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
NICE-SUGAR was a randomized, parallel-group phase 4 treatment trial evaluating intensive insulin therapy versus conventional insulin therapy in patients with hyperglycemia and critical illness. The registry records 6,104 enrolled participants and one primary endpoint: all-cause mortality at 90 days.
| Feature | NICE-SUGAR |
|---|---|
| Trial acronym | NICE-SUGAR |
| Full title | Normoglycaemia in Intensive Care Evaluation and Survival Using Glucose Algorithm Regulation (NICE - SUGAR STUDY) |
| Phase | Phase 4 |
| Condition | Hyperglycemia; Critical Illness |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | None |
| Primary purpose | Treatment |
| Enrollment | 6,104 |
| Number of arms | 2 |
| Interventions | Intensive Insulin Therapy (other); Conventional Insulin therapy (other) |
| Lead sponsor | The George Institute |
| Sponsor type | OTHER |
| Status | Completed |
| Start date | 2005-04 |
| Primary completion date | 2008-11 |
| ClinicalTrials.gov | NCT00220987 |
2. Clinical Question
The central statistical question is whether assignment to intensive insulin therapy versus conventional insulin therapy is associated with a difference in all-cause mortality at 90 days among patients with hyperglycemia and critical illness.
Population
Patients represented by the registry condition categories of hyperglycemia and critical illness.
Intervention
Intensive Insulin Therapy, recorded in the registry as an "other" intervention type.
Comparator
Conventional Insulin therapy, also recorded in the registry as an "other" intervention type.
Primary question
Does randomized assignment to intensive insulin therapy versus conventional insulin therapy affect all-cause mortality at 90 days?
The registry structure is important statistically because the treatment comparison is randomized and the design is parallel-group. Randomization establishes the framework for comparing outcomes according to assigned treatment, while the parallel design means that the two treatment strategies are evaluated concurrently rather than sequentially in the same participants.
3. Trial Design
Intensive Insulin Therapy
- Randomized treatment strategy
- Intervention type recorded as "other"
- Evaluated within the parallel-group design
Conventional Insulin therapy
- Randomized comparator strategy
- Intervention type recorded as "other"
- Evaluated concurrently with intensive insulin therapy
Because the registry records no masking, the trial is best understood as an open-label randomized comparison at the design level. That does not prevent statistical analysis, but it is relevant when considering outcomes that could be influenced by knowledge of treatment assignment. For an endpoint such as all-cause mortality, however, the outcome itself is less dependent on subjective assessment than outcomes such as symptom scores or investigator-assessed response.
4. Endpoints
| Endpoint | Registry definition | Time frame |
|---|---|---|
| Primary endpoint | all-cause mortality | 90 days |
The registry identifies only one registered primary endpoint: all-cause mortality, assessed over a 90-day time frame. No additional primary endpoint is listed in the registry information represented here.
5. Statistical Methodology
The registry identifies the primary endpoint but does not post a statistical analysis for it. For a randomized trial with all-cause mortality measured at 90 days, several standard approaches are possible depending on the prespecified statistical analysis plan and the information collected during follow-up.
Binary 90-day mortality analysis
If the endpoint is analyzed strictly as mortality status at day 90, each participant can be classified as having died by 90 days or as not having died by 90 days. The treatment groups can then be compared using an appropriate two-group method for binary outcomes, with the treatment effect expressed through a risk ratio, risk difference, odds ratio, or another prespecified measure.
An absolute risk difference compares the two probabilities directly, whereas a risk ratio expresses the relative mortality probability between treatment groups.
Time-to-event analysis
If the analysis retains the exact timing of death and accounts for censoring, all-cause mortality can instead be treated as a time-to-event endpoint. Kaplan-Meier estimation can describe survival over time, a log-rank test can compare survival distributions, and a Cox proportional-hazards model can estimate a hazard ratio.
Here, S(t) represents the probability of remaining alive beyond time t. A time-to-event analysis uses the timing of events rather than reducing every participant to a single yes/no outcome at one time point.
Which approach is preferable?
That depends on the prespecified analysis plan. If the scientific question is specifically mortality status at 90 days, a binary analysis directly targets that question. If information about the timing of death and censoring is important, survival analysis can use more of the observed follow-up information.
The registry information available here does not specify which statistical test, effect measure, covariate adjustment strategy, stratification factors, or survival model was prespecified. Those details should therefore not be treated as established features of the trial design.
