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
EPaNIC was a randomized phase 4 clinical trial evaluating the effect of withholding parenteral nutrition during the first week of ICU stay versus providing parenteral nutrition to complete enteral nutrition in adult critically ill patients.
| Feature | EPaNIC |
|---|---|
| Trial acronym | EPaNIC |
| NCT identifier | NCT00512122 |
| Phase | Phase 4 |
| Status | Active, not recruiting |
| Condition | Critical Illness; Starvation |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Single |
| Primary purpose | Treatment |
| Enrollment | 4640 |
| Number of arms | 2 |
| Lead sponsor | KU Leuven |
| Sponsor type | Other |
2. Clinical Question
The central clinical question is whether withholding parenteral nutrition during the first week of ICU stay differs from providing parenteral nutrition to complete enteral nutrition with respect to the registered primary endpoint: length of stay in ICU and length of stay in the hospital.
Population
Adult critically ill patients, with Critical Illness and Starvation listed as the trial conditions.
Intervention
Withholding PN during the first week of ICU stay.
Comparator
Oliclinomel N71000 OR N71000E, or Clinimix N17G35 OR N17G35E.
Primary question
How do the randomized treatment strategies compare with respect to ICU and hospital length of stay?
3. Trial Design
Withholding parenteral nutrition
- Withholding PN during the first week of ICU stay.
- Intervention type: other.
Parenteral nutrition completing enteral nutrition
- Oliclinomel N71000 OR N71000E.
- Clinimix N17G35 OR N17G35E.
- Intervention type: drug.
The registry describes the study as randomized, parallel, single-masked, and treatment-focused. It does not report additional design features such as a crossover procedure, factorial structure, stratification factors, or a Bayesian component.
4. Endpoints
| Endpoint | Registry definition | Time frame |
|---|---|---|
| Primary endpoint | Length of stay in ICU and length of stay in the hospital. | 2 years |
The wording of the registered primary endpoint combines two related length-of-stay outcomes: length of stay in ICU and length of stay in the hospital. The record does not specify in the information reported here whether these were intended to function as co-primary endpoints, a composite endpoint, or two components of one primary endpoint family.
5. Statistical Methodology
The primary endpoint consists of two duration-based outcomes. For a randomized comparison of length of stay, the statistical analysis should preserve the distinction between time spent in the ICU and time spent in the hospital rather than treating either quantity as a simple binary response.
Descriptive analysis of length of stay
Length-of-stay data are often right-skewed: many patients may have relatively short stays while a smaller number remain hospitalized substantially longer. For that reason, descriptive summaries should include measures of central tendency and dispersion that remain informative when the distribution is not approximately symmetric. Medians and interquartile ranges are often useful, while means can also be informative when interpreted alongside the distribution.
Two related primary outcomes
ICU length of stay and hospital length of stay answer related but distinct questions. ICU length of stay measures time in the intensive-care setting, whereas hospital length of stay measures time in the hospital overall. A patient can therefore contribute differently to these two outcomes.
Potential comparative methods
For an ordinary continuous-outcome analysis, a two-group comparison could use a parametric model when its distributional assumptions are reasonable, or a rank-based method when the data are substantially non-normal. Regression modeling can also adjust for prespecified baseline variables when justified by the design and analysis plan.
For length-of-stay outcomes specifically, time-to-discharge methods may sometimes be appropriate because discharge is an event occurring over time. The appropriate model depends on how death, discharge alive, competing risks, censoring, and the precise estimand are defined. Those details are not reported in the registry information available for this record.
Here, Y represents a length-of-stay outcome. The choice of mean difference, median difference, ratio, rank-based effect, or time-to-event estimand changes the scientific question being answered.
Multiplicity
Because the registered primary endpoint contains both ICU length of stay and hospital length of stay, multiplicity is an important design consideration. If both outcomes are tested separately as confirmatory endpoints, the overall type I error may require prespecified control. The registry information does not specify an alpha-allocation or multiplicity strategy.
Missing data
Length-of-stay analysis also requires clear rules for participants whose follow-up is incomplete or whose disposition prevents a conventional stay from being observed. The registry information does not report a missing-data or imputation strategy.
6. Planned Analysis
What would normally be analysed?
The registered primary endpoint would normally be analysed by comparing the two randomized groups with respect to ICU length of stay and hospital length of stay. Because length-of-stay variables are frequently skewed, the analysis should first characterize their empirical distributions and then use a prespecified method appropriate to those distributions.
If a mean-based estimand were selected, the treatment effect could be expressed as a mean difference. If a median-based or rank-based estimand were selected, the resulting measure would answer a different question. A time-to-discharge analysis would instead describe the rate at which patients reach discharge, requiring explicit treatment of censoring and competing events.
