This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record. Numerical results on this page are restricted to the ClinicalTrials.gov record.
1. Trial at a Glance
COAT was a randomized, parallel, single-masked phase 4 trial in people with cryptococcal meningitis, HIV infections, or AIDS. The registered primary endpoint was 26-week mortality, analyzed according to the intention-to-treat principle with a Cox proportional-hazards model.
| Feature | COAT |
|---|---|
| Brief title | Cryptococcal Optimal ART Timing Trial |
| Phase | Phase 4 |
| Status | Completed |
| Therapeutic area | Infectious Disease |
| Conditions | Cryptococcal Meningitis; HIV Infections; AIDS |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Single |
| Primary purpose | Treatment |
| Enrollment | 177 |
| Lead sponsor | University of Minnesota |
| Study period | 2010-11 to primary completion in 2012-10 |
| ClinicalTrials.gov | NCT01075152 |
2. Clinical Question
The central question was whether the timing of HIV therapy affected survival and other clinical and microbiologic outcomes in people with cryptococcal meningitis.
Population
Participants enrolled in the COAT trial with cryptococcal meningitis, HIV infections, or AIDS.
Earlier HIV Therapy
The randomized arm designated Earlier HIV Therapy. The ClinicalTrials.gov record identifies efavirenz and nucleoside interventions.
Deferred HIV Therapy
The randomized comparator arm designated Deferred HIV Therapy.
Primary question
Does assignment to earlier rather than deferred HIV therapy change 26-week survival?
3. Trial Design
Earlier HIV Therapy
- Randomized earlier HIV therapy strategy.
- The intervention record identifies efavirenz as a drug and nucleoside as a biological intervention.
Deferred HIV Therapy
- Randomized deferred HIV therapy strategy.
- The ClinicalTrials.gov record identifies this as the comparator arm.
4. Primary Endpoint
| Endpoint | Registered definition | Time frame | Analysis |
|---|---|---|---|
| Mortality | Intention to treat analysis of 26 week survival of all subjects enrolled. Reported below are the numbers of participants who died by Week 26. | 26 weeks from study entry | Cox proportional-hazards model |
The endpoint is a time-to-event outcome. That distinction matters: the analysis is concerned not simply with whether death occurred, but with survival over time and the timing of events during follow-up.
5. Statistical Methodology
Intention-to-treat analysis
The primary mortality analysis followed the intention-to-treat principle. In practical terms, participants are compared according to their randomized treatment assignment rather than being reclassified according to what treatment they ultimately received. This preserves the treatment comparison created by randomization.
Cox proportional-hazards model
The registry reports Regression, Cox, normalized as a Cox proportional-hazards model, for the primary mortality analysis. The model produces a hazard ratio comparing the instantaneous event rates between the randomized groups over the analyzed follow-up.
For this trial, the reported HR is 1.73. An HR above 1 indicates a higher estimated hazard in the numerator group, here Earlier HIV Therapy, relative to Deferred HIV Therapy, under the fitted model.
Two-sided hypothesis testing
The primary analysis notes specify a two-sided type-I error with alpha=0.05. The trial was statistically powered to detect a 25% relative survival benefit corresponding to a 15% absolute benefit, with 90% power and overall two-sided alpha=0.05.
Secondary statistical methods
The registry reports a broader statistical toolkit for secondary endpoints: Fisher exact tests for categorical outcomes, Gray's method of cumulative incidence for competing-risk outcomes, repeated-measures analysis using an MMRM framework, and linear regression for microbiologic clearance. The analysis notes provide additional detail for several of these methods.
