This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record.
1. Trial at a Glance
FinIP was a phase 3, randomized, parallel-group, double-blind prevention trial evaluating GSK Biologicals' pneumococcal conjugate vaccine GSK1024850A against invasive disease. The registry reports 41188.0 enrolled participants across 7 arms and two primary endpoints focused on culture-confirmed invasive pneumococcal disease caused by any of the 10 pneumococcal vaccine serotypes.
Primary question
Does GSK1024850A reduce the person-year rate of culture-confirmed invasive pneumococcal disease due to any of the 10 pneumococcal vaccine serotypes in children beginning vaccination within 7 months of life?
Primary comparison
The registry reports separate primary analyses for children assigned to a 3-dose primary vaccination course and children assigned to a 2-dose primary vaccination course.
Follow-up
The primary endpoint follow-up extended from administration of the first vaccine dose through the end of the blinded invasive disease follow-up period, which lasted at least 30 months.
Statistical framework
The registry identifies linear regression in the posted method field, while the analysis description specifies negative binomial log-linear modeling with correction for dispersion group- and cluster-related effects for the primary comparisons.
2. Clinical Question
Population
The primary endpoint population was the Infant Vaccinated cohort. This included vaccinated subjects in the 10PN-PD-DIT-043 study and vaccinated subjects of the 10PN-PD-DIT-053 study who contributed to the effectiveness analysis, with first dose of study vaccine below 7 months of age and assignment to either a 3-dose or 2-dose primary vaccination course.
Intervention
The intervention was pneumococcal conjugate vaccine GSK1024850A. The trial also included GSK Biologicals' Engerix TM hepatitis B vaccine and Havrix TM hepatitis A vaccine as biological interventions.
Comparator
The primary comparisons were between the corresponding 10Pn vaccine group and control group. The registry identifies control groups using the study-specific labels 10Pn3+1-6W-6M/043+053 versus Ctrl-6W-6M/043+053 for the 3-dose analysis and 10Pn2+1-6W-6M/043+053 versus Ctrl-6W-6M/043+053 for the 2-dose analysis.
Primary statistical question
The primary effectiveness measure was vaccine effectiveness, defined in the registry as VE = 1 − RR, based on comparison of the numbers of culture-confirmed IPD cases and the corresponding person-year rates. The hypothesis was that vaccine-type IPD vaccine effectiveness was 0%, with superiority demonstrated when the two-sided p-value was below 5%.
3. Trial Design
The trial enrolled 41188.0 participants and used 7 arms. The registered design therefore created a multi-group randomized comparison rather than a simple two-arm trial, although the primary statistical questions were expressed as specific vaccine-versus-control comparisons within the infant vaccination schedules.
GSK1024850A
- 3-dose primary vaccination schedule in the primary endpoint analysis
- 2-dose primary vaccination schedule in the primary endpoint analysis
- 12–18 month catch-up schedule represented in secondary analyses
Hepatitis A / Hepatitis B vaccines
- Control groups were used in the corresponding randomized schedule comparisons
- Control vaccination also defined the comparison context for indirect-effect analyses in non-vaccinated persons living in study cluster areas in which study participants received vaccine.
4. Endpoints
Primary endpoint 1 — 3-dose primary vaccination course
Person Year Rate as Regards Subjects With Culture-confirmed IPD Due to Any of the 10 Pneumococcal Vaccine Serotypes. This endpoint applied to children starting vaccination within 7 months of life and assigned to a 3-dose primary vaccination course.
The 10 vaccine serotypes were 1, 4, 5, 6B, 7F, 9V, 14, 18C, 19F and 23F. The person-year rate was calculated as the number of subjects reported with culture-confirmed IPD divided by the sum of follow-up time expressed in years, per 1000, with a corresponding two-sided 95% confidence interval.
Primary endpoint 2 — 2-dose primary vaccination course
Person Year Rate as Regards Subjects With Culture-confirmed IPD Due to Any of the 10 Pneumococcal Vaccine Serotypes. This endpoint applied to children starting vaccination within 7 months of life and assigned to a 2-dose primary vaccination course.
