parallel advice is used for a handful of products a year.

Screening trials are judged on deaths, which take fifteen years to count, and on cancers found, which is the wrong direction. Preventing a man from turning up with cancer already in his bones happens three times as often as preventing a death, arrives years earlier, and is the outcome he cares about.

A blood test at six weeks can show whether a treatment is doing anything. Trials should use it to drop failing combinations fast and move patients on.

Multi-cancer early detection blood tests take a decade to prove they save lives. Regulators could accept a fall in late-stage cancers as the first answer, validated against pooled data from completed screening trials, provided approval is conditional on continued mortality follow-up.

As results come in, the trial sends more new patients to the arms that are working and fewer to those that are not, so more people benefit and bad arms die faster.

Companies sometimes choose the biomarker threshold that makes their trial look best after seeing the data. Requiring the threshold to be fixed and published before the big trial starts prevents this.

A drug that shrinks a tumour by half but leaves the resistant sub-population untouched will fail. Trials should measure whether every sub-population is cleared, not just overall size.

Most cancer patients are over 65 and a large share have other illnesses, yet trials routinely exclude them on organ function, prior cancers, HIV or brain metastases. Regulators should require sponsors to justify each exclusion, include patients with controlled comorbidities by default, drop upper age limits, and report enrolment over 75 in labels.

Trials and hospital records describe the same things in different languages. Publish the translation so trial patients can be followed for life in routine data and trial results compared with routine care.

Lung cancer spreads to the brain more than any other common cancer, and the newest drugs seem to stop it happening. Almost no trial is designed to prove that, so the claim stays a footnote.

Metastasis causes most cancer deaths, yet no drug is developed to stop cells spreading. Create a formal approval route for drugs that prevent metastasis, tested in people at high risk of it.

Several drugs are approved for the same cancer, but nobody tests which order works best because no company benefits from the answer. Public multi-arm trials could settle these questions efficiently.

Rare cancers are collectively common but each is too rare for normal trials. Link every rare cancer patient worldwide into one network with registries and trial designs built for small numbers.

Prevention drug trials run separately in each precancer and are slow. One master protocol with shared infrastructure across Barrett's oesophagus, oral leukoplakia, lung nodules and pancreatic cysts, using regression as an intermediate endpoint and adding or dropping arms adaptively, would test several drugs at once.

Rare cancers together make up a fifth of all cancers, but no single one supports its own trial, so most patients get off-label therapy with no data capture. One permanent national umbrella trial, with molecular screening for all comers and arms opened by mechanism, would give every rare cancer a route.

Whether finding cancer earlier saves lives depends on the cancer. Public models, one per cancer, would let trials and payers predict the mortality benefit from a stage shift honestly.

Every trial writes its protocol and analysis plan from scratch. A shared library of well-written templates and ready-to-run analysis code would let teams start from the best version rather than a blank page.

A drug that adds years of life would earn extra years of market protection; one that adds a few weeks would earn none. Extensions would be lost if the promised benefit is not confirmed.

Before a trial is funded, patients who have had the disease sign off on how many visits, scans and blood draws it demands, and on whether the endpoints measure things that matter to them.

When two accepted treatments are equally reasonable, the computer system would offer to randomise the choice and track the result, turning ordinary care into a continuous trial.

A trial can be designed to detect a tiny improvement that is statistically real but too small to matter. Protocols should state up front what size of benefit would be worth having, and be built to detect that.

For a frail 80-year-old, staying independent may matter more than living a few months longer. Trials designed for older patients should measure what they care about.

Patients join a long-term cohort once and agree in advance that they may be offered new treatments as they appear, while others in the cohort serve as the comparison group. No new trial has to start from zero.

When control-arm patients switch to the new drug after their cancer grows, the survival comparison gets muddied. Trials should plan in advance how they will correct for this, and report both raw and corrected numbers.

Give drug combinations before surgery and measure how much tumour remains at resection; that answer arrives in months. A standing neoadjuvant platform with a shared control arm, as I-SPY 2 runs in breast cancer, would test combinations quickly in lung, bladder, melanoma, head and neck and oesophago-gastric cancer.

Some cancers reach the brain in up to a quarter of patients. Prevention trials could aim specifically at stopping that, instead of treating it once it has happened.

Nobody knows what a cancer trial should cost because budgets are secret. Publishing anonymised cost per patient by trial type would expose waste and let funders set targets.

Trials often have independent radiologists check whether tumours grew. How often they disagree with the treating doctors is a measure of how trustworthy the result is, and it is rarely published.

Trials record why each screened patient did not join, but that data is never shared. Publishing it would show which rules block the most people and which are pointless.

Regulators are unsure what to accept as proof that an anti-wasting drug helps. Agreeing on a simple measure such as stair climbing would unblock the whole field.

Before a big trial starts, ask hundreds of patients how much extra survival they would trade for a given side-effect, so the trial is designed to test something patients would actually want.

National cancer registries already collect the outcome data. Adding a randomisation button lets doctors compare two standard treatments across thousands of patients at a fraction of the usual cost.

Rare cancer patients are already tracked in registries. Offering randomisation inside the registry makes trials far cheaper and lets almost anyone take part.

Half of rectal cancer patients given all their chemotherapy and radiotherapy first can keep their rectum. Nobody has ever randomised that against having the operation, so the trade-off between avoiding a stoma and the risk of the cancer regrowing is still guesswork.

