A clinical trial is a research study in which human subjects are prospectively assigned to one or more interventions (which may include placebo or other controls) to evaluate the effects of those interventions on health-related biomedical or behavioural outcomes. The International Committee on Harmonisation (ICH) defines it as "any investigation in human subjects intended to discover or verify the clinical, pharmacological and/or other pharmacodynamic effects of an investigational product, and/or to identify any adverse reactions, and/or to study absorption, distribution, metabolism, and excretion, with the object of ascertaining the safety and/or efficacy of the product."
Clinical trials are necessary for several fundamental reasons:
In the late 1950s, thalidomide was marketed in Europe as a safe sedative for pregnant women without adequate clinical trial evidence. It was prescribed to over 20,000 patients before it was discovered that the drug caused severe birth defects (phocomelia — limb malformations) in approximately 10,000 babies. This catastrophe, one of the worst drug disasters in history, directly led to the 1962 Kefauver–Harris Amendment in the US, which mandated that all new drugs must demonstrate both safety and efficacy through rigorous clinical trials before approval. The thalidomide tragedy remains the most powerful illustration of why clinical trials are not merely desirable but absolutely essential — without them, the consequences can be devastating and irreversible.
The SOLVD trial (Studies of Left Ventricular Dysfunction) investigated whether enalapril (an ACE inhibitor) could reduce mortality in patients with heart failure. In the randomised, double-blind, placebo-controlled trial, 2,569 patients were assigned to enalapril or placebo. After an average follow-up of 41 months, the mortality rate was 35.2% in the placebo group and 30.2% in the enalapril group — a statistically significant 16% relative risk reduction (p = 0.0036). Without the clinical trial, this benefit could not have been reliably established because heart failure patients on ACE inhibitors may differ from those not on ACE inhibitors in many ways (severity, comorbidities, treatment preferences) — all of which confound observational comparisons. The randomised design eliminates these confounders, providing trustworthy evidence.
Ethics in clinical trials is founded on the principle that the rights, safety, and well-being of individual trial subjects must take precedence over all other interests, including scientific and societal interests. No matter how important a research question may be, it cannot be pursued at the expense of a participant's welfare or informed consent.
From 1932 to 1972, the US Public Health Service conducted a study of 399 African American men with syphilis in Tuskegee, Alabama. The study observed the natural progression of untreated syphilis without informing the participants of their diagnosis or providing treatment — even after penicillin became the standard cure in 1947. The men were told they were receiving free healthcare when in fact they were being deliberately denied effective treatment. This study violated every ethical principle: no informed consent, no beneficence, racial targeting (justice), and continued long after a cure was available. It led to the Belmont Report (1979) and the establishment of IRBs to prevent such abuses. The Tuskegee case remains the most infamous example of research ethics failure in history.
A Phase III trial compares a new immunotherapy drug with standard chemotherapy for advanced lung cancer. The informed consent form includes: (a) Purpose: "To determine whether Drug X improves overall survival compared to standard chemotherapy." (b) Procedures: "You will receive either Drug X or standard chemotherapy, assigned randomly. You will not know which treatment you receive (double-blind). You will visit the clinic every 3 weeks for 2 years." (c) Risks: "Drug X may cause severe immune-related side effects including colitis, hepatitis, and pneumonitis, which can be life-threatening. Standard chemotherapy may cause hair loss, nausea, and increased infection risk." (d) Benefits: "Drug X may shrink your tumour and extend your life, but this is not guaranteed." (e) Alternatives: "You may choose not to participate and receive standard treatment from your doctor." (f) Right to withdraw: "You may leave the trial at any time without affecting your future medical care." The form is written at an 8th-grade reading level and is available in the participant's language. An independent witness is present during consent if the participant is illiterate.
Bias is a systematic deviation of the estimated treatment effect from the true effect, caused by flaws in the design, conduct, or analysis of a study. It does not decrease with increasing sample size. Random error is the deviation due to chance variability — it is inherent in any sampling process and does decrease with increasing sample size. The total error in a clinical study is: \(\text{Total Error} = \text{Bias} + \text{Random Error}\).
Selection bias: Systematic differences between the groups being compared due to non-random allocation. Example: If a physician assigns sicker patients to the experimental treatment and healthier patients to the control, the experimental treatment will appear less effective than it truly is. Prevention: Randomisation (especially blocked or stratified randomisation).
Information (measurement) bias: Systematic errors in measuring outcomes or exposures. Subtypes include:
Confounding bias: A third variable associated with both the treatment and the outcome distorts the apparent treatment effect. Example: If older patients are more likely to receive Treatment A and also more likely to have poor outcomes, age confounds the treatment–outcome relationship. Prevention: Randomisation, stratification, and statistical adjustment.