The same 10 vaccine serotypes defined the endpoint. The registry specifies the person-year rate as the number of culture-confirmed IPD subjects divided by total follow-up time in years, per 1000, with a two-sided 95% confidence interval calculated using a profile log-likelihood ratio approach.
Secondary endpoints
The registry also reports person-year rates for culture-confirmed invasive disease, hospital-diagnosed pneumonia, and tympanostomy tube placement in children beginning vaccination at 12–18 months, together with indirect-effect analyses among non-vaccinated persons living in study cluster areas.
5. Statistical Methodology
Person-year rates
The central quantity in the FinIP analyses is the person-year rate. Conceptually, the rate is the number of observed events divided by the accumulated time that participants contribute to follow-up. This is important because participants can contribute different amounts of observation time.
where n is the number of subjects with the specified event and T is the sum of follow-up time expressed in years. The registry expresses rates per 1000 person-years for the posted analyses.
Profile log-likelihood confidence intervals
The primary endpoint definition specifies a two-sided 95% confidence interval calculated as a profile log-likelihood ratio confidence interval using a classical log-linear Poisson regression with strata. This approach is suited to rate-based outcomes because it models event counts in relation to accumulated exposure time.
Negative binomial regression
The primary analysis description states that the comparison of culture-confirmed IPD between vaccine and control groups used a negative binomial log-linear model with correction for dispersion group- and cluster-related effects. Several secondary analyses likewise use a negative binomial model with strata, while some analyses use a negative binomial model without strata.
Event counts accumulated over person-time can show more variability than a simple Poisson model allows. A negative binomial model introduces an additional dispersion component, making it useful when the observed count variability is greater than the Poisson assumption would permit. In FinIP, the registry specifically describes corrections related to dispersion and cluster effects in the primary comparison.
Linear regression in the registry method field
The registry's the statistical method identifies linear regression for the two primary analyses and for several secondary analyses. The detailed primary analysis description, however, specifies a negative binomial log-linear model for the vaccine-effectiveness comparison. The two descriptions therefore represent different levels of the registry's statistical-method reporting: the method field gives a broad normalized label, while the analysis description provides the more specific modeling information.
Risk ratio and vaccine effectiveness
The registry reports the primary effect measure as VE (1-RR). A vaccine effectiveness estimate of 100 therefore corresponds to a relative risk estimate of 0, while an estimate of 91.8 corresponds to a relative risk of 0.082 when applying the registry's stated definition. The clinically important point is that VE is a transformed relative measure, not an absolute difference in event rates.
6. Primary Results
Primary endpoint 1 — 3-dose primary vaccination course
Vaccine effectiveness against culture-confirmed IPD due to the 10 vaccine serotypes
95% CI 82.8–100 · two-sided p-value <0.0001
Comparison: 10Pn3+1-6W-6M/043+053 Group vs Ctrl-6W-6M/043+053 Group
The analysis was performed on the Infant Vaccinated cohort, including vaccinated subjects from the 10PN-PD-DIT-043 study and vaccinated subjects from the 10PN-PD-DIT-053 study who contributed to the effectiveness analysis, with first dose below 7 months of age and assignment to a 3-dose primary vaccination course.
What the estimate means: The reported VE of 100 is the registry's estimate from the transformation VE = 1 − RR. It indicates that the estimated relative risk corresponding to this analysis was 0 under that transformation.
What it does not mean: It does not mean that every vaccinated child is protected, nor does it mean that the probability of disease for an individual patient is literally zero. It is a group-level estimate from the specified randomized comparison and analysis model.
Precision: The two-sided 95% CI extends from 82.8 to 100. The interval describes uncertainty around the estimated effect under the statistical framework; it is not a range containing individual treatment responses.
The p-value: The p-value of <0.0001 addresses the registry's null hypothesis of vaccine-type IPD VE = 0%. It does not measure the size of the effect. Effect size is conveyed by the VE estimate and its confidence interval.