6. Statistical Methods Explained
Why is randomization statistically important?
Randomization creates treatment groups through an allocation process that is not determined by participants' baseline characteristics or investigators' treatment preferences. In expectation, this balances both measured and unmeasured prognostic factors across groups. The treatment comparison can therefore be interpreted as a comparison of outcomes under the randomized strategies, subject to adherence, missing data, crossover, and other post-randomization issues.
What does a 90-day mortality endpoint actually measure?
A 90-day all-cause mortality endpoint asks whether a participant has died from any cause within the specified 90-day period. It is an absolute clinical outcome rather than a surrogate measurement. The time frame is essential: mortality by 90 days and mortality over a longer period are different estimands and should not be treated as interchangeable.
Why might mortality be analyzed with Kaplan-Meier methods?
Deaths do not necessarily occur at the same time. Kaplan-Meier estimation uses the observed event times and retains information from participants who remain alive at their last observed follow-up. This is especially useful when follow-up times differ or when some observations are censored before the event occurs.
What does a hazard ratio represent?
If a Cox model were used, the hazard ratio would describe the relative instantaneous event rate between the treatment groups under the fitted model. An HR below 1 would correspond to a lower estimated hazard in the numerator treatment group; an HR above 1 would correspond to a higher estimated hazard. It would not mean that the same percentage of participants experienced a proportional change in their individual probability of death.
Why is an absolute risk difference useful?
An absolute risk difference directly compares the probabilities of death by the specified time point. It can be easier to translate into the number of additional or fewer events per population of interest than a relative effect measure. Relative measures and absolute measures therefore provide complementary information.
What does a confidence interval tell us?
A confidence interval describes statistical uncertainty around an estimated treatment effect under the specified analysis framework. A narrower interval indicates greater statistical precision than a wider interval. It does not describe the range of effects experienced by individual patients, nor does it give the probability that a particular hypothesis is true.
What does a p-value tell us?
A p-value measures the compatibility of the observed data with a specified null hypothesis under the assumptions of the statistical test. It is not a measure of effect size, clinical importance, or the probability that one treatment is better than another. Effect estimates and their confidence intervals are needed to understand the magnitude and precision of a treatment comparison.
7. Planned Analysis
The registry identifies all-cause mortality at 90 days as the primary endpoint. No posted statistical analysis is present in the ClinicalTrials.gov record represented here, and results have not been posted on ClinicalTrials.gov.
Primary endpoint
Time frame: 90 days
The primary comparison would ordinarily evaluate mortality between the randomized intensive-insulin and conventional-insulin groups using a prespecified binary or time-to-event method.
How the endpoint would typically be analyzed
For a fixed 90-day mortality endpoint, the principal analysis would generally compare the proportion of participants who died by day 90 between the randomized treatment groups. The analysis could report an absolute risk difference and a relative effect such as a risk ratio or odds ratio, together with a confidence interval.
If the prespecified analysis treats death as a time-to-event outcome, Kaplan-Meier survival curves would provide estimates of survival over time. A log-rank comparison could test the survival distributions, and a Cox proportional-hazards model could estimate a hazard ratio with a confidence interval. The proportional-hazards assumption would need to be considered when interpreting a single Cox hazard ratio.
Estimand
The registered endpoint is mortality from any cause within the 90-day time frame. The treatment comparison is anchored to randomized assignment.
Analysis population
The registry information represented here does not specify the exact analysis population or any deviations from the randomized population.
Effect measure
The registry does not specify whether the primary treatment effect was to be expressed as a risk difference, risk ratio, odds ratio, hazard ratio, or another measure.
Adjustment
The registry information represented here does not specify covariate adjustment, stratification, or a prespecified regression model for the primary endpoint.
8. Interpreting the Trial Design Statistically
The NICE-SUGAR design combines several features that are important when evaluating a randomized treatment comparison. The allocation is randomized, the design is parallel, and the primary purpose is treatment. Together, these features create a straightforward framework for comparing outcomes between two concurrently assigned strategies.
Randomization separates treatment assignment from prognosis
In an observational comparison, patients receiving intensive versus conventional treatment might differ systematically before treatment begins. Randomization is intended to prevent treatment selection from being driven by those prognostic differences. This is the fundamental statistical advantage of the trial design.