The key statistical issue is therefore not simply whether one group has a smaller average stay. It is defining the estimand first and then selecting an analysis that matches that estimand, the distribution of the outcome, and the clinical events that can prevent or alter discharge.
| Primary outcome | Typical statistical focus | Important considerations |
|---|---|---|
| ICU length of stay | Two-group comparison of duration | Skewness, extreme stays, discharge process, competing events |
| Hospital length of stay | Two-group comparison of duration | Skewness, prolonged hospitalization, discharge process, competing events |
7. Statistical Methods Explained
Why is length of stay not automatically analysed with a t-test?
A t-test compares means under assumptions about the sampling distribution and independence of observations. Length-of-stay data can be strongly right-skewed, so a simple mean comparison may not adequately describe the distribution. A t-test may still be reasonable under some circumstances, particularly with a sufficiently large randomized sample and appropriate robustness, but the choice should be justified rather than automatic.
Why might the median be useful?
The median identifies the middle observation and is less influenced by a small number of very long stays than the mean. If a minority of patients experience prolonged hospitalization, the mean and median can tell different stories. Reporting both can make the distribution easier to understand.
Why might a rank-based test be considered?
A rank-based comparison does not require the same normality assumptions as a conventional mean-based test. It can be useful when the length-of-stay distribution is substantially skewed. However, a rank-based result is not automatically a test of a difference in means; its interpretation depends on the underlying estimand and distributional assumptions.
Why might a time-to-event model be useful for discharge?
Length of stay can be viewed as the time until discharge. A time-to-event framework can therefore model the timing of discharge while accommodating incomplete observation in ways that ordinary continuous-outcome methods may not. The choice becomes more complicated when death or another event prevents the discharge event from occurring in the same way for every patient.
Why are ICU stay and hospital stay separate outcomes?
They represent different stages of healthcare utilization. ICU length of stay concerns the intensive-care setting specifically, while hospital length of stay encompasses the broader hospitalization. A treatment strategy could affect one duration differently from the other, so the two outcomes should not be silently collapsed into a single measure.
Why does multiplicity matter here?
If ICU length of stay and hospital length of stay are both tested as primary outcomes, each statistical test creates an opportunity for a false-positive finding. A prespecified multiplicity strategy can control the overall type I error. The registry information does not identify such a strategy for EPaNIC.
8. Interpretation of a Potential Length-of-Stay Effect
A treatment effect for length of stay must be interpreted in the units and estimand used by the analysis. For example, a mean difference describes a difference in average stay, while a ratio describes relative duration and a time-to-discharge hazard ratio describes the relative instantaneous rate of discharge under the specified model.
A shorter average or median stay does not by itself establish why the difference occurred. Length of stay is influenced by the clinical course as well as the discharge process, and its interpretation depends on how deaths and other events are handled statistically.
If a treatment-effect estimate were reported, its confidence interval would describe statistical uncertainty around that estimate under the chosen model and sampling framework. A narrow interval indicates greater precision than a wide interval, but the interval does not describe the range of outcomes experienced by individual patients.
A p-value evaluates compatibility of the observed data with a specified null hypothesis under the statistical model. It does not quantify how large a treatment effect is. For length-of-stay outcomes, the magnitude and clinical meaning of the estimated difference should therefore be examined separately from statistical significance.
9. Analysis Population and Randomization
EPaNIC was randomized, with 4640 participants enrolled across two parallel arms. Randomization is important statistically because it establishes treatment groups through an allocation process rather than allowing treatment selection to be determined by observed clinical characteristics.
| Design feature | Registry information | Statistical implication |
|---|---|---|
| Allocation | Randomized | Supports comparison of outcomes according to randomized treatment assignment. |
| Design model | Parallel | Participants are assigned to one of two treatment strategies rather than sequentially receiving both. |
| Number of arms | 2 | The primary comparison is between two randomized strategies. |
| Masking | Single | The registry identifies masking, but does not specify the masked party in the information reported here. |
| Enrollment | 4640 | A large randomized population can provide substantial information about treatment differences, although precision depends on the outcome distribution and event structure. |
The registry does not report an analysis-population definition, such as an explicit intention-to-treat population or an as-treated safety population. It also does not report baseline stratification factors.
10. Missing Data, Censoring, and Competing Events
Length-of-stay endpoints require particular attention to how patient disposition is handled. If all participants have a fully observed discharge date, a conventional continuous analysis may be possible. If some observations are incomplete, the analysis needs an explicit rule for censoring or missingness.
Death presents an additional conceptual issue for a time-to-discharge analysis because a patient who dies has not experienced discharge in the same way as a patient who leaves the hospital alive. Treating death as ordinary censoring can answer a different question from one that treats death as a competing event.
The registry information does not specify how missing length-of-stay observations, censoring, death, or competing risks are handled statistically. These choices can materially affect the interpretation of a length-of-stay analysis.
Continuous-outcome perspective
The observed duration is analysed directly, with assumptions made about its distribution and the handling of incomplete observations.
