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
JiVitA-3 was a randomized, parallel-group, quadruple-masked phase 3 prevention trial evaluating antenatal iron-folic acid supplementation versus multiple micronutrient supplementation. The registered primary endpoint was infant mortality through 6 months of age, defined as infant mortality to age 6 months, or 180 days from birth.
| Feature | JiVitA-3 |
|---|---|
| Phase | Phase 3 |
| Therapeutic area | Endocrinology & Metabolism |
| Additional therapeutic area | Women's Health & Pregnancy |
| Conditions | Infant Mortality; Preterm Birth; Low Birth Weight; Neonatal Mortality; Perinatal Mortality |
| Design | Randomized, parallel |
| Masking | Quadruple |
| Primary purpose | Prevention |
| Enrollment | 44567 |
| ClinicalTrials.gov | NCT00860470 |
| Lead sponsor | Johns Hopkins Bloomberg School of Public Health |
2. Clinical Question
The central question was whether antenatal multiple micronutrient supplementation differed from iron and folic acid supplementation with respect to infant mortality through 6 months of age. The primary efficacy comparison reported in the registry is expressed as the risk ratio for Iron and Folate vs Multiple Micronutrient, with the iron-folic acid supplemented group as the referent.
Population
Women recruited into the randomized trial, recognizing that not all women recruited had pregnancies yielding live births. The primary analysis represents all live births by arm.
Intervention
Multiple micronutrient dietary supplementation.
Comparator
Iron (27 mg) plus folic acid (600 ug) dietary supplementation.
Primary question
How does infant mortality through 6 months of age compare between the iron-folic acid group and the multiple micronutrient group?
3. Trial Design
Iron and folic acid
- Iron (27 mg)
- Folic acid (600 ug)
- Dietary supplement
Multiple micronutrient
- Multiple micronutrient
- Dietary supplement
The quadruple-masked, randomized structure is important statistically because randomization establishes the basis for a comparative treatment effect, while masking is intended to reduce the influence of knowledge of assignment on trial conduct and outcome assessment. The parallel design means participants remain associated with their randomized arm rather than crossing between interventions as part of the study design.
4. Endpoints
| Endpoint | Definition / time frame | Type |
|---|---|---|
| Infant Mortality Through 6 mo of Age | Infant Mortality to Age 6 months (180 days from birth); 6-months post-birth | Primary, binary |
| Neonatal Mortality | 1 month post-birth | Secondary, binary |
| Post-neonatal Mortality | 1-6 months post-birth | Secondary, binary |
| Still Birth Rates | 24 weeks gestation to delivery | Secondary, binary |
| Preterm Birth | Up to 37 weeks of gestation | Secondary, binary |
| Extremely Pre-term | Up to 28 weeks of gestation | Secondary, binary |
| Very Pre-term | Between 27 and 33 weeks of gestation | Secondary, binary |
| Moderate to Late Preterm | Between 31 and 38 weeks of gestation | Secondary, binary |
| Low Birth Weight | Measured at delivery/birth | Secondary, binary |
| Small for Gestation Age | At delivery/birth | Secondary, binary |
The registry lists one registered primary endpoint and ten posted outcome measures. All posted statistical analyses use binary endpoints and report risk ratios. This makes the analysis framework fundamentally different from a time-to-event analysis: the central statistical question is the relative probability of an event over a defined observation window, rather than the instantaneous event rate over time.
5. Statistical Methodology
Logistic regression with a risk-ratio effect measure
The registry identifies logistic regression as the statistical method used for the posted analyses. The reported effect measure is the risk ratio (RR). For a binary endpoint, the risk ratio compares the probability of the event between the two groups.
An RR of 1 represents equal estimated risk in the two groups. An RR below 1 indicates a lower estimated risk in the numerator group, while an RR above 1 indicates a higher estimated risk in the numerator group.
