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
STRESS was a randomized, parallel-group, quadruple-masked phase 3 trial evaluating methylprednisolone versus isotonic saline in children with congenital heart disease and an inflammatory response after infant heart surgery. The registry reports enrollment of 1263 participants and a binary primary endpoint based on a severity-ranked global outcome.
| Feature | STRESS |
|---|---|
| Phase | Phase 3 |
| Population | Children with congenital heart disease and inflammatory response after infant heart surgery |
| Design | Randomized, parallel-group, quadruple-masked |
| Allocation | Randomized |
| Primary purpose | Treatment |
| Enrollment | 1263 |
| Primary endpoint | Number of participants at each global rank endpoint based upon their most-severe outcome |
| Primary endpoint type | Binary |
| ClinicalTrials.gov | NCT03229538 |
| Status | Completed |
2. Clinical Question
The central statistical question was whether methylprednisolone, compared with isotonic saline, was associated with a different distribution of severe postoperative outcomes through a composite global rank endpoint incorporating mortality, major morbidity, and length of stay.
Population
Children with congenital heart disease and inflammatory response undergoing the clinical setting represented by the trial.
Intervention
Methylprednisolone.
Comparator
Isotonic saline.
Primary question
Does methylprednisolone change the occurrence of increasingly severe postoperative outcomes compared with the saline group?
3. Trial Design
Methylprednisolone
- Methylprednisolone as the study intervention.
- 599 participants were at risk for the serious-adverse-event summary reported by arm.
Placebo / saline
- Isotonic saline as the comparator intervention.
- 601 participants were at risk for the serious-adverse-event summary reported by arm.
The statistical structure is important because the primary endpoint is not simply a single event such as death. Each randomized participant receives a rank according to the most severe outcome experienced, allowing several clinically important postoperative outcomes to be incorporated into one ordered severity framework.
4. Primary Endpoint
| Endpoint | Registry definition | Time frame |
|---|---|---|
| Number of Participants at Each Global Rank Endpoint Based Upon Their Most-severe Outcome | A composite mortality, major morbidity and length of stay global rank endpoint with endpoints ranked according to severity. Each randomized patient is assigned a rank based upon their most-severe outcome. | Until hospital discharge, up to 4 months |
How the global rank works
The registry defines ranks 1 through 90 according to postoperative length of stay in days. Higher ranks represent progressively more severe clinical outcomes. Rank 91 corresponds to postoperative length of stay greater than 90 days; rank 92 includes post-op cardiac arrest, multi-system organ failure, renal failure with temporary dialysis, or prolonged ventilator support; rank 93 includes reoperation for bleeding, unplanned delayed sternal closure, or post-op unplanned interventional cardiac catheterization; rank 94 includes post-operative mechanical circulatory support or unplanned cardiac reoperation exclusive of reoperation for bleeding.
The highest categories are rank 95, comprising renal failure with permanent dialysis, neurologic deficit persistent at discharge, or respiratory failure requiring tracheostomy; rank 96, heart transplant during hospitalization; and rank 97, operative mortality.
5. Statistical Methodology
The registry identifies logistic regression as the principal the statistical method and reports both odds ratios and risk differences as effect measures. The primary endpoint is classified as binary, and the posted analyses compare methylprednisolone with placebo/saline.
| Statistical component | Registry-reported approach |
|---|---|
| Primary endpoint type | Binary |
| Primary model | Logistic regression |
| Primary effect measure | Adjusted odds ratio |
| Additional effect measure | Risk difference / percent difference |
| Confidence interval | 95%, two-sided |
| Primary endpoint analyses posted | 8 |
| Primary analyses with estimate + CI | 8 |
| Statistical analyses posted | 14 |
For a binary outcome, logistic regression models the relationship between treatment assignment and the odds of the specified outcome. An adjusted odds ratio below 1 indicates lower estimated odds in the methylprednisolone group relative to the comparator, whereas an odds ratio above 1 indicates higher estimated odds. The registry does not identify the adjustment covariates in the statistical-analysis fields presented here, so the adjustment set should not be inferred from the effect estimate alone.
