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
SALT-ED was a randomized, parallel-group, single-masked clinical trial comparing 0.9% saline with physiologically-balanced isotonic crystalloid in the emergency department setting. The registered primary endpoint was hospital-free days to day 28.
| Feature | SALT-ED |
|---|---|
| Study acronym | SALT-ED |
| Registry identifier | NCT02614040 |
| Phase | NA |
| Status | COMPLETED |
| Study start | 2016-01-01 |
| Primary completion | 2017-06-30 |
| Enrollment | 14,000 |
| Allocation | RANDOMIZED |
| Design model | PARALLEL |
| Masking | SINGLE |
| Primary purpose | TREATMENT |
| Conditions | Critical Illness; Acute Kidney Injury |
| Lead sponsor | Vanderbilt University |
| Sponsor type | OTHER |
2. Clinical Question
The registered comparison asks whether the choice between 0.9% saline and a physiologically-balanced isotonic crystalloid is associated with a different number of days alive and free of hospitalization during the first 28 days after enrollment.
Population
Patients enrolled in the SALT-ED trial in the emergency department setting, with the registry identifying critical illness and acute kidney injury as the study conditions.
Intervention
Physiologically-balanced isotonic crystalloid.
Comparator
0.9% saline.
Primary question
How does the randomized choice of crystalloid relate to hospital-free days through day 28?
3. Trial Design
0.9% Saline
- Intervention type: other
- Randomized treatment arm
Physiologically-balanced isotonic crystalloid
- Intervention type: other
- Randomized treatment arm
The registry describes the study as randomized, parallel, and single-masked. A parallel design means that participants are assigned to one of the treatment strategies and the randomized groups are compared with one another rather than having the same participant receive both randomized strategies in a crossover sequence.
4. Trial Timeline
Study start
The registry lists January 1, 2016 as the study start date.
Primary completion
The registry lists June 30, 2017 as the primary completion date.
Study status
The ClinicalTrials.gov record lists the study status as completed.
5. Endpoints
| Endpoint | Time frame | Registry definition |
|---|---|---|
| Hospital-free days to day 28 | 28 days after enrollment | The number of days alive and free of hospitalization in the first 28 days after study enrollment. Patients alive at the time of discharge will be presumed to be alive at 28 days. A patient who dies before hospital discharge will receive zero hospital-free days. A patient who remains in the hospital 28 days after enrollment will receive zero hospital-free days. |
Why the endpoint is statistically unusual
Hospital-free days is not simply a measure of length of stay. It combines survival and hospitalization status into a single number. The registry definition creates a particularly important ordering: death before discharge produces zero hospital-free days, and remaining hospitalized through day 28 also produces zero hospital-free days.
This means that the endpoint contains information from different clinical states. A patient discharged early can accumulate many hospital-free days, whereas a patient who dies before discharge receives zero. The resulting distribution can therefore differ substantially from a conventional normally distributed continuous outcome.
6. Statistical Methodology
Randomized treatment comparison
The randomized allocation is the foundation of the treatment comparison. In a randomized parallel-group trial, the treatment assignment is determined before the outcome is observed, allowing the comparison between randomized groups to estimate the effect of assignment under the trial's design assumptions.
The registry identifies two randomized treatment strategies: 0.9% saline and physiologically-balanced isotonic crystalloid. The primary outcome is then evaluated over the same 28-day timeframe for the randomized groups.
Hospital-free days as a composite outcome
Hospital-free days can be viewed as a hierarchical clinical outcome because both mortality and hospitalization contribute to the final value. The definition prevents a patient who dies early from receiving a large number of hospital-free days merely because the observation period continues after death.
The registry definition assigns zero hospital-free days to a patient who dies before hospital discharge or remains hospitalized at day 28.
Distribution of the endpoint
A key statistical issue is that hospital-free days can be bounded and may contain a concentration of observations at zero. That makes ordinary methods based on an approximately normal outcome potentially inappropriate unless their assumptions are adequately supported.
For outcomes of this type, analysts may consider rank-based methods, ordinal models, count-based models, or other approaches that account for the bounded and potentially skewed distribution. The appropriate choice depends on the prespecified statistical analysis plan, the observed distribution, and the exact inferential target.
What would normally be reported
A complete statistical analysis of hospital-free days would normally report a treatment-group summary of the endpoint, the prespecified between-group effect measure, a confidence interval where applicable, and the hypothesis-testing procedure specified before unblinding. For a bounded outcome, it is also useful to describe the distribution rather than relying only on a single mean.