Generalized estimating equations for the cluster-randomized design
The registry states that a GEE with a log link function and exchangeable correlation was used to adjust confidence intervals for the cluster randomized design. The exchangeable correlation structure treats the pairwise correlation within a cluster as common when estimating the working correlation.
This is an important distinction from simply applying an ordinary regression model as if every observation were statistically independent. Cluster randomization can create within-cluster dependence, so the variance estimation must account for that dependence to produce appropriate uncertainty estimates.
Referent group
The iron-folic acid supplemented group was the referent for the analyses. Thus, an estimated risk ratio below 1 is interpreted as a lower estimated risk in the iron-and-folate group relative to the multiple-micronutrient group.
Two-sided confidence intervals
The primary estimate is accompanied by a two-sided 95% confidence interval. A confidence interval describes the statistical uncertainty around the estimated effect under the model and sampling framework. It does not describe the range of effects that individual infants would experience.
Multiplicity adjustment
The registry identifies multiplicity adjustment in the analysis text for the primary endpoint and selected secondary endpoints. Specifically, the relevant analyses state Bonferroni-adjusted alpha=0.01 (for 5 comparisons).
The registry explicitly reports alpha=0.01 for 5 comparisons. The purpose of this adjustment is to make the decision threshold more stringent when several comparisons are being considered within the specified family.
6. Primary Result: Infant Mortality Through 6 mo of Age
The primary analysis compares Iron and Folate vs Multiple Micronutrient among live births by arm. The endpoint was infant mortality through 6 months of age, defined as infant mortality to age 6 months, or 180 days from birth.
Risk ratio for infant mortality
95% CI: 0.86–1.06 · P = 0.36
Two-sided 95% confidence interval; superiority hypothesis.
| Primary endpoint | Iron and Folate vs Multiple Micronutrient |
|---|---|
| Outcome | Infant Mortality Through 6 mo of Age |
| Time frame | 6-months post-birth |
| Analysis population | All live births by arm |
| Method | Logistic regression |
| Effect measure | Risk Ratio (RR) |
| Estimate | 0.95 |
| 95% CI | 0.86–1.06 |
| P-value | 0.36 |
| Multiplicity | Bonferroni-adjusted alpha=0.01 (for 5 comparisons) |
| Correlation adjustment | GEE with log link function and exchangeable correlation for the cluster randomized design |
The estimated risk ratio of 0.95 means that the estimated risk of infant mortality through 6 months in the iron-and-folate group was 0.95 times the estimated risk in the multiple-micronutrient group, using the iron-folic acid group as the referent in the registry analysis.
The estimate itself does not establish that the intervention caused a 5% reduction in infant mortality. It is a relative estimate from the specified regression analysis, and the associated uncertainty is substantial enough that the confidence interval includes values below and above 1.
The 95% CI of 0.86–1.06 indicates that the estimate is compatible with a range of relative effects under the statistical model. Because the interval includes 1, the data do not exclude equal risk at the conventional two-sided 5% level.
The P-value of 0.36 measures the compatibility of the observed result with the null hypothesis under the specified testing framework; it does not measure the size or clinical importance of the effect. A P-value is not the probability that the null hypothesis is true.
Multiplicity is especially important here. The registry specifies a Bonferroni-adjusted alpha of 0.01 for 5 comparisons. Consequently, the primary analysis should be interpreted within that prespecified multiple-comparison framework rather than by treating an arbitrary unadjusted threshold as the sole decision rule.