Attrition bias: Systematic differences between groups due to differential loss to follow-up. If more patients drop out of the treatment group due to side effects, and these patients would have had poor outcomes, the treatment appears more effective than it truly is. Prevention: Intention-to-treat (ITT) analysis.
Publication bias: Studies with positive results are more likely to be published than those with negative results, distorting the published literature. Prevention: Clinical trial registration (e.g., ClinicalTrials.gov) before the trial begins.
For a treatment effect estimate \(\hat{\delta}\) (e.g., difference in means or proportions):
\[ \text{Random Error} \approx SE(\hat{\delta}) = \frac{\sigma}{\sqrt{n}} \]
The random error decreases as \(\frac{1}{\sqrt{n}}\) — to halve the random error, the sample size must be quadrupled. This is why sample size determination is critical in clinical trials: the sample must be large enough to detect a clinically meaningful treatment effect with adequate statistical power, but not so large that it exposes more participants than necessary to experimental treatments.
A non-randomised study compares the effect of Drug A (given to patients at Hospital X) versus Drug B (given to patients at Hospital Y) on survival after a heart attack. The 1-year survival rate is 85% for Drug A and 72% for Drug B, suggesting Drug A is superior. However, Hospital X is a specialised cardiac centre that receives younger, healthier patients, while Hospital Y is a general hospital that receives older patients with more comorbidities. After adjusting for age, sex, and comorbidity score using logistic regression, the adjusted survival difference shrinks from 13% to 2% and is no longer statistically significant (p = 0.38). The apparent superiority of Drug A was entirely due to selection bias — the groups were not comparable because the treatment assignment was confounded with hospital and patient characteristics. A properly randomised multi-centre trial would eliminate this bias.
A trial evaluates a new topical cream for eczema. In an open-label (unblinded) design, the investigator assesses improvement on a 0–4 scale. The mean improvement score is 2.8 for the treatment group and 2.1 for the control group (p = 0.04). However, because the investigator knows which patients received the new cream, they may unconsciously rate them more favourably (observer bias). When the same trial is repeated with double-blinding (neither the patient nor the investigator knows which cream is active, and the control cream has an identical appearance and smell), the mean improvement scores are 2.4 for treatment and 2.2 for control (p = 0.32). The apparent treatment effect in the unblinded trial was largely due to observer bias, not a true drug effect. This demonstrates why blinding is essential for subjective outcome measures.
The conduct of a clinical trial follows a rigorous, pre-specified protocol that governs every aspect from participant recruitment to data analysis. The key principle is that the protocol must be finalised before the trial begins, and any changes during the trial must be documented and justified as protocol amendments.
A pharmaceutical company develops a protocol for a Phase III trial comparing a new antihypertensive (Drug X) with the standard drug (lisinopril). Key protocol elements include:
For a Phase II trial of a new antidepressant:
Inclusion criteria: (1) Age 18–65; (2) DSM-5 diagnosis of major depressive disorder; (3) Hamilton Depression Rating Scale (HAM-D) score ≥ 20 (moderate-to-severe depression); (4) Able to give informed consent; (5) Fluent in the study language.
Exclusion criteria: (1) Current use of other antidepressants; (2) History of bipolar disorder or psychosis; (3) Active suicidal ideation; (4) Substance abuse in the past 6 months; (5) Pregnant or breastfeeding women; (6) Liver or kidney impairment; (7) Known hypersensitivity to the study drug class.
Well-defined inclusion and exclusion criteria ensure that the study population is homogeneous enough to detect a treatment effect, but representative enough that the results are generalisable. Too broad criteria introduce heterogeneity; too narrow criteria limit generalisability and recruitment.
Clinical trials are conducted in a sequential series of phases, each with a distinct purpose. A new drug must successfully pass through each phase before proceeding to the next. The entire process from pre-clinical testing to market approval typically takes 10–15 years and costs approximately $1–2 billion.
| Feature | Phase I | Phase II | Phase III | Phase IV |
|---|---|---|---|---|
| Purpose | Safety & dose finding | Efficacy & safety | Confirm efficacy & safety | Post-marketing surveillance |
| Population | Healthy volunteers (20–80) | Patients (100–300) | Patients (1000–5000) | General population |
| Design | Open / dose-escalation | Randomised, controlled | Randomised, double-blind | Observational |
| Duration | Months | Months–2 years | 1–4 years | Ongoing |
| Success Rate | ~70% | ~33% | ~25–30% | N/A |
| Primary Endpoint | MTD, PK/PD | Response rate | Clinical outcomes | Long-term safety |
Phase I trials are the first stage of testing in humans. The primary objective is to assess safety and determine the Maximum Tolerated Dose (MTD). They typically involve 20–80 healthy volunteers (or patients, for oncology drugs). Common designs include:
Phase II trials evaluate the efficacy of the drug at the MTD determined in Phase I, while continuing to assess safety. They involve 100–300 patients with the target disease. Phase II can be single-arm (comparing the response rate to a historical control) or randomised (comparing two or more doses or regimens). A negative Phase II trial (insufficient efficacy) typically terminates the drug's development.
