From bench to bedside: clinical trials and approval

A new therapy reaches approval through a sequence of investigations of increasing scale, increasing cost, and increasing human exposure. The canonical structure:

• Preclinical — laboratory and animal studies. Identify candidate molecules, characterize mechanism, assess toxicity in cells and animals.
• Phase I — first administration to humans. Healthy volunteers (in most cases). 20–100 subjects. Primary endpoints: safety, tolerability, pharmacokinetics (PK), dose-finding.
• Phase II — patients with the target disease. 100–300 subjects. Primary endpoints: efficacy signal, dose-response, more safety data.
• Phase III — large randomized controlled trials in patients. 300–3,000+ subjects. Primary endpoints: efficacy at a fixed dose, comparison against placebo or standard of care, definitive safety. The data supporting regulatory approval.
• Phase IV — post-marketing surveillance. After approval, in real-world use. Monitor rare adverse events, long-term outcomes, off-label patterns, additional indications.

The total time from first preclinical work to approval is typically 10–15 years. The total cost — including the cost of failures — is estimated at $1.5–2.5 billion per approved drug, with the figure varying by methodology and timeframe.

The rest of this lesson examines each phase, where the failures concentrate, and what the regulatory output actually certifies.

Phase I is the first administration of a candidate drug to humans. For most small-molecule drugs the subjects are healthy volunteers; for some classes (oncology, certain biologics, gene therapies) the subjects are patients with the disease because the risk-benefit calculation does not justify exposing healthy volunteers.

The primary objectives:

• Safety and tolerability. Identify dose-limiting toxicities. What dose causes what side effects? Are they reversible?
• Pharmacokinetics (PK). How does the body absorb, distribute, metabolize, and excrete the drug? Half-life, peak concentration, bioavailability, drug-drug interactions.
• Pharmacodynamics (PD). What does the drug do to its target? Often measured via biomarkers (a substrate for the enzyme, a receptor occupancy assay).
• Maximum tolerated dose (MTD) or recommended phase II dose (RP2D) — the dose to use going forward.

Phase I trials are typically open-label and dose-escalating. A small cohort gets a low dose, safety is assessed, the next cohort gets a higher dose, and so on until a tolerability ceiling is reached or the target exposure is achieved.

Phase I is the cheapest phase but the most ethically constrained. Many candidates die here from unacceptable toxicity, insufficient PK to support meaningful PD at safe doses, or unexpected mechanism-related side effects.

Phase II moves from healthy volunteers to patients with the disease. The trial is larger (typically 100–300 patients) and usually randomized — patients get the candidate at one or more doses, others get placebo or standard of care.

The primary objectives:

• Efficacy signal. Does the drug show the expected biological effect in patients? Are biomarkers moving in the right direction?
• Dose-response. Across multiple doses, what is the relationship between dose and effect? Which dose offers the best efficacy-vs-tolerability balance?
• More safety data. Patients with the disease may metabolize the drug differently, have organ dysfunction that healthy volunteers lacked, or take concomitant medications that interact.
• Outcome measures. Often a surrogate endpoint (HbA1c for diabetes, blood pressure for hypertension, viral load for HIV) rather than a hard clinical endpoint. Phase II rarely runs long enough to detect mortality or major morbidity directly.

Phase II is the highest-attrition phase: roughly 70% of phase II trials fail to meet their efficacy endpoint sufficiently to support phase III. The reasons concentrate in:

• The drug's effect in patients is smaller than the biomarker-based prediction.
• The effect varies more across patients than expected, so the mean signal does not reach significance.
• The dose-response is not what the mechanism predicted.
• Tolerability constraints push the usable dose below the efficacious range.

A successful phase II program produces a dose, a target population, and a primary endpoint for the pivotal phase III trial.

Phase III trials are large, randomized, often double-blind, and designed to produce the evidence regulators will use to grant or deny approval.

Standard features:

• Randomization of patients to treatment or comparator (placebo, standard of care, or active comparator).
• Double-blinding — neither patients nor investigators know who is on what — to remove bias from outcome assessment and patient behavior.
• A pre-specified primary endpoint with statistical power calculations indicating the sample size needed to detect the expected effect with high probability.
• Intent-to-treat (ITT) analysis — every randomized patient is counted in the analysis of their assigned arm regardless of whether they completed treatment.
• Pre-registration of the trial design and statistical plan, to prevent post-hoc selection of favorable analyses.

The sample size for a phase III trial is determined by the effect size the drug is expected to produce, the variance of the primary endpoint, and the significance and power thresholds (usually $\alpha = 0.05$, power $\geq 80\%$). A drug with a large expected effect needs fewer patients than one with a small effect. Cardiovascular outcome trials often need 10,000+ patients because the event rates are low and the effect sizes per-patient are modest.

Two adequate and well-controlled trials are typically required for full FDA approval (the 'substantial evidence' standard from 1962 amendments to the FD&C Act). In some cases, a single very-compelling trial with confirmatory mechanism evidence suffices.

Phase III attrition is lower than phase II (~30–40% of phase III trials fail) but the costs of failure here are much larger because the trials are so much bigger. A failed phase III erases a substantial fraction of a biotech's market capitalization in a day.

An endpoint is a measurable outcome that defines whether a trial succeeded. Endpoints fall on a spectrum.

Hard clinical endpoints. Death, hospitalization, major irreversible events (heart attack, stroke, blindness, kidney failure). What patients ultimately care about.

Surrogate endpoints. Measurements presumed to predict hard clinical outcomes. HbA1c for diabetes-related complications. Blood pressure for stroke. LDL cholesterol for cardiovascular events. Viral load for HIV-related morbidity.