7. Secondary Results
The registry contains formal statistical analyses for nine secondary outcome measures. The same broad analytical framework was used across these binary outcomes: logistic regression with a risk-ratio effect measure, with GEE and an exchangeable correlation structure used to adjust confidence intervals for the cluster randomized design.
| Secondary endpoint | Time frame | RR | 95% CI | P-value |
|---|---|---|---|---|
| Neonatal Mortality | 1 month post-birth | 0.98 | 0.88–1.20 | 0.78 |
| Post-neonatal Mortality | 1-6 months post-birth | 0.81 | 0.63–1.04 | 0.11 |
| Still Birth Rates | 24 weeks gestation to delivery | 0.89 | 0.81–0.99 | 0.02 |
| Preterm Birth | Up to 37 weeks of gestation | 0.85 | 0.80–0.91 | <0.001 |
| Extremely Pre-term | Up to 28 weeks of gestation | 0.75 | 0.57–0.97 | 0.03 |
| Very Pre-term | Between 27 and 33 weeks of gestation | 0.73 | 0.62–0.86 | <0.001 |
| Moderate to Late Preterm | Between 31 and 38 weeks of gestation | 0.87 | 0.81–0.93 | 0.87 |
| Low Birth Weight | Measured at delivery/birth | 0.88 | 0.85–0.91 | <0.001 |
| Small for Gestation Age | At delivery/birth | 0.98 | 0.96–1.01 | 0.13 |
Still Birth Rates
The reported risk ratio was 0.89, with a 95% CI of 0.81–0.99 and P-value 0.02. The point estimate is below 1, but the registry specifies a Bonferroni-adjusted alpha of 0.01 for 5 comparisons for this analysis. The P-value of 0.02 therefore should not be read as meeting that stated adjusted threshold.
Preterm Birth
For preterm birth up to 37 weeks of gestation, the reported risk ratio was 0.85, with a 95% CI of 0.80–0.91 and P-value <0.001. The confidence interval lies below 1, and the reported P-value is below the registry's stated Bonferroni-adjusted alpha of 0.01 for this comparison family.
Extremely Pre-term
For extremely pre-term birth up to 28 weeks of gestation, the risk ratio was 0.75, with a 95% CI of 0.57–0.97 and P-value 0.03. The confidence interval is below 1, but the registry's multiplicity-adjusted alpha for analyses identified as part of the 5-comparison family is 0.01.
Very Pre-term
For very pre-term birth between 27 and 33 weeks of gestation, the reported risk ratio was 0.73, with a 95% CI of 0.62–0.86 and P-value <0.001. The estimate indicates a lower estimated risk in the iron-and-folate group relative to the multiple-micronutrient group under the reported analysis.
Low Birth Weight
For low birth weight measured at delivery/birth, the reported risk ratio was 0.88, with a 95% CI of 0.85–0.91 and P-value <0.001. The narrow confidence interval indicates substantially greater statistical precision around this relative estimate than for several of the less frequent mortality outcomes.
Small for Gestation Age
The registry reports a risk ratio of 0.98, with a 95% CI of 0.96–1.01 and P-value 0.13. Notably, the registry's "Groups Compared" field for this analysis states Iron and Folate, while the analysis notes identify the iron-folic acid supplemented group as the referent. The displayed registry information therefore does not provide a conventional two-group label in the groups-compared field for this endpoint, and the result is best read using the stated referent and the reported RR.
Neonatal and Post-neonatal Mortality
Neonatal mortality had a reported RR of 0.98 (95% CI 0.88–1.20; P=0.78), while post-neonatal mortality had an RR of 0.81 (95% CI 0.63–1.04; P=0.11). Both confidence intervals include 1. The post-neonatal estimate is further from 1 than the neonatal estimate, but its confidence interval still includes equal estimated risk.
Moderate to Late Preterm
The registry reports an RR of 0.87, with a 95% CI of 0.81–0.93 and P-value 0.87. The numerical combination of this P-value with the reported confidence interval is unusual: the interval is entirely below 1 while the P-value is reported as 0.87. The appropriate statistical reporting practice is to preserve the registry values rather than silently recompute or replace them. The discrepancy also illustrates why a registry result should be interpreted from the full analysis specification rather than from one statistic in isolation.