Time-to-discharge perspective
Discharge is treated as an event occurring over time, making censoring and competing events central to the estimand.
11. Interim Analysis, Crossover, and Bayesian Methods
The registry information does not report an interim efficacy analysis, a crossover procedure, a factorial design, or Bayesian statistical methods.
This distinction matters because these features would change how the primary comparison should be interpreted. For example, repeated interim looks can require type I error control, crossover can complicate interpretation of randomized treatment effects, and Bayesian analyses require explicit prior and posterior specifications. None of those details are reported for this record.
12. Trial Timeline
Trial start
The registered study start date is 2007-08.
Primary completion
The registered primary completion date is 2011-02.
Active, not recruiting
The ClinicalTrials.gov record currently identifies the study status as ACTIVE_NOT_RECRUITING.
13. Limitations
- No posted statistical results: ClinicalTrials.gov does not report treatment-effect estimates, confidence intervals, or p-values for the registered primary endpoint.
- Endpoint structure: the registry groups ICU length of stay and hospital length of stay under one primary endpoint without specifying whether they are co-primary outcomes, components of a composite, or a primary endpoint family.
- Analysis method: the registry information does not specify the statistical test or regression model used for length of stay.
- Multiplicity: no alpha-allocation or multiplicity strategy is reported for the two length-of-stay outcomes.
- Missing data: no imputation or missing-data strategy is reported.
- Censoring and competing events: the registry information does not specify how incomplete observation or events such as death would be handled in an analysis of length of stay.
- Analysis populations: the registry information does not provide explicit definitions of intention-to-treat, per-protocol, or safety analysis populations.
- Stratification: no randomization stratification factors are reported in the information available for this record.
- Masking detail: the record identifies the study as single-masked but does not specify the masked party in the reported information.
- Baseline characteristics: detailed baseline characteristics are not reported in the registry information used here.
14. Why This Trial Matters Statistically
EPaNIC is a useful statistical teaching case because its primary endpoint looks simple but raises several important questions about estimands and outcome distributions. Length of stay is not merely another continuous variable: it reflects a clinical process that unfolds over time and can be affected by discharge decisions, prolonged hospitalization, incomplete observation, and events that alter the possibility of discharge.
| Concept | How it appears in EPaNIC |
|---|---|
| Randomization | 4640 participants were allocated randomly to two treatment strategies. |
| Parallel-group design | The trial has two parallel arms. |
| Continuous duration outcome | The primary endpoint includes ICU and hospital length of stay. |
| Skewed outcomes | Length-of-stay variables commonly require attention to non-normal distributions. |
| Estimand selection | Mean difference, median difference, relative duration, and time-to-discharge effects answer different questions. |
| Multiplicity | Two related length-of-stay outcomes create an important type I error consideration if tested separately. |
| Missing data | Incomplete observation requires explicit handling rules. |
| Competing events | A time-to-discharge framework must distinguish discharge from events that prevent discharge. |
| Analysis populations | The registry does not provide detailed definitions of the populations used for analysis. |
| Interpretability | A statistical difference in duration must be translated into a clearly defined clinical estimand. |
15. Statistical Reasoning: From Endpoint to Estimand
A strong analysis begins by asking exactly what treatment effect is intended to be estimated. The phrase "length of stay" can support several distinct statistical questions.
Mean difference
How much shorter or longer is the average stay under one randomized strategy compared with the other?
Median difference
How much do the middle observed stays differ between the randomized groups?
Relative duration
What is the ratio of typical or modeled stay duration between the two strategies?
Time to discharge
Does one randomized strategy lead to discharge at a different rate over time?
These are not interchangeable. Two treatment groups can have similar medians but different means if one group contains a small number of very prolonged stays. Likewise, a time-to-discharge analysis can produce a different interpretation from an analysis of observed durations.
For this reason, the statistical method should follow the estimand rather than the other way around. The registry identifies the endpoint but does not specify the estimand or statistical model in the information reported here.
16. Sources
- ClinicalTrials.gov: EPaNIC — NCT00512122.
Continue through the Clinical Biostats knowledge graph
Explore clinical-trial methodology, statistical concepts, and analysis workflows across the Clinical Biostats site.
17. Record Summary
EPaNIC is a randomized phase 4, parallel-group trial with 4640 enrolled participants and two treatment arms. Its registered primary endpoint is length of stay in ICU and length of stay in the hospital, with a time frame of 2 years. Statistically, the central challenge is defining the appropriate estimand for these duration outcomes and then selecting an analysis that accounts for their distribution, incomplete observation, and the clinical events that can affect discharge.
The registry establishes the randomized design and the primary endpoint, but does not provide posted formal statistical analyses or numerical outcome results for the endpoint. Consequently, the most informative statistical interpretation at the registry level is methodological: understanding how ICU and hospital length of stay should be defined, summarized, compared, and interpreted in a randomized clinical trial.