Trials usually study one treatment at a time, so nobody knows the best order. Deciding the next step in advance, by lottery, answers the sequencing question at little extra cost.

Metastatic prostate cancer now has six classes of treatment that work, and no trial has ever compared the orders they can be given in. Every trial adds a drug to the front; none asks what should follow it, so the sequence a man receives is decided by habit and by what was licensed first.

Immunotherapy is often given for two years or until it stops working, but responses can last long after stopping. Trials that randomly assign responders to stop or continue would show whether the extra year is needed.

Use the cancer registry itself as the trial machine: randomise patients at diagnosis, then let the registry collect the outcomes for a fraction of the usual cost.

Few companies develop cancer prevention drugs because trials take decades. If regulators accepted validated precancer endpoints, as they do cholesterol for heart disease, industry would return.

Every drug gallbladder cancer patients receive was tested in mixed bile duct trials where they were a subgroup. Requiring trials to pre-specify and publish gallbladder results, with a minimum number enrolled, would give the disease its own evidence at almost no cost.

The best explanation for why one second-line CAR-T trial failed when two succeeded is how long the cells took to make. That interval is almost never a reported endpoint.

A treatment that adds two months of life but takes up most of those days in hospitals and clinics may not be worth it to an older patient. Trials should report how many days treatment consumes.

Phase 3 trials are often launched on a small single-arm response-rate study with no comparison group, and most then fail. Requiring a randomised phase 2 with a concurrent control first, with exceptions only for large effects in refractory settings, would filter out weak drugs earlier.

Once two drugs of a kind exist, a third should have to prove itself against the best of them, not against an outdated comparison, so patients and payers learn which is actually better.

The traditional way of finding a dose looks only at severe side effects in the first month. Modern statistical designs can use all patients' data and count the grumbling, long-lasting problems that make people quit months later.

Small-cell lung cancer has had two real advances in twenty-five years. It is probably four diseases being tested as one, in separate small trials that each need their own control group.

Instead of stopping after the mid-sized trial, waiting a year, then starting the big one, run them as one study with a clear pre-agreed rule for continuing. This saves a year or more per drug.

Wasting has several causes. Measuring the specific hormone in each patient's blood would put the right patients into the right trial instead of mixing everyone together.

Ninety-five percent of advanced bowel cancers ignore immunotherapy, and the only real signal so far came in patients without active liver secondaries. Trials keep enrolling by treatment line rather than by immune biology, which guarantees the responders are diluted away.

Real treatment is a series of decisions: start with this, switch to that if it fails. Sequential multiple-assignment randomised trials test whole strategies by randomising patients again at each decision point.

When a trial has no comparison group, the comparison is sometimes built from old patient records. Set rules for how that is done so the answer is not rigged.

One in five people with advanced cancer has brain spread, yet most trials refuse them. Requiring brain cohorts would give those patients evidence instead of guesswork.

Large phase 3 trials often continue for years after interim data show the drug is unlikely to work. Pre-registering futility boundaries at 30 to 50 percent of the information, judged by independent committees and reported publicly, would stop them earlier, sparing patients and freeing money for better ideas.

Giving men with castration-resistant prostate cancer large doses of the hormone the treatment has spent years removing makes a third of them respond, and makes half of them respond again to the drug that had stopped working. It has never been taken to a definitive trial, partly because the endpoint that shows the benefit is not the one trials usually use.

Trials that only enrol patients with a positive biomarker can never prove the biomarker matters. Including a smaller biomarker-negative group would show whether the test is really needed.

Guidelines send everyone whose chance-found gallbladder cancer has just reached the muscle layer back for liver and lymph node surgery, yet the largest international series found 95 in 100 alive without the disease at five years whether or not they had it. A prospective study could spare thousands of operations, or confirm they are needed.

For treatments that will not cure, what matters is how long the treatment keeps working without becoming unbearable, and how the person feels. Trials should measure both of those as their main results.

New cancer drugs should have to prove not only that they extend life but how they affect how people feel and function, with that result printed in the label like efficacy.

Nearly half of women with metastatic triple-negative breast cancer develop brain metastases and survive under five months after the diagnosis, yet the trials that set the standard mostly exclude active brain disease. Requiring a brain metastasis cohort in every phase 3, with intracranial response as an endpoint, would answer whether the new drugs reach the brain.

Frail older patients are often given full doses and then have them cut after bad side effects, or are given nothing. Trials should test starting at a lower dose from the beginning.

Scans at three months often cannot tell whether immunotherapy is working. A blood test at six weeks may give a clearer, earlier answer.

A blood test detects lymphoma left behind better than a scan does, and no trial has yet used it to change anyone's treatment.

Giving immunotherapy for a few weeks before surgery produces a tumour sample that shows exactly what the drug did. That is the fastest way to learn who responds.

If a blood test shows no tumour DNA after months of treatment, a trial could test pausing the drug and restarting only when the DNA reappears, giving patients time off without waiting for scans to show growth.

Pragmatic trials want to use the progression dates recorded in ordinary clinic notes instead of expensive protocol scans. Checking how well those routine records match formal trial measurements would show when that shortcut is safe.

Every patient on the new drug is compared with every patient on the old one: who lived longer, and if equal, who had fewer serious side effects, and if still equal, who felt better. The share of 'wins' becomes the result.