7. Planned Analysis
The registry identifies hospital-free days to day 28 as the primary endpoint and defines the outcome over the first 28 days after enrollment. The endpoint is the number of days alive and free of hospitalization during that period, with zero hospital-free days assigned to patients who die before hospital discharge or remain hospitalized at day 28.
For an endpoint with this structure, a typical analysis would compare the distributions of hospital-free days between the randomized treatment groups. A rank-based comparison can be appropriate when the outcome is skewed or bounded, while ordinal or count-based regression approaches may be considered when their assumptions and interpretation match the prespecified estimand. If the endpoint contains substantial zero inflation, that feature should be incorporated into the statistical reasoning rather than treated as an ordinary continuous measurement.
Primary estimand
The clinically relevant target is the difference between randomized crystalloid strategies in hospital-free days through day 28.
Time horizon
The registered endpoint is explicitly limited to the first 28 days after enrollment.
Outcome structure
The outcome simultaneously reflects survival and freedom from hospitalization.
Registry reporting
No formal statistical analyses are posted to ClinicalTrials.gov for this endpoint in the record represented here.
8. Statistical Methods Explained
Why is hospital-free days different from ordinary length of stay?
Length of stay measures time spent in the hospital. Hospital-free days instead counts the portion of a fixed 28-day period during which a patient is alive and not hospitalized. Its definition therefore incorporates information about both hospitalization and mortality.
Why does the zero value matter?
Zero is not simply a low value on a continuous scale. Under the registry definition, zero can arise because a patient dies before hospital discharge or because the patient remains hospitalized through day 28. This creates an important concentration of clinically meaningful observations at the lower boundary.
Why might a rank-based analysis be considered?
Rank-based methods do not require the outcome to follow a normal distribution. That can be useful for bounded, skewed outcomes such as hospital-free days. A rank-based comparison evaluates the relative ordering of observations rather than relying on the arithmetic mean alone.
Could an ordinal model be used?
Potentially. Hospital-free days has a natural ordering from fewer to more days, so an ordinal approach can sometimes be used when its proportional-odds or related assumptions are appropriate. The resulting effect measure would have a different interpretation from a simple mean difference, so the estimand must be specified clearly.
Could a count model be used?
Hospital-free days is numerically a count of days, but it is not an unrestricted count outcome: it is bounded by the 28-day observation period and has clinically defined zero values. A Poisson or negative-binomial model therefore should not be selected merely because the endpoint is expressed as a number of days. Its assumptions and estimand would need to fit the actual outcome distribution.
Why is mortality embedded in the endpoint?
The endpoint prevents death before discharge from being treated as if the patient simply had a very long hospitalization followed by no further observation. Assigning zero hospital-free days incorporates death directly into the outcome definition and makes mortality part of the clinical interpretation of the endpoint.
9. Interpreting the Primary Endpoint
A higher number of hospital-free days would indicate that, over the 28-day period, patients spent more of the observed time both alive and outside the hospital. The interpretation is therefore broader than a statement about hospital length of stay alone.
A difference in hospital-free days would not by itself identify whether the difference arose primarily from mortality, duration of hospitalization, or a combination of the two. Those components would need to be examined separately if the relevant data and analyses were available.
Two treatment groups could have the same mean hospital-free days while having very different distributions. One group could contain many patients clustered around intermediate values, while another could contain more patients at zero and more at the upper end. Distributional summaries therefore provide important context.
If a formal hypothesis test is reported, its p-value addresses compatibility with the null hypothesis under the specified statistical model. It does not quantify the clinical magnitude of a difference. The effect estimate and its uncertainty are needed to understand the size and precision of the treatment comparison.
10. Confidence Intervals and Statistical Precision
A confidence interval is useful because a point estimate alone does not communicate how precisely the treatment effect has been estimated. For a hospital-free-days analysis, the interval should correspond to the chosen effect measure, such as a difference in means, a median difference, an odds ratio, or another prespecified estimand.
The interpretation of the interval depends on that effect measure. For example, a confidence interval around a mean difference describes uncertainty in the estimated difference in average hospital-free days, whereas a confidence interval around an ordinal-model odds ratio describes uncertainty in a different quantity.
The estimate describes the observed treatment contrast under the chosen analysis. The confidence interval describes uncertainty around that estimate under the corresponding statistical framework.
11. Analysis Population and Randomization
The registry identifies the study as randomized but does not provide a separate analysis-population definition in the record summarized here. In randomized clinical-trial analysis, the population used for the primary efficacy comparison is important because excluding randomized participants after treatment assignment can affect the protection reported by randomization.
An intention-to-treat analysis generally compares participants according to their randomized assignment. This preserves the treatment-assignment contrast created by randomization. Per-protocol or as-treated analyses answer different questions and can be more vulnerable to post-randomization differences between groups.
