This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record.
1. Trial at a Glance
PARADIGM-2 was a randomized, parallel, single-masked phase 3 study of nonacog beta pegol in patients with congenital bleeding disorder and haemophilia B. The registry reports 74 enrolled participants across three treatment arms, with two prophylaxis regimens and an on-demand regimen.
| Feature | PARADIGM-2 |
|---|---|
| Phase | Phase 3 |
| Condition | Congenital Bleeding Disorder; Haemophilia B |
| Design | Randomized, parallel, single-masked |
| Allocation | Randomized |
| Primary purpose | Treatment |
| Enrollment | 74 |
| Number of arms | 3 |
| Intervention | Nonacog beta pegol |
| Lead sponsor | Novo Nordisk A/S |
| Sponsor type | Industry |
| Status | Completed |
| ClinicalTrials.gov | NCT01333111 |
2. Clinical Question
The central statistical question concerns the occurrence of inhibitory antibodies against factor IX during treatment with nonacog beta pegol in patients with haemophilia B. The registry defines the primary endpoint using a factor IX inhibitor titre equal to or above 0.6 BU (Bethesda Units).
Population
Patients with congenital bleeding disorder and haemophilia B enrolled in a phase 3 treatment study.
Intervention
Nonacog beta pegol was used in all three treatment arms represented in the registry.
Comparator structure
The three-arm design included prophylaxis at low dose, prophylaxis at high dose, and on-demand treatment. The registry does not describe these as separate active drugs.
Primary question
How many subjects developed inhibitory antibodies against factor IX, defined by a titre equal to or above 0.6 BU, during the specified treatment periods?
3. Trial Design
Low Dose
- 10 U/kg
- 52 weeks
- Serious adverse events: 1/30
High Dose
- 40 U/kg
- 52 weeks
- Serious adverse events: 3/29
On-Demand Treatment
- On-demand treatment
- 28 weeks
- Serious adverse events: 0/15
4. Primary Endpoints
| Endpoint | Registry definition | Time frame |
|---|---|---|
| Incidence of inhibitory antibodies against factor IX defined as titre equal to or above 0.6 BU | Number of subjects who developed inhibitory antibodies against factor IX. Inhibitors were analysed with either the Nijmegen modified factor IX Bethesda assay or a heat/cold Nijmegen modified factor IX Bethesda assay. | 52 weeks after treatment start for patients on prophylaxis |
| Incidence of inhibitory antibodies against factor IX defined as titre equal to or above 0.6 BU | Number of subjects who developed inhibitory antibodies against factor IX. Inhibitors were analysed with either the Nijmegen modified factor IX Bethesda assay or a heat/cold Nijmegen modified factor IX Bethesda assay. | 28 weeks after treatment start on on-demand treatment |
These are incidence endpoints expressed as the number of subjects