Model and clustering: The detailed analysis description uses a negative binomial log-linear model with correction for dispersion group- and cluster-related effects. Because the endpoint is based on person-time and events occurring within study clusters, the cluster structure is an important part of the analysis rather than a nuisance detail.
Primary endpoint 2 — 2-dose primary vaccination course
Vaccine effectiveness against culture-confirmed IPD due to the 10 vaccine serotypes
95% CI 58.3–99.6 · two-sided p-value = 0.0009
Comparison: 10Pn2+1-6W-6M/043+053 Group vs Ctrl-6W-6M/043+053 Group
This analysis used the Infant Vaccinated cohort for children whose first dose occurred below 7 months of age and who were assigned to a 2-dose primary vaccination course.
What the estimate means: The reported VE of 91.8 corresponds to a substantial estimated reduction in relative risk under the registry's definition VE = 1 − RR.
What it does not mean: The estimate is not an absolute reduction of 91.8 percentage points in the probability of IPD. It is a relative measure derived from the modeled comparison of event rates.
Precision: The 95% CI of 58.3–99.6 is considerably wider than the point estimate itself. The lower endpoint shows that the data are compatible with a materially smaller vaccine-effectiveness estimate than 91.8, while the upper endpoint approaches 100.
The p-value: The two-sided p-value of = 0.0009 is evidence against the registry's null hypothesis of vaccine-type IPD VE = 0%. It is not a measure of clinical magnitude and should be interpreted alongside the effect estimate and confidence interval.
Comparison with the 3-dose result: The two primary estimates should not be treated as if they were a simple head-to-head comparison of two vaccine doses. They arise from different assigned vaccination-course groups and represent distinct primary analyses.
7. Secondary Results: Culture-Confirmed Invasive Disease
The registry reports a series of secondary person-year rate analyses in the 12–18 month catch-up vaccination schedule. These analyses examine indirect effects among non-vaccinated persons living in study cluster areas. The registry states that the non-vaccinated persons were not enrolled in the study and that events occurring around 6 months or more after study start were compared with treated-group numbers.
2010 — Culture-confirmed IPD, any serotype
| Population | Method | PYAR | 95% CI |
|---|---|---|---|
| 10Pn12-18M/043+053-associated non-vaccinated population | Linear regression | 14.657 | 13.229–16.197 |
| Control-associated non-vaccinated population | Linear regression | 13.582 | 11.691–15.693 |
| 10Pn12-18M/043+053-associated non-vaccinated population | Negative binomial regression | 14.487 | 13.071–16.015 |
| Control-associated non-vaccinated population | Negative binomial regression | 13.740 | 11.841–15.857 |
2010 — Culture-confirmed IPD, vaccine serotype
| Population | Method | PYAR | 95% CI |
|---|---|---|---|
| 10Pn12-18M/043+053-associated non-vaccinated population | Linear regression | 8.452 | 7.376–9.639 |
| Control-associated non-vaccinated population | Linear regression | 7.603 | 6.206–9.221 |
| 10Pn12-18M/043+053-associated non-vaccinated population | Negative binomial regression | 7.813 | 6.782–8.955 |
| Control-associated non-vaccinated population | Negative binomial regression | 7.789 | 6.377–9.420 |
2010 — Culture-confirmed IPD, vaccine related
| Population | Method | PYAR | 95% CI |
|---|---|---|---|
| 10Pn12-18M/043+053-associated non-vaccinated population | Linear regression | 1.637 | 1.185–2.205 |
| Control-associated non-vaccinated population | Linear regression | 1.845 | 1.194–2.724 |
| 10Pn12-18M/043+053-associated non-vaccinated population | Negative binomial regression | 2.313 | 1.770–2.972 |
| Control-associated non-vaccinated population | Negative binomial regression | 1.984 | 1.307–2.886 |
2010 — Culture-confirmed IPD, non-vaccine and non-vaccine related
| Population | Method | PYAR | 95% CI |
|---|---|---|---|
| 10Pn12-18M/043+053-associated non-vaccinated population | Negative binomial regression | 3.997 | 3.269–4.839 |
| Control-associated non-vaccinated population | Negative binomial regression | 3.322 | 2.423–4.445 |
| 10Pn12-18M/043+053-associated non-vaccinated population | Negative binomial regression | 4.172 | 3.429–5.028 |
| Control-associated non-vaccinated population | Negative binomial regression | 3.968 | 2.981–5.177 |
The repeated estimates illustrate an important statistical feature of the FinIP registry results: the same broad outcome can be summarized under different serotype classifications and with different model specifications. Those estimates should therefore be interpreted as separate analyses rather than collapsed into one overall rate.
