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 registry-reported ATEMPT trial data.
1. Trial at a Glance
ATEMPT was a randomized, parallel, unmasked phase 2 trial in breast cancer comparing trastuzumab emtansine with a regimen containing paclitaxel plus trastuzumab. The ClinicalTrials.gov record reports 512 enrolled participants, two arms, two registered primary endpoints, and three posted statistical analyses for secondary endpoints.
| Feature | ATEMPT |
|---|---|
| Trial name | ATEMPT |
| Brief title | T-DM1 vs Paclitaxel/Trastuzumab for Breast (ATEMPT Trial) |
| Phase | 2 |
| Status | COMPLETED |
| Condition | Breast Cancer |
| Allocation | RANDOMIZED |
| Design model | PARALLEL |
| Masking | NONE |
| Primary purpose | TREATMENT |
| Enrollment | 512.0 |
| Lead sponsor | Dana-Farber Cancer Institute |
| Sponsor type | OTHER |
| Start | 2013-05 |
| Primary completion | 2019-08-26 |
| ClinicalTrials.gov | NCT01853748 |
2. Clinical Question
The trial addresses a randomized treatment comparison in breast cancer between trastuzumab emtansine and a regimen containing paclitaxel plus trastuzumab. The ClinicalTrials.gov record identifies superiority as the hypothesis type for the posted statistical analyses.
Population
Participants enrolled in the ATEMPT phase 2 randomized breast cancer trial.
Intervention
Trastuzumab emtansine (T-DM1).
Comparator
Paclitaxel plus trastuzumab.
Primary questions
How does the trial assess clinically relevant toxicities and 3-year disease-free survival for trastuzumab emtansine?
3. Trial Design
Trastuzumab Emtansine
- Trastuzumab emtansine (T-DM1)
- Serious adverse events reported as 62/383 affected / at risk in the ClinicalTrials.gov record
Paclitaxel + Trastuzumab
- Paclitaxel
- Trastuzumab
- Serious adverse events reported as 26/114 affected / at risk in the ClinicalTrials.gov record
4. Registered Primary Endpoints
ATEMPT has two registered primary endpoints in the ClinicalTrials.gov record: one binary endpoint concerning clinically relevant toxicities and one time-to-event endpoint concerning 3-year disease-free survival.
5. Statistical Methodology
The ClinicalTrials.gov record identifies three the statistical methods: the Cochran-Mantel-Haenszel test, the Fisher exact test, and linear regression. All three posted statistical analyses are secondary analyses.
| Method | Data structure | Role in reported ATEMPT analyses |
|---|---|---|
| Cochran-Mantel-Haenszel test | Categorical data | Secondary analysis of patients with neuropathy, stratified by baseline neuropathy status. |
| Fisher exact test | Categorical data | Secondary analysis of grade 2–3 thrombocytopenia incidence. |
| Linear regression | Linear models | Secondary analysis of alopecia. |
Because the statistical analyses posted on ClinicalTrials.gov are all secondary, they should not be presented as the formal statistical analyses of the two registered primary endpoints. The data posted on ClinicalTrials.gov for ATEMPT do not provide primary-endpoint effect estimates, confidence intervals, or formal primary-endpoint P-values.
Binary endpoints
For a binary endpoint, the basic statistical object is a proportion: the number of participants meeting an outcome definition divided by the relevant analysis population. A treatment comparison can be expressed through a risk difference, risk ratio, odds ratio, or a hypothesis test. The appropriate choice depends on the prespecified analysis plan and the structure of the data.
Time-to-event endpoints
Disease-free survival is a time-to-event endpoint because participants can have different follow-up times and some may be censored before an event occurs. A typical analysis uses Kaplan-Meier estimation to describe the event-free survival function and a time-to-event comparison such as a log-rank test or Cox proportional-hazards model when specified by the analysis plan.
The survival function represents the probability of remaining free of the defined event beyond time t. For a disease-free survival endpoint, the exact event definition must come from the protocol or registry endpoint definition.
6. Primary Endpoint Results
The ClinicalTrials.gov record states that results are posted for both registered primary endpoints, but the three registry-reported formal statistical analyses are all secondary analyses. No primary-endpoint statistical estimate, confidence interval, or P-value is included in the ClinicalTrials.gov record.