Odds ratio interpretation
For the primary adjusted odds ratio of 0.86, the estimated odds ratio is below 1. In simple relative terms, 0.86 corresponds to an estimated 14% lower odds, but the confidence interval extends from 0.71 to 1.05. Because that interval includes 1, the estimate is compatible with both a reduction and a modest increase in the odds on the odds-ratio scale.
Risk difference provides a complementary scale
The registry also reports percent differences for increasingly severe global-rank thresholds. These are expressed as methylprednisolone minus placebo. Unlike an odds ratio, a risk difference is an absolute difference in event probability. It is therefore directly interpretable in percentage-point terms when the underlying binary threshold is understood.
6. Primary Results
The registry reports eight primary-endpoint analyses. One is an adjusted odds-ratio analysis of the global rank endpoint, while seven are percent-difference analyses at progressively higher severity thresholds.
Overall global rank analysis
Adjusted odds ratio
95% CI: 0.71–1.05 · P = 0.14
Methylprednisolone arm versus placebo arm.
An adjusted odds ratio of 0.86 indicates that the estimated odds under the fitted logistic model were lower in the methylprednisolone group than in the placebo group. The point estimate alone does not establish that methylprednisolone reduced the underlying probability of a severe outcome, because an odds ratio is not itself a risk difference and the endpoint is a composite severity-ranking construct.
The 95% confidence interval of 0.71–1.05 describes the statistical precision of the estimate. It includes 1, so the interval includes values corresponding to lower odds as well as values corresponding to slightly higher odds. The two-sided p-value of 0.14 is a measure of compatibility with the statistical null hypothesis under the specified analysis; it is not a measure of the size or clinical importance of the estimated effect.
Because this is a logistic-regression analysis, interpretation also depends on the model specification and the analysis population. The registry identifies the method as logistic regression but does not provide the adjustment variables in the posted analysis summary presented here.
Severity thresholds
| Global-rank threshold | Effect measure | Estimate | 95% CI | P-value |
|---|---|---|---|---|
| Rank = 97 | Risk difference / percent difference | -0.8 | -2.6 to 0.9 | Not reported |
| Rank ≥ 96 | Risk difference / percent difference | -1.5 | -3.5 to 0.5 | Not reported |
| Rank ≥ 95 | Risk difference / percent difference | -2.2 | -4.4 to 0.1 | Not reported |
| Rank ≥ 94 | Risk difference / percent difference | -0.8 | -4.3 to 2.7 | Not reported |
| Rank ≥ 93 | Risk difference / percent difference | -3.8 | -7.8 to 0.2 | Not reported |
| Rank ≥ 92 | Risk difference / percent difference | -3.4 | -7.8 to 0.9 | Not reported |
| Rank ≥ 91 | Risk difference / percent difference | -3.1 | -7.5 to 1.3 | Not reported |
All seven risk-difference estimates are negative, meaning the estimated percentage difference favored fewer participants reaching the specified threshold in the methylprednisolone group when the difference is defined as methylprednisolone minus placebo. However, every corresponding 95% confidence interval crosses zero. The estimates therefore vary in magnitude without providing an interval that excludes no difference at any of the reported thresholds.
7. Secondary Endpoint Results
Mortality through 30 days
Adjusted odds ratio for mortality
95% CI: 0.34–1.57 · P = 0.428
Mortality included in-hospital mortality or mortality after hospital discharge but within 30 days of the last dose of study drug.
The point estimate of 0.74 corresponds to lower estimated odds of mortality in the methylprednisolone group, but the 95% confidence interval of 0.34–1.57 is wide and crosses 1. The result therefore contains substantial uncertainty about the direction and magnitude of the underlying odds ratio.
The p-value of 0.428 should not be interpreted as a 42.8% probability that the treatment has no effect. It is a test statistic expressed as a probability under the specified null model. It also does not describe the clinical magnitude of the observed odds ratio.
Death or major complication
Adjusted odds ratio
95% CI: 0.61–1.13 · P = 0.228
Death or major complication defined by an outcome in one of the 7 highest global-ranking categories.
The estimated odds ratio is below 1, indicating lower estimated odds of the specified composite outcome in the methylprednisolone group. The confidence interval, however, extends from 0.61 to 1.13 and therefore includes both a reduction and a possible increase in odds.
