Comparative Effectiveness of Readmission Reduction Interventions for Individuals With Sepsis or Pneumonia: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Comparative Effectiveness of Readmission Reduction Interventions for Individuals With Sepsis or Pneumonia (NCT04829188), a randomized na study. The primary outcome is Post-discharge Home Days.

2. Clinical Question

3. Trial Design

The registry describes a randomized, double-masked, parallel-group adaptive platform trial sponsored by the University of Pittsburgh. The record lists the study as completed and reports enrollment of 1296. It gives a start date of 2021-03-30 and a primary completion date of 2024-12-15. Phase is listed as not applicable.

The care models combine different approaches to post-discharge contact and monitoring. Structured Telephone Support involves a nurse or health plan case manager providing assessment, education, and medication review by telephone, with home care as needed and follow-up with a primary care provider within seven days after discharge. Remote monitoring questions are sent twice weekly for up to 90 days after discharge. Responses can trigger alerts screened by a nurse-staffed call center. The descriptions distinguish low- and high-intensity monitoring and standard and enhanced response teams, although the registry record does not provide the complete operational details for every team combination.

The five reported primary-outcome groups are structured telephone support; low-intensity monitoring with the standard team; high-intensity monitoring with the standard team; low-intensity monitoring with the enhanced team; and high-intensity monitoring with the enhanced team. The record does not report the randomization procedure, allocation ratio, masking procedures, or how the platform’s adaptive features operated. It also does not report whether or when treatment arms were added, stopped, or changed.

These design features matter when interpreting the comparisons. A parallel-group randomized design assigns participants to care models for comparison, but the word “adaptive” alone does not specify which adaptations were planned or how they were handled analytically. Those details are not reported in the registry record. The findings described by the primary endpoint are therefore best read as the posted summaries and model estimates for the named groups, without assuming additional design or analysis procedures.

4. Participant Flow and Accounting

The registry reports how participants moved through each stage of the study. All counts below are exactly as posted; missing counts are not inferred from other stages.

Overall Study

5. Baseline Characteristics

The table reproduces the baseline characteristics posted in the registry, which describe the participants analysed at baseline and help show whether the randomized groups were comparable.

6. Primary Endpoint

The registry lists the primary outcome as Post-discharge Home Days, measured over 90 days after discharge to home.

The primary endpoint, post-discharge home days, combines survival and time spent alive at home during the 90 days after discharge to home. A non-survivor is assigned -1. For survivors, the measure reflects the number of days alive and at home; time in hospital, skilled nursing, rehabilitation, and long-term ventilatory care facilities is excluded. Because the survivor and non-survivor values do not share a conventional unit, the registry characterizes the overall endpoint as a unitless ordinal variable.

This composite addresses a limitation of measuring readmission alone: someone who dies cannot be treated as simply having avoided readmission, and days away from home can include care settings other than a hospital. Combining survival with days at home allows the endpoint to distinguish outcomes across that ordered range. It should not, however, be interpreted as a direct measure of readmission frequency, quality of life, or the reason a participant spent time away from home.

The posted arm summaries give a median of 90.0 in each group. The interquartile ranges differ: 83.0 to 90.0 for structured telephone support, 84.0 to 90.0 for low-intensity monitoring with the standard team, 85.0 to 90.0 for high-intensity monitoring with the standard team, 83.0 to 90.0 for low-intensity monitoring with the enhanced team, and 84.0 to 90.0 for high-intensity monitoring with the enhanced team. The common median therefore does not mean the full distributions are identical. Interpretation also depends on the model-based ordinal comparisons, whose uncertainty intervals and p-values are not reported in the registry record.

7. Why No Re-analysis Is Shown

8. Statistical Interpretation

9. Statistical Methods Explained

The primary endpoint is analyzed as an ordered outcome rather than as a conventional continuous count. That choice is consistent with the registry’s description: survivors are represented by days at home, while non-survivors receive -1, making the endpoint unitless and ordinal overall. The posted summaries use medians and interquartile ranges, and the reported primary analyses use Bayesian cumulative logit or cumulative logistic regression.

What does a cumulative odds ratio describe?