Randomized comparison
Randomization establishes the treatment-assignment comparison that underlies causal interpretation.
Post-randomization events
Treatment exposure, discharge, hospitalization, and death can all affect the observed hospital-free-days distribution.
12. Missing Data and Censoring
The hospital-free-days definition itself fixes the observation horizon at 28 days and specifies how two clinically important situations are handled: death before hospital discharge and continued hospitalization through day 28 both result in zero hospital-free days.
That is statistically useful because those events do not become ordinary missing observations simply because the patient does not complete a conventional outpatient follow-up assessment. However, any other missing information would still require a prespecified handling strategy.
13. Multiplicity
The ClinicalTrials.gov record represented here identifies one registered primary endpoint: hospital-free days to day 28. No additional primary endpoints, multiplicity-adjustment procedure, or hierarchical testing strategy is reported in the provided registry information.
Multiplicity becomes important when several hypotheses are tested and interpreted as confirmatory. Testing many outcomes can increase the chance of obtaining a statistically significant result by chance alone. When a trial has a single primary endpoint, the multiplicity problem is different from a design built around multiple co-primary or multiple primary endpoints.
14. Interim Analysis
The registry information represented here does not report an interim statistical analysis, interim efficacy boundary, alpha-spending procedure, or stopping rule. These design elements are important in trials that examine accumulating outcome data before the planned final analysis because repeated unadjusted testing can alter the overall type I error rate.
When interim monitoring is part of a prespecified design, the final inferential procedure should account for the interim looks. Without a documented interim-analysis framework, a statistical reader should not assume that an ordinary final-analysis p-value has been adjusted for interim monitoring.
15. Non-Inferiority, Crossover, and Bayesian Methods
| Design topic | Registry information | Statistical implication |
|---|---|---|
| Non-inferiority margin | Not reported in the provided registry information. | No non-inferiority margin can be identified from the registry record represented here. |
| Crossover | The design model is PARALLEL. | The registered design is parallel rather than a crossover design. |
| Factorial design | Not reported; the design model is PARALLEL with two arms. | No factorial treatment structure is identified. |
| Bayesian methods | Not reported. | No Bayesian analysis can be identified from the registry information represented here. |
16. Safety and Secondary Endpoints
The registry information represented here identifies one registered primary endpoint: hospital-free days to day 28. It does not provide posted statistical analyses for secondary endpoints or a set of serious adverse-event counts by randomized arm.
Safety analysis is conceptually separate from the primary efficacy endpoint. For randomized trials, safety outcomes are generally summarized by treatment exposure, adverse-event category, severity, seriousness, and relevant time period. The appropriate denominator and analysis population should be stated explicitly because they determine how treatment-group safety rates are interpreted.
17. What Hospital-Free Days Captures
The primary endpoint is particularly useful for illustrating why clinical outcomes sometimes combine several dimensions of patient experience into one prespecified measure.
Survival component
A patient who dies before hospital discharge receives zero hospital-free days under the registry definition.
Hospitalization component
A patient who remains hospitalized 28 days after enrollment also receives zero hospital-free days.
Discharge component
A patient alive at discharge is presumed to remain alive at day 28 for purposes of the endpoint.
Fixed horizon
The endpoint is evaluated over a fixed 28-day period rather than an open-ended follow-up interval.
This construction means that hospital-free days can be clinically intuitive while remaining statistically nonstandard. A statistical analysis needs to respect the way the endpoint was defined rather than treating it as an arbitrary count of hospital days.
18. Statistical Interpretation of the Study Design
| Concept | How it appears in SALT-ED |
|---|---|
| Randomization | Participants were assigned through a randomized allocation to one of two treatment arms. |
| Parallel design | The registry identifies a parallel-group design rather than a crossover design. |
| Single masking | The registry identifies the study as single-masked. |
| Two-arm comparison | The trial has two registered interventions: 0.9% saline and physiologically-balanced isotonic crystalloid. |
| Fixed follow-up horizon | The primary endpoint is measured over 28 days after enrollment. |
| Composite clinical information | Hospital-free days incorporates survival and hospitalization status. |
| Bounded outcome | The endpoint is constrained by the 28-day observation period. |
| Zero-valued clinical state | Death before discharge and hospitalization through day 28 both produce zero hospital-free days. |
19. Limitations
- Endpoint complexity: hospital-free days combines information about survival and hospitalization, so a single summary measure may not identify which component drives a treatment-group difference.
- Bounded distribution: the outcome is restricted by the 28-day horizon and can have a concentration of observations at zero.