This endpoint illustrates why the confidence interval should be read together with the point estimate. The point estimate provides one model-based summary, whereas the interval describes the range of effect sizes reasonably compatible with the analysis at the stated confidence level.
Post-operative length of stay greater than 90 days
Risk difference
95% CI: -4.0 to 0.4
Difference defined as methylprednisolone minus placebo.
Prolonged mechanical ventilation
Adjusted odds ratio
95% CI: 0.50–1.25 · P = 0.309
Prolonged mechanical ventilation was defined as greater than 7 days.
Post-operative low cardiac output syndrome
Adjusted odds ratio
95% CI: 0.52–1.57 · P = 0.723
Any other post-operative complications
Adjusted odds ratio
95% CI: 0.67–1.11 · P = 0.256
Any other post-operative complications from the start of study drug administration until hospital discharge.
| Secondary endpoint | Effect | 95% CI | P-value |
|---|---|---|---|
| Mortality, up to 30 days | OR 0.74 | 0.34–1.57 | 0.428 |
| Death or major complication, highest 7 global categories | OR 0.83 | 0.61–1.13 | 0.228 |
| Post-operative length of stay > 90 days | Risk difference -1.8 | -4.0 to 0.4 | Not reported |
| Prolonged mechanical ventilation > 7 days | OR 0.79 | 0.50–1.25 | 0.309 |
| Post-operative low cardiac output syndrome | OR 0.91 | 0.52–1.57 | 0.723 |
| Any other post-operative complications | OR 0.86 | 0.67–1.11 | 0.256 |
The registry describes the secondary analyses involving odds ratios as superiority analyses. For the sample-size rationale attached to these analyses, it states that 1200 patients, with 600 per trial group, were estimated to provide more than 90% power to detect superiority of methylprednisolone.
8. Safety
The registry reports serious adverse events by treatment arm using affected participants over participants at risk.
| Arm | Serious adverse events | At risk |
|---|---|---|
| Methylprednisolone | 2 | 599 |
| Placebo / saline | 0 | 601 |
These counts describe the number of participants affected and the corresponding number at risk in each arm. The registry summary does not provide a formal comparative effect estimate or confidence interval for this serious-adverse-event comparison, so the counts should not be converted into an implied hypothesis test.
9. Statistical Methods Explained
Why was logistic regression used?
The posted analyses classify the primary endpoint and several secondary endpoints as binary outcomes. Logistic regression is designed for binary outcomes and estimates the relationship between treatment assignment and the odds of the outcome. It can also incorporate covariates when an adjusted analysis is specified.
What does an odds ratio of 0.86 mean?
An odds ratio of 0.86 means that the estimated odds in the methylprednisolone group are 0.86 times those in the comparator group under the fitted model. Expressed as a simple relative comparison, this is a 14% lower estimated odds. It is not equivalent to saying that the probability of the outcome is 14% lower.
Why report risk difference as well as odds ratio?
Odds ratios and risk differences describe effects on different scales. An odds ratio is relative and compares odds, while a risk difference describes an absolute difference in probabilities. For a severity threshold, a risk difference such as -3.8 directly describes a percentage-point difference when interpreted using the registry's stated definition of methylprednisolone minus placebo.
How should a confidence interval be interpreted?
A 95% confidence interval quantifies statistical precision around an estimated effect under the analysis framework. For the primary odds ratio, the interval is 0.71–1.05. Because it crosses 1, the interval does not exclude the null value on the odds-ratio scale. For risk differences, the corresponding null value is 0; all seven reported primary risk-difference intervals cross zero.
Why does the p-value not measure effect size?
A p-value reflects how compatible the observed result is with a specified null hypothesis under the statistical model. It depends on both the estimated effect and the amount of information available. Consequently, a small p-value is not a measure of clinical magnitude, and a larger p-value does not establish that the treatment effects are exactly equal.
What does the global rank endpoint add?
The global rank approach allows mortality, major complications, and prolonged hospitalization to contribute to a common severity ordering. This can retain information that would be discarded by analyzing mortality alone. At the same time, the clinical meaning of an overall odds ratio depends on how the ranks are defined and how the binary analysis maps that ordered clinical structure into a statistical comparison.
