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
SVI was a randomized, parallel-group, phase 3 trial evaluating whether rectal indomethacin without prophylactic pancreatic stent placement could be used for prevention of post-ERCP pancreatitis compared with rectal indomethacin combined with prophylactic pancreatic stent placement.
| Feature | SVI |
|---|---|
| Phase | Phase 3 |
| Condition | Post-ERCP Pancreatitis |
| Brief title | Stent vs. Indomethacin for Preventing Post-ERCP Pancreatitis |
| Design | Randomized, parallel-group |
| Masking | Quadruple |
| Primary purpose | Prevention |
| Enrollment | 1,950 |
| Primary endpoint | Proportion of subjects in each study group with post-ERCP pancreatitis |
| Primary endpoint time frame | Within 48 hours after ERCP |
| Primary effect measure | Risk difference |
| ClinicalTrials.gov | NCT02476279 |
| Lead sponsor | Medical University of South Carolina |
2. Clinical Question
The central question was whether indomethacin 100 mg administered rectally immediately after ERCP, without prophylactic pancreatic stent placement, could be considered noninferior to the combination of indomethacin and prophylactic pancreatic stent placement for preventing post-ERCP pancreatitis.
Population
Participants enrolled in the phase 3 randomized trial for prevention of post-ERCP pancreatitis.
Intervention
Indomethacin 100 mg rectally immediately after ERCP, with no prophylactic pancreatic stent placement.
Comparator
Indomethacin 100 mg rectally immediately after ERCP plus prophylactic pancreatic stent placement.
Primary question
Is the risk of post-ERCP pancreatitis with indomethacin alone sufficiently close to that with indomethacin plus stent to satisfy the prespecified noninferiority criterion?
3. Trial Design
Indomethacin Alone
- Indomethacin 100 mg rectally immediately after ERCP
- No prophylactic pancreatic stent placement
Indomethacin + Pancreatic Stent
- Indomethacin 100 mg rectally immediately after ERCP
- Prophylactic pancreatic stent placement
The trial began in September 2015 and had a primary completion date of January 25, 2023. The randomized parallel structure is important for interpreting the primary comparison because treatment assignment, rather than subsequent clinical management, defines the principal efficacy comparison.
4. Primary Endpoint
| Endpoint | Definition | Time frame | Type |
|---|---|---|---|
| The Proportion of Subjects in Each Study Group With Post-ERCP Pancreatitis | Post-ERCP pancreatitis was based on a widely validated consensus definition used as a diagnostic framework. It involved new onset or increase of upper-abdominal pain, pancreatic enzyme elevation of at least three times the upper limit of normal 24 hours after the procedure, and hospitalization for at least two nights. The outcome was independently adjudicated by three ERCP experts at non-enrolling centers. PEP was declared when at least two of the three adjudicators agreed. | Within 48 hours after ERCP | Binary |
The adjudication framework is statistically important because the endpoint is binary: each participant is classified according to whether post-ERCP pancreatitis occurred within the specified time frame. The resulting treatment comparison is therefore naturally expressed as an absolute difference in event risk rather than a time-to-event hazard ratio.
5. Statistical Methodology
Risk difference
The primary effect measure reported by the registry was the risk difference. For this trial, the risk difference was defined as the risk of post-ERCP pancreatitis in the indomethacin-alone group minus the risk in the indomethacin-plus-stent group.
A positive risk difference means that the observed risk of post-ERCP pancreatitis was higher in the indomethacin-alone group. A negative value would mean that the observed risk was lower in that group.
Intention-to-treat and per-protocol populations
The primary endpoint was analyzed in both the intention-to-treat and per-protocol populations. The intention-to-treat analysis preserves the comparison created by randomization by analyzing participants according to their randomized study group. The per-protocol analysis instead focuses on participants who remained within the protocol-defined analysis population.
Using both populations is particularly relevant in a noninferiority trial. A treatment can appear similar to its comparator after exclusions or protocol deviations, so agreement between intention-to-treat and per-protocol analyses provides an important check on the robustness of the noninferiority conclusion.
Formal analysis method
The ClinicalTrials.gov record reports the effect measure as risk difference and provides two-sided 95% confidence intervals, but it does not report the specific statistical method used to calculate the risk difference or its confidence interval.
For a binary endpoint, a risk difference is typically estimated from the observed event proportions in the two randomized groups, with a confidence interval constructed using an appropriate method for comparing independent binomial proportions. The particular confidence-interval construction matters because different methods can behave differently, especially when event rates are low or sample sizes are unequal. Here, the two groups were of equal randomized size.
Noninferiority framework
The registry specifies a noninferiority margin of 5 percentage points for the primary endpoint. The relevant contrast was:
The criterion had to be satisfied in both the intention-to-treat and per-protocol analysis populations.
This is different from a conventional superiority test. The purpose is not to demonstrate that indomethacin alone produces a lower pancreatitis risk than indomethacin plus stent. Instead, the question is whether any increase in risk remains sufficiently small that it stays below the prespecified 5% margin.
6. Results: Primary Endpoint
The registry posts two primary analyses of the proportion of participants with post-ERCP pancreatitis within 48 hours after ERCP. One is based on the intention-to-treat population and the other on the per-protocol population.
Intention-to-Treat Analysis
Risk difference: indomethacin alone minus indomethacin + stent
Two-sided 95% CI: 0.006–0.066
Analysis population: Intention-to-treat
The risk difference of 0.036 means that the estimated risk of post-ERCP pancreatitis was 3.6 percentage points higher with indomethacin alone than with indomethacin plus pancreatic stent, using the intention-to-treat analysis population and the registry's stated direction of subtraction.
The estimate does not mean that every individual participant had a 3.6-percentage-point increase in risk. It is an average between-group absolute difference in the binary outcome.
The two-sided 95% confidence interval extends from 0.006 to 0.066. The upper bound is 6.6 percentage points, which is above the prespecified 5% noninferiority margin. Thus, for the intention-to-treat analysis, the confidence interval itself extends beyond the allowable noninferiority boundary.
The confidence interval describes statistical uncertainty around the estimated risk difference; it is not a range containing the individual risks of participants. There is no p-value reported for this primary analysis in the registry data, and a p-value would not by itself determine noninferiority. The relevant question is whether the confidence interval remains within the prespecified noninferiority margin.
Per-Protocol Analysis
Risk difference: indomethacin alone minus indomethacin + stent
Two-sided 95% CI: −0.003–0.060
Analysis population: Per protocol
The per-protocol risk difference of 0.028 corresponds to an estimated 2.8-percentage-point higher risk of post-ERCP pancreatitis with indomethacin alone than with indomethacin plus stent, using the registry's stated direction of subtraction.
The two-sided 95% confidence interval ranges from −0.003 to 0.060. The interval includes zero, so the data are compatible with a small lower risk, no difference, or a higher risk in the indomethacin-alone group within the uncertainty represented by this interval.
For noninferiority, however, the key feature is the upper confidence limit. The upper limit of 0.060 is 6.0 percentage points, which is above the 5% margin. The per-protocol confidence interval therefore also extends beyond the prespecified noninferiority boundary.
Again, the confidence interval addresses precision around the estimated risk difference; it does not describe individual treatment responses. No p-value for this primary analysis is reported in the registry data.
7. Noninferiority Logic in Detail
Noninferiority trials require a different interpretation from ordinary superiority trials. The null hypothesis is not simply "the two groups are equal." Instead, the analysis asks whether the experimental strategy could be worse than the comparator by more than a prespecified clinically acceptable amount.
| Component | SVI specification | Interpretation |
|---|---|---|
| Effect measure | Risk difference | Absolute difference in the proportion with post-ERCP pancreatitis |
| Direction | Indomethacin alone minus indomethacin + stent | Positive values indicate greater risk with indomethacin alone |
| Confidence interval | Two-sided 95% | Quantifies uncertainty around the estimated risk difference |
| Noninferiority margin | 5% | Maximum acceptable excess risk under the registry's stated criterion |
| Required analysis populations | Intention-to-treat and per protocol | Both populations were required for the stated noninferiority criterion |
| Criterion | Upper CI bound less than 5% | The entire two-sided 95% CI must remain below the margin |
The distinction between the point estimate and the upper confidence limit is central. A point estimate below the margin does not establish noninferiority if the uncertainty interval extends above the margin. In this trial, both point estimates were below 0.05, but both upper confidence limits were above 0.05.
This is also why the p-value is not the central quantity for the noninferiority decision. A p-value can address a particular null hypothesis, but the prespecified noninferiority criterion is expressed directly through the confidence interval and its relationship to the margin.
8. Secondary Endpoint: Moderate-Severe Post-ERCP Pancreatitis
The registry also reports the proportion of participants in each study group with moderate-severe post-ERCP pancreatitis within one month of ERCP. Outcome data were missing for two participants, one in each treatment arm, and those participants were excluded from the percentage denominator.
| Analysis population | Risk difference | 95% CI | Direction |
|---|---|---|---|
| Intention-to-treat | 0.021 | −0.002–0.043 | Indomethacin alone minus indomethacin + stent |
| Per protocol | 0.020 | −0.004–0.044 | Indomethacin alone minus indomethacin + stent |
Intention-to-treat
The estimated risk difference was 0.021, with a two-sided 95% CI of −0.002 to 0.043.
Per protocol
The estimated risk difference was 0.020, with a two-sided 95% CI of −0.004 to 0.044.
These estimates are smaller than the corresponding primary-endpoint point estimates. Both confidence intervals include zero, so the reported uncertainty includes the possibility of little or no difference in moderate-severe pancreatitis risk between the study strategies.
The registry does not identify a specific statistical method beyond the reported risk difference and confidence interval. For a binary secondary endpoint, the same general risk-difference framework provides an intuitive absolute comparison, while the missing outcome observations require attention to the denominator and analysis-population definition.
9. Safety: Serious Adverse Events
The registry reports serious adverse events by study group using affected participants over the number at risk.
| Study group | Serious adverse events affected / at risk |
|---|---|
| Indomethacin Alone | 355 / 975 |
| Indomethacin + Pancreatic Stent | 352 / 975 |
The affected-participant counts are close between the two randomized groups: 355 of 975 participants in the indomethacin-alone group and 352 of 975 in the indomethacin-plus-stent group. These counts describe the reported serious-adverse-event experience; they do not by themselves establish a causal difference in safety because the registry data presented here do not provide a comparative statistical test or confidence interval for this safety measure.
10. Statistical Methods Explained
Why was risk difference used?
The primary endpoint is binary: each participant either developed post-ERCP pancreatitis within the specified 48-hour period or did not. A risk difference expresses the treatment contrast directly in absolute percentage-point terms. This is especially useful for a noninferiority question because the acceptable excess risk can be defined on the same absolute scale.
What does a risk difference of 0.036 mean?
Because the registry defines the contrast as indomethacin alone minus indomethacin plus stent, an estimate of 0.036 means the estimated probability of post-ERCP pancreatitis was 0.036 higher in absolute terms in the indomethacin-alone group. Equivalently, the estimated excess risk was 3.6 percentage points. This is an absolute risk difference, not a relative risk or odds ratio.
Why is the confidence interval compared with the noninferiority margin?
The noninferiority margin represents the largest excess risk that the trial's design permits while still calling the intervention noninferior. If the upper confidence limit remains below that margin, the uncertainty around the treatment difference is sufficiently bounded. If the upper limit extends above the margin, clinically important excess risk has not been excluded at the prespecified confidence level.
Why are both intention-to-treat and per-protocol analyses important?
Intention-to-treat analysis maintains the advantages of randomized assignment, whereas per-protocol analysis focuses more closely on participants who followed the study protocol. In noninferiority trials, protocol deviations can bias the groups toward appearing similar, making interpretation of only one analysis population potentially misleading. The SVI registry criterion explicitly required the upper confidence bound to be below the margin in both populations.
What does it mean that the per-protocol confidence interval includes zero?
The per-protocol estimate was 0.028 with a 95% CI of −0.003 to 0.060. Because zero lies inside this interval, the data are compatible with no absolute difference between the groups as well as with modest differences in either direction. For noninferiority, however, whether the interval crosses zero is not the principal criterion. The upper limit of 0.060 is compared with the 0.05 margin.
Why does the p-value not answer the noninferiority question by itself?
A p-value is tied to a particular null hypothesis and does not directly state how close the observed treatment difference is to a clinically important boundary. In a noninferiority analysis, the margin and confidence interval provide the direct framework for determining whether an unacceptable excess risk has been excluded. The registry's stated rule is based on the upper bound of the two-sided 95% confidence interval.
11. Endpoint Adjudication and Blinding
The primary endpoint was independently adjudicated by three ERCP experts at non-enrolling centers. They reviewed medical records for participants who were hospitalized with any adverse event within two days of ERCP. Records were redacted to remove information that could potentially reveal study-group assignment, including radiology reports. PEP was declared when at least two of the three adjudicators agreed.
The registry also documents several potential sources of unmasking. Potential unmasking events could have influenced care. Blinding of secondary outcomes could decay because an unblinded clinician could resume care of a patient after 48 hours. Adjudicators received radiographic information when assessing PEP severity, which could have influenced decision making. Unblinded endoscopists made procedural decisions that could have differed according to study arm.
12. Interpreting the Primary Result Without Overreading It
The two primary risk-difference estimates, 0.036 for intention-to-treat and 0.028 for per protocol, are both below the prespecified 0.05 noninferiority margin. Point estimates, however, are only one part of a noninferiority analysis.
The intention-to-treat 95% CI was 0.006–0.066, while the per-protocol 95% CI was −0.003–0.060. The upper bounds therefore exceed the 0.05 margin in both analyses. The intervals show that the uncertainty around the point estimates includes excess-risk values above the prespecified boundary.
The risk difference does not establish that the two treatment strategies are biologically identical, nor does it mean that the absolute difference for every participant is fixed at the point estimate. It summarizes the between-group difference in the trial population under the specified analysis.
Randomization supports a causal comparison between the assigned strategies, but interpretation still depends on adherence, masking, outcome adjudication, and the analysis population. The registry specifically identifies potential unmasking and procedural decision-making issues that should be considered when interpreting the findings.
13. Missing Data and Analysis Population
The secondary moderate-severe post-ERCP pancreatitis endpoint had outcome data missing for two participants, one in each treatment arm. The registry states that these two participants were excluded from the percentage denominator.
This detail matters because risk estimates depend on the denominator. Excluding participants with missing outcome data is not mathematically equivalent to treating them as having or not having the event. The reported risk differences should therefore be interpreted using the analysis-population description and missing-data handling stated in the registry.
| Endpoint | Analysis population | Missing-data detail |
|---|---|---|
| Post-ERCP pancreatitis within 48 hours | Intention-to-treat and per protocol | No additional missing-data handling is specified in the reported analysis text. |
| Moderate-severe post-ERCP pancreatitis within one month | Intention-to-treat and per protocol | Outcome data missing for two participants, one in each treatment arm; excluded from the percentage denominator. |
14. Limitations
- Potential unmasking: the registry reports potential unmasking events that could have influenced care.
- Decay in secondary-outcome blinding: an unblinded clinician could resume care of a patient after 48 hours.
- Adjudication information: adjudicators received radiographic information when assessing PEP severity, which may have influenced decision making.
- Procedural decisions: unblinded endoscopists made procedural decisions that could have been altered according to study arm.
- Noninferiority uncertainty: the point estimates were below the 5% margin, but the upper bounds of both primary two-sided 95% confidence intervals exceeded the margin.
- Unreported statistical procedure: the registry reports the risk-difference estimates and confidence intervals but does not identify the specific method used to calculate them.
- Secondary endpoint missingness: two participants had missing outcome data for the moderate-severe pancreatitis endpoint and were excluded from its percentage denominator.
- Generalizability: the registry notes that findings may not be applicable to community endoscopists because all sites were tertiary referral centers.
15. Why This Trial Matters Statistically
SVI is a useful teaching example because it illustrates how a randomized clinical trial can be built around an absolute binary endpoint and a noninferiority question rather than a conventional superiority hypothesis.
| Statistical concept | How it appears in SVI |
|---|---|
| Randomization | Participants were randomized to two parallel study strategies. |
| Binary endpoint | Post-ERCP pancreatitis was assessed as an occurrence/non-occurrence outcome within a defined time frame. |
| Risk difference | The primary treatment contrast was the absolute risk difference between study groups. |
| Confidence interval | Two-sided 95% confidence intervals quantify uncertainty around the reported risk differences. |
| Noninferiority margin | The upper bound of the two-sided 95% CI had to be below 5% in both primary analysis populations. |
| Intention-to-treat analysis | The primary endpoint was analyzed according to randomized treatment assignment. |
| Per-protocol analysis | A second primary analysis evaluated the endpoint in the per-protocol population. |
| Masking | The trial used quadruple masking, while the registry also documents potential unmasking events. |
| Outcome adjudication | Three ERCP experts independently adjudicated PEP, with agreement from at least two required. |
| Missing data | Two participants had missing data for the moderate-severe PEP endpoint and were excluded from its percentage denominator. |
16. Statistical Takeaways
Absolute effect
The primary analysis used risk difference, making the treatment contrast directly interpretable as an absolute difference in pancreatitis risk.
Two populations
The primary endpoint was analyzed in both intention-to-treat and per-protocol populations, which is particularly relevant to noninferiority interpretation.
Margin matters
The prespecified noninferiority margin was 5%. The upper confidence limit, not merely the point estimate, determines whether the stated criterion is met.
Precision matters
The intention-to-treat CI was 0.006–0.066 and the per-protocol CI was −0.003–0.060. Both upper limits extend above 0.05.
17. Clinical Biostats Interpretation of the Evidence
The SVI statistical story is best understood by separating the estimated effect from the noninferiority decision rule. The intention-to-treat point estimate of 0.036 and per-protocol point estimate of 0.028 are both numerically below the 5% margin. If only the point estimates were considered, the observed differences would appear to fall within the prespecified boundary.
Confidence intervals add the essential second layer. The intention-to-treat upper confidence limit was 0.066, and the per-protocol upper confidence limit was 0.060. Because the registry's noninferiority criterion required the upper limit to be below 0.05 in both populations, neither reported confidence interval remains entirely below the margin.
This distinction illustrates why noninferiority trials cannot be interpreted by looking only at whether the observed point estimate is smaller than the margin. The analysis must incorporate uncertainty. A point estimate can be below the margin while the confidence interval still permits an excess risk that exceeds the margin.
The secondary moderate-severe pancreatitis endpoint provides a related but distinct perspective. Its risk differences were 0.021 in the intention-to-treat analysis and 0.020 in the per-protocol analysis, with confidence intervals of −0.002–0.043 and −0.004–0.044, respectively. Those intervals include zero and have upper bounds below 0.05, but this secondary endpoint was not identified in the registry data as having the same stated noninferiority hypothesis as the primary endpoint. It therefore should not be substituted for the prespecified primary analysis when interpreting the primary noninferiority question.
Finally, statistical interpretation should remain connected to trial conduct. Potential unmasking, clinician decision-making, adjudication procedures, missing secondary-endpoint observations, and the tertiary-referral setting all affect how broadly the findings can be applied and how confidently the observed treatment contrast can be attributed to the randomized strategies.
18. Sources
- ClinicalTrials.gov: NCT02476279.
- PubMed: PubMed record for PMID 38219767.
- PubMed: PubMed record for PMID 26941086.
19. Related Tutorials
Learn more about the methods used in this trial:
20. Related Calculators
Continue through the Clinical Biostats statistical library
Explore tutorials and statistical calculators covering the concepts used to design, analyze, and interpret randomized clinical trials.
21. Record Summary
SVI provides a focused example of noninferiority methodology for a binary clinical endpoint. The randomized phase 3 trial compared indomethacin alone with indomethacin plus prophylactic pancreatic stent placement, using post-ERCP pancreatitis within 48 hours as the primary endpoint and risk difference as the effect measure.
The primary analyses reported risk differences of 0.036 in the intention-to-treat population and 0.028 in the per-protocol population. Their corresponding two-sided 95% confidence intervals were 0.006–0.066 and −0.003–0.060. Because the registry's stated noninferiority criterion required the upper bound of the confidence interval to be below 5% in both populations, the upper limits are the central quantities for interpreting the prespecified noninferiority framework.
The trial also reported moderate-severe post-ERCP pancreatitis within one month as a secondary endpoint, with risk differences of 0.021 and 0.020 in the intention-to-treat and per-protocol analyses, respectively. Serious adverse events were reported in 355/975 participants in the indomethacin-alone group and 352/975 in the indomethacin-plus-stent group.