The core idea
Efficacy compares outcome rates between study groups. It is not the percentage of recipients who receive a permanent guarantee against illness.
1. Identify the study and its question
Raches Ella and colleagues reported interim results of a randomised, double-blind, controlled phase 3 trial in The Lancet, published online on 11 November 2021, DOI 10.1016/S0140-6736(21)02000-6. The peer-reviewed paper studied BBV152, an inactivated whole-virus vaccine developed in India. It examined clinical outcomes rather than assuming that a laboratory immune response alone demonstrated prevention of disease.
The trial recruited adults at 25 Indian sites. Its primary efficacy question concerned first laboratory-confirmed symptomatic COVID-19 beginning at least 14 days after the second study dose. The per-protocol analysis included participants without detectable SARS-CoV-2 antibodies at enrolment who received both doses, had no major protocol deviations and had no confirmed infection before efficacy follow-up began. Symptoms, laboratory confirmation, timing and eligibility determine which events count. The study’s historical schedule is part of its design, not an instruction for current use.
Sources: Ella and colleagues: BBV152 phase 3 trial, Lancet 2021, full article archive ↗ · PubMed: BBV152 publication record and abstract ↗
2. Why randomise and mask?
If people choose their own study group, the groups may differ in occupation, exposure or health. Random assignment reduces systematic selection differences by allocating participants by chance. It does not make every characteristic exactly equal, especially in small samples. Researchers still describe baseline characteristics so readers can assess the comparison. Randomisation addresses allocation bias; it cannot repair every later problem.
Masking keeps relevant participants and investigators unaware of assignment, reducing differences in reporting or assessment caused by expectations. A control group experiences the same observation process and supplies a comparison. Consistent contact and outcome testing remain essential: if one group is tested much more often, recorded case rates may differ for reasons other than the intervention.
Sources: NIH: clinical trials, randomisation and masking ↗ · NHLBI: assessing the quality of controlled intervention studies ↗
3. Worked example: relative and absolute effects
Consider an illustrative trial with 1,000 people per group and equal follow-up. Suppose 80 control participants and 20 intervention participants develop the defined outcome. Their risks are 80/1,000 = 8% and 20/1,000 = 2%. Relative risk is intervention risk divided by control risk: 2% ÷ 8% = 0.25. Relative efficacy is (1 − 0.25) × 100 = 75%.
The absolute risk difference is 8% − 2% = 6 percentage points, equivalent to 60 fewer cases per 1,000 participants during this period. Both descriptions are correct and answer different questions. If risks were instead 0.8% and 0.2%, relative efficacy would still be 75%, but the absolute difference would be 0.6 percentage points. Exposure intensity and follow-up duration affect the absolute numbers.
Sources: CDC Principles of Epidemiology: measures of public-health impact ↗
4. Worked example: account for follow-up time
Not everyone is observed for exactly the same duration. Person-time adds time contributed while participants are at risk and under observation; for a first-event analysis, time after that event is excluded. Ten people each contributing half a year provide five person-years. An incidence rate divides new cases by person-time. Its units differ from risk: cases per person-year describe a rate, whereas a proportion of people affected over a stated interval describes risk.
In a second illustrative trial, 12 cases occur over 600 person-years in the intervention arm and 30 over 500 person-years in the control arm. Rates are 0.02 and 0.06 cases per person-year. Their ratio is one-third, giving about 66.7% relative efficacy. Simply comparing 12 with 30 would give 60% and ignore unequal observation time. A sound calculation follows the study's specified denominator.
Sources: Ella and colleagues: BBV152 phase 3 trial, Lancet 2021, full article archive ↗ · CDC Principles of Epidemiology: measures of public-health impact ↗
5. Read the actual result with its uncertainty
The BBV152 paper's primary analysis reported 24 cases among 8,471 vaccine recipients and 106 among 8,502 placebo recipients. It estimated efficacy at 77.8%, with a reported 95% confidence interval of 65.2–86.4%. Its method uses person-time incidence rates, so dividing the two raw case counts does not reproduce the published estimate. These numbers describe the paper's endpoint and historical observation period.
