Vaccine Efficacy vs Effectiveness: How Each Is Measured

Vaccine efficacy and vaccine effectiveness are both estimates of relative reduction in a defined outcome. Efficacy usually comes from a randomized controlled trial conducted under specified conditions. Effectiveness usually comes from observational studies after a vaccine is used in routine practice. Neither term means a universal percentage of protection against every infection, symptom, hospitalization, transmission event, and death.

An “80 percent effective vaccine” statement is incomplete unless it names the outcome, comparison, population, circulating strain, dose schedule, time since vaccination, calendar period, and uncertainty. The same vaccine can have one estimate against laboratory-confirmed infection and a different estimate against hospitalization. Careful reading begins with the endpoint, not the percentage.

The basic efficacy calculation#

In a randomized vaccine trial, participants are assigned to vaccine or control. Investigators count a prespecified outcome after the immune-response window begins, and if 100 of 10,000 control participants and 20 of 10,000 vaccinated participants develop the endpoint, risks are 1 percent and 0.2 percent.

The risk ratio is 0.2 divided by 1.0, or 0.20. Vaccine efficacy is one minus that ratio, multiplied by 100, which equals 80 percent. The absolute risk reduction is 0.8 percentage points, and over that study period, about 125 people would need vaccination to prevent one endpoint under those simplified conditions, calculated as one divided by 0.008. The relative and absolute statements are both correct. The relative estimate tends to transport more consistently across baseline risks, while the absolute estimate tells how many events were prevented in that setting.

Why 80 percent does not mean 20 percent become ill#

Efficacy compares groups. It does not assign a protected or unprotected label to each person. In the example, 0.2 percent of vaccinated participants had the outcome, not 20 percent.

Nor does 80 percent mean you personally have exactly 80 percent protection. The estimate is an average causal effect in the trial population under the trial's definitions. Immune response varies with age, health, prior immunity, dose timing, and other factors. The outcome must also be named. An estimate against symptomatic disease cannot be handed to you as the same estimate against infection of any kind, or against death.

Randomization strengthens causal inference#

Random assignment tends to balance measured and unmeasured baseline factors between groups. Blinding can reduce differences in symptom reporting, care seeking, testing, and outcome assessment. A placebo control helps separate vaccine reactions from background symptoms.

Randomization does not make a trial flawless. Participants may differ from the population that later receives the vaccine. Cases can be missed, adherence can differ, and loss to follow-up can bias results. A trial conducted during one strain and one transmission period may not predict later performance.

The analysis population matters. Per-protocol analyses ask about participants who received the planned schedule without major deviations. Intention-to-treat analyses preserve assignment and can better represent the consequence of offering the strategy. Modified definitions often begin counting after a specified immune-response interval.

Endpoints create different efficacy estimates#

A vaccine trial can measure laboratory-confirmed infection, symptomatic disease, medically attended disease, hospitalization, severe disease, or death. Severe endpoints matter greatly but occur less often, requiring larger trials or longer follow-up.

Case definitions affect counts. Requiring a positive polymerase chain reaction test plus several symptoms creates a different endpoint from any positive test. Active weekly testing detects asymptomatic infections that symptom-triggered testing misses. An endpoint hierarchy should be prespecified, because selecting the most favorable definition after results are known inflates false-positive risk; what you want to see is the protocol, the statistical analysis plan, and the multiplicity control.

Effectiveness moves into routine practice#

After authorization or licensure, randomized placebo trials may no longer be ethical or feasible for every question. Public-health programs need estimates in older adults, pregnant people, people with immune compromise, children, and those with multiple conditions. They also need evidence against new strains, after different dose intervals, and over time.

Vaccine effectiveness studies use health records, surveillance networks, registries, cohorts, case-control designs, and test-negative designs. They reflect real storage, access, scheduling, and population diversity. That realism is valuable, but vaccination is no longer randomly assigned.

People who choose vaccination can differ from those who do not in health, behavior, occupation, prior infection, prevention practices, and care use. Analysis must measure and adjust for enough of those differences to make the comparison credible.

The test-negative design#

In a test-negative study, people seek care with a compatible illness and receive a laboratory test. Those positive for the target pathogen are cases; those testing negative are controls. Investigators compare prior vaccination between the groups, often using an adjusted odds ratio. Effectiveness is estimated as one minus that odds ratio.

Restricting both groups to people who sought care for similar symptoms helps reduce bias from different willingness to seek care or testing. The design has been used extensively for influenza and respiratory-virus vaccines.

It does not eliminate all bias. Vaccination can affect whether another infection causes symptoms severe enough for care. Test sensitivity varies with timing and specimen quality. Prior infection, access, and clinician testing decisions can differ. The method's assumptions must be examined for the pathogen, outcome, and setting.

Cohort and case-control approaches#

A cohort study follows vaccinated and unvaccinated people and compares outcome risks or rates. Large linked databases can produce rapid estimates and examine rare severe outcomes. Confounding, misclassified vaccination, incomplete capture of tests, and migration between health systems can distort results.

A traditional case-control study compares vaccination histories in people with the outcome and selected controls without it. Control selection is critical. Controls should represent the population that produced the cases, not a convenient group with a different chance of vaccination.

Self-controlled designs compare different time windows within the same person and are useful for some safety questions. Screening methods compare the vaccination proportion among cases with population coverage, but are sensitive to coverage accuracy and confounding. No study label guarantees validity; what you have to judge is the specific implementation.

