Why so many new drugs stall in trials, and how flexible designs may save them

The four questions a new drug must answer before approval

A new medicine does not enter the market by proving one thing at a time. It has to survive four separate questions, each asked in a different setting and with a different kind of evidence.

In Phase I trials, usually 20 to 100 healthy volunteers take part, often starting with very small doses that are increased step by step. Doctors watch for side effects, measure how quickly the drug appears in blood and tissue, and learn what dose range seems tolerable. If the body clears the compound too fast or if dangerous reactions appear early, the project may stop before it reaches patients who need treatment.

The second question is whether there is any sign that the drug helps people with the disease being treated. Phase II trials usually include 100 to 300 patients, and they are often run in two linked steps. The earlier step looks for biological changes or early responses that can guide dose selection. The later step tests how well the treatment works in a larger group and helps choose the dose that will be used in the next stage.

By the third question, the trial has to show whether the benefit holds up outside a carefully chosen group. Phase III studies usually enroll 300 to 3,000 participants across several centers and compare the new treatment with existing options or with a placebo. The population is broader, so it includes people who differ in age, sex, other illnesses, and background treatments. This stage supplies much of the evidence that regulators need when deciding whether a medicine can be approved.

After approval, a fourth question remains: what happens when many more people use the drug over a longer period? Phase IV monitoring watches for rare side effects and long term outcomes that smaller trials may miss because they are shorter or more tightly controlled. It can also test whether the medicine works in new diseases, behaves well with other drugs, and still offers benefits that outweigh risks in everyday care.

Why so many trials stall, and the flexible designs meant to rescue them

Trials stall for reasons that are often practical before they are scientific. A drug may look promising in early tests, yet the people who need it cannot be found quickly enough. Eligibility rules can exclude patients with other illnesses, older adults, or those taking common medicines. Trial sites may sit far from where eligible families live, and some patients hesitate to join a study that asks them to take an unproven treatment. These barriers matter because a trial that enrolls slowly loses momentum: costs rise, staff turnover changes routines, and the disease can evolve while the calendar slips.

Failure rates show how expensive this stall is. About half of Phase II trials do not succeed, and 40% of Phase III trials fail despite large investment. That means many drugs die after companies have already spent years on chemistry, safety testing, site setup, and patient visits. The issue is that fixed trial designs often keep running a poor arm long enough to waste patients’ time.

Flexible designs try to cut the waste by letting a trial learn while it runs. In response-adaptive randomization, more patients are assigned to treatments that appear to be working better. A breast cancer example showed fewer patients exposed to weaker options and faster identification of effective therapies. The statistical rules are set in advance, so the design can bend without turning into an ad hoc experiment.

Seamless Phase II/III designs merge dose finding with confirmatory testing. Because earlier data can feed the final analysis, development may shorten by 12 to 18 months while preserving statistical power. Platform trials take a different route: one master protocol tests several treatments against a common control, and new arms enter or exit according to pre-set rules. During the COVID-19 pandemic, the RECOVERY trial used this approach to identify dexamethasone’s benefit quickly.

Bayesian methods add another layer of flexibility. They begin with prior knowledge from earlier studies or expert judgment, then update the estimate as new data arrive. This can make probability statements more intuitive and reduce penalties for interim looks at the data. The US FDA has grown more accepting of such designs, especially in rare diseases where recruiting a traditional sample size may be impractical.

Sensors, algorithms, and real-world records as trial witnesses

A trial is not only a building where patients sit in chairs while staff check boxes. Increasingly, it is also a stream of signals that follows people through their ordinary days. A wrist sensor can measure how steadily someone walks, how deeply they sleep, or how much their heart rate swings from one beat to the next. Those numbers arrive between clinic visits, when symptoms may be shifting too slowly for a short appointment to catch.

In one Parkinson’s disease example, smartphone based assessments detected changes in motor function more sensitively than quarterly in person clinic visits. Such tools matter because a drug may improve walking speed on some days and not others, or it may ease tremor only after repeated use. Continuous records can reveal those patterns without asking patients to remember every stumble. The catch is that these digital measures still need stronger standards before regulators fully trust them as primary evidence. Guidance for such tools is evolving, which means a sensor reading can be useful context even when it cannot yet carry the whole case.

Algorithms add another layer of witness. They can scan thousands of entries for patterns a human reviewer might miss, such as dosing errors, missing visits, or values that clash with the rest. Central statistical monitoring uses this kind of analysis to flag sites where data look suspicious, not because every entry must be checked by hand. But the method depends on good models and skilled teams, capabilities that many organizations still lack.

Real world records bring a different kind of proof. Electronic health records, claims databases, patient registries, and wearable devices can show how treatment behaves in routine care. Those settings include people who would never qualify for a tightly controlled trial. They also reveal whether benefits persist when patients refill prescriptions, change doctors, or live with the treatment for years.