- Interpretation of averages: a mean difference can conceal important differences in the distribution of hospital-free days.
- Analysis-method dependence: the interpretation of an effect estimate depends on the statistical method used, such as a rank-based comparison, ordinal model, or count-based model.
- Registry detail: the record represented here does not report a formal statistical analysis for the primary endpoint.
- Safety information: arm-specific serious-adverse-event counts are not reported in the registry information represented here.
- Additional design details: the registry information represented here does not specify a non-inferiority margin, interim-analysis framework, multiplicity procedure, or Bayesian method.
20. Why This Trial Matters Statistically
SALT-ED is a useful statistical teaching case because it illustrates a problem that appears frequently in clinical research: the most clinically meaningful outcome is not always a conventional continuous, binary, or time-to-event variable.
Hospital-free days has an intuitive interpretation, but its statistical structure requires careful attention. The endpoint has a fixed upper time horizon, clinically meaningful zeros, and an interaction between survival and hospitalization. These characteristics affect both descriptive summaries and the choice of inferential method.
| Statistical lesson | Why it matters |
|---|---|
| Define the estimand first | The treatment effect must be tied to a clearly defined contrast in hospital-free days. |
| Respect the outcome distribution | Bounded and skewed outcomes may violate assumptions of ordinary parametric methods. |
| Distinguish clinical states | Zero hospital-free days can represent different clinical pathways with the same numerical value. |
| Separate efficacy from safety | The primary endpoint does not summarize all potential treatment benefits and harms. |
| Report uncertainty | An effect estimate without a measure of precision provides incomplete statistical information. |
| Use the registered timeframe | The primary endpoint is specifically defined for the first 28 days after enrollment. |
21. A Deeper Look at the Estimand
An estimand describes precisely what treatment effect a trial is trying to quantify. For SALT-ED, the registered primary endpoint gives an important part of that definition: the outcome is hospital-free days measured during the first 28 days after enrollment.
The endpoint definition also establishes how death and persistent hospitalization are incorporated. This is important because different ways of handling those outcomes could lead to different statistical questions. A simple length-of-stay analysis, for example, would not be equivalent to the registered hospital-free-days endpoint.
The statistical estimator used to quantify that contrast can vary. The clinical interpretation should remain anchored to the registered endpoint definition.
22. Why Endpoint Definition Should Precede Statistical Testing
Statistical testing is meaningful only after the outcome and treatment contrast have been defined. For SALT-ED, the registry provides a detailed operational definition of hospital-free days, including explicit rules for death before discharge, continued hospitalization at day 28, and survival at discharge.
Those rules matter because changing them would change the endpoint itself. For example, treating death as missing rather than assigning zero hospital-free days would produce a different estimand and potentially a different statistical conclusion.
23. Registry Record and Evidence Status
The ClinicalTrials.gov record identifies SALT-ED as a completed randomized, parallel-group, single-masked treatment trial with 14,000 enrolled participants and one registered primary endpoint. The primary endpoint is hospital-free days to day 28, defined over the 28 days after enrollment.
The registry record represented here does not contain posted statistical analyses for that endpoint. Consequently, the available statistical evidence in the record consists of the study design and prespecified endpoint definition rather than a posted treatment-effect estimate with an associated confidence interval and p-value.
24. Sources
- ClinicalTrials.gov: NCT02614040 — SALT-ED.
- PubMed: PubMed record.
- PubMed: PubMed record.
- PubMed: PubMed record.
Continue through the Clinical Biostats knowledge graph
Clinical trials provide practical examples of how endpoint definitions, randomized designs, and statistical estimands shape the interpretation of medical evidence.
25. Record Summary
SALT-ED is a randomized, parallel-group, single-masked clinical trial with 14,000 enrolled participants comparing 0.9% saline with physiologically-balanced isotonic crystalloid. Its registered primary endpoint, hospital-free days to day 28, is a particularly instructive outcome because it integrates survival and hospitalization into a fixed 28-day measure. The registry explicitly assigns zero hospital-free days to patients who die before hospital discharge or remain hospitalized at day 28, while patients alive at discharge are presumed to remain alive at day 28.
Statistically, the endpoint highlights the importance of defining the estimand before selecting a model. Hospital-free days is bounded, potentially skewed, and contains clinically meaningful zero values. Rank-based, ordinal, or count-based approaches may be considered depending on the prespecified analysis and the observed distribution, but their effect measures answer different questions. The ClinicalTrials.gov record represented here establishes the endpoint and design but does not post formal statistical analyses, effect estimates, confidence intervals, or p-values for the primary comparison.