2011 — Culture-confirmed IPD, any serotype
| Population | Method | PYAR | 95% CI |
|---|---|---|---|
| 10Pn12-18M/043+053-associated non-vaccinated population | Negative binomial regression | 14.017 | 12.628–15.516 |
| Control-associated non-vaccinated population | Negative binomial regression | 15.066 | 13.079–17.269 |
2011 — Culture-confirmed IPD, vaccine serotype
| Population | Method | PYAR | 95% CI |
|---|---|---|---|
| 10Pn12-18M/043+053-associated non-vaccinated population | Negative binomial regression | 5.916 | 5.026–6.917 |
| Control-associated non-vaccinated population | Negative binomial regression | 8.557 | 7.077–10.255 |
2011 — Culture-confirmed IPD, vaccine related
| Population | Method | PYAR | 95% CI |
|---|---|---|---|
| 10Pn12-18M/043+053-associated non-vaccinated population | Negative binomial regression | 7.813 | 6.782–8.955 |
| Control-associated non-vaccinated population | Negative binomial regression | 7.789 | 6.377–9.420 |
2011 — Culture-confirmed IPD, non-vaccine and non-vaccine related
| Population | Method | PYAR | 95% CI |
|---|---|---|---|
| 10Pn12-18M/043+053-associated non-vaccinated population | Negative binomial regression | 5.011 | 4.196–5.939 |
| Control-associated non-vaccinated population | Negative binomial regression | 4.315 | 3.285–5.566 |
2012 — Culture-confirmed IPD, any serotype
| Population | Method | PYAR | 95% CI |
|---|---|---|---|
| 10Pn12-18M/043+053-associated non-vaccinated population | Linear regression | 2.977 | 2.357–3.710 |
| Control-associated non-vaccinated population | Linear regression | 2.048 | 1.361–2.960 |
2012 — Culture-confirmed IPD, vaccine serotype
| Population | Method | PYAR | 95% CI |
|---|---|---|---|
| 10Pn12-18M/043+053-associated non-vaccinated population | Negative binomial regression | 5.916 | 5.026–6.917 |
| Control-associated non-vaccinated population | Negative binomial regression | 8.557 | 7.077–10.255 |
2012 — Culture-confirmed IPD, vaccine related
| Population | Method | PYAR | 95% CI |
|---|---|---|---|
| 10Pn12-18M/043+053-associated non-vaccinated population | Negative binomial regression | 2.313 | 1.770–2.972 |
| Control-associated non-vaccinated population | Negative binomial regression | 1.984 | 1.307–2.886 |
2012 — Culture-confirmed IPD, non-vaccine and non-vaccine related
| Population | Method | PYAR | 95% CI |
|---|---|---|---|
| 10Pn12-18M/043+053-associated non-vaccinated population | Negative binomial regression | 4.172 | 3.429–5.028 |
| Control-associated non-vaccinated population | Negative binomial regression | 3.968 | 2.981–5.177 |
8. Secondary Results: Hospital-Diagnosed Pneumonia
The registry reports person-year rates for hospital-diagnosed pneumonia in the 12–18 month vaccination schedule. Follow-up for these analyses extended from administration of the first vaccine dose through December 31, 2011, with mean follow-up time of 27 months.