Confounding can move the estimate in either direction#

Healthy-vaccinee bias occurs when vaccinated people are healthier or more prevention-oriented, making the vaccine appear more protective, and confounding by indication can work in the other direction when people at highest risk are prioritized for vaccination.

Frailty can produce complex patterns. A very frail person may be less likely to receive vaccination during acute decline yet more likely to be hospitalized; access, income, race and ethnicity as socially patterned variables, occupation, and geography can affect both vaccination and infection risk.

Regression adjustment and matching help only for measured factors represented adequately in the model. Negative-control outcomes, sensitivity analyses, and comparison across designs can reveal residual bias. A narrow confidence interval does not correct a biased estimate.

Prior infection is difficult to measure#

Prior infection can protect against subsequent disease and is related to vaccination decisions. Many infections are never tested or recorded. If unvaccinated people have more unmeasured prior immunity, effectiveness can be underestimated. If vaccinated people differ in prior infection or testing, the bias can reverse.

Antibody testing does not perfectly solve the problem because antibodies wane, vaccines generate some of the same markers, and assays differ. Health-record history is specific but incomplete; a good study defines how prior infection was measured, adjusts or stratifies where it can, and tells you whether the conclusion changes under plausible amounts of missing history.

Calendar time and variants matter#

Vaccination rollout often occurs over calendar time. Early recipients can have longer time since dose, different risk, and a different circulating variant than later recipients. Comparing them without close calendar adjustment can confuse waning with variant change or seasonal intensity.

Variant classification can come from sequencing, representative surveillance, or calendar dominance. Each has uncertainty. A study during a mixed-variant transition should not label every case as the dominant strain without support. Vaccine product, dose number, interval, and prior doses also matter. Pooling unlike schedules can obscure meaningful differences.

Waning requires a fair comparison#

Protection can decline as immune responses change and the pathogen evolves. A simple comparison of recently vaccinated and long-ago vaccinated groups can be biased because they were vaccinated in different periods and may differ in age, health, and behavior.

Strong waning analyses align calendar time, measure time since dose precisely, account for prior infection, and compare the same outcome, and they show confidence intervals rather than treating a series of point estimates as an exact downward curve. Waning against infection can occur faster than waning against hospitalization. Saying a vaccine “stopped working” because infection protection declined erases outcome severity.

Confidence intervals show statistical uncertainty#

An estimate of 60 percent with a 95 percent confidence interval from 40 to 74 percent is compatible with a range of effects under the model, and a small subgroup can produce a point estimate that looks dramatic but has a wide interval crossing zero.

Intervals do not capture every uncertainty. Misclassification, unmeasured confounding, outcome definition, and transport to another population can matter more than sampling error. Subgroup comparisons need interaction tests. One subgroup having a statistically significant estimate and another not significant does not prove the effects differ.

Relative effectiveness is not absolute effectiveness#

Some studies compare a new booster with people who completed older doses rather than with completely unvaccinated people, and the result is relative vaccine effectiveness, the added protection of the new strategy against the comparator.

If a report says “50 percent effectiveness” without naming the comparison, you may assume it compares vaccinated with unvaccinated. The methods should state reference group, prior doses, and time windows.

Incremental effectiveness can still answer a useful policy question: what benefit does another dose add now? It simply answers a different question from original trial efficacy.

Program impact includes coverage#

A highly efficacious vaccine has little population impact if few people receive it, delivery misses the highest-risk groups, or doses arrive after the epidemic peak. Conversely, moderate individual effectiveness can prevent many events when disease burden and coverage are high.

Impact compares observed disease burden with what would have occurred without the program. It depends on coverage, indirect protection, strain match, timing, health-system capacity, and population structure. So keep efficacy, effectiveness and impact as three separate words. One describes controlled relative performance, one real-world relative performance, and one the program's population consequence.

Safety is measured on a different framework#

An efficacy estimate does not describe safety. Trials compare solicited reactions, unsolicited events, serious adverse events, and prespecified conditions. Rare events may become detectable only after millions of doses.

Post-authorization systems use passive reports, active surveillance, record linkage, observed-versus-expected analyses, and formal epidemiologic studies. A report after vaccination is a safety signal, not automatic proof of causation.

Benefit-risk decisions compare outcome-specific vaccine benefits with confirmed or plausible harms by age, sex, health, dose, and current disease burden. A strong efficacy percentage cannot substitute for that analysis, and a safety signal cannot be evaluated without a valid comparator.

A checklist for any vaccine percentage#

Identify the vaccine, schedule, and comparator. Name the population, place, calendar period, strain, and time since dose. Find the exact endpoint and case definition. Determine whether the estimate is efficacy, effectiveness, relative effectiveness, or impact.

Then inspect absolute event risks, confidence intervals, adjustment variables, prior-infection handling, missing data, and subgroup evidence. Ask whether testing and health-care use differed. Check funding and protocol registration without assuming either determines validity.

Finally, translate the result into the decision you actually face. An estimate against mild infection in young adults cannot answer hospitalization benefit in older adults. Matching the evidence to the question is the central skill.

References#

1. WHO explanation of vaccine efficacy, effectiveness, and protection
2. CDC guide to measuring vaccine efficacy and effectiveness
3. CDC review of biases in vaccine-effectiveness studies
4. CDC vaccine-effectiveness terminology
5. WHO guidance on observational vaccine-effectiveness evaluation
6. CDC methods review of the inpatient test-negative design

For your own health, talk with your clinician.*