Parallel-group allocation preserves a contemporaneous comparator
Because the two interventions are evaluated in parallel, each treatment group provides a contemporaneous reference for the other. This reduces the dependence on historical control data, which can be affected by changes in patient populations, supportive care, diagnostic practice, and other factors over time.
Open-label treatment has different implications for different endpoints
The absence of masking is relevant because knowledge of treatment assignment can affect behavior, co-interventions, assessment, and decisions made during follow-up. The magnitude of that concern depends on the endpoint. An objective endpoint such as all-cause mortality is generally less vulnerable to subjective measurement than an endpoint requiring judgment or patient self-report.
9. Missing Data and Censoring
Missingness and censoring are important considerations for any mortality analysis, particularly when the endpoint is tied to a fixed follow-up period. A participant whose vital status is known through 90 days can be classified directly for a binary 90-day endpoint. In a time-to-event analysis, a participant who has not experienced death by the last known follow-up may contribute censored information rather than being treated as if death had occurred.
The registry information represented here does not specify the missing-data strategy, censoring rules, imputation method, or sensitivity analyses used for the primary endpoint. Those choices can affect statistical inference and should be distinguished from the underlying randomized comparison.
10. Analysis Population and the Intention-to-Treat Principle
Randomized clinical trials commonly distinguish the treatment effect associated with assignment from the effect associated with actually receiving or adhering to treatment. An intention-to-treat analysis preserves the randomized comparison by analyzing participants according to their assigned group, whereas an as-treated or per-protocol analysis can answer a different question and may reintroduce selection bias.
The registry information represented here confirms randomized allocation but does not specify the exact primary analysis population. Therefore, the statistical distinction is more important than assigning an unsupported label to the trial's analysis set.
The closer an analysis remains to randomized assignment, the more directly it estimates the effect of being assigned to the intervention strategy rather than the effect among a selected subset of participants who adhered to treatment.
11. Multiplicity and Multiple Endpoints
The registry identifies one primary endpoint: all-cause mortality at 90 days. A single registered primary endpoint simplifies the interpretation of the principal confirmatory question because there is no second registered primary endpoint in the information represented here that must share a familywise type I error budget.
That does not mean multiplicity is irrelevant to the broader trial. Secondary endpoints, subgroup comparisons, repeated analyses, and multiple exploratory questions can each create opportunities for chance findings. Their interpretation depends on how they were prespecified and whether multiplicity was formally controlled.
No multiplicity adjustment strategy is specified in the registry information represented here. Consequently, the existence of the single registered primary endpoint should not be taken as evidence about the treatment of every other statistical comparison conducted during the study.
12. Interim Analysis and Adaptive Features
No interim analysis plan is specified in the registry information represented here. Similarly, the registry information does not establish an alpha-spending approach, stopping boundary, adaptive randomization rule, or other interim-monitoring method.
If a trial performs an interim efficacy analysis, the statistical framework ordinarily needs to account for repeated looks at accumulating data. Without adjustment, repeatedly testing the same null hypothesis can increase the overall probability of a false-positive conclusion. Group-sequential methods and alpha-spending procedures are common ways to control that error while permitting prespecified interim decisions.
What would be prespecified?
An interim framework would normally identify information times, decision boundaries, type I error allocation, and the circumstances under which enrollment or treatment might stop early.
Why it matters
The nominal p-value from a single analysis may not have its usual interpretation if the data were repeatedly examined without an appropriate sequential design.
13. Stratification and Covariate Adjustment
Randomized trials can stratify randomization or adjust analyses for important baseline variables. Stratification can improve balance on selected prognostic factors and can increase precision when the analysis is aligned with the randomization scheme.
The registry information represented here does not specify any stratification factors or prespecified covariates for the primary analysis. It therefore would not be appropriate to assign particular adjustment variables to the NICE-SUGAR statistical model without additional documentation.
14. Non-Inferiority, Equivalence, and Superiority Logic
The registered information does not identify a non-inferiority or equivalence design. The primary endpoint is all-cause mortality at 90 days, but the registry information represented here does not provide a non-inferiority margin or an equivalence margin.
This distinction matters because non-inferiority trials are interpreted against a prespecified clinically acceptable margin rather than simply asking whether a two-sided test produces a p-value below a conventional threshold. Without a documented margin, a non-inferiority conclusion cannot be constructed from the registry information alone.