Surrogate endpoints are used because they:

• Move faster (weeks to months) than hard endpoints (years).
• Have higher signal-to-noise ratios per patient, so trials can be smaller.
• Allow trials to be feasible when the hard endpoint is rare.

The trade-off: a surrogate endpoint that moves doesn't guarantee the hard endpoint will follow. The history of medicine includes drugs that improved surrogate measures but harmed hard outcomes:

• The CAST trial (1989) showed antiarrhythmic drugs that reduced ventricular arrhythmias (surrogate) increased mortality (hard endpoint).
• The ILLUMINATE trial (2007) showed a cholesterol-modifying drug (torcetrapib) that raised HDL (surrogate) increased mortality.

The FDA's policy on surrogate endpoints distinguishes:

• Validated surrogates with strong evidence linking them to clinical outcomes. Permitted for standard approval.
• Reasonably likely to predict clinical benefit. Permitted under accelerated approval, but require confirmatory trials with hard endpoints post-approval.

An 'approved on a surrogate' drug carries the asterisk that the hard-endpoint evidence is still being collected. The two CAST and ILLUMINATE examples are why the post-approval confirmation matters.

Regulatory approval is a specific decision: the regulator concludes that the drug's demonstrated benefits outweigh its known risks for the indicated population.

The FDA's primary pathways (with parallels at EMA, MHRA, PMDA):

• Standard approval. Substantial evidence of effectiveness from two adequate and well-controlled trials with hard or validated-surrogate endpoints.
• Accelerated approval. For drugs that address a serious condition and demonstrate effect on a surrogate 'reasonably likely to predict clinical benefit.' Approval comes with a requirement to complete confirmatory trials post-approval.
• Priority review. A faster review timeline (6 months instead of 10) for drugs that would represent significant advances.
• Breakthrough therapy designation. A status that grants intensive FDA guidance and rolling review for drugs with preliminary evidence of substantial improvement on clinically significant endpoints.
• Orphan drug designation. For drugs targeting rare diseases (<200,000 patients in the US). Provides tax credits, fee waivers, and seven years of market exclusivity.
• Emergency use authorization (EUA). For public health emergencies, granted under specific statutory criteria, distinct from full approval.

What approval certifies:

• The trial evidence supports a favorable benefit-risk assessment for the indicated population on the indicated regimen.
• The labeling describes the demonstrated effects, the known side effects, the populations studied, and any post-approval requirements.

What approval does not certify:

• That the drug works in populations not represented in trials (children, pregnant patients, very elderly, multiple comorbidities — often excluded from pivotal trials).
• That the drug is the best treatment for the indication (regulators do not generally do comparative-effectiveness assessment; that is left to payers, guidelines, and individual prescribers).
• That long-term outcomes match short-term effects.
• That the price is justified (the FDA's mandate excludes pricing considerations).

The 90% attrition rate from preclinical to approval is one of the most-cited numbers in pharmaceutical development. The structural breakdown:

• Roughly ~10,000 compounds are screened or designed in early discovery.
• ~250 of these enter preclinical testing.
• ~5 of those advance to phase I clinical trials.
• ~1 of those is approved.

The phase-by-phase numbers vary by therapeutic area but typical estimates for the post-IND probability of reaching approval:

• Phase I → approval: about 10–15%.
• Phase II → approval: about 30–35%.
• Phase III → approval: about 50–60%.
• NDA/BLA submission → approval: about 80–90%.

The failure modes concentrate in:

• Lack of efficacy. The drug does not produce a clinically meaningful effect at tolerated doses. The largest single category.
• Safety issues. Side effects unacceptable for the indication's benefit, or a previously-undetected toxicity emerges.
• Commercial-strategic. Sponsor decides not to pursue further development due to market analysis, competitor moves, or pipeline prioritization.
• Strategic regulatory issues. Manufacturing problems, clinical-hold issues, design problems in pivotal trials.

The cost of failures is built into the cost of the successes. A biotech firm that takes 10 candidates through phase I expects 1–2 approvals on average, and the prices of those approved products reflect the combined R&D costs of all the failures.

After approval, phase IV activities continue to refine what is known about the drug.

• Mandatory post-approval studies. Often required as a condition of approval, especially under accelerated approval. The sponsor must complete pre-specified trials to confirm the surrogate-based approval or to characterize specific safety signals.
• Pharmacovigilance. Spontaneous reporting systems (FDA's FAERS, EMA's EudraVigilance) collect adverse event reports from clinicians, patients, and sponsors. Signals from these systems can trigger label changes, boxed warnings, restricted distribution, or withdrawal.
• Real-world evidence (RWE). Analyses of electronic health records, claims data, and registries to characterize how a drug performs outside controlled trials. RWE has become increasingly important to regulators in some indications and supplements but does not replace randomized trial evidence.
• Indication expansion. Sponsors run additional trials to support new indications (the drug works in a different disease) or new populations (children, additional age groups, etc.).
• Comparative effectiveness. Studies comparing the new drug to existing alternatives — usually funded by payers, guideline bodies, or academic researchers rather than the sponsor.

Withdrawals from market post-approval do occur — Vioxx (cardiovascular safety, 2004), pemoline (hepatotoxicity), and others — usually triggered by accumulation of phase IV safety signals that outweigh the original benefit-risk assessment. The asymmetry of approvals vs withdrawals is structural: approval requires affirmative evidence of benefit-risk; withdrawal requires affirmative evidence of unacceptable risk. The bar for the second is higher than for the first, and the timeline for withdrawal can extend over years of debate and additional studies.

The final lesson moves from the trial-and-regulation framework to how to read the evidence: how to assess effect sizes, confidence intervals, and the difference between statistical significance and clinical importance.