General statistical explanation: an ordinal model considers the odds of being at or beyond a series of ordered outcome thresholds. A cumulative odds ratio condenses the group contrast across those thresholds, subject to the model’s structure and coding. It is not a ratio of medians, a difference in days, or a direct risk ratio. The interpretation of a value above or below 1 depends on which group is the reference and which direction is treated as a higher outcome. Although the registry identifies the groups in each analysis, it does not provide all the coding details needed to translate every estimate into an unambiguous direction of benefit.

General statistical explanation: a proportional-odds formulation assumes that the group contrast is represented consistently across the outcome thresholds. If that assumption is unsuitable, a model may allow the contrast to vary across thresholds. The registry names cumulative logit and cumulative logistic regression, but it does not report diagnostics, assumption checks, or enough model specification to determine exactly how these features were handled in each primary analysis.

What does Bayesian estimation add?

General statistical explanation: Bayesian estimation combines a statistical model with prior information to produce a probability distribution for an unknown parameter. A point estimate alone does not show how uncertain that estimate is. Intervals from the posterior distribution, along with details about priors and model specification, help readers assess the range of values compatible with the analysis. For the posted primary estimates, the registry record does not report intervals, p-values, priors, or computational diagnostics. Consequently, their precision and sensitivity cannot be evaluated from the recorded estimates alone.

The arm-level medians and interquartile ranges answer a descriptive question: where are the middle of the outcomes and the central half of each group’s values? They do not replace an adjusted or model-based comparison. Conversely, a model estimate does not reveal the full distribution or show how many participants occupy each particular outcome value. The summaries and estimates serve different purposes and should be considered together.

How should the secondary recalculations be read?

General statistical explanation: the posted recalculations for binary outcomes express a group contrast as a risk difference, with a confidence interval and an unadjusted p-value. A risk difference is an absolute contrast between event proportions. Its confidence interval describes a range of values consistent with the data and method; if it includes zero, the data do not rule out no absolute difference at the corresponding confidence level. An unadjusted p-value assesses compatibility with a null contrast under the stated test, without correcting for the many outcomes or comparisons.

These recalculations are simple, unadjusted comparisons of posted summary counts. They are not the sponsor’s Bayesian model analyses and may differ in estimand, adjustment, and modeling assumptions. For example, the sponsor’s reported odds ratios for mortality and survivor readmissions are not the same quantity as the recalculated risk differences. They should not be compared as though they were interchangeable measures.

What do the planning illustrations mean?

General statistical explanation: sample-size illustrations show how many participants per arm a future study might require under stated assumptions. The registry’s illustrations use the observed effect or a stated fraction of it, a two-sided significance level of 0.05, equal allocation, and a normal approximation. They are planning illustrations, not observed power and not the trial’s own sample-size justification. Because they are anchored to observed effects, which can be imprecise, they should not be taken as a guarantee that a future study will achieve a particular result. The registry record does not report the trial’s original sample-size justification.

For time-to-event planning, the stated convention is to give total events; the other planning illustrations give participants per arm. No time-to-event planning illustration is needed to interpret the binary recalculations described in the record. The key methodological caution is that a planning calculation depends on its assumptions and does not retroactively determine whether the completed trial was adequately powered.

10. Analyses Reported by the Sponsor

The sponsor’s primary analyses use Bayesian cumulative ordinal regression, whereas the posted arm summaries give medians and interquartile ranges. The point estimates for the four pairwise primary comparisons with structured telephone support are 0.96, 0.86, 1.01, and 0.96. No confidence or credible intervals or p-values accompany those estimates in the registry record. The record also includes further cumulative odds ratio entries, but does not make every comparison and estimand sufficiently clear for a detailed interpretation.

For mortality and survivor readmissions, the sponsor reports Bayesian odds ratios. The recalculations use unadjusted risk differences derived from the posted counts, with their own confidence intervals and p-values. These estimates answer related but not identical statistical questions: an odds ratio contrasts odds, while a risk difference contrasts event proportions on an absolute scale. Differences in modeling and group coding also limit direct comparison.

The recalculations are not a replacement for the sponsor’s prespecified analysis. The registry record does not provide sufficient detail to establish that they use the same adjustment, analysis population, or estimand as the sponsor’s models. The absence of uncertainty measures for the sponsor estimates also means the evidence cannot be ranked by apparent point-estimate size alone. Both forms of results are best described according to the method that produced them, without treating a difference between their numerical values as a contradiction.