| Year | Associated population | PYAR | 95% CI |
|---|---|---|---|
| 2010 | 10Pn12-18M/043+053 | 9.218 | 9.103–9.335 |
| 2010 | Ctrl12-18M/043+053 | 9.212 | 9.052–9.375 |
| 2011 | 10Pn12-18M/043+053 | 10.500 | 10.378–10.624 |
| 2011 | Ctrl12-18M/043+053 | 10.429 | 10.259–10.601 |
| 2012 | 10Pn12-18M/043+053 | 10.118 | 9.997–10.239 |
| 2012 | Ctrl12-18M/043+053 | 9.921 | 9.755–10.088 |
All six posted analyses used a negative binomial model with strata. The registry's indirect-effect description again identifies the comparison as one involving non-vaccinated persons living in study cluster areas rather than a conventional individual-level randomized comparison.
The hospital-diagnosed pneumonia estimates are person-year rates, not risks over a fixed individual follow-up interval. Consequently, differences between rates should be interpreted in the context of accumulated observation time and the modeling approach. The confidence intervals quantify uncertainty around each estimated rate; they do not establish causality independently of the underlying cluster-level design.
9. Secondary Results: Tympanostomy Tube Placement
The registry reports person-year rates for prevention of all tympanostomy tube placements among children starting vaccination in the 12–18 month schedule, again including indirect effects on the unvaccinated population.
| Year | Associated population | PYAR | 95% CI |
|---|---|---|---|
| 2009 | 10Pn12-18M/043+053 | 22.624 | 22.020–23.240 |
| 2009 | Ctrl12-18M/043+053 | 22.747 | 21.912–23.606 |
| 2010 | 10Pn12-18M/043+053 | 24.503 | 23.852–25.166 |
| 2010 | Ctrl12-18M/043+053 | 26.236 | 25.308–27.189 |
| 2011 | 10Pn12-18M/043+053 | 27.502 | 26.810–28.207 |
| 2011 | Ctrl12-18M/043+053 | 26.100 | 25.171–27.055 |
| 2012 | 10Pn12-18M/043+053 | 28.661 | 27.958–29.377 |
| 2012 | Ctrl12-18M/043+053 | 29.835 | 28.843–30.850 |
These analyses used negative binomial regression with strata. The endpoint is a count/rate outcome, making the negative binomial framework conceptually appropriate when variability in event counts exceeds what a simple Poisson model would accommodate.
The estimates should be read as rates accumulated over person-time, not as percentages of children undergoing tympanostomy tube placement. A difference between two PYAR estimates therefore reflects a difference in modeled event occurrence per unit of accumulated follow-up, not necessarily a corresponding difference in individual-level probability.
10. Safety Results
The registry reports serious adverse events by randomized arm as affected participants divided by participants at risk.
| Study group | Serious adverse events | At risk |
|---|---|---|
| 10Pn3+1-6W-6M/043 Group | 6 | 8427 |
| 10Pn2+1-6W-6M/043 Group | 7 | 9112 |
| Ctrl-6W-6M/043 Group | 8 | 8872 |
| 10Pn7-11M/043 Group | 3 | 3689 |
| Ctrl7-11M/043 Group | 2 | 1812 |
| 10Pn12-18M/043 Group | 2 | 6249 |
| Ctrl12-18M/043 Group | 2 | 3020 |
These are descriptive safety counts by arm. The registry does not provide a comparative inferential analysis for these serious adverse-event counts in the posted information summarized here. Accordingly, the counts should not be interpreted as adjusted risk ratios or formal evidence of a between-group safety difference.
11. Statistical Methods Explained
Why use person-year rates?
Participants can contribute different amounts of follow-up. A person-year rate accounts for this unequal observation time by placing the number of events over total accumulated follow-up. This is especially useful for relatively uncommon invasive disease events, where the amount of time under observation is an important component of the information available for analysis.
Why use a negative binomial model?
A Poisson model assumes that the conditional variance of the event count equals its mean. The negative binomial model relaxes this restriction by allowing additional dispersion. FinIP explicitly describes negative binomial modeling for the primary comparison and many secondary rate analyses, including versions with and without strata.
What does VE = 1 − RR mean?
Vaccine effectiveness is a transformed relative-risk measure. If RR represents the relative risk in the vaccine group compared with the control group, then VE = 1 − RR expresses the corresponding proportional reduction in relative risk. It is therefore a relative measure, not an absolute risk difference.
Why does the confidence interval matter?
A point estimate alone does not describe statistical uncertainty. The 95% confidence interval gives a range of effect estimates compatible with the specified model and sampling framework at the stated confidence level. For the two primary analyses, the intervals are 82.8–100 and 58.3–99.6, respectively, showing that the second primary estimate is less precise.
Why is cluster structure important?
The primary analysis description explicitly refers to correction for dispersion group- and cluster-related effects. Secondary indirect-effect analyses also compare non-vaccinated populations living in study cluster areas. Ignoring the cluster structure could understate or otherwise misrepresent uncertainty because observations associated with the same cluster are not necessarily statistically independent.
Why are the indirect-effect analyses different from the primary analyses?
The primary analyses compare randomized vaccine and control groups among vaccinated participants. The indirect-effect analyses concern non-vaccinated persons who were not enrolled in the trial but lived in areas where participants received vaccine or control vaccine. The causal question therefore changes from direct randomized vaccine effectiveness to a population-level association involving the surrounding vaccinated study clusters.
12. Primary Analysis Logic
The registry states that vaccine effectiveness was demonstrated if the two-sided p-value for this null hypothesis was lower than 5%.
The primary endpoint was constructed around culture-confirmed IPD due to any of the 10 pneumococcal vaccine serotypes. This endpoint is particularly well suited to a rate-based analysis because invasive disease events are uncommon relative to the large number of participants and because follow-up time differs among participants.
The analysis combines three related statistical ideas:
- Event counting: identify subjects with culture-confirmed IPD meeting the serotype definition.
- Exposure time: accumulate follow-up time to form a person-year denominator.
- Relative comparison: compare vaccine and control rates and transform the resulting relative risk into vaccine effectiveness.
13. Confidence Intervals and P-values
VE 100
The two-sided 95% CI is 82.8–100. The corresponding p-value is <0.0001.
VE 91.8
The two-sided 95% CI is 58.3–99.6. The corresponding p-value is = 0.0009.
The two primary p-values address the same broad null hypothesis—zero vaccine effectiveness—but apply to different assigned primary vaccination courses. The p-value tells us how incompatible the observed result is with the specified null hypothesis under the model; it does not tell us how large or clinically important the vaccine effect is.
The confidence intervals provide the complementary information. They show the precision of the effect estimate and make it possible to distinguish a highly precise estimate from an estimate that is statistically significant but comparatively uncertain.
14. Limitations
- Registry method labeling: the registry's the statistical method identifies linear regression for the primary analyses, while the detailed analysis description specifies negative binomial log-linear modeling with dispersion and cluster-related correction. Readers should therefore rely on the detailed method description when interpreting the primary model.
- Clustered design: the primary analysis explicitly accounts for group- and cluster-related effects. This makes the cluster structure integral to interpretation of the confidence intervals.
- Indirect-effect interpretation: analyses in non-vaccinated persons are not direct randomized individual-level treatment comparisons. They describe populations living in study cluster areas and therefore answer a different question from the primary vaccine-effectiveness analyses.
- Multiple analyses: the registry contains 33 posted outcome measures and 56 posted statistical analyses. The two primary analyses are clearly designated, but many secondary analyses are also reported. Their p-value and confidence-interval interpretations should not automatically be treated as if they were part of a single confirmatory hypothesis-testing family.
- Heterogeneous endpoint definitions: the secondary analyses distinguish any serotype, vaccine serotype, vaccine-related disease, and non-vaccine/non-vaccine-related disease. These are distinct endpoints and should not be pooled without an explicitly defined statistical framework.
- Rate interpretation: PYAR is a rate, not a cumulative incidence proportion. It depends on accumulated follow-up time and should not be converted into an individual-level probability without additional assumptions.
- Safety interpretation: serious adverse-event counts are presented descriptively by arm in the registry information. A formal comparative safety conclusion requires an appropriate statistical analysis of those data.