| Endpoint type | Registry method | Statistical purpose |
|---|---|---|
| Time-to-event | Cox proportional-hazards model | Compare event-time distributions using a hazard ratio |
| Competing risks | Gray's test | Compare cumulative incidence while accounting for competing events |
| Binary | Fisher exact test | Compare categorical outcome frequencies |
| Continuous longitudinal | MMRM | Compare repeated functional-status measurements |
| Continuous microbiologic outcome | Linear regression / linear mixed-effects regression | Estimate the rate of microbiologic clearance |
6. Primary Result: 26-Week Mortality
Hazard ratio for mortality
95% CI: 1.06–2.82 · P = 0.03
Earlier HIV Therapy vs Deferred HIV Therapy · 26 weeks from study entry
The primary analysis compared the randomized arms for survival using time-to-event methods under the intention-to-treat principle. The reported hazard ratio was 1.73, with a two-sided 95% confidence interval of 1.06–2.82 and a reported P-value of 0.03.
The HR of 1.73 means that, under the fitted Cox model, the estimated instantaneous mortality hazard was 1.73 times as high in the Earlier HIV Therapy group as in the Deferred HIV Therapy group over the analyzed follow-up. Equivalently, 1.73 is a 73% higher estimated hazard on that relative hazard scale.
The HR does not mean that 73% of participants died, that mortality probability was exactly 73% higher, or that every individual participant experienced the same relative increase in risk. A hazard ratio is a model-based time-to-event measure, not an absolute risk difference.
The 95% CI of 1.06–2.82 describes statistical uncertainty around the estimated hazard ratio under the analysis framework. It does not describe the range of effects experienced by individual patients. Because the interval remains above 1, the reported estimate is consistent with a higher mortality hazard in the Earlier HIV Therapy group under the prespecified two-sided testing framework.
The P-value of 0.03 addresses evidence against the null hypothesis in the specified testing framework; it is not a measure of effect size and does not indicate the probability that the treatment hypothesis is true. It should be read alongside the HR and its confidence interval.
Finally, the registry's early-stopping caveat is important. Enrollment was stopped early by the Data and Safety Monitoring Board because mortality differences were sufficiently large, and the registry specifically notes that historically, trials stopped early can over-estimate the magnitude of benefit or harm. The Cox model also relies on the proportional-hazards framework, so the HR should not be interpreted as a simple fixed risk ratio at every point in time.
7. Secondary Results
46-Week Survival
Hazard ratio for survival
95% CI: 1.03–2.68 · P = 0.04
Earlier HIV Therapy vs Deferred HIV Therapy · 46 weeks
The 46-week survival analysis used a Cox proportional-hazards model to compare the randomized groups. The reported HR was 1.66, with a 95% CI of 1.03–2.68 and P = 0.04.
An HR of 1.66 represents a higher estimated hazard in the Earlier HIV Therapy group relative to the Deferred HIV Therapy group under the reported Cox model. The confidence interval quantifies uncertainty around that estimate; it extends from 1.03 to 2.68 rather than identifying a single known underlying effect. The P-value of 0.04 is evidence against the null under the reported superiority-testing framework, but it does not quantify the magnitude or clinical importance of the difference.
Incidence of Immune Reconstitution Inflammatory Syndrome
| Outcome | Time frame | Analysis population | Method | P-value |
|---|---|---|---|---|
| Incidence of Immune Reconstitution Inflammatory Syndrome | 46 weeks | Persons who survived to initiate HIV therapy | Gray's test (competing risks) | 0.32 |
The registry explains that the cumulative-incidence analysis accounted for the competing risk of death when comparing IRIS, relapse, and adverse-event endpoints. The reported P-value for IRIS was 0.32.
Incidence of Cryptococcal Relapse
| Outcome | Time frame | Method | P-value |
|---|---|---|---|
| Incidence of Cryptococcal-relapse | 46 weeks | Gray's test (competing risks) | 0.06 |
The reported P-value was 0.06. Because death can prevent a participant from subsequently experiencing relapse, the registry used a competing-risks cumulative-incidence approach rather than treating death as ordinary noninformative censoring.
Safety of ART Initiation
| Outcome | Time frame | Method | P-value |
|---|---|---|---|
| Safety of ART Initiation | 46 weeks | Fisher exact test | 0.98 |
The registry states that categorical secondary endpoints were compared with Fisher's exact tests. The reported P-value for Safety of ART Initiation was 0.98.
HIV-1 Viral Suppression
| Outcome | Time frame | Analysis population | Method | P-value |
|---|---|---|---|---|
| HIV-1 Viral Suppression | 26 weeks | Among persons alive at 26 weeks | Fisher exact test | 0.26 |
The registry specifies that this analysis was conducted among people alive at 26 weeks. It also records that one participant in the early ART arm had consent withdrawn by their family on day 2 after study entry and that three participants in the deferred ART arm were missing their 26-week viral result in the posted analysis description.
Antiretroviral Therapy Tolerability
| Outcome | Time frame | Analysis population | Method | P-value |
|---|---|---|---|---|
| Antiretroviral Therapy Tolerability | 26 weeks | Persons who survived to initiate HIV therapy | Fisher exact test | 0.23 |
The reported P-value was 0.23. The restricted analysis population is important: this is not an all-randomized-participant comparison because the registry defines the population as people who survived to initiate HIV therapy.
Karnofsky Functional Status
| Outcome | Time frame | Analysis population | Method | P-value |
|---|---|---|---|---|
| Karnofsky Functional Status | 46 weeks | Persons alive at the time point | MMRM (mixed model for repeated measures) | 0.34 |
The registry describes a repeated-measures analysis using an unstructured covariance matrix and adjustment for the baseline value. The reported P-value was 0.34.
Microbiologic Clearance
| Outcome | Time frame | Analysis population | Method | P-value |
|---|---|---|---|---|
| Microbiologic Clearance | 4 weeks | All participants with >2 quantitative CSF cultures obtained | Linear regression; analysis notes describe linear mixed-effects regression | 0.44 |
The outcome unit was log10 CFU/mL/day. The registry's analysis description states that a linear mixed-effects regression model with a random intercept and slope estimated the rate of clearance of log10 colony-forming units of Cryptococcus per mL of CSF per day among participants with more than 2 cultures.
8. Safety Results
The registry provides serious adverse-event counts by randomized arm as affected participants over those at risk.
| Safety measure | Earlier HIV Therapy | Deferred HIV Therapy |
|---|---|---|
| Serious adverse events | 49/88 | 53/89 |
These counts are presented as affected participants over the number at risk reported by the registry. They should not be treated as an estimate of a causal treatment effect without the corresponding prespecified safety analysis framework.
9. Statistical Methods Explained
Why was a Cox proportional-hazards model used for mortality?
Mortality was registered as a time-to-event endpoint. A Cox model uses both whether an event occurred and the time at which it occurred, while accommodating participants whose event time is not observed during the relevant follow-up. The result is a hazard ratio rather than a simple proportion difference.
What does the primary hazard ratio of 1.73 mean?
The hazard ratio compares estimated instantaneous event hazards between the randomized groups. Here, 1.73 indicates a higher estimated mortality hazard for Earlier HIV Therapy relative to Deferred HIV Therapy under the fitted model. It does not mean that the probability of death was exactly 1.73 times higher at every time point.
Why does the confidence interval matter?
The estimate alone, 1.73, gives one point on the hazard-ratio scale. The 95% CI of 1.06–2.82 communicates the statistical uncertainty around that estimate. A wider interval would indicate less precision; a narrower interval would indicate greater precision. The interval is not a range containing the individual treatment effects experienced by participants.
Why was Fisher's exact test used for categorical outcomes?
Fisher's exact test is designed for comparing categorical outcomes between two groups when the sample size or cell counts may make large-sample approximations less attractive. COAT used Fisher exact testing for Safety of ART Initiation, HIV-1 Viral Suppression, and Antiretroviral Therapy Tolerability.
Why was Gray's test used for IRIS and relapse?
For IRIS, relapse, and adverse-event endpoints, death can act as a competing event: once a participant dies, that participant can no longer experience the endpoint in question. Gray's method compares cumulative incidence while explicitly accounting for such competing risks. Treating the competing event as ordinary censoring can answer a different statistical question.
Why was an MMRM used for Karnofsky Functional Status?
Karnofsky Functional Status is a repeated measurement rather than a single binary event. The registry describes an MMRM with an unstructured covariance matrix and baseline adjustment. This framework allows the analysis to account for correlation among repeated measurements from the same participant rather than treating every measurement as independent.
Why use a linear mixed-effects model for microbiologic clearance?
The microbiologic endpoint tracks a quantitative measure over time. The registry's analysis notes describe a linear mixed-effects regression model with a random intercept and slope, allowing each participant to contribute repeated measurements while modeling both participant-specific starting levels and rates of change. The outcome was the rate of change in log10 CFU/mL/day.
10. Interpreting P-values in COAT
| Endpoint | P-value | What the P-value addresses |
|---|---|---|
| 26-week Mortality | 0.03 | Evidence against the null under the reported two-sided superiority analysis |
| 46-week Survival | 0.04 | Evidence against the null for the reported Cox comparison |
| IRIS | 0.32 | Evidence against the null in the competing-risks comparison |
| Cryptococcal relapse | 0.06 | Evidence against the null in the competing-risks comparison |
| Safety of ART Initiation | 0.98 | Evidence against the null in the Fisher exact comparison |
| HIV-1 Viral Suppression | 0.26 | Evidence against the null in the Fisher exact comparison |
| ART Tolerability | 0.23 | Evidence against the null in the Fisher exact comparison |
| Karnofsky Functional Status | 0.34 | Evidence against the null in the repeated-measures comparison |
| Microbiologic Clearance | 0.44 | Evidence against the null in the reported regression analysis |
A P-value is not an effect-size measure. For example, the primary mortality analysis combines a P-value of 0.03 with an HR of 1.73 and a 95% CI of 1.06–2.82. The HR describes the estimated relative hazard; the confidence interval describes uncertainty around that estimate; and the P-value addresses compatibility with the null hypothesis under the specified statistical framework.
11. Competing Risks: Why the Analysis Changes
IRIS and cryptococcal relapse illustrate an important distinction between ordinary time-to-event analysis and competing-risk analysis. If death occurs before a participant experiences IRIS or relapse, that participant is no longer capable of experiencing the endpoint being studied.
Ordinary survival analysis
A conventional time-to-event analysis can treat an observation as censored when the event has not occurred by the end of follow-up.
Competing-risk setting
When another event prevents the endpoint from occurring, that event changes the probability structure of the endpoint. Gray's method directly addresses the cumulative-incidence comparison.
The COAT registry explicitly states that the competing risk of death was accounted for when comparing randomized groups for IRIS, relapse, and adverse events. This is particularly relevant because the primary endpoint itself was mortality.
12. Longitudinal Analysis: Karnofsky Functional Status
The Karnofsky analysis is statistically different from mortality. Mortality is a time-to-event endpoint, whereas Karnofsky Functional Status is a continuous score observed longitudinally among people alive at the relevant time point.
The registry normalizes this method as an MMRM (mixed model for repeated measures). The analysis compared overall change from baseline in Karnofsky performance status scores between groups.
The unstructured covariance specification is important conceptually because repeated measurements from the same participant are expected to be correlated. Rather than assuming all observations are independent, the model allows the covariance structure across repeated observations to be estimated.
The reported P-value was 0.34. That value should be interpreted in the context of the specified longitudinal model and analysis population of persons alive at the time point, not as a direct comparison of mortality or survival.
13. Microbiologic Clearance Analysis
Microbiologic clearance was measured over 4 weeks in log10 CFU/mL/day. The analysis population consisted of all participants with >2 quantitative CSF cultures obtained.
The registry's analysis description states that a linear mixed-effects regression model with a random intercept and slope estimated the rate of clearance.
This approach is useful because a participant can contribute multiple microbiologic measurements. The random intercept represents participant-specific differences in the starting level, while the random slope allows the rate of change to vary between participants.
The reported P-value for the comparison was 0.44. The registry does not provide an effect estimate and confidence interval for this endpoint in the ClinicalTrials.gov record, so no additional numerical effect measure is presented here.
14. Prespecified CSF White Cell Count Subgroup Analysis
The registry includes a prespecified subgroup analysis examining the percentage of participants, per CSF WBC subgroup, who died by Week 26. The analysis was restricted to people with a measured CSF white cell count at randomization, defined as Day 7–11 of amphotericin treatment.
| Outcome | Time frame | Method | Reported P-value |
|---|---|---|---|
| Percentage of Participants, Per CSF WBC Subgroup, Who Died by Week 26 | 26 weeks | Cox proportional-hazards model with an interaction term between treatment arm and subgroup | 0.02 |
The registry states that prespecified subgroups formed by baseline characteristics were compared for 26-week survival using models containing an interaction term between treatment arm and subgroup. This is an important distinction from simply performing separate analyses within each subgroup.
An interaction term asks whether the relationship between randomized treatment assignment and the outcome varies according to the subgroup variable. That is statistically different from asking whether a treatment comparison is significant within one subgroup. The reported P-value of 0.02 is associated with the posted subgroup analysis, but the ClinicalTrials.gov record does not provide the subgroup-specific hazard-ratio estimates needed to describe the magnitude of any interaction in more detail.
15. Trial Design Features That Affect Interpretation
Randomization
Randomized allocation is the foundation of the primary causal comparison. The mortality analysis preserved this design through an intention-to-treat approach.
Early stopping
Enrollment was stopped early by the Data and Safety Monitoring Board because mortality differences were sufficiently large. The registry cautions that early-stopped trials can over-estimate benefit or harm.
Different populations
Secondary analyses were not all conducted in the same population. Some were limited to survivors or people who initiated HIV therapy, while microbiologic clearance required more than 2 quantitative CSF cultures.
Different estimands
Mortality, relapse, IRIS, viral suppression, tolerability, functional status, and microbiologic clearance answer different questions and appropriately require different statistical methods.
16. Limitations
- Early stopping: the registry states that mortality differences were sufficiently large for the Data and Safety Monitoring Board to stop enrollment early. It also specifically notes the historical tendency for trials stopped early to over-estimate the magnitude of benefit or harm.
- Different analysis populations: several secondary endpoints were analyzed among restricted populations, including people alive at 26 weeks, people who survived to initiate HIV therapy, and participants with more than 2 quantitative CSF cultures.
- Hazard-ratio interpretation: the Cox model produces a relative hazard measure rather than an absolute probability or risk difference. Interpretation also depends on the proportional-hazards framework.
- Secondary P-values: the ClinicalTrials.gov record does not specify a multiplicity-adjustment procedure for the multiple secondary endpoints, so these P-values should not automatically be treated as independently confirmatory.
- Incomplete effect measures for secondary endpoints: several posted secondary analyses provide P-values without an effect estimate and confidence interval in the ClinicalTrials.gov record.
- Competing risks: endpoints such as IRIS and relapse are affected by death as a competing event, making their interpretation different from ordinary time-to-event endpoints.
- Generalizability: the ClinicalTrials.gov record identifies the trial population through the registered conditions but do not provide a detailed baseline-characteristics table. No additional baseline claims are made here.
17. Why This Trial Matters Statistically
COAT is a useful statistical teaching case because a single randomized trial incorporates several distinct families of clinical-trial methods. The primary endpoint is a time-to-event outcome analyzed with a Cox model and an intention-to-treat population, while secondary endpoints require competing-risks, exact categorical, repeated-measures, and mixed-effects approaches.
| Concept | How it appears in COAT |
|---|---|
| Randomization | Randomized allocation to earlier versus deferred HIV therapy in a parallel design |
| Intention-to-treat analysis | Primary 26-week survival analysis of all subjects enrolled according to randomized assignment |
| Time-to-event endpoints | 26-week mortality and 46-week survival |
| Hazard ratio | Primary mortality HR 1.73 and 46-week survival HR 1.66 |
| Confidence intervals | 95% CIs reported for the primary and 46-week survival hazard ratios |
| Cox proportional-hazards model | Used for mortality and survival analyses, including the CSF WBC subgroup analysis |
| Competing risks | Gray's method used for IRIS and cryptococcal relapse |
| Fisher exact test | Used for categorical safety, viral suppression, and tolerability endpoints |
| MMRM | Used for longitudinal Karnofsky Functional Status |
| Linear mixed-effects regression | Used to estimate microbiologic clearance rates |
| Early stopping | Enrollment was stopped early because mortality differences were sufficiently large |
18. Statistical Interpretation vs Clinical Interpretation
Statistical interpretation
The primary mortality analysis reported an HR of 1.73 with a 95% CI of 1.06–2.82 and P = 0.03. The 46-week survival analysis reported an HR of 1.66 with a 95% CI of 1.03–2.68 and P = 0.04.
Clinical interpretation
The statistical results describe differences between the randomized strategies in mortality and survival under the trial's analysis framework. They do not, by themselves, provide an individualized prediction of outcome or eliminate uncertainty about the magnitude of the observed effect.
The distinction is especially important because the trial was stopped early. A statistically detectable difference and a precisely estimated long-term treatment effect are not synonymous. The registry itself identifies early stopping as a caveat when interpreting the magnitude of the observed mortality difference.
19. What the Primary Hazard Ratio Does — and Does Not — Mean
The primary HR of 1.73 is a model-based estimate comparing the mortality hazards for Earlier HIV Therapy and Deferred HIV Therapy over the analyzed follow-up. On the relative hazard scale, the estimated hazard in the Earlier HIV Therapy group was 73% higher than the comparator group's estimated hazard.
It does not mean that 73% of participants died, that the probability of death increased by exactly 73%, or that every participant had a 73% increase in individual risk. It also does not provide an absolute difference in mortality probabilities.
The 95% CI of 1.06–2.82 indicates uncertainty around the point estimate. The upper limit is substantially farther from 1 than the lower limit, illustrating that the estimated magnitude is not known with perfect precision.
The P-value of 0.03 is determined by the observed data and the specified null and alternative hypotheses under the statistical model. It does not tell us whether an effect is large, small, clinically important, or likely to replicate in another population.
20. The Importance of Analysis Population
One of the most instructive features of COAT is that the statistical population changes across endpoints.
| Endpoint | Analysis population | Why it matters |
|---|---|---|
| 26-week Mortality | All subjects enrolled under ITT | Preserves the randomized comparison for the primary endpoint |
| HIV-1 Viral Suppression | Persons alive at 26 weeks | Participants who died before 26 weeks cannot contribute a 26-week viral-suppression measurement |
| ART Tolerability | Persons who survived to initiate HIV therapy | Conditions on survival and treatment initiation |
| Karnofsky Functional Status | Persons alive at the time point | Functional status at the time point requires the participant to be alive |
| Microbiologic Clearance | Participants with >2 quantitative CSF cultures | Requires sufficient longitudinal microbiologic measurements to estimate a clearance rate |
These are not interchangeable populations. Conditioning an analysis on survival or treatment initiation can change the interpretation of the comparison relative to the all-randomized ITT analysis. Consequently, the secondary results should not be read as if they were simply alternative measurements of the primary endpoint.
21. Longitudinal Trial History
Trial start
The COAT trial began enrollment in November 2010.
Primary completion
The registered primary completion date was October 2012.
Results posted
The registry record is marked completed and contains 10 posted outcome measures and 10 posted statistical analyses.