15. Bayesian Methods
The registry information represented here does not specify a Bayesian analysis. No prior distribution, posterior probability, Bayesian decision threshold, or other Bayesian component is identified.
A Bayesian mortality analysis, if prespecified, would combine observed data with a prior distribution to produce a posterior distribution for the treatment effect. That is conceptually different from a frequentist confidence interval and should not be described using frequentist probability language.
16. Safety and Serious Adverse Events
The requested statistical interpretation of safety outcomes depends on the safety data available for the trial. The registry information represented here identifies the two treatment strategies and the primary mortality endpoint, but it does not provide serious adverse-event counts by treatment arm.
17. Clinical Interpretation vs Statistical Interpretation
Statistical interpretation
The trial provides a randomized parallel-group framework for comparing intensive insulin therapy with conventional insulin therapy on all-cause mortality at 90 days. The registry identifies the endpoint and time frame but does not post its statistical results.
Clinical interpretation
The clinical question concerns mortality in a population characterized by hyperglycemia and critical illness. The relevance of the trial therefore centers on whether the two therapeutic strategies differ with respect to the prespecified 90-day mortality outcome.
These two levels of interpretation should remain separate. Statistical methodology describes how a treatment comparison is estimated and tested. Clinical interpretation considers what that endpoint represents in the context of critical illness. Neither level can be completed responsibly by substituting an unsupported numerical result for the actual trial outcome.
18. Important Limitations and Interpretation Issues
- No posted statistical results: the ClinicalTrials.gov record represented here does not contain a posted statistical analysis for the registered primary endpoint.
- Limited registry methodology: the registry information does not specify the primary statistical test, effect measure, analysis population, covariate adjustment, or censoring strategy.
- Open-label design: the registry records no masking, so treatment assignment was not masked according to the registry record. This can matter more for subjective outcomes than for all-cause mortality.
- Fixed time point versus survival analysis: mortality at 90 days can be framed as a binary endpoint, whereas a time-to-event analysis incorporates the timing of death and censoring.
- Missing data: the registry information does not specify the handling of missing 90-day vital-status information or the sensitivity analyses used to evaluate assumptions about missingness.
- Unspecified stratification: no primary-analysis stratification factors are identified in the registry information represented here.
- Unspecified interim monitoring: the registry information does not establish whether interim efficacy analyses or alpha-spending procedures were used.
- Unspecified multiplicity procedures: the registry identifies one primary endpoint, but no broader multiplicity strategy is specified for other possible analyses.
- Analysis population: the registry confirms randomized allocation but does not provide the exact analysis-population definition for the primary endpoint.
19. Why This Trial Matters Statistically
NICE-SUGAR is a useful statistical teaching case because the underlying question is clinically direct while the choice of analysis has meaningful consequences. A single endpoint—90-day all-cause mortality—can be represented as a binary outcome or as a time-to-event outcome, and those representations answer related but different questions.
| Concept | How it appears in NICE-SUGAR |
|---|---|
| Randomization | Participants were randomized between two treatment strategies. |
| Parallel-group design | The intensive and conventional insulin strategies were evaluated concurrently. |
| Primary endpoint | All-cause mortality. |
| Time frame | 90 days. |
| Binary outcome analysis | Mortality status can be compared at the prespecified 90-day time point. |
| Survival analysis | If event timing is available, mortality can be analyzed as a time-to-event endpoint. |
| Kaplan-Meier estimation | Can describe survival probabilities over time when time-to-event data are analyzed. |
| Hazard ratio | Can summarize a relative event-rate comparison under a Cox proportional-hazards model. |
| Confidence interval | Quantifies uncertainty around an estimated treatment effect. |
| Intention-to-treat principle | Provides a framework for preserving the randomized treatment comparison. |
| Missing data | Potentially important for determining 90-day mortality status and censoring. |
| Multiplicity | Relevant to secondary, subgroup, and exploratory comparisons even with one registered primary endpoint. |
20. The Statistical Meaning of a Mortality Comparison
If mortality proportions are available for the two randomized groups, the risk difference is obtained by subtracting the mortality probability in one group from that in the other. This expresses the treatment contrast directly on the probability scale.
A risk ratio compares the mortality probabilities multiplicatively. A hazard ratio, by contrast, compares instantaneous event rates within a time-to-event model. These measures should not be treated as interchangeable.