8. Safety
The registry reports serious adverse events by arm as affected participants over participants at risk.
| Safety measure | Iron and Folic Acid (600 ug) | Multiple Micronutrient |
|---|---|---|
| Serious adverse events, affected / at risk | 0/22114 | 0/22346 |
There were 0/22114 serious adverse events in the iron-and-folic-acid group and 0/22346 in the multiple-micronutrient group according to the registry's affected/at-risk representation. These are event counts, not comparative effect estimates. In particular, when no events are observed in either arm, a conventional risk ratio based on observed event counts is not reported by the registry and should not be constructed here.
9. Statistical Methods Explained
Why use a risk ratio for these endpoints?
The outcomes are binary: an infant either experiences the specified mortality outcome within the defined time frame or does not; similarly, birth outcomes such as preterm birth and low birth weight are represented as binary endpoints. A risk ratio directly expresses the relative probability of the event between the two randomized groups. This is often easier to interpret clinically than an odds ratio when the outcome is not rare.
What does an RR of 0.95 mean?
An RR of 0.95 means that the estimated risk in the iron-and-folate group is 0.95 times the estimated risk in the multiple-micronutrient group. It is a relative measure, not an absolute risk difference. Without the underlying event probabilities, the RR alone cannot tell us how many events were prevented or caused.
Why is the confidence interval important?
The point estimate is only one possible estimate from the observed data. The 95% confidence interval of 0.86–1.06 for the primary endpoint communicates uncertainty around the RR of 0.95. Because the interval crosses 1, equal risk remains compatible with the statistical model and sampling framework.
Why was GEE used?
The registry explicitly states that GEE with a log link function and exchangeable correlation was used to adjust confidence intervals for the cluster randomized design. The key statistical issue is dependence: observations within the same randomized cluster may be correlated. GEE provides a framework for estimating regression effects while accounting for that correlation in the variance structure.
Why does multiplicity matter?
When several hypotheses are evaluated, the probability of obtaining at least one apparently unusual result can increase if each comparison is judged independently at the same nominal threshold. The registry addresses this for specified analyses with a Bonferroni-adjusted alpha of 0.01 for 5 comparisons. That adjustment changes how the corresponding P-values should be interpreted.
Does a P-value measure the size of an effect?
No. The P-value and effect estimate answer different questions. The RR describes the magnitude and direction of the relative association under the model, while the P-value describes how compatible the observed result is with a specified null hypothesis under the testing framework. Precision is conveyed separately by the confidence interval.
Why does the primary analysis use live births?
The registry explicitly states that not all women recruited into the trial had pregnancies yielding live births and that the primary analysis represents all live births by arm. This defines the analysis population for the infant mortality endpoint and distinguishes the population entering the mortality analysis from the full enrollment population.
10. Interpreting Relative Effects and Absolute Outcomes
A risk ratio compares the probability of an outcome between two groups. An RR below 1 means that the estimated risk is lower in the numerator group relative to the denominator group. The magnitude of that difference is relative, not absolute.
A confidence interval provides a range of values expressing uncertainty around the estimated relative effect. A narrow interval indicates greater precision than a wide interval, but precision and clinical importance are different concepts.
A P-value is not an effect-size measure. For example, two endpoints can have similarly sized relative effects but different P-values because their estimates have different precision. Conversely, a small P-value does not by itself establish that an effect is large or clinically important.
This distinction is particularly useful in JiVitA-3 because the registry contains outcomes with confidence intervals of different widths. Mortality outcomes generally have less statistical precision than outcomes such as low birth weight, where the reported interval is comparatively narrow. Statistical interpretation should therefore consider the estimate and its interval together rather than ranking endpoints by P-value alone.