Clinically Relevant Toxicities
Registered primary endpoint
Number of Participants of Clinically Relevant Toxicities (CRT)
Time frame: 5 years after completion of study treatment or until death, whichever occurs first.
This is a composite binary endpoint: each participant is classified according to whether the prespecified clinically relevant toxicity criteria are met during the defined assessment period.
The ClinicalTrials.gov record does not provide the primary comparison's effect estimate, confidence interval, or P-value. Therefore, no numerical treatment effect can be calculated or attributed to the registry from the ClinicalTrials.gov record.
For a binary endpoint, a P-value—when reported—would address evidence against a prespecified null hypothesis; it would not itself measure the magnitude of the treatment difference. The clinically important quantities would also include the event proportions and an effect measure with its confidence interval.
3-year Disease-Free Survival Rate of Trastuzumab Emtansine
Registered primary endpoint
3-year Disease Free Survival (DFS) Rate of Trastuzumab Emtansine (TDM-1)
Time frame: 3 years
The registry definition states that DFS is evaluated from randomization until the first specified disease-related event. The complete event list is not included in the registry-reported extract.
A 3-year DFS rate is a time-specific survival quantity: it describes the estimated proportion remaining free of the registry-defined DFS event at 3 years. A proper analysis must account for censoring and the exact event definition.
The ClinicalTrials.gov record does not provide a DFS estimate, confidence interval, hazard ratio, or P-value for this primary endpoint. Consequently, this page does not infer or reconstruct a numerical 3-year DFS result.
For a time-to-event comparison, the confidence interval around an effect estimate is important because it describes statistical precision. A P-value, if reported, would address a hypothesis test and would not by itself quantify the magnitude or clinical importance of the treatment effect.
7. Secondary Statistical Results
Three formal statistical analyses are reported in the ClinicalTrials.gov record. All are secondary analyses and are therefore distinct from the two registered primary endpoints.
Neuropathy
Number of Patients Has Neuropathy
Method: Cochran-Mantel-Haenszel test
Hypothesis type: Superiority
| Feature | Supplied registry result |
|---|---|
| Endpoint role | Secondary |
| Outcome measure | Number of Patients Has Neuropathy |
| Time frame | Surveys are took at week 3, 12 and month 6 and 12 during treatment, and month 18. The data cutoff date is 18 month. |
| Outcome unit | Participants |
| Endpoint type | Binary |
| Groups compared | T-DM1 With no Baseline Neuropathy vs TH With no Baseline Neuropathy vs T-DM1 With Any Baseline Neuropathy vs TH With Any Baseline Neuropathy |
| Analysis population | The PNQ analysis was done on patients that completed at least baseline questionnaires and at least one follow-up questionnaire. Some patients did not complete this questionnaire at either baseline or follow-up. |
| Method as reported | Mantel Haenszel |
| P-value | 0.0012 |
The reported P-value of 0.0012 indicates evidence against the null hypothesis under the registry-reported Cochran-Mantel-Haenszel analysis. The test is particularly relevant when a categorical comparison is made while accounting for a stratification variable; here the registry-reported groups explicitly distinguish participants by baseline neuropathy status.
The P-value does not tell us the size of the neuropathy difference, the probability that one treatment is superior, or the clinical importance of the finding. The ClinicalTrials.gov record does not provide the corresponding event proportions, effect estimate, or confidence interval.
The analysis population is also important: the registry description says the PNQ analysis included participants with a baseline questionnaire and at least one follow-up questionnaire. That population is not automatically identical to all randomized participants.
Alopecia
Number of Patients Have Alopecia
Method: Linear regression
Hypothesis type: Superiority
| Feature | Supplied registry result |
|---|---|
| Endpoint role | Secondary |
| Outcome measure | Number of Patients Have Alopecia |
| Time frame | Median follow-up was 3.9 years. The AE data cutoff date is 21 April 2020. |
| Outcome unit | Participants |
| Endpoint type | Binary |
| Groups compared | Trastuzumab Emtansine (T-DM1) vs Paclitaxel + Trastuzumab |
| Method as reported | Regression, Linear |
| P-value | 0.0001 |
The registry reports a linear regression analysis for this binary outcome, with P = 0.0001. This page preserves the registry's reported method rather than replacing it with an assumed alternative analysis.
Because the outcome is classified as binary in the ClinicalTrials.gov record, interpretation requires attention to how the binary outcome was represented in the actual model. The registry-reported extract does not provide the regression coefficient, confidence interval, coding scheme, or model specification.
