Safety Study Of Tofacitinib Versus Tumor Necrosis Factor (TNF) Inhibitor In Subjects With Rheumatoid Arthritis: Complete Statistical Analysis

1. Trial at a Glance

This page covers the registry results for Safety Study Of Tofacitinib Versus Tumor Necrosis Factor (TNF) Inhibitor In Subjects With Rheumatoid Arthritis (NCT02092467), a randomized phase 4 study. The primary outcome is Incidence Rate of Adjudicated Malignancies Excluding Non-melanoma Skin Cancers (NMSC).

2. Clinical Question

This was a post-marketing safety study in rheumatoid arthritis. Its stated focus included major cardiovascular adverse events and malignancies excluding non-melanoma skin cancers. The brief summary also identifies other safety events and efficacy parameters as matters to be collected and evaluated. The posted primary outcome concerns malignancy incidence; it should not be treated as a complete account of every safety or efficacy question addressed by the study.

3. Trial Design

The registry describes a randomized, parallel-group, unmasked Phase 4 study. The official title calls it a Phase 3B/4 randomized safety endpoint study. Enrollment was 4372. The listed start date is 2014-03-14, the primary completion date is 2020-07-22, and the study status is completed.

The treatment groups were tofacitinib 5 mg BID, tofacitinib 10 mg BID, and TNFi. For the TNFi group, adalimumab was used in the US, Canada and Puerto Rico, and etanercept was used in all other countries. The registry lists tofacitinib twice among the interventions, alongside adalimumab and etanercept. The arm descriptions for the tofacitinib groups are blank, although the posted outcome data identify the two doses.

The primary analysis population was the Safety Analysis Set: participants randomized in the study who received at least one dose of the randomized investigational drug, tofacitinib or TNFi. For the primary endpoint, the registry describes follow-up from baseline up to the last contact date, with a maximum up to 72 months. The risk period ended at the last contact date, defined using the latest of specified event, visit, withdrawal or contact dates; for a participant who died, it ended at the death date.

The registry record does not report the randomization procedure, allocation concealment, the sample-size justification, or a detailed statistical analysis plan. The record labels the study unmasked, so participants and study personnel were not described as blinded. No further protocol details should be inferred from the parallel-group and randomized design labels alone.

4. Participant Flow and Accounting

The registry reports how participants moved through each stage of the study. All counts below are exactly as posted; missing counts are not inferred from other stages.

Before assignment: Study started from 14 March 2014 and completed on 22 July 2020.

Overall Study

Reasons for not completing this period:

Analysis population for the primary outcome, as defined in the registry: SAS included all participants who were randomized in the study and received at least one dose of the randomized investigational drug (tofacitinib or TNFi).

5. Baseline Characteristics

The table reproduces the baseline characteristics posted in the registry, which describe the participants analysed at baseline and help show whether the randomized groups were comparable.

Population: Safety Analysis Set (SAS) included all participants who were randomized in the study and received at least one dose of the randomized investigational drug (tofacitinib or TNFi).

6. Primary Endpoint

The registry lists the primary outcome as Incidence Rate of Adjudicated Malignancies Excluding Non-melanoma Skin Cancers (NMSC), measured over Baseline up to last contact date (maximum up to 72 months).

The primary endpoint is the incidence rate of adjudicated malignancies excluding non-melanoma skin cancers. It is expressed as participants with an event per 100 participant years. The registry defines the rate as the number of participants with admissible events divided by the total time at risk among qualifying participants in the treatment group. Malignancy events were adjudicated by a steering committee, and first events within the risk period were counted.

This rate uses follow-up time as well as the number of participants with events. That matters when participants have different lengths of observation: participant years represent accumulated time at risk rather than simply a count of people. The endpoint is therefore intended to describe the pace at which first qualifying malignancy events occurred over observed time, rather than the proportion of participants who ever had an event.

The registry says participants without an event, and those whose event occurred outside the risk period, were censored at the end of the risk period. Censoring means that their observed time contributes up to that endpoint, without treating them as having a qualifying event during the period. The listed timeframe runs from baseline to last contact, up to 72 months.

The endpoint excludes non-melanoma skin cancers, so it does not describe all malignancies. Its interpretation also depends on the event-adjudication process, the risk-period definition, and the follow-up available for each group.

7. Why No Re-analysis Is Shown

8. Statistical Interpretation

9. Statistical Methods Explained

The registry presents a time-to-event safety endpoint as an incidence rate and also reports hazard-ratio comparisons. These measures describe different aspects of event occurrence. The endpoint definition, follow-up rules, analysis population, and comparative model all affect what can be concluded from the results.

What does an incidence rate describe?