When the patient is rare or the treatment is a living cell

When a disease is rare, the hardest part of a trial may be finding enough people who can safely take part. A gene therapy for inherited retinal dystrophy caused by biallelic RPE65 mutations had only 31 participants in its confirmatory test because so few patients carried that exact genetic defect. The small group did not weaken the question being asked. It sharpened it: could a one-time treatment restore useful vision, and would that benefit last long enough to matter? Traditional eye charts were not always the right measure, so the trial looked at functional vision, the kind of seeing that helps a person move through daily life. Long follow-up became essential because a single infusion can leave effects behind for years, and late problems may appear only after the first visits end.

A living cell therapy adds another difficult layer to the trial. CAR-T treatments such as Kymriah and Yescarta begin with a patient’s own immune cells, which are removed, engineered in the lab, and returned to attack cancer. That manufacturing step takes about three weeks, so timing becomes part of medicine. The treatment can trigger cytokine release syndrome, an intense inflammatory reaction that may require close monitoring and intensive care support. For that reason, trials need sites with special equipment, trained staff, and a clear plan for managing severe reactions. A patient who responds well in one center may still face practical barriers if the nearest qualified site is far away.

Zolgensma, a one-time gene therapy for spinal muscular atrophy, offers another example of why timing matters. Infants with the disease can lose motor function before they reach milestones such as sitting or walking. A single dose may change that course, but only if it arrives early enough and is followed closely over time. Gene therapy trials also have to ask whether the new instruction reaches the right cells, stays active long enough, and avoids an immune response that could undo the benefit.

Rules that move faster than drugs: approval paths and data quality

Approval rules can now move faster than the medicines they govern. The US FDA has granted Breakthrough Therapy Designation to more than 300 drug programs since 2012. It gives intensive guidance and rolling review when early evidence looks strong. Accelerated approval lets a medicine reach patients earlier, based on surrogate endpoints. A surrogate endpoint is a measurable sign expected to reflect real clinical benefit. Confirmatory trials must still follow, so the early claim remains provisional. About a quarter of novel oncology approvals use this route in the US. The EMA offers a similar Priority Medicines scheme for promising treatments in Europe. Speed alone does not protect patients if the evidence behind it is weak. Data quality acts as a regulatory gatekeeper for every trial result. Poor data can invalidate a well designed trial, waste money, and delay access to useful treatments. Missed visits, wrong doses, eligibility errors, entry mistakes, unclear instructions, and weak site training can all corrupt a dataset. More than 40 percent of trials need protocol amendments before the first participant visit. That early churn suggests many protocols launch too complex or unstable. Telemedicine visits, wearables, and direct to patient delivery add new compliance gaps. International trials also face different readings of the same quality standards across countries. Regulators are shifting from checking every data point to watching the risks that matter most. Design teams should identify which safety and endpoint data need closest monitoring before a trial starts. Central statistical monitoring uses algorithms to flag fraud, systematic errors, or poor site performance. It is more efficient than verifying every source record, but it needs strong analytics skills. Build risk based controls into the design from the start. Use electronic capture with built in checks and real time review of entries. Simplify protocols by collecting only the safety and efficacy data that truly matter. Train sites with role specific instruction before start up, then again during the trial. Create a collaborative quality culture instead of relying on compliance alone. Ethical compliance, recruitment, diversity, and data quality are deeply linked in practice. Improving one area can help the others, while failure can spread through a trial.

A faster path to medicines that patients can actually receive

The journey from a promising molecule to an approved medicine is becoming less like a fixed ladder and more like a living system. Each phase still asks its own question, but the answers now arrive through adaptive rules, wearable sensors, real-world records, and regulatory shortcuts that were rare even a decade ago. The opportunity is clear: fewer wasted trials, faster access for patients with rare diseases, and better evidence about who benefits from a treatment before it reaches millions of people.

The obstacles are just as concrete. Recruitment remains difficult when eligibility rules are narrow or trial sites are far away. Data quality can collapse an otherwise clever design if visits are missed, doses are wrong, or records do not match source documents. Digital tools promise continuous observation, yet their validation standards and regulatory acceptance are still developing. Gene and cell therapies add further strain because they require precise patient selection, specialized manufacturing, long-term follow-up, and centers capable of managing serious immune reactions.

What matters now is whether these pieces can be joined into one coherent practice. Adaptive designs need reliable data, and real-world evidence needs trustworthy records. Expedited approval also needs confirmatory monitoring that does not disappear after launch. Diverse enrollment needs protocols simple enough to run well in many settings. The coming test will not be another single technology. It will be the ability of sponsors, sites, regulators, and patients to share a common standard: speed without shortcuts, flexibility without drift, and innovation without leaving out the people who need it most. Can trial systems become as adaptive as the therapies they are meant to approve?