A p-value addresses compatibility with a null hypothesis under a specified test. It does not tell the reader whether a treatment effect is clinically important or how large that effect is.
A confidence interval provides an interval description of statistical uncertainty around an effect estimate. It should be considered alongside the point estimate rather than used as a substitute for it.
21. Binary Mortality vs Time-to-Event Mortality
| Feature | 90-day binary analysis | Time-to-event analysis |
|---|---|---|
| Primary quantity | Whether death occurred by 90 days | Time until death or censoring |
| Typical effect measures | Risk difference, risk ratio, odds ratio | Hazard ratio; survival probability |
| Timing of death | Collapsed into the 90-day status | Retained in the analysis |
| Censoring | Depends on how unresolved 90-day status is handled | Explicitly incorporated through survival methods |
| Clinical interpretation | Difference in mortality by a fixed time point | Difference in survival experience over follow-up |
The distinction is not merely technical. Suppose two treatment groups had the same proportion of deaths by day 90 but deaths occurred substantially earlier in one group. A binary endpoint could regard those groups as identical at day 90, whereas a time-to-event analysis could detect a difference in the timing of events. Conversely, if the scientific question is specifically mortality status at 90 days, the fixed-time endpoint may be the most direct estimand.
22. What Would Make a Mortality Result Persuasive?
A statistically informative report of the NICE-SUGAR primary endpoint would ideally present the number of participants analyzed in each randomized group, the number who died by 90 days, the corresponding mortality proportions, an effect estimate on an interpretable scale, and a confidence interval. If time-to-event analysis were prespecified, survival estimates and the corresponding hazard-ratio analysis would provide additional information.
The analysis should also identify the population used, describe how missing vital-status information was handled, and state whether any adjustment or stratification was prespecified. These details determine exactly what quantity the reported estimate represents.
Magnitude
How large is the estimated difference between intensive and conventional insulin therapy?
Precision
How much statistical uncertainty surrounds the treatment-effect estimate?
Analysis population
Which randomized participants contribute to the primary comparison, and how were deviations handled?
Clinical meaning
Does the observed mortality difference represent a clinically meaningful change in outcomes, rather than merely a statistically detectable difference?
23. Trial Timeline
Trial start
The NICE-SUGAR study began in April 2005 according to the registry record.
Primary completion
The registry records November 2008 as the primary completion date.
Registry status
The ClinicalTrials.gov record identifies the study status as completed.
24. Statistical Questions for Readers
Question 1
Is the scientific estimand mortality status at exactly 90 days, or is the entire time-to-death experience the intended target?
Question 2
Would an absolute mortality difference be more clinically interpretable than a relative effect measure?
Question 3
If a Cox model is used, is the proportional-hazards assumption reasonable over the relevant follow-up period?
Question 4
How were participants with incomplete 90-day vital-status information handled?
Question 5
Was the primary analysis performed according to randomized assignment, and how were treatment deviations handled?
Question 6
Were any secondary or subgroup analyses adjusted for multiplicity or treated as exploratory?
25. Sources
- ClinicalTrials.gov: NCT00220987.
Continue through Clinical Biostats
Explore additional clinical trial analyses, statistical tutorials, and quantitative tools for understanding clinical research methods.
26. Record Summary
NICE-SUGAR is a randomized phase 4, parallel-group treatment trial enrolling 6,104 participants with hyperglycemia and critical illness. Its registered primary endpoint is all-cause mortality at 90 days, comparing intensive insulin therapy with conventional insulin therapy. Statistically, the trial provides a clear framework for examining how randomized treatment assignment can be connected to a clinically meaningful mortality endpoint.
The most important analytical distinction is between a fixed-time binary mortality analysis and a time-to-event survival analysis. The first focuses directly on whether death occurred by 90 days; the second retains information about when deaths occurred and can accommodate censoring. Whichever approach is prespecified, the interpretation should include the treatment-effect estimate, its uncertainty, the analysis population, and the assumptions underlying the method.
The registry identifies the randomized design, two treatment arms, lack of masking, 90-day primary endpoint, enrollment of 6,104 participants, and the trial's 2005-04 start and 2008-11 primary completion dates. It does not post a statistical analysis of the primary endpoint, so the appropriate statistical reading of the registry is a description of the planned mortality question and the methods ordinarily used to analyze such an endpoint rather than a numerical conclusion about the treatment effect.