15. Why This Trial Matters Statistically
FinIP is a useful teaching case because it demonstrates how randomized vaccine trials can move beyond simple event proportions and use person-time, cluster-aware count models, and indirect-effect analyses to address several related epidemiologic questions.
| Concept | How it appears in FinIP |
|---|---|
| Randomization | Randomized allocation with a parallel-group design |
| Blinding | Double masking |
| Person-time | Primary and secondary outcomes expressed as person-year rates |
| Relative risk | Primary effect measure expressed through VE = 1 − RR |
| Negative binomial regression | Used for primary detailed analysis and numerous secondary rate analyses |
| Dispersion | Primary analysis includes correction for dispersion group- and cluster-related effects |
| Stratification | Secondary negative binomial analyses include models with strata |
| Confidence intervals | Two-sided 95% intervals accompany primary and secondary rate estimates |
| Hypothesis testing | Primary analyses test the null hypothesis of 0% vaccine-type IPD VE |
| Indirect effects | Non-vaccinated populations living in study cluster areas are analyzed separately |
| Safety analysis | Serious adverse events are reported by arm as affected/at-risk counts |
16. Reading the Primary Results Correctly
The two primary analyses compare vaccine and control groups within children beginning vaccination below 7 months of age. Their effect measure is vaccine effectiveness, defined as one minus the relative risk.
The reported VE estimates are relative measures. They do not tell the reader the absolute number of IPD cases prevented per 1000 children without also considering the underlying event rate and follow-up time.
The 3-dose estimate has a 95% CI from 82.8 to 100, while the 2-dose estimate has a 95% CI from 58.3 to 99.6. The latter therefore carries more uncertainty around its point estimate.
Both primary p-values are below the 5% threshold specified in the registry analysis description. The p-values support rejection of the stated null hypothesis in those analyses, while the confidence intervals describe the corresponding uncertainty in effect magnitude.
17. Direct and Indirect Effects Are Different Estimands
One of the most important statistical distinctions in FinIP is the separation between direct randomized vaccine effectiveness and indirect effects.
| Analysis type | Population | Question | Primary statistical object |
|---|---|---|---|
| Direct primary analysis | Vaccinated participants | How does vaccine assignment affect culture-confirmed IPD? | VE = 1 − RR |
| Indirect-effect analysis | Non-vaccinated persons living in study cluster areas in which study participants received vaccine. | How does living in a vaccine-exposed cluster relate to disease rates among non-vaccinated persons? | PYAR |
These estimands should not be merged. The direct analysis is anchored by randomization of vaccinated participants. The indirect analysis uses the geographic and cluster structure created by the vaccination program and examines persons who were not enrolled in the trial.
This distinction is more than terminology. It changes the interpretation of the comparison, the population to which the estimate applies, and the assumptions required for causal interpretation.
18. Related Tutorials
Learn more about the methods used in this trial:
19. Related Calculators
20. Sources
- ClinicalTrials.gov: FinIP — NCT00861380.
- PubMed: PubMed record for PMID 34986178.
- PubMed: PubMed record for PMID 27125273.
Continue with Clinical Biostats
Explore the statistical methods behind randomized trials, rate outcomes, regression models, confidence intervals, and comparative effect measures.
21. Record Summary
FinIP illustrates a statistical framework built around a relatively uncommon infectious-disease endpoint observed over substantial person-time. The two primary analyses use vaccine effectiveness defined as 1 minus relative risk, with two-sided 95% confidence intervals and hypothesis testing against zero vaccine effectiveness. The detailed primary analysis description uses a negative binomial log-linear model with correction for dispersion group- and cluster-related effects.
The secondary analyses broaden the statistical question. They examine indirect effects among non-vaccinated persons living in study cluster areas and report person-year rates for culture-confirmed invasive disease, hospital-diagnosed pneumonia, and tympanostomy tube placement. The use of negative binomial regression, stratification, and person-time denominators demonstrates why the choice of statistical model should follow the structure of the endpoint and the way observation time is accumulated.