11. Multiplicity, Missing Data and Analysis Populations

The registry record lists 16 outcome measures and 122 analyses. It does not report a multiplicity strategy specifying which outcomes or comparisons were primary for statistical testing, whether secondary analyses were exploratory, or how the possibility of chance findings across repeated analyses was addressed. The record also does not report interim analyses, stopping boundaries, or the statistical rules used to adapt the platform. The word “adaptive” does not by itself identify any of these procedures.

Missing-data handling is not reported in the registry record. There is no description of how unavailable outcome information, withdrawal, loss to follow-up, or incomplete monitoring responses were treated. The analysis populations are also not reported in enough detail to determine whether the posted group summaries use an intention-to-treat, per-protocol, or other population definition. The numbers contributing to some outcomes differ from those in the primary endpoint summaries, but the record does not explain those differences; they should not be attributed to a particular cause without additional information.

The primary sponsor estimates are presented without confidence or credible intervals and without p-values. Model specifications, prior distributions, diagnostics, and sensitivity analyses are not reported. The secondary recalculations provide unadjusted uncertainty intervals and p-values, but those do not address multiplicity or replace a prespecified analysis plan. The registry record does not report the protocol-level details needed to assess how the many reported analyses were prioritized.

These omissions do not show that the analyses were conducted incorrectly. They limit what can be assessed from the registry record. In particular, the precision of the primary estimates, the influence of missing outcomes, the robustness of findings to analytic assumptions, and the interpretation of repeated comparisons remain uncertain.

12. Safety Results

The registry reports serious adverse events as affected participants divided by participants at risk.

Deaths

Other (non-serious) adverse events

Participants with at least one non-serious event at or above the registry's 0% reporting threshold.

The safety record reports zero serious adverse events and zero other adverse events in each listed group. It separately reports deaths: 26 for structured telephone support, 18 for low-intensity monitoring with the standard team, 7 for high-intensity monitoring with the standard team, 24 for low-intensity monitoring with the enhanced team, and 14 for high-intensity monitoring with the enhanced team. These death counts are also consistent with the mortality outcome counts listed for the corresponding groups.

Deaths should not be interpreted as evidence that an intervention caused harm. The registry record does not report causes of death, attribution to the care models, or a detailed safety analysis. It also does not explain the relation between the adverse-event risk groups and the analysis populations for other outcomes. Zero reported adverse events therefore means that none were recorded in those adverse-event categories in this entry, not that no health events occurred or that all safety concerns were excluded.

Safety figures are descriptive. Differences in counts cannot be interpreted as comparative risk without considering the group sizes, follow-up, definitions, and uncertainty. The registry record does not provide a comparative adverse-event model or uncertainty estimates for these counts.

13. What the Result Does, and Does Not, Mean

The posted evidence describes a trial comparing several post-discharge care models on an outcome that combines survival with days alive at home during the 90 days after discharge. The primary group medians are all 90.0, while the lower quartiles vary. The sponsor’s Bayesian cumulative odds ratio estimates are available for some contrasts, but the registry record does not report intervals or p-values for them. On the information recorded, the estimates do not support a confident declaration that a particular model is superior, equivalent, or inferior.

The secondary recalculations add simple comparisons for mortality and survivor readmissions. Their confidence intervals include zero for the reported risk differences, but this does not establish equivalence or rule out potentially relevant differences. The recalculations are unadjusted and do not incorporate a multiplicity correction. The sponsor’s model-based odds ratios answer a different statistical question and are not a direct replacement for those risk differences.

The findings also do not show that the models prevent readmissions, improve health, or have no effect. The primary composite includes time at home and survival, and the secondary outcomes cover only some elements of the broader clinical picture. The record does not report quality-of-life results, detailed causes of death, or a complete set of results for every listed secondary endpoint.

What the registry supports is a cautious description of the available distributions, model estimates, event counts, and uncertainty in the recalculated comparisons. Broader conclusions require the complete analysis plan, clearer model details, uncertainty estimates for the sponsor analyses, and fuller outcome reporting. No care decision should be based on these registry summaries alone.

14. Why This Trial Matters Statistically

This trial raises a statistical challenge common to post-discharge research: outcomes may involve survival, readmission, and time spent in different care settings. A single readmission count cannot capture all of these dimensions. The composite endpoint places survival and time at home on an ordered scale, which can retain information about more than whether a participant was readmitted. At the same time, its unitless ordinal structure requires careful model interpretation.