As with the other secondary analyses, the P-value is evidence from a hypothesis test; it is not an effect-size measure and does not indicate the magnitude of the difference between treatment groups.
Grade 2–3 Thrombocytopenia
Incidence of T-DM1 Induced Grade 2-3 Thrombocytopenia
Method: Fisher exact test
Hypothesis type: Superiority
| Feature | Supplied registry result |
|---|---|
| Endpoint role | Secondary |
| Outcome measure | Number of Incidence of T-DM1 Induced Grade 2-3 Thrombocytopenia |
| Time frame | AE is reported every 3 weeks for the first 12 weeks, then every 9 weeks for the subsequent 39 weeks, then follow-up up to 5 years after completion of study treatment or until death, whichever occurs first. Median follow-up 3.1 years. |
| Outcome unit | Participants |
| Endpoint type | Binary |
| Groups compared | Trastuzumab Emtansine (T-DM1) vs Paclitaxel + Trastuzumab |
| Method as reported | Fisher Exact |
| P-value | 0.0001 |
Fisher's exact test evaluates a categorical comparison using the observed cell counts in a contingency table. The registry-reported analysis reports P = 0.0001.
This P-value does not quantify the absolute incidence difference or its clinical magnitude. The ClinicalTrials.gov record does not provide the treatment-group event counts for this endpoint, a risk difference, an odds ratio, or a confidence interval.
8. Safety
The ClinicalTrials.gov record reports serious adverse events by treatment arm as affected participants divided by participants at risk. These figures provide an arm-specific safety count, but they do not by themselves describe the full adverse-event profile.
| Arm | Serious adverse events | Affected / at risk |
|---|---|---|
| Trastuzumab Emtansine (T-DM1) | Serious AEs | 62/383 |
| Paclitaxel + Trastuzumab | Serious AEs | 26/114 |
The ClinicalTrials.gov record does not provide confidence intervals, formal hypothesis tests, exposure-adjusted rates, treatment-emergent adverse-event tables, or discontinuation rates for the overall safety profile.
9. Statistical Methods Explained
Why use a Cochran-Mantel-Haenszel test for neuropathy?
The registry-reported neuropathy analysis compares categorical outcomes across treatment groups while explicitly separating participants by baseline neuropathy status. The Cochran-Mantel-Haenszel framework provides a way to assess an association between treatment and outcome while accounting for a categorical stratification variable.
What does P = 0.0012 mean for the neuropathy analysis?
Under the null hypothesis specified for that analysis, the reported P-value of 0.0012 represents the probability, under the statistical testing framework, of observing evidence at least as inconsistent with the null as the observed result. It does not equal the probability that the null hypothesis is true and does not quantify the magnitude of the neuropathy difference.
Why use Fisher's exact test?
Fisher's exact test is a categorical-data method based directly on the observed contingency-table counts. It is particularly useful when sample sizes or expected cell counts make large-sample approximations less appropriate. The registry-reported ATEMPT registry analysis used Fisher's exact test for grade 2–3 thrombocytopenia.
What does P = 0.0001 mean for thrombocytopenia?
The reported P-value provides evidence against the null hypothesis under the Fisher exact test used for that secondary analysis. It does not provide the incidence difference, an odds ratio, or an absolute risk measure. Those quantities would be needed to describe the size of the observed treatment difference.
Why is linear regression listed for a binary endpoint?
The ClinicalTrials.gov record classifies the alopecia endpoint as binary but report “Regression, Linear” as the method. This page preserves that registry-reported method rather than assuming a different model. Without the coefficient, coding scheme, and model specification, the ClinicalTrials.gov record is insufficient to determine exactly how the binary outcome entered the regression model.
Why is the analysis population important?
The neuropathy analysis explicitly states that the PNQ analysis included patients who completed at least a baseline questionnaire and at least one follow-up questionnaire. This means the analyzed population is defined partly by questionnaire completion. Such an analysis can answer a question about participants with the required questionnaire data, but its result should not automatically be interpreted as though every randomized participant contributed outcome data.
Why does a P-value not measure effect size?
A P-value is a measure arising from a hypothesis-testing procedure. Its magnitude depends on the effect, variability, and amount of information in the analysis. A small P-value can accompany either a modest or a large effect. To understand the magnitude of a treatment difference, investigators generally need the underlying event proportions or means together with an effect estimate and confidence interval.
10. Multiplicity and Interpretation of Secondary Analyses
The ClinicalTrials.gov record identifies two registered primary endpoints and three posted statistical analyses, all of which are secondary. The extract does not provide an alpha-allocation scheme, multiplicity-adjustment procedure, or formal hierarchy connecting these secondary analyses.
| Analysis | Role in the ClinicalTrials.gov record | P-value | Multiplicity information reported |
|---|---|---|---|
| Clinically relevant toxicities | Primary endpoint | Not reported | Not reported |
| 3-year DFS rate | Primary endpoint | Not reported | Not reported |
| Neuropathy | Secondary analysis | 0.0012 | Not reported |
| Alopecia | Secondary analysis | 0.0001 | Not reported |
| Grade 2–3 thrombocytopenia | Secondary analysis | 0.0001 | Not reported |
11. Non-Inferiority, Crossover, Factorial, Interim, and Bayesian Design Features
The ClinicalTrials.gov record identifies the hypothesis type as Superiority for each posted statistical analysis. They do not identify a non-inferiority margin.
Non-inferiority margin
No non-inferiority margin is reported. The listed hypothesis type is superiority.
Crossover
No crossover procedure is reported in the ClinicalTrials.gov record.
Factorial design
The design model is parallel. No factorial structure is reported.
Interim analysis
No interim-analysis plan is reported in the ClinicalTrials.gov record.
Bayesian methods
No Bayesian method is reported.
Stratification
The registry-reported statistical record explicitly identifies baseline neuropathy status in the neuropathy group comparison, but no broader randomization-stratification scheme is reported.
12. Missing Data and Analysis Populations
The registry-reported neuropathy analysis contains the clearest information about missing outcome data. It states that the PNQ analysis included patients who completed at least baseline questionnaires and at least one follow-up questionnaire and notes that some patients did not complete the questionnaire at either baseline or follow-up.
This is an analysis-population definition, not a statement that missing questionnaires were statistically imputed.
The ClinicalTrials.gov record does not identify an imputation method such as multiple imputation, last observation carried forward, baseline observation carried forward, or a specific missing-at-random model. No such method is inferred here.
13. What the Reported Secondary P-values Do — and Do Not — Mean
The reported Cochran-Mantel-Haenszel test provides evidence against its null hypothesis under the specified categorical-data analysis. It does not state how much neuropathy differed between the groups, because the ClinicalTrials.gov record does not provide the corresponding effect estimate or confidence interval.
The reported linear-regression analysis yields P = 0.0001. The P-value alone cannot establish the magnitude of the treatment difference or its clinical relevance, and the ClinicalTrials.gov record does not include the model coefficient or confidence interval.
The Fisher exact test yields P = 0.0001 in the ClinicalTrials.gov record. This indicates evidence against the tested null hypothesis under that categorical analysis, but it does not provide an absolute event-rate difference or relative effect estimate.
14. Serious Adverse Events: How to Read the Counts
The serious-adverse-event data are reported as affected participants divided by participants at risk: 62/383 for T-DM1 and 26/114 for paclitaxel plus trastuzumab.
| Measure | T-DM1 | Paclitaxel + Trastuzumab |
|---|---|---|
| Affected participants | 62 | 26 |
| Participants at risk | 383 | 114 |
| Reported format | 62/383 | 26/114 |
The denominator is essential because the arm sizes represented in the serious-AE data are not the same. A comparison of 62 versus 26 alone would ignore that difference. A complete comparative safety analysis would ordinarily report the corresponding proportions and, depending on the prespecified analysis, an appropriate measure of uncertainty.
15. Why This Trial Matters Statistically
ATEMPT is a useful statistical teaching case because its the ClinicalTrials.gov record combines a randomized parallel-group design, a binary toxicity endpoint, a time-to-event disease-free survival endpoint, categorical secondary analyses, an explicitly defined analysis population for questionnaire-based neuropathy assessment, and arm-specific serious-adverse-event counts.
| Concept | How it appears in ATEMPT |
|---|---|
| Randomization | The trial is classified as RANDOMIZED. |
| Parallel design | The design model is PARALLEL with two arms. |
| Binary endpoint | Clinically relevant toxicities are a registered primary endpoint, and neuropathy, alopecia, and thrombocytopenia are listed as binary secondary outcomes. |
| Time-to-event analysis | 3-year disease-free survival is a registered primary endpoint. |
| Cochran-Mantel-Haenszel test | Used in the registry-reported secondary neuropathy analysis. |
| Fisher exact test | Used in the registry-reported secondary thrombocytopenia analysis. |
| Linear regression | Reported for the registry-reported secondary alopecia analysis. |
| Analysis populations | The PNQ analysis has a questionnaire-completion-based population definition. |
| P-values | 0.0012 for neuropathy and 0.0001 for both alopecia and thrombocytopenia. |
| Safety denominators | Serious adverse events are reported as affected / at risk by arm. |
| Multiplicity | Multiple secondary analyses are reported, but no adjustment procedure is reported. |
16. Limitations
- Primary-endpoint numerical results are incomplete in the ClinicalTrials.gov record: the ClinicalTrials.gov record identifies results as posted for both primary endpoints but do not provide their effect estimates, confidence intervals, or formal P-values.
- Secondary-analysis emphasis: all three registry-reported formal statistical analyses are secondary analyses. Their P-values should therefore be interpreted in the context of the overall analysis plan.
- Missing effect estimates: the registry-reported secondary analyses provide P-values but not corresponding risk differences, odds ratios, regression coefficients, or confidence intervals.
- Analysis-population restriction: the neuropathy analysis required baseline and follow-up questionnaire completion, so its population is defined by available questionnaire data.
- Method specification: the registry extract identifies methods but does not provide full model specifications, covariates, coding rules, or variance-estimation details.
- Multiplicity: no multiplicity-adjustment procedure is posted on ClinicalTrials.gov for the three secondary analyses.
- Safety interpretation: serious-adverse-event counts use different denominators by arm and should not be interpreted from raw affected counts alone.
- No inferred results: no numerical primary endpoint result is reconstructed from publications or external sources because the ClinicalTrials.gov record rules restrict numerical trial facts to the ClinicalTrials.gov record.
17. Statistical Interpretation vs Clinical Interpretation
Statistical interpretation
The registry-reported secondary analyses report P = 0.0012 for neuropathy and P = 0.0001 for both alopecia and grade 2–3 thrombocytopenia. These are hypothesis-test results, not measures of effect size. The primary-endpoint estimates are not reported.
Clinical interpretation
The clinical meaning of these findings depends on the underlying event rates, magnitude of differences, confidence intervals, endpoint definitions, follow-up, and analysis populations. Those details are not all provided in the registry-reported extract.
18. A Practical Framework for Reading ATEMPT
A rigorous reading of the trial separates four distinct questions:
The trial randomized participants between T-DM1 and paclitaxel plus trastuzumab in a parallel-group design.
Clinically relevant toxicities and 3-year disease-free survival.
Cochran-Mantel-Haenszel, linear regression, and Fisher exact testing.
Primary endpoint effect estimates and confidence intervals, several model specifications, and complete long-form endpoint definitions.
This separation prevents a common statistical error: treating a small P-value from a secondary endpoint as though it were itself a complete description of the trial's primary evidence.
19. Related Tutorials
Learn more about the methods used in this trial:
20. Related Calculators
21. Sources
- ClinicalTrials.gov: ATEMPT, NCT01853748.
- PubMed: PMID 38935923.
- PubMed: PMID 36450763.
- PubMed: PMID 35173164.
- PubMed: PMID 34077270.
Continue through the Clinical Biostats statistical pathway
Use the trial's endpoints and statistical methods as a starting point for deeper tutorials and practical statistical tools.
22. Record Summary
ATEMPT provides a useful example of how randomized clinical-trial evidence can combine binary toxicity outcomes, time-to-event disease-free survival, categorical secondary analyses, questionnaire-defined analysis populations, and safety counts with arm-specific denominators. The ClinicalTrials.gov record reports three formal secondary statistical analyses—Cochran-Mantel-Haenszel testing, linear regression, and Fisher exact testing—with P-values of 0.0012, 0.0001, and 0.0001, respectively. At the same time, the registry-reported extract does not contain the numerical estimates and confidence intervals for the two registered primary endpoints. A statistically responsible interpretation therefore distinguishes the reported hypothesis-test results from effect size, precision, analysis population, endpoint definition, and the broader multiplicity context.