Skip to main content

Trial Phases (I/II/III)

Updated Jun 24, 2026 at 8:22pm

Research Draft High 1,182 words

A drug candidate's progression through Phase I, II, and III human clinical trials is the central value-creation (and value-destruction) engine of a development-stage biotech. Each phase is a sequential gate that answers a different question — is it safe? does it work? does it work better than what already exists, reliably? — and each gate carries its own probability of failure. For investors, the phases are not just a regulatory timeline; they are a re-pricing schedule. A clinical-stage biotech with no revenue is essentially a portfolio of options on these binary readouts, and the entire equity story rises or collapses on data events that the FDA's process defines in advance. The core tension: the value created by advancing a drug is enormous, but the base-rate probability of any single Phase I candidate reaching market is well under 10%, so most of these options expire worthless.

How the phases are structured

The FDA's clinical research framework (the "Step 3" of its drug-development process) defines three pre-approval phases, conducted under an active Investigational New Drug (IND) application:

  • Phase I — first-in-human safety and pharmacology. Typically 20–80 participants (healthy volunteers, or patients in oncology where dosing healthy people is unethical). Goal: establish a safe dose range, characterize pharmacokinetics/pharmacodynamics, and identify acute side effects. Duration commonly several months to ~1 year. Per FDA, roughly 70% of drugs advance past Phase I.
  • Phase II — preliminary efficacy and dose-finding in 100–300 patients with the target condition. This is where a drug must show a real biological/clinical signal. Per FDA, only about 33% advance to Phase III — historically the highest-attrition gate. Often split into Phase IIa (proof-of-concept) and IIb (dose-ranging).
  • Phase III — pivotal, confirmatory efficacy and safety in 300–3,000+ patients, usually randomized, controlled, and powered for statistical significance against placebo or standard of care. Duration commonly 1–4 years. Per FDA, roughly 25–30% advance. Success here generally supports an NDA (small molecule) or BLA (biologic) filing.

(A Phase IV post-marketing/safety-surveillance phase exists but occurs after approval and is outside the I/II/III development gate.)

How it's used in practice

Investors and analysts translate the phases into a risk-adjusted NPV (rNPV) for each asset: projected peak sales discounted both for time and for the cumulative probability of reaching market. The key input is the Probability of Success (PoS) or Likelihood of Approval (LOA) at the asset's current phase — which rises sharply as the drug advances. A typical chain implies that the same asset is worth far more the morning after a positive Phase III readout than the morning before, because the conditional probability of approval jumps from a low base to a high one once a successful pivotal trial is in hand (an IQVIA 2024 analysis put approval at ~92% for trials that meet their primary objective — though a meaningful minority of filings still fail on safety, manufacturing, or a demand for a second study).

This makes biotech an event-driven sector. The tradeable catalysts are the data readouts themselves: topline Phase II/III results, FDA meeting outcomes (End-of-Phase-2, pre-NDA), and the PDUFA decision date. Stocks routinely gap 30–80% (in either direction) on a single binary. Practitioners watch the trial design closely — endpoints, control arm, statistical powering, use of a selection biomarker to enrich the patient population — because a well-designed trial materially shifts the odds.

Adoption, debate & evidence

The I/II/III framework is the universal, FDA-codified standard — there is no competing taxonomy. What is empirically measured (and worth citing precisely) is the attrition:

  • The BIO / Informa / QLS study covering 2011–2020 (12,728 phase transitions) found an overall Likelihood of Approval from Phase I of 7.9% (figure independently corroborated). That report also cites a Phase I→II transition of ~52% and a Phase II→III transition of ~29% — confirming Phase II as the steepest single gate — though those exact per-phase transition values rest on that single primary report and should be treated as report-specific, not universal constants.
  • An earlier BIO study (2006–2015) put Phase-I LOA at 9.6%, illustrating that the figure drifts and is methodology-dependent.
  • The widely cited Wong, Siah & Lo (2019, Biostatistics) analysis of ~406,000 entries estimated overall success at ~13.8% across all indications — higher partly because of methodology and inclusion of vaccines.

The honest takeaway: the overall Phase-I-to-approval rate is somewhere in the high-single-digits to low-teens, and the exact number depends on the dataset, time window, and whether oncology and vaccines are included. Two robust, repeatedly confirmed findings: oncology has the lowest success (BIO 2006–2015 oncology LOA ~5.1%) and biomarker-selected programs roughly triple the odds (BIO: ~25.9% with a selection biomarker vs ~8.4% without). Modality also matters — biologics (~9.1% LOA) outperform small-molecule NMEs (~5.7%) in the 2011–2020 data.

Strengths & limitations

Where the framework helps: it gives a disciplined, sequential way to price risk and to know which catalyst matters. Treating each readout as a conditional probability gate (and demanding biomarker enrichment, a real control arm, and a registrational endpoint) is a defensible analytical anchor.

Where it misleads: base rates are population averages — applying "~30% Phase II→III" to a specific asset ignores indication, mechanism novelty, prior human data, and trial quality, which dominate the real odds. The #1 misuse is anchoring on the headline phase label while ignoring trial design: a "Phase III" with a soft surrogate endpoint, an underpowered sample, or no active comparator is far riskier than its label implies, and conversely a single positive Phase III does not guarantee approval (FDA may demand a second, or reject on manufacturing/safety). Phase labels also blur — adaptive, seamless Phase I/II and II/III designs, and accelerated-approval pathways using surrogate endpoints, break the clean three-gate model. Finally, survivorship and publication bias flatter success-rate datasets; quietly abandoned programs are underrepresented.

Sources

Dispute flagged: the overall Phase-I-to-approval rate ranges from ~7.9% (BIO 2011–2020) to ~13.8% (Wong/Siah/Lo) depending on dataset, time window, and inclusion of oncology/vaccines — treat any single figure as an estimate, not a constant.