Phase III trials are the pivotal, large-scale, randomised, double-blind, controlled studies that provide the definitive evidence of efficacy and safety required for regulatory approval. They involve 1,000–5,000 patients and are the most expensive and time-consuming phase. Phase III trials must be conducted at multiple centres (often internationally) and follow patients for months to years.
Phase IV trials are conducted after the drug has been approved and marketed. They monitor long-term safety in the general population, detect rare adverse effects that were not apparent in the smaller Phase III trials, and explore new indications. They are typically observational (not randomised) and may involve tens of thousands of patients.
A Phase I trial tests 5 dose levels of a new chemotherapy drug. The 3+3 escalation proceeds as follows:
The Phase II trial will test the drug at 100 mg to evaluate efficacy in a larger patient population.
The Pfizer-BioNTech COVID-19 vaccine Phase III trial enrolled 43,548 participants at 152 sites worldwide. Participants were randomised 1:1 to receive the BNT162b2 vaccine or placebo (saline injection), in a double-blind design. Two doses were administered 21 days apart. The primary endpoint was confirmed COVID-19 with onset ≥7 days after the second dose. Results: 8 COVID-19 cases in the vaccine group vs. 162 in the placebo group out of 36,523 evaluable participants. Vaccine efficacy:
\[ VE = 1 - RR = 1 - \frac{8/18198}{162/18325} = 1 - \frac{0.000439}{0.00884} = 1 - 0.0497 = 95.03\% \]
with a 95% CI of [90.3%, 97.6%]. This far exceeded the pre-specified success criterion of VE > 30%. The trial was unblinded early by the DSMB due to overwhelming efficacy, and the vaccine received Emergency Use Authorization from the FDA within weeks.
A multi-center trial is a clinical trial conducted simultaneously at several clinical sites (hospitals, research centres) following a common protocol. Multi-center trials are essential for Phase III studies because no single centre can recruit enough patients within a reasonable timeframe, and they enhance the generalisability of results across diverse populations and practice settings.
The ISIS-2 (Second International Study of Infarct Survival) was a randomised, placebo-controlled trial of streptokinase and aspirin in patients with suspected acute myocardial infarction. It enrolled 17,187 patients at 417 hospitals in 16 countries over 3 years. The trial demonstrated that both streptokinase and aspirin independently reduced mortality, and the combination was even more effective. The multi-center design was essential: (a) no single hospital could recruit 17,000 heart attack patients; (b) the results were generalisable across countries, healthcare systems, and patient populations; (c) the large sample size provided definitive evidence with narrow confidence intervals. The ISIS-2 trial fundamentally changed the standard of care for heart attacks worldwide.
A multi-center trial of a new asthma drug involves 10 centres. Centre A (a large urban hospital) contributes 200 patients and shows a strong treatment effect (odds ratio = 0.55). Centre J (a small rural clinic) contributes only 15 patients and shows a slight negative effect (OR = 1.10). If the centre effect is ignored and data are simply pooled, Centre A dominates the analysis due to its large sample size, potentially masking heterogeneity. The proper approach is to: (a) test for treatment-by-centre interaction using a Breslow-Day test or a random effects model; (b) if significant interaction exists, report treatment effects separately by centre or use a random effects meta-analysis; (c) if no significant interaction, pool the data with centre as a stratification factor. In this example, the interaction test is non-significant (p = 0.28), so the pooled OR = 0.68 with 95% CI [0.52, 0.89] is reported, adjusted for centre as a fixed effect.
Clinical data management (CDM) is the process of collecting, cleaning, and managing data generated from clinical trials. The goal is to produce a high-quality, statistically sound database that supports reliable analysis. Poor data management can introduce errors that undermine the validity of the entire trial.
Before data collection begins, every variable must be precisely defined:
A Case Report Form (CRF) is a printed or electronic document used to record all protocol-required data for each trial participant. It is the primary data collection instrument in a clinical trial.
Key design principles for CRFs:
The clinical trial database must be designed before data collection begins. It should:
ICH-GCP requires that clinical trial data be collected, handled, and stored in a way that ensures accuracy, completeness, and traceability. Key requirements include:
A CRF for a Phase III diabetes trial includes the following pages:
Each variable has a defined valid range: e.g., HbA1c must be between 4.0% and 15.0%, BMI between 15 and 60, age between 18 and 80. The eCRF will not accept out-of-range values without a documented override reason.
During data review, the data manager identifies the following discrepancies in a hypertension trial:
All queries and resolutions are documented. Once all queries are resolved, the database is locked and the analysis dataset is generated.