A confidence interval expresses uncertainty from the estimation procedure under its assumptions. Narrower intervals generally indicate greater statistical precision; they do not remove bias or guarantee future performance. In repeated comparable studies, a procedure producing 95% confidence intervals is designed to cover the true parameter about 95% of the time. An interval is not a list of protection percentages assigned to different individuals.
Use the observation time in the denominator
| Illustrative group | Cases | Person-years | Cases/person-year |
|---|---|---|---|
| Intervention | 12 | 600 | 0.02 |
| Control | 30 | 500 | 0.06 |
Sources: Ella and colleagues: BBV152 phase 3 trial, Lancet 2021, full article archive ↗ · PubMed: BBV152 publication record and abstract ↗ · NHLBI: assessing the quality of controlled intervention studies ↗
6. Follow the people and distinguish safety questions
A trial can have different analysis populations. The safety population includes people according to exposure and safety follow-up rules. A per-protocol efficacy population includes those meeting specified eligibility and follow-up conditions. Readers should trace exclusions rather than assume every randomised person appears in every calculation. Exclusions chosen after seeing outcomes can distort results; prespecified rules make the analysis more transparent.
An adverse event is an unfavourable occurrence after an intervention; timing alone does not establish causation. Researchers compare event patterns, severity, timing and background occurrence, with appropriate assessment. Even a large trial has limited ability to identify very rare events or effects appearing after its follow-up ends. Efficacy and safety therefore require related but distinct evidence and continuing investigation.
Sources: NIH: clinical trials, randomisation and masking ↗ · NHLBI: assessing the quality of controlled intervention studies ↗ · Ella and colleagues: BBV152 phase 3 trial, Lancet 2021, full article archive ↗
7. A responsible research conclusion
The achievement was the production of controlled clinical evidence through an Indian multicentre trial, not merely a favourable headline percentage. To interpret any such study, name the population, comparator, outcome, period and uncertainty. Changes in circulating pathogens, prior exposure or population composition can change later results. Historical efficacy should remain attached to its historical question.
A useful learner's conclusion explains both what was measured and why the comparison was credible. It also identifies unresolved questions without pretending the study answered them. Reading methods carefully is a practical skill: the denominator, observation window and case definition often explain more than the largest number in a summary.
Sources: Ella and colleagues: BBV152 phase 3 trial, Lancet 2021, full article archive ↗ · NHLBI: assessing the quality of controlled intervention studies ↗
PUT IT INTO PRACTICE
Practice: analyse a fictional trial summary
- Write a precise fictional endpoint and observation period. Keep them identical for both groups.
- For 15 cases among 500 intervention participants and 45 among 500 controls, calculate both risks, relative efficacy and absolute risk difference.
- Describe one bias randomisation reduces and one problem it cannot solve by itself.
- Write a four-sentence interpretation that names the endpoint, gives both effect measures and states one uncertainty. Use only fictional data, without collecting anyone's medical information.
Check your understanding
What are the practice results?
Risks are 3% and 9%. Relative efficacy is 1 − 3/9 = 66.7% approximately; the absolute difference is 6 percentage points.
Does 75% efficacy mean 75 out of 100 recipients can never become ill?
No. It is a relative reduction in a defined outcome compared with the control under the study conditions.
Why cannot 24 and 106 alone reproduce the BBV152 estimate?
The published calculation accounts for the groups' person-time. Raw case counts omit those denominators.
Would a narrow confidence interval eliminate measurement bias?
No. Precision concerns random uncertainty; a consistently biased measurement can still look precise.
Why state the endpoint instead of saying only 'effective'?
Preventing symptomatic illness, severe illness and infection are different outcomes. Evidence for one does not automatically quantify the others.
What makes this a historical methods lesson?
It analyses a specified 2021 publication and its trial design. It does not infer current recommendations, availability or protection against present variants.