BiostatisticsTier 1Medical Sciences concept

Clinical trial phases (I-IV) and their purposes

Core concept

Phase	n	Population	Primary question	Design
Pre-clinical	-	In vitro, animal	Toxicology, PK, mechanism	-
I	~20-100	Healthy volunteers (patients in oncology and for cytotoxic agents)	Safety, tolerability, PK/PD, maximum tolerated dose	Dose escalation (3+3, CRM); usually open-label, uncontrolled
II	~100-300	Target disease	Does it work at all? Which dose? Short-term safety	Randomised, placebo-controlled, dose-ranging; often a surrogate endpoint. IIa = proof of concept, IIb = dose-finding
III	~1,000-3,000+	Target disease, broad	Confirmatory efficacy vs standard of care; common adverse effects	Large RCT, clinical endpoints, powered for the primary outcome; the registration trial
IV	Population-scale	Real world	Rare and long-term harms, effectiveness, new indications, drug interactions	Post-marketing surveillance, registries, pharmacovigilance, cohort/case-control

The most examined point: rare adverse effects (~1 in 10,000) cannot be detected before marketing
- Rule of three - to have a 95% chance of seeing at least one event of frequency 1/n, you need ~3n patients
  - 1 in 10,000 -> ~30,000 exposures - an order of magnitude more than any phase III programme
- -> phase IV is where rare harms are found (rofecoxib, rosiglitazone, cerivastatin, troglitazone)
Phase 0 - microdosing, sub-therapeutic, human PK only, no therapeutic intent

Endpoints and how the phases differ

Phase II uses surrogate endpoints (tumour response rate, HbA1c, BP, LDL, biomarker); phase III should use clinical endpoints (death, MI, stroke, fracture, hospitalisation)
- The commonest reason a phase III fails after a positive phase II is that the surrogate did not translate - torcetrapib raised HDL and increased mortality
Non-inferiority vs superiority - a non-inferiority design needs a pre-specified margin justified against historical placebo-controlled data; analyse per-protocol AND intention-to-treat, because sloppiness biases toward non-inferiority
Attrition of the pipeline - only ~10% of agents entering phase I reach approval; oncology is worse

Design features the exam asks about

Adaptive designs - pre-planned modification of dose, sample size or arms at interim (seamless phase II/III, platform trials such as RECOVERY and REMAP-CAP)
Data Safety Monitoring Board with pre-specified stopping rules for harm, benefit or futility
- Trials stopped early for benefit systematically overestimate effect size
Registration before enrolment (ANZCTR, ClinicalTrials.gov) to prevent outcome-switching
Basket trial = one biomarker, many tumour types; umbrella trial = one tumour type, many biomarkers
Special populations - children, pregnancy, elderly and renal impairment are usually excluded from phase III, so evidence there is generated post-marketing

Regulatory pathway in Australia - TGA registers the drug (ARTG); PBAC then decides reimbursement on cost-effectiveness, so a TGA-approved drug may be unavailable on the PBS
- "Registered" and "subsidised" are different questions - a common source of confusion in prescribing
Accelerated/provisional approval is granted on a surrogate endpoint with a mandated confirmatory trial; withdrawal follows if the confirmatory trial fails
Adverse event reporting - report suspected reactions to the TGA (Blue Card / online); mandatory for sponsors, voluntary for clinicians
- Reporting is spontaneous and incomplete, so it detects signals, not incidence; a denominator is never available
Interpreting a new drug for a patient
- Was the trial population like this patient?
- Was the comparator the real standard of care, or placebo where an active drug exists?
- Clinical or surrogate endpoint? Composite endpoints driven by the softest component?
- Absolute risk reduction and NNT, not just relative risk reduction
- How long was follow-up relative to the duration of intended therapy?

3 more sections, plus exam facts

Premium unlocks every note across every specialty, and the full exam fact library behind it.