An incidence rate relates the number of qualifying events to the amount of time participants were observed and at risk. The registry expresses the primary result as participants with an event per 100 PY. Participant time is accumulated during follow-up, so a rate is not the same as a simple proportion of participants with an event. It can be useful when follow-up duration varies between participants or groups.

The endpoint counts first events within the risk period. Participants without a qualifying event are censored at the end of that period. In general, censoring records the time during which an event could be observed and then stops contributing follow-up; it does not mean the participant experienced the event. The record defines last contact using the latest of specified event, study visit, withdrawal, or telephone-contact dates, and uses the death date when a participant died. It does not give further detail on how dates that could not be determined were handled.

What does the hazard ratio add?

A hazard ratio compares event hazards between groups over follow-up under a time-to-event analysis. A value around 1 indicates similar estimated hazards in the compared groups under that model; a value above or below 1 indicates a higher or lower estimated hazard in the first group relative to the second, respectively. This is a general explanation of the measure, not a claim that a particular treatment caused an outcome. The registry gives hazard ratios and confidence intervals but does not report the model method or its assumptions.

The primary endpoint’s point estimates are rates per participant time, whereas the reported hazard ratios are comparative time-to-event estimates. The hazard ratios cannot be recovered reliably from the rate estimates alone. A full interpretation would require details such as the model specification and how the analysis addressed the timing of events; these are not reported in the registry record.

How should a confidence interval be read?

A confidence interval describes uncertainty in an estimate using the statistical procedure applied. It is not a guarantee that the true value lies within its bounds, nor does it account for all design limitations, measurement issues, or analysis choices. A narrow interval generally indicates greater statistical precision than a wide interval, while an interval spanning values with different practical interpretations signals uncertainty about the size and direction of the comparative effect.

The primary rates are reported with 95% confidence intervals. The hazard-ratio comparisons likewise have 95% confidence intervals. The registry does not report p-values for those hazard-ratio analyses, and the intervals should be read as estimates from the sponsor’s analysis rather than as independently recalculated results.

What does the non-inferiority label establish?

The registry labels the all-tofacitinib versus TNFi comparison and the comparison between tofacitinib doses as non-inferiority analyses. In general, a non-inferiority analysis evaluates whether an estimate is compatible with a prespecified margin for an acceptable loss relative to a comparator. The margin and the detailed decision rule are essential to interpret that question. Neither is reported in the registry record, so the label alone is not enough to independently assess the non-inferiority claim.

The registry also lists dose-specific comparisons with TNFi as other analyses. It does not state that these were non-inferiority analyses. No multiplicity procedure or adjustment is reported for the set of comparisons. Consequently, the individual estimates should be considered in the context of the full group of analyses, rather than treated as isolated definitive findings.

What do the planning illustrations mean?

Sample-size illustrations based on an observed effect are planning scenarios, not observed power. The posted assumptions specify a two-sided significance level of 0.05, equal allocation between two arms, and a normal approximation, with the effect to detect expressed as a fraction of the observed effect. For time-to-event tables, the illustration gives total events using the Schoenfeld approach; other illustrations give participants per arm. These scenarios show how planned sample size would vary with the assumed effect and requested power under those assumptions. They do not describe the trial’s actual design justification, which is not reported in the registry record.

General statistical considerations also matter. Over-dispersion, when present, means that variation is greater than a model assumes; this can make conventional uncertainty estimates too optimistic if not addressed. The registry does not report whether over-dispersion was assessed or relevant to the primary time-to-event analysis. No such adjustment should be presumed.

The primary analysis population is stated as randomized participants who received at least one dose. The record also specifies the endpoint’s risk period and censoring rule. Those definitions help identify who contributed to the analysis and how follow-up was counted, but they do not supply a complete account of missing data, protocol deviations, or sensitivity analyses.

10. Analyses Reported by the Sponsor

The registry reports the sponsor’s comparative hazard-ratio analyses directly. For all tofacitinib compared with TNFi, the reported estimate is 1.48 with a 95% confidence interval from 1.04 to 2.09, and the analysis is labeled non-inferiority. The comparison between the two tofacitinib doses is reported as 1.00, with a 95% confidence interval from 0.70 to 1.43, and is also labeled non-inferiority. Dose-specific comparisons with TNFi are listed separately as other analyses.

No independent recalculation of the primary hazard-ratio analyses is available from the posted arm-level incidence rates. A rate per participant time is not enough to recreate the time-to-event model, and the registry does not give the model method. The sponsor’s reported hazard ratios therefore remain the relevant comparative estimates in the registry record; they should not be replaced with a comparison calculated from the displayed rates.

The record does not report p-values, the non-inferiority margin, the detailed decision rule, or the statistical method for these hazard ratios. As a result, readers can describe the estimates and their confidence intervals, but cannot independently verify the non-inferiority assessment from the registry record alone. The record does not identify a discrepancy between sponsor-reported and recalculated primary analyses.

11. Multiplicity, Missing Data and Analysis Populations

The primary analysis population is defined in the registry as participants randomized in the study who received at least one dose of the randomized investigational drug. This definition identifies the Safety Analysis Set used for the primary outcome. The record does not describe other analysis populations, such as a per-protocol set, or give details on how participants with protocol deviations were handled.

For the primary endpoint, the registry specifies a risk period ending at last contact, with the last contact date determined from the latest of listed dates and the death date used for participants who died. It states that participants without an event, or with an event outside the risk period, were censored at the end of the risk period. This gives important information about follow-up and event counting. It does not fully explain how missing dates, incomplete event information, or other missing data were addressed. Those details are not reported in the registry record.

The registry records several primary hazard-ratio comparisons and a number of secondary outcomes. No multiplicity adjustment, testing hierarchy, or control of the overall type I error rate is reported. In general, when many hypotheses are tested, unadjusted results can overstate the strength of evidence because chance findings become more likely across the set. The non-inferiority margin and the detailed non-inferiority decision rule are also not reported, limiting independent assessment of the analyses carrying that label.

The record does not report interim analyses, stopping rules, sensitivity analyses for missing data, or the statistical analysis plan. These are not details that can be inferred from the study’s completed status or from the posted outcome summaries. Without them, readers can describe the registered population and censoring rules, but cannot determine whether additional analyses were planned or how robust the findings were to alternative assumptions.

12. Safety Results

The registry reports serious adverse events as affected participants divided by participants at risk.

Deaths

Other (non-serious) adverse events

Participants with at least one non-serious event at or above the registry's 5% reporting threshold.

Most frequent serious adverse event terms

Most frequent non-serious adverse event terms

Serious events with the highest arm-level rate: pneumonia (3.8%), osteoarthritis (2.0%), pulmonary embolism (1.5%), cellulitis (1.0%), urinary tract infection (1.0%).

The registry reports serious adverse events for each treatment group as numbers of participants affected: 373 for tofacitinib 5 mg BID, 420 for tofacitinib 10 mg BID, and 339 for TNFi. The corresponding numbers at risk were 1455, 1456, and 1451. These are participant counts, not incidence rates adjusted for follow-up time. The registry also reports other adverse events for 1108, 1127, and 1056 participants in the same groups, respectively.

Deaths are reported for 49 participants in the tofacitinib 5 mg BID group, 66 in the tofacitinib 10 mg BID group, and 38 in the TNFi group. The most frequently listed serious adverse-event terms include pneumonia (3.8%), osteoarthritis (2.0%), pulmonary embolism (1.5%), cellulitis (1.0%), urinary tract infection (1.0%), and rheumatoid arthritis (0.8%).

These counts are descriptive safety summaries. They do not, by themselves, establish whether an event was caused by treatment or allow a fair comparison of risk over time. The registry record does not provide exposure-adjusted rates for these serious-event counts in this summary. Differences in observed counts should therefore not be interpreted as causal treatment effects without fuller analysis.

13. What the Result Does, and Does Not, Mean

The primary results describe adjudicated malignancy events excluding non-melanoma skin cancers during the registry-defined risk period. The incidence rates and reported hazard ratios provide estimates of event occurrence and comparative hazard in the analyzed population. Their confidence intervals show that the estimates have uncertainty, and the registry’s limited reporting of the comparative methods constrains how fully they can be evaluated.

These results do not describe every malignancy, because non-melanoma skin cancers are excluded. They also do not answer every safety question identified by the study. The registry lists outcomes concerning cardiovascular events, infections, hepatic events, adverse events, laboratory parameters, and deaths, but those outcomes require their own endpoint definitions and analyses. The primary endpoint cannot stand in for all of them.

The randomized design supports a treatment-group comparison, but the record’s open-label designation, missing analysis details, and unreported multiplicity approach are relevant to interpretation. Randomization does not remove uncertainty in estimates, make every outcome equally reliable, or establish causation for individual adverse events. In general, results from a comparison should be interpreted in light of how events were defined, observed, analyzed, and reported.

The hazard-ratio analyses labeled non-inferiority cannot be independently judged against a non-inferiority margin because that margin and the detailed decision rule are not reported in the registry record. The reported estimate and confidence interval can be stated, but the label alone does not establish what margin was used or whether all prespecified requirements were met.

Accordingly, the posted results support a description of observed event rates and sponsor-reported comparative estimates in the defined Safety Analysis Set. They do not provide a complete basis for conclusions about all long-term risks, every secondary outcome, the mechanisms behind events, or how the results apply beyond the studied population and follow-up.

14. Why This Trial Matters Statistically

This trial illustrates why safety outcomes with different follow-up times are often expressed as incidence rates or analyzed with time-to-event methods. A count alone does not account for how long participants were observed, while a participant-time rate incorporates the observed time at risk. The registry’s endpoint definition also makes the censoring and last-contact rules part of the statistical meaning of the result, rather than minor administrative details.

It also demonstrates why the same clinical question can be represented by more than one statistic. The primary endpoint is reported as an incidence rate, while comparative analyses are reported as hazard ratios. These measures complement one another, but they are not interchangeable. The hazard-ratio estimate depends on the time-to-event analysis, and cannot be derived simply by comparing the displayed rates.

The non-inferiority labels make the reporting of design details especially important. A non-inferiority conclusion depends on a prespecified margin and a defined decision rule. Without those details, readers can report the sponsor’s estimates and confidence intervals but cannot independently reconstruct the inferential claim from the registry record.

Finally, the registry lists multiple secondary safety outcomes and analyses. This makes multiplicity and the distinction between confirmatory and exploratory evidence central to interpretation. The absence of a reported adjustment or hierarchy means that secondary estimates should be read with appropriate caution. The study is a useful example of how endpoint definitions, analysis populations, follow-up, uncertainty, and transparent statistical reporting jointly shape the meaning of a clinical trial result.

15. Limitations

• The posted summary statistics are the only data these results rest on; patient-level data, full protocols and statistical analysis plans are not available.
• Recalculated results are simple, unadjusted calculations from posted summaries. They do not replace the sponsor's prespecified analysis.

Several trial-specific details needed for a fuller statistical assessment are not reported in the registry record. These include the sample-size justification, the randomization procedure, allocation concealment, the statistical analysis plan, the method used for the primary hazard-ratio analyses, and the non-inferiority margin. Without the margin and decision rule, the non-inferiority labels cannot be independently evaluated.

The registry describes the study as unmasked. Lack of masking can affect how events are reported, assessed, or managed, although the malignancies were adjudicated by a steering committee. The record does not describe the committee’s procedures or provide details that would establish whether adjudication was blinded to treatment assignment.

The primary endpoint excludes non-melanoma skin cancers and counts first events within the risk period. As a result, it does not summarize all malignancies or recurrent events. The record defines censoring and last contact but does not report sensitivity analyses addressing alternative risk-period definitions or incomplete event information.

There are multiple reported treatment comparisons and a broad set of secondary outcomes. The registry does not report an approach to multiplicity, and it does not describe interim analyses or stopping rules. Secondary findings should therefore not be read as though each were a separately prespecified, confirmatory test.

The TNFi comparator was not a single drug across all locations: adalimumab was used in the US, Canada and Puerto Rico, while etanercept was used elsewhere. The record does not provide location-specific outcome results in the information summarized for the primary endpoint. The posted overall TNFi result therefore combines a comparator group whose specific agent varied by location.

16. Related Tutorials

17. Related Statistical Calculators

18. Sources

• ClinicalTrials.gov: Safety Study Of Tofacitinib Versus Tumor Necrosis Factor (TNF) Inhibitor In Subjects With Rheumatoid Arthritis — NCT02092467.
• Registry record last updated 2021-08-17; status last verified 2021-07.
• Results first posted 2021-08-17.
• Data retrieved from ClinicalTrials.gov on 2026-11-09.
• Group sizes, means, event counts and sponsor-reported p-values are reproduced as posted. Effect estimates, confidence intervals and p-values not attributed to the sponsor are independently calculated from the posted summary statistics.

19. Record Summary

NCT02092467 is a completed randomized, parallel-group, unmasked Phase 4 study in rheumatoid arthritis. Its official title describes a Phase 3B/4 safety endpoint study. The record lists enrollment of 4372 and a study period beginning 2014-03-14, with primary completion on 2020-07-22.

The primary endpoint was the incidence rate of adjudicated malignancies excluding non-melanoma skin cancers, expressed per 100 participant years. The registry reports arm-level rates and confidence intervals, plus sponsor-reported hazard-ratio comparisons. It labels the all-tofacitinib versus TNFi and tofacitinib-dose comparisons as non-inferiority analyses, but the margin and detailed statistical method are not reported.

The Safety Analysis Set included randomized participants who received at least one dose. The registry also lists secondary cardiovascular, infection, hepatic, adverse-event, laboratory, and mortality outcomes. Multiplicity procedures, interim analyses, sample-size justification, and a detailed statistical analysis plan are not reported in the registry record.