11. Multiplicity and the Five-Comparison Framework
The registry explicitly identifies a Bonferroni-adjusted alpha of 0.01 (for 5 comparisons) in the primary analysis and selected secondary analyses. Bonferroni adjustment is a simple familywise error-control strategy: if a family contains 5 comparisons and the desired overall nominal alpha is 0.05, each comparison is evaluated against 0.01.
| Endpoint | P-value | Registry notes |
|---|---|---|
| Infant Mortality Through 6 mo of Age | 0.36 | Bonferroni-adjusted alpha=0.01 (for 5 comparisons) |
| Still Birth Rates | 0.02 | Bonferroni-adjusted alpha=0.01 (for 5 comparisons) |
| Preterm Birth | <0.001 | Bonferroni-adjusted alpha=0.01 (for 5 comparisons) |
| Low Birth Weight | <0.001 | Bonferroni-adjusted alpha=0.01 (for 5 comparisons) |
| Small for Gestation Age | 0.13 | Bonferroni-adjusted alpha=0.01 (for 5 comparisons) |
The adjustment is a design-level issue rather than a property of an individual P-value. A P-value of 0.02 means the same numerical thing whether or not multiplicity adjustment was planned, but its evidentiary interpretation changes when the prespecified decision threshold is 0.01 rather than 0.05.
12. Cluster Randomization and Correlated Data
The registry states that the study was cluster randomized and that GEE with a log link and exchangeable correlation was used to adjust confidence intervals. This is an important statistical feature of the trial.
Why clustering matters
If individuals within the same cluster tend to have more similar outcomes than individuals from different clusters, treating every observation as independent can produce an inappropriate variance estimate.
What exchangeable correlation means
An exchangeable working correlation assumes a common correlation among observations within a cluster. The model therefore incorporates the cluster structure rather than assuming independence across all observations.
The practical consequence is that the reported confidence intervals are not simply the intervals that would result from treating every live birth as an independent observation. The registry specifically identifies the GEE approach as the mechanism used to account for the cluster randomized design.
13. Why Logistic Regression Can Report a Risk Ratio
The registry labels the method as logistic regression while separately identifying the effect measure as a risk ratio. These concepts should not be conflated. The regression model specifies how the binary outcome is modeled, whereas the reported effect measure describes how the treatment contrast is expressed.
The model, outcome type, and effect measure describe different aspects of the analysis. The registry identifies all three separately.
For a reader interpreting the JiVitA-3 registry, the most important point is that the reported number such as 0.85 for preterm birth is a risk ratio, not an odds ratio and not a regression coefficient on the log-odds scale.
14. Primary Analysis Population and the Meaning of Enrollment
The trial enrolled 44567 participants, but the registry makes an explicit distinction for the primary mortality analysis: not all women recruited into the trial had pregnancies yielding live births. The primary analysis represents all live births by arm.
This distinction matters because an enrollment number and an endpoint-specific analysis population answer different questions. Enrollment describes the scale of the randomized study, whereas the analysis population determines which observations contribute to a particular endpoint estimate.
15. Limitations
- Low iron-deficiency setting: the registry notes that the study was run in an area where iron deficiency rates were low. This contextual feature is relevant when considering how broadly the findings may apply to populations with different underlying iron-deficiency rates.
- Clustered design: the cluster randomized structure requires appropriate variance adjustment. The registry reports GEE with an exchangeable correlation structure for the confidence intervals.
- Endpoint multiplicity: the registry explicitly identifies Bonferroni-adjusted alpha=0.01 for 5 comparisons for selected analyses. Secondary endpoints therefore should not all be interpreted as though they were isolated tests at a nominal 0.05 threshold.
- Different analysis populations: the primary mortality analysis is described in terms of live births by arm rather than all women recruited, while safety is reported using separate affected/at-risk counts.
- Registry-level information: the ClinicalTrials.gov record provides the reported estimates, confidence intervals, P-values, and methods summarized here, but the information presented does not include a baseline characteristics table, subgroup analyses, or a more detailed individual-level model specification.
- Interpretation of secondary outcomes: several pregnancy and birth outcomes are statistically related to the same randomized comparison. Their interpretation therefore benefits from attention to the prespecified multiplicity framework rather than isolated consideration of individual P-values.
16. Why This Trial Matters Statistically
JiVitA-3 is a useful teaching example because it combines randomized trial design with binary clinical endpoints, cluster-level dependence, regression modeling, relative-risk interpretation, and multiplicity adjustment. The primary endpoint is also a useful illustration of why the headline estimate should never be interpreted without its confidence interval and analysis framework.
| Concept | How it appears in JiVitA-3 |
|---|---|
| Randomization | The trial uses randomized allocation to two parallel intervention groups. |
| Blinding | The registry identifies quadruple masking. |
| Binary outcomes | Infant mortality, neonatal mortality, preterm birth, low birth weight, and other endpoints are analyzed as binary outcomes. |
| Logistic regression | The registry identifies logistic regression as the posted analysis method. |
| Risk ratio | The reported effect measure is the risk ratio / relative risk. |
| GEE | GEE with a log link and exchangeable correlation adjusts confidence intervals for the cluster randomized design. |
| Confidence intervals | Two-sided 95% confidence intervals accompany the reported risk ratios. |
| Multiplicity | Selected analyses specify Bonferroni-adjusted alpha=0.01 for 5 comparisons. |
| Analysis population | The primary mortality analysis represents all live births by arm rather than all women recruited. |
| Safety reporting | Serious adverse events are reported as affected participants over participants at risk by arm. |
17. What the Primary Risk Ratio Does — and Does Not — Mean
The primary RR of 0.95 is a relative comparison of infant mortality through 6 months between the iron-and-folate group and the multiple-micronutrient group. Under the reported analysis, the estimated risk in the iron-and-folate group was 0.95 times the estimated risk in the multiple-micronutrient group.
The RR does not mean that 95% of infants survived, that 5% of infants benefited, or that the absolute probability of mortality differed by 5 percentage points. Those interpretations require absolute event risks, which are not reported in the statistical-analysis information summarized here.
The 95% CI of 0.86–1.06 is more informative than the point estimate alone because it communicates uncertainty. It includes 1, so the statistical analysis is compatible with both a lower and a higher relative risk for the iron-and-folate group compared with the multiple-micronutrient group.
The P-value of 0.36 is evidence about compatibility with the specified null hypothesis under the statistical model. It should not be used as a measure of effect magnitude, clinical relevance, or probability that one treatment is superior to the other.
18. Statistical Concepts in This Trial
Learn more about the methods used in this trial:
19. Related Statistical Calculators
20. Sources
- ClinicalTrials.gov: JiVitA-3, NCT00860470.
- PubMed: PMID 40653272.
- PubMed: PMID 39145520.
- PubMed: PMID 39062451.
- PubMed: PMID 35155983.
- PubMed: PMID 32844185.
Continue through Clinical Biostats
Connect the statistical methods used in JiVitA-3 with deeper tutorials and practical statistical calculators.
21. Record Summary
JiVitA-3 provides a compact but statistically rich example of randomized clinical-trial analysis. The registered primary endpoint was infant mortality through 6 months of age, analyzed as a binary outcome using logistic regression with a risk-ratio effect measure. The reported primary RR was 0.95 with a two-sided 95% CI of 0.86–1.06 and P-value 0.36.
The broader set of posted secondary analyses illustrates why interpretation cannot rely on P-values alone. The registry reports risk ratios, confidence intervals, and P-values for neonatal mortality, post-neonatal mortality, still birth, several categories of preterm birth, low birth weight, and small for gestation age. The analysis framework also explicitly accounts for the cluster randomized design through GEE and identifies Bonferroni-adjusted alpha=0.01 for 5 comparisons for selected endpoints.
From a statistical perspective, the central lessons are the distinction between relative effect and absolute risk, the role of the confidence interval in quantifying uncertainty, the difference between an effect estimate and a P-value, the importance of accounting for within-cluster correlation, and the need to interpret multiple endpoints within their prespecified multiplicity framework.