The medians illustrate why distributional summaries matter. Each group has a median of 90.0, yet the lower quartiles differ. A shared median does not imply identical outcome distributions, and a model-based contrast may capture aspects of the distribution that are not visible in the median alone. However, model estimates need uncertainty intervals and clear coding to show how much confidence to place in the apparent contrasts.

The multiple care models also create an interpretive challenge. Monitoring intensity and response-team type vary, so a comparison can represent a contrast in a bundled care approach rather than the isolated effect of a single component. The registry record does not provide a factorial analysis or other information that separates component effects. Readers should therefore understand the named intervention combinations as the comparison units.

Finally, the registry lists many outcomes and analyses but provides incomplete uncertainty and multiplicity details. This makes transparency especially important: estimates, outcome definitions, analysis populations, missing-data handling, and adaptation rules all affect interpretation. The trial is statistically informative as an example of a complex, multi-arm clinical question, while the limits of the recorded analysis details constrain the strength of conclusions that can be drawn from the posted results.

15. Limitations

• The posted summary statistics are the only data these results rest on; patient-level data, full protocols and statistical analysis plans are not available.
• Recalculated results are simple, unadjusted calculations from posted summaries. They do not replace the sponsor's prespecified analysis.

Several limitations are specific to the information recorded for this trial. First, the primary endpoint is a composite with a non-survivor value of -1 and survivor values based on days at home. The registry describes it as ordinal and unitless overall, but does not report the detailed category distribution or clarify how every cumulative odds ratio maps to outcome ordering. This limits interpretation of the primary estimates beyond their numerical values.

Second, although the trial is described as an adaptive platform, the registry does not report adaptation rules, interim analyses, stopping criteria, or whether any changes to the set of comparisons occurred. The randomization and masking procedures are also not described in the outcome information. These missing design details make it difficult to assess the operational context of the reported comparisons.

Third, the primary Bayesian estimates are not accompanied by uncertainty intervals, p-values, or model diagnostics. The secondary results available for recalculation are limited to certain outcomes and comparisons, and the unadjusted results do not address the large number of outcomes and analyses in the record. The record does not report multiplicity adjustment, missing-data methods, or analysis-population definitions.

Fourth, the registry lists more secondary outcomes than have detailed results in the record. The detailed results described include mortality and survivor readmissions at 7-day and 30-day time points, but not a full set of results for all listed outcomes. The record also does not explain why outcome-specific group counts differ. These reporting gaps prevent a complete assessment of consistency across endpoints and follow-up definitions.

Finally, the care arms combine several components. The registry descriptions do not provide enough detail to isolate the contribution of monitoring intensity, response-team type, or other care elements. The findings should therefore be understood as comparisons among the named care models, not as separate estimates of each component’s effect.

16. Related Tutorials

17. Related Statistical Calculators

18. Sources

• ClinicalTrials.gov: Comparative Effectiveness of Readmission Reduction Interventions for Individuals With Sepsis or Pneumonia — NCT04829188.
• Registry record last updated 2026-10-26; status last verified 2026-07.
• Results first posted 2026-10-26.
• Data retrieved from ClinicalTrials.gov on 2026-11-09.
• Group sizes, means, event counts and sponsor-reported p-values are reproduced as posted. Effect estimates, confidence intervals and p-values not attributed to the sponsor are independently calculated from the posted summary statistics.

19. Record Summary

NCT04829188 is a completed, randomized, double-masked, parallel-group adaptive platform trial sponsored by the University of Pittsburgh. It enrolled 1296 people hospitalized with sepsis or lower respiratory tract infection. The primary outcome was post-discharge home days, an ordinal composite of survival and days alive at home during the 90 days after discharge.

The primary summaries report a median of 90.0 in each of five care-model groups, with interquartile ranges extending from 83.0 to 85.0 at the lower quartile and 90.0 at the upper quartile. Bayesian cumulative odds ratios are reported, but uncertainty intervals and p-values are not. Available secondary results include mortality and survivor readmissions at 7-day and 30-day time points; simple recalculations are unadjusted and do not account for multiplicity.

The registry record does not report key details on missing data, analysis populations, adaptation rules, or multiplicity. The posted results are therefore descriptive and incomplete for judging the precision and robustness of comparative effects.