10. Understanding the Primary Result
The primary adjusted odds ratio of 0.86 is the central statistical summary of the posted global-rank analysis. Its interpretation should begin with the endpoint itself rather than with the numerical estimate in isolation. The endpoint incorporates outcomes ranging from postoperative length of stay through progressively severe complications and operative mortality.
Point estimate
The estimate of 0.86 is below 1, corresponding to lower estimated odds in the methylprednisolone group under the reported logistic model.
Precision
The 95% CI of 0.71–1.05 spans the null value of 1, indicating uncertainty about whether the underlying odds are lower or higher.
Statistical test
The two-sided p-value is 0.14. It is evidence from a hypothesis test, not a direct estimate of treatment magnitude.
Absolute scale
The reported risk differences range from -0.8 to -3.8 across the specified global-rank thresholds, with all corresponding confidence intervals crossing zero.
A useful way to read the complete primary analysis is therefore to consider three pieces together: the direction of the point estimate, the width and location of the confidence interval, and the clinical definition of the endpoint. Looking only at the odds ratio would obscure the fact that the registry also provides absolute-scale differences at clinically meaningful severity thresholds.
The threshold results are directionally consistent in the sense that all seven reported risk differences are negative. However, their confidence intervals all include zero. The estimates therefore do not provide an interval excluding no difference at any of the reported thresholds.
11. Multiplicity and Interpretation Across Thresholds
The primary endpoint generates multiple statistical summaries because the global rank can be examined at several severity cut points. That structure matters when interpreting the collection of estimates as a whole.
The registry reports one registered primary endpoint, but the posted statistical analyses include eight primary analyses: one adjusted odds-ratio analysis and seven risk-difference analyses at specific global-rank thresholds. The registry summary does not provide a multiplicity-adjustment procedure for these eight analyses.
12. Limitations
Composite severity endpoint
The primary outcome combines postoperative length of stay, major morbidity, and mortality. This increases the amount of clinical information represented by the endpoint, but it also makes the overall effect less directly interpretable than an effect on one specific outcome.
Binary summaries of a ranked outcome
The registry classifies the primary endpoint as binary for the posted statistical analyses. The risk-difference analyses similarly evaluate whether participants fall at or above particular severity thresholds. These summaries are useful but do not preserve every detail of the original rank ordering in a single numerical effect estimate.
Confidence-interval overlap with the null
The primary odds-ratio confidence interval includes 1, while every reported primary risk-difference confidence interval includes 0. This limits the precision with which the registry results distinguish a treatment effect from no difference at the stated confidence level.
Limited adjustment detail in the posted summary
The registry identifies logistic regression and adjusted odds ratios but does not specify the adjustment variables in the statistical-analysis fields represented here. The numerical estimates should therefore be interpreted as the reported adjusted estimates without assuming a particular covariate set.
Secondary endpoints
The secondary analyses cover mortality, major complications, prolonged hospitalization, prolonged ventilation, low cardiac output syndrome, and other postoperative complications. These endpoints address clinically distinct questions and should not be collapsed into one overall conclusion.
13. Why This Trial Matters Statistically
STRESS illustrates a recurring challenge in clinical-trial biostatistics: important postoperative outcomes are multidimensional. Mortality is unequivocally severe, but a patient can also experience prolonged hospitalization, mechanical ventilation, organ failure, neurologic deficit, reoperation, or other major complications without dying.
A global-rank endpoint attempts to preserve that clinical hierarchy. Instead of treating every nonfatal complication as equivalent, the registry explicitly orders outcomes by severity, with postoperative length of stay occupying ranks 1 through 90 and increasingly severe clinical events occupying ranks 91 through 97.
The trial also demonstrates why reporting both relative and absolute effect measures is valuable. The adjusted odds ratio provides a model-based relative comparison, while risk differences at defined thresholds show the estimated absolute percentage-point separation between treatment groups.
Finally, the trial provides a practical example of why statistical interpretation should extend beyond a single p-value. The primary estimate, confidence interval, effect scale, endpoint construction, number of related analyses, and model specification all contribute to understanding what the result actually says.
Learn more about clinical trial statistics
Explore the statistical concepts that help explain randomized trials, binary endpoints, effect measures, and uncertainty.
14. Related Tutorials
Learn more about the methods used in this trial: