This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record. Numerical results and trial characteristics on this page are restricted to the ClinicalTrials.gov record.
1. Trial at a Glance
ARTEMIS-1 was a randomized, parallel, triple-masked phase 3 trial evaluating the safety and efficacy of USL261 (intranasal midazolam) in patients with seizure clusters. The registry reports a superiority framework and four posted statistical analyses: one primary binary endpoint analysis and three secondary analyses, including two time-to-event analyses.
| Feature | ARTEMIS-1 |
|---|---|
| Trial | ARTEMIS-1 |
| ClinicalTrials.gov identifier | NCT01390220 |
| Brief title | Study to Evaluate the Safety and Efficacy of USL261 (Intranasal Midazolam) in Patients With Seizure Clusters |
| Phase | Phase 3 |
| Condition | Epilepsy |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Triple |
| Primary purpose | Treatment |
| Enrollment | 292 |
| Interventions | USL261 (drug); Placebo (drug) |
| Primary endpoint type | Binary |
| Hypothesis type | Superiority |
| Results posted | Yes |
| Statistical analyses posted | 4 |
| Lead sponsor | UCB Biopharma S.P.R.L. |
| Sponsor type | Industry |
| Status | Terminated |
2. Clinical Question
The primary statistical question was whether USL261 was superior to placebo for achieving the registered definition of treatment success after administration of the double-blind dose in the Comparative Phase (CP), assessed over 6 hours.
Population
Patients with seizure clusters enrolled in the ARTEMIS-1 phase 3 trial.
Intervention
USL261, identified in the registry as a drug intervention.
Comparator
Placebo, identified in the registry as a drug intervention.
Primary question
Among randomized participants who received the double-blind dose in the CP, was treatment success more frequent with USL261 CP than with Placebo CP?
3. Trial Design
USL261
- USL261 drug intervention.
- Comparative Phase analysis group identified in the statistical analysis record as USL261 CP.
- Primary and secondary efficacy analyses compared this group with Placebo CP.
Placebo
- Placebo drug intervention.
- Comparative Phase analysis group identified in the statistical analysis record as Placebo CP.
- Primary and secondary efficacy analyses compared this group with USL261 CP.
4. Primary Endpoint
| Endpoint | Registry definition / time frame | Type | Analysis |
|---|---|---|---|
| Participants Who Met the Criteria for Treatment Success After Administration of the Double-blind Dose in the Comparative Phase (CP) | 6 hours. Treatment Success is defined as achieving both of the following: 1) termination of seizure(s) within 10 minutes after double-blind study drug administration, and 2) no recurrence of seizure(s) beginning 10 minutes after study drug administration to 6 hours after study drug administration. Participants who received the open-label second dose within 6 hours of administration of the double-blind dose are addressed in the registered endpoint definition. | Binary | Fisher exact test; 2-sided |
This endpoint is binary because each participant is classified according to whether the prespecified treatment-success criteria were met. The definition combines two clinical conditions: rapid termination of seizure(s) and absence of recurrence during the subsequent observation window.
The statistical comparison is therefore a comparison of proportions of participants meeting a compound binary definition, rather than a comparison of a continuous seizure-count measure.
5. Results: Primary Endpoint
The registry reports a formal statistical analysis for the primary endpoint among randomized participants who received the double-blind dose in the CP. The comparison was USL261 CP versus Placebo CP using a 2-sided Fisher exact test.
Fisher exact test
Hypothesis type: superiority · Analysis population: randomized participants who received the double-blind dose in the CP
Groups compared: USL261 CP vs Placebo CP
The reported P = 0.0109 is the 2-sided Fisher exact test result for the primary binary treatment-success endpoint. Under the superiority framework reported in the registry, it represents evidence against the null hypothesis of no association between randomized treatment group and the binary treatment-success outcome under the test's assumptions.
The ClinicalTrials.gov record does not provide the treatment-success counts or percentages by arm, an effect estimate such as a risk ratio or odds ratio, or a confidence interval. Therefore, the statistical result should not be converted into a numerical treatment effect that is not reported.
The P-value also does not measure the magnitude of the treatment effect. A P-value is a measure of compatibility of the observed data with the null hypothesis under the specified testing framework; it is not the probability that the treatment works, the probability that the null hypothesis is true, or a measure of clinical importance.
Because the endpoint is binary, an effect estimate and confidence interval would ordinarily add important information about the size and precision of the treatment difference. Their absence from the ClinicalTrials.gov record limits how quantitatively the primary finding can be interpreted.
6. Secondary Endpoint Results
Participants With Seizure(s) >10 Minutes to 4 Hours After Administration of the Double-blind Dose
This secondary endpoint was binary and was evaluated among randomized participants who received the double-blind dose in the CP. The groups compared were USL261 CP and Placebo CP.
Fisher exact test
Time frame: 4 hours · Hypothesis type: superiority · Analysis: 2-sided
Fisher's exact test was used to compare the occurrence of the binary seizure outcome between USL261 CP and Placebo CP. The reported P = 0.0043 is a hypothesis-test result, not an effect-size estimate.
The ClinicalTrials.gov record does not provide the number or percentage of participants experiencing the event in each arm, nor a confidence interval or relative-effect estimate. Consequently, the result cannot be used to state the absolute or relative magnitude of the between-group difference from the ClinicalTrials.gov record alone.
Occurrence of Seizure With a Start Time >10 Minutes After Administration of the Double-blind Dose
This secondary endpoint was a time-to-event outcome evaluated over 24 hours. The analysis population was randomized participants who received the double-blind dose in the CP, with USL261 CP compared with Placebo CP.
Log-rank test
Time frame: 24 hours · Hypothesis type: superiority
The log-rank test compares the observed time-to-event experience between the randomized groups over the analysis period. Here, the event is the occurrence of a seizure with a start time >10 minutes after administration of the double-blind dose.
The reported P = 0.0124 does not tell us how much longer the event-free time was in one group, nor does it provide a hazard ratio. The registry analysis does not report a Kaplan-Meier estimate, median time to event, hazard ratio, confidence interval, or event count for this endpoint.
As with the primary endpoint, the P-value should not be interpreted as an effect-size measure. A time-to-event analysis also depends on how censoring and follow-up are handled; the ClinicalTrials.gov record identifies the log-rank method but does not provide enough detail to reconstruct those features.
Time to Next Seizure With a Start Time >10 Minutes After Administration of the Double-blind Dose
This secondary endpoint measured time in hours to the next seizure with a start time >10 minutes after administration of the double-blind dose, with a 24-hour time frame. The registry reports Kaplan-Meier estimates in the analysis notes and a log-rank test for the comparison.
Log-rank test
Time frame: 24 hours · Outcome unit: hours · Hypothesis type: superiority
Analysis note: Kaplan-Meier estimates.
For this endpoint, the Kaplan-Meier method provides a way to estimate the distribution of time until the next qualifying seizure while accounting for participants whose event time may not be observed within the analysis period. The log-rank test then compares the survival distributions between the randomized groups.
The reported P = 0.0124 indicates the result of that time-to-event hypothesis test. It does not itself give the difference in hours, a median time, a hazard ratio, or a confidence interval. None of those numerical effect estimates are reported in the ClinicalTrials.gov record.
It is also important not to equate a time-to-event P-value with a statement that one group had a particular percentage reduction in seizure risk. Such a claim would require an effect estimate that is not reported here.
| Endpoint | Type | Time frame | Method | P-value |
|---|---|---|---|---|
| Primary treatment success after double-blind dose in CP | Binary | 6 hours | Fisher exact | 0.0109 |
| Participants with seizure(s) >10 minutes to 4 hours after double-blind dose were analyzed as having had a seizure. | Binary | 4 hours | Fisher exact | 0.0043 |
| Occurrence of seizure with a start time >10 minutes after double-blind dose were analyzed as having had a seizure. | Time-to-event | 24 hours | Log-rank | 0.0124 |
| Time to next seizure with a start time >10 minutes after double-blind dose were analyzed as having had a seizure. | Time-to-event | 24 hours | Log-rank; Kaplan-Meier estimates | 0.0124 |
7. Statistical Methodology
Fisher exact test
The primary endpoint and one secondary endpoint are binary outcomes. The registry reports the Fisher exact test for both comparisons. Fisher's exact test evaluates the association between treatment group and a binary outcome using the observed contingency table and calculates an exact significance probability rather than relying on a large-sample approximation.
For this trial, the binary outcome is whether a participant met the endpoint definition. The exact test compares the observed allocation of participants across treatment groups and outcome categories under the null hypothesis of no treatment-group association.
Why Fisher exact testing fits a binary endpoint
A binary endpoint reduces each participant's outcome to one of two categories. The primary treatment-success endpoint is explicitly defined this way. Fisher's exact test is therefore a direct method for comparing the outcome distributions of the two treatment groups without requiring a continuous outcome model.
Log-rank test
The registry reports log-rank testing for both secondary time-to-event endpoints. Unlike a simple binary comparison at a fixed time, a time-to-event analysis uses the timing of events during follow-up and can accommodate right-censored observations.
For the ARTEMIS-1 endpoint measuring time to the next qualifying seizure, the registry specifically notes Kaplan-Meier estimates and reports a log-rank test.
Kaplan-Meier estimation
Kaplan-Meier estimation describes the probability of remaining event-free over time. Its key advantage is that participants who do not experience the event during observed follow-up can still contribute information up to their censoring time.
Here, di represents events at time ti, while ni represents participants at risk immediately before that time.
Superiority testing
The registry identifies the hypothesis type as superiority. In a superiority framework, the statistical question is whether the observed data provide evidence that the treatment groups differ in the prespecified direction of interest rather than whether a treatment is merely not worse than a comparator by a predefined non-inferiority margin.
Two-sided testing
The primary Fisher exact analysis is explicitly marked as 2-sided, as is the secondary Fisher exact analysis. A two-sided test considers departures from the null in either direction when defining the extremeness of the observed result.
8. Statistical Methods Explained
Why was Fisher's exact test used?
The registry classifies the primary endpoint as binary and reports Fisher exact testing. This method is designed for categorical comparisons between groups and is especially useful when the exact configuration of counts in the contingency table is the basis for inference. The ClinicalTrials.gov record does not provide the underlying 2 × 2 counts, so the exact calculation cannot be reconstructed from the record alone.
What does a binary treatment-success endpoint mean?
It means that each participant is classified according to whether both components of the registered Treatment Success definition were satisfied: seizure termination within 10 minutes and no recurrence from 10 minutes through 6 hours. The endpoint therefore summarizes a compound clinical criterion rather than measuring only the time to the first seizure termination.
Why is the log-rank test different from Fisher's exact test?
Fisher's exact test compares categorical outcomes, whereas the log-rank test compares time-to-event distributions. A time-to-event endpoint retains information about when the event occurs and can account for censoring. ARTEMIS-1 uses Fisher exact testing for binary endpoints and log-rank testing for its posted time-to-event endpoints.
What does the P = 0.0124 mean for the time-to-event endpoints?
It is the reported log-rank test result for each of the two 24-hour time-to-event analyses. It does not mean that the probability of treatment benefit is 0.0124, and it does not quantify the difference in event-free time. An effect estimate such as a hazard ratio, together with its confidence interval, would provide additional information about magnitude and precision, but those quantities are not reported here.
Why does the primary P-value not tell us the size of the treatment effect?
The primary P = 0.0109 summarizes the hypothesis-test evidence from the Fisher exact procedure. The magnitude of a binary treatment effect would normally be described using quantities such as a risk difference, risk ratio, or odds ratio, accompanied by an interval estimate. Because those values are not provided in the ClinicalTrials.gov record, the P-value should not be used as a substitute for them.
Why is a Kaplan-Meier method useful for time to next seizure?
The time-to-next-seizure endpoint records when the qualifying event occurs rather than simply whether it occurred by one fixed time. Kaplan-Meier estimation can represent the changing proportion remaining event-free over the observation period, while the log-rank test provides a group comparison of those time-to-event distributions.
9. Analysis Populations
The registry identifies the analysis population for all four posted statistical analyses as randomized participants who received the double-blind dose in the Comparative Phase (CP).
| Analysis | Population | Groups compared |
|---|---|---|
| Primary treatment success | Randomized participants who received the double-blind dose in the CP | USL261 CP vs Placebo CP |
| Seizure(s) >10 minutes to 4 hours | Randomized participants who received the double-blind dose in the CP | USL261 CP vs Placebo CP |
| Occurrence of seizure >10 minutes | Randomized participants who received the double-blind dose in the CP | USL261 CP vs Placebo CP |
| Time to next seizure >10 minutes | Randomized participants who received the double-blind dose in the CP | USL261 CP vs Placebo CP |
10. Safety Results
The ClinicalTrials.gov record reports serious adverse events by arm. These figures are presented as affected participants divided by participants at risk, with the detailed arm labels preserved from the ClinicalTrials.gov record.
| Safety analysis group | Serious adverse events | Affected / at risk |
|---|---|---|
| USL261 TDP | Serious adverse events | 2/292 |
| USL261 CP, USL261 5 mg Only | Serious adverse events | 0/91 |
| USL261 CP, USL261 5 mg + 5 mg | Serious adverse events | 0/43 |
| Placebo CP, Placebo Only | Serious adverse events | 0/26 |
| Placebo CP, Placebo + USL261 5 mg | Serious adverse events | 1/41 |
The serious-adverse-event data show the counts and denominators reported in the ClinicalTrials.gov record. They should not be collapsed into a single treatment-group rate without additional information about the relationship among the listed treatment periods and analysis populations.
11. Results in Statistical Context
ARTEMIS-1 illustrates an important distinction between testing for a difference and estimating the size of a difference. The registry posts four formal hypothesis tests, but the ClinicalTrials.gov record contains P-values without corresponding effect estimates and confidence intervals.
What is directly reported
The primary Fisher exact test has P = 0.0109. The secondary binary Fisher exact test has P = 0.0043. Both secondary log-rank tests have P = 0.0124.
What is not reported in the analyses posted on ClinicalTrials.gov
Arm-specific event counts for the efficacy endpoints, risk differences, risk ratios, odds ratios, hazard ratios, and confidence intervals are not reported.
Why that distinction matters
A small P-value can provide evidence against a null hypothesis without describing how large or clinically important the underlying treatment difference is.
Time-to-event perspective
The two log-rank analyses address when qualifying seizures occur, not merely whether an event was observed by a fixed cutoff.
12. P-values: What They Do and Do Not Tell Us
The primary analysis reports P = 0.0109 from a 2-sided Fisher exact test. This is evidence from the specified hypothesis test, but it is not an estimate of treatment benefit.
The reported P = 0.0043 is the 2-sided Fisher exact test result for participants with seizure(s) >10 minutes to 4 hours after administration of the double-blind dose.
Both posted log-rank analyses report P = 0.0124. These tests compare time-to-event distributions; they do not themselves report the magnitude of the difference in event timing.
A P-value is conditional on the statistical model or test procedure and the data being analyzed. It should therefore be read together with the endpoint definition, analysis population, direction of the hypothesis, and the design of the trial. In particular, a P-value should not be converted into a probability that the treatment is effective or into a percentage treatment benefit.
13. Multiplicity and Multiple Endpoints
The ClinicalTrials.gov record identifies 1 primary endpoint and 3 secondary statistical analyses. The primary hypothesis type is reported as superiority, and the posted analyses include one primary and three secondary tests.
| Endpoint role | Number in the ClinicalTrials.gov record | Statistical method |
|---|---|---|
| Primary | 1 | Fisher exact test |
| Secondary | 3 | Fisher exact test; log-rank test; log-rank test |
Because multiple endpoints are tested, the interpretation of the collection of P-values depends on the prespecified multiplicity strategy. The ClinicalTrials.gov record identifies the endpoint roles and methods but do not provide an alpha-allocation or multiplicity-adjustment procedure. Accordingly, the secondary P-values should not be relabeled as independently confirmatory results on the basis of the ClinicalTrials.gov record alone.
14. Censoring and Time-to-Event Analysis
Two of the four posted analyses are time-to-event analyses with a 24-hour time frame. The registry reports the log-rank test for both, and specifically notes Kaplan-Meier estimates for the endpoint measuring time to next seizure.
The statistical advantage of a time-to-event endpoint is that it retains information about event timing. Participants who remain event-free through their observed follow-up may be censored rather than treated as if their event time were known. Kaplan-Meier estimation is designed to accommodate this structure under its usual assumptions.
For ARTEMIS-1, the relevant event is a qualifying seizure according to the particular registered time-to-event definition. The registry's reported log-rank analysis compares the resulting event-time distributions between USL261 CP and Placebo CP.
The ClinicalTrials.gov record does not provide enough information to evaluate the detailed censoring rules, the number censored, or the shape of the Kaplan-Meier curves. Those elements should therefore not be reconstructed or assumed.
15. Blinding and Randomization as Statistical Design Features
Randomization
The trial is explicitly described as randomized. Randomization is important because, in expectation, it balances both measured and unmeasured baseline characteristics between treatment groups, supporting a causal interpretation of the randomized comparison when the trial is appropriately conducted and analyzed.
Triple masking
The registry identifies the masking as triple. Masking can reduce the influence of knowledge of treatment assignment on participant behavior, investigator conduct, and other aspects of trial implementation or assessment. The ClinicalTrials.gov record does not specify which three parties were masked, so the term should not be expanded beyond the registry wording.
Parallel design
The parallel design means that participants are assigned to treatment groups that are compared across the trial rather than each participant serving as their own randomized crossover sequence. This is relevant when interpreting the between-group hypothesis tests.
16. Limitations
- Limited effect-size reporting in the ClinicalTrials.gov record: the statistical analyses provide P-values but do not provide arm-specific efficacy event counts, effect estimates, or confidence intervals.
- Analysis population differs from enrollment: the posted analyses use randomized participants who received the double-blind dose in the CP, while total enrollment was 292. The ClinicalTrials.gov record does not provide the number analyzed for each endpoint.
- Secondary multiplicity: three secondary statistical analyses are posted, but the ClinicalTrials.gov record does not identify a multiplicity-adjustment strategy or hierarchical testing procedure.
- Time-to-event detail: the registry identifies log-rank testing and, for one endpoint, Kaplan-Meier estimates, but the ClinicalTrials.gov record does not include event counts, censoring counts, Kaplan-Meier estimates, medians, hazard ratios, or confidence intervals.
- Binary endpoints: the primary endpoint is a compound binary definition. A binary result does not by itself show which component of the definition drove a treatment-group difference.
- Safety interpretation: serious-adverse-event counts are reported under detailed arm labels, but the ClinicalTrials.gov record does not establish a single directly comparable safety population across all listed groups.
- Trial status: the registry profile lists the study as terminated. The ClinicalTrials.gov record does not provide a reason for termination, so no causal explanation should be inferred.
- Generalizability: the ClinicalTrials.gov record identifies the condition and trial design but does not provide a baseline-characteristics table or detailed eligibility information sufficient for a broader assessment of external validity.
17. Why This Trial Matters Statistically
ARTEMIS-1 is a useful teaching case because its posted analyses illustrate how different endpoint structures lead to different statistical methods within the same randomized trial. The primary endpoint is binary and is analyzed with Fisher's exact test, while two secondary endpoints are time-to-event outcomes analyzed with the log-rank test, with Kaplan-Meier estimates specifically identified for one of them.
| Concept | How it appears in ARTEMIS-1 |
|---|---|
| Randomization | Randomized parallel-group phase 3 design |
| Blinding | Triple-masked trial |
| Binary endpoint | Primary treatment-success endpoint at 6 hours |
| Fisher exact test | Primary endpoint and one secondary binary endpoint |
| Superiority testing | Registered hypothesis type for the posted statistical analyses |
| Two-sided testing | Explicitly reported for the primary and secondary Fisher exact analyses |
| Time-to-event analysis | Two secondary endpoints assessed over 24 hours |
| Log-rank test | Used for both posted time-to-event analyses |
| Kaplan-Meier estimation | Specifically identified for the time to next seizure endpoint |
| Analysis population | Randomized participants who received the double-blind dose in the CP |
| Multiplicity | One primary and three secondary statistical analyses are posted |
| Safety reporting | Serious adverse events reported by detailed treatment-period groups |
The strongest statistical lesson is that the endpoint determines the analysis. A participant-level binary classification calls for a categorical-data method, whereas an endpoint defined by the timing of an event calls for a time-to-event framework. Reading the P-value without first understanding that structure would lose much of the statistical meaning of the trial.
18. A Practical Reading Guide to ARTEMIS-1
Step 1: Identify the estimand-like question
Start with the exact endpoint definition. The primary question concerns meeting a compound treatment-success criterion within 6 hours after the double-blind dose.
Step 2: Identify the data type
The primary endpoint is binary. Two secondary endpoints are time-to-event outcomes. This determines why different tests appear in the registry.
Step 3: Identify the analysis population
The statistical-analysis records specify randomized participants who received the double-blind dose in the CP.
Step 4: Read the P-value correctly
The P-value addresses the specified hypothesis test. It is not a treatment-effect estimate and cannot substitute for an absent confidence interval.
This approach prevents a common mistake in trial interpretation: treating a statistically significant P-value as if it were a complete description of the treatment effect. The registry-reported ARTEMIS-1 record supports a more disciplined interpretation in which the endpoint, analysis population, method, and inferential result are considered together.
19. Sources
- ClinicalTrials.gov: NCT01390220 — ARTEMIS-1.
- PubMed: PMID 36410152.
- PubMed: PMID 31140596.
Continue through the Clinical Biostats statistical pathway
Explore the underlying concepts used to analyze randomized clinical-trial endpoints, including categorical tests, survival methods, randomization, blinding, and P-values.
20. Record Summary
ARTEMIS-1 provides a compact example of how clinical-trial statistical methodology follows endpoint structure. The randomized, parallel, triple-masked phase 3 design evaluated USL261 versus placebo in patients with seizure clusters. Its registered primary endpoint was a binary treatment-success measure assessed over 6 hours, analyzed with a 2-sided Fisher exact test and reported with P = 0.0109. Three secondary analyses were also posted: a 4-hour binary seizure endpoint analyzed with Fisher's exact test and reported with P = 0.0043, plus two 24-hour time-to-event endpoints analyzed with log-rank tests and reported with P = 0.0124.
The statistical record is informative but incomplete for effect-size interpretation. It supplies formal hypothesis-test results without the corresponding efficacy event counts, effect estimates, or confidence intervals. For that reason, the most defensible analysis is to explain what Fisher exact testing and log-rank testing establish, distinguish binary from time-to-event endpoints, identify the specified analysis population, and avoid turning P-values into unreported measures of treatment magnitude.