22. Secondary Endpoint Analysis Map
| Endpoint | Type | Method | Estimate / P-value reported |
|---|---|---|---|
| Mortality | Time-to-event | Cox proportional-hazards model | HR 1.73; 95% CI 1.06–2.82; P = 0.03 |
| 46-week Survival | Time-to-event | Cox proportional-hazards model | HR 1.66; 95% CI 1.03–2.68; P = 0.04 |
| Incidence of Immune Reconstitution Inflammatory Syndrome | Time-to-event | Gray's test | P = 0.32 |
| Incidence of Cryptococcal-relapse | Time-to-event | Gray's test | P = 0.06 |
| Safety of ART Initiation | Binary | Fisher exact test | P = 0.98 |
| HIV-1 Viral Suppression | Binary | Fisher exact test | P = 0.26 |
| Antiretroviral Therapy Tolerability | Binary | Fisher exact test | P = 0.23 |
| Karnofsky Functional Status | Continuous | MMRM | P = 0.34 |
| Microbiologic Clearance | Continuous | Linear regression / linear mixed-effects regression | P = 0.44 |
| CSF WBC subgroup mortality | Time-to-event | Cox model with treatment-by-subgroup interaction | P = 0.02 |
23. Why the COAT Statistical Story Is More Than One P-value
The primary result is statistically important because it integrates several layers of design and analysis: randomization, intention-to-treat analysis, time-to-event methodology, a prespecified two-sided alpha of 0.05, and a Cox-model hazard ratio with a confidence interval.
But the broader statistical story requires additional distinctions. IRIS and relapse require competing-risk methods because death prevents subsequent occurrence of those endpoints. Binary outcomes use Fisher exact tests. Functional status is longitudinal and therefore uses an MMRM. Microbiologic clearance requires repeated quantitative measurements and is modeled using a linear mixed-effects framework.
Finally, the interpretation of the primary effect must account for the registry's early-stopping caveat. The fact that a trial stopped early because mortality differences were sufficiently large is part of the evidence-generation process itself. The registry specifically warns that early-stopped trials have historically over-estimated the magnitude of benefit or harm.
24. Related Tutorials
Learn more about the methods used in this trial:
25. Related Statistical Calculators
26. Sources
- ClinicalTrials.gov: COAT — NCT01075152.
- PubMed: PMID 24963568.
- PubMed: PMID 39972326.
- PubMed: PMID 32843994.
- PubMed: PMID 31504335.
- PubMed: PMID 31319498.
Continue through the Clinical Biostats statistical tutorials
Explore the survival, competing-risk, categorical, longitudinal, and regression methods that appear across randomized clinical-trial analyses.
27. Record Summary
COAT provides a compact but statistically rich example of randomized clinical-trial analysis. The primary endpoint was 26-week mortality, analyzed by intention to treat with a Cox proportional-hazards model. The reported hazard ratio was 1.73 with a 95% CI of 1.06–2.82 and P = 0.03. A 46-week survival analysis produced an HR of 1.66 with a 95% CI of 1.03–2.68 and P = 0.04.
The secondary analyses demonstrate why endpoint-specific methodology matters. IRIS and cryptococcal relapse were evaluated with Gray's competing-risks method; categorical outcomes used Fisher exact tests; Karnofsky Functional Status used an MMRM with baseline adjustment and an unstructured covariance matrix; and microbiologic clearance used a linear mixed-effects regression approach with random intercept and slope. A prespecified CSF WBC subgroup analysis used a treatment-by-subgroup interaction in a Cox model.
The most important interpretive caveat is the trial's early stopping. The registry states that mortality differences were sufficiently large for the Data and Safety Monitoring Board to stop enrollment and explicitly cautions that historically, early-stopped trials can over-estimate the magnitude of benefit or harm. Accordingly, the COAT statistical record is best understood by considering the effect estimate, confidence interval, P-value, analysis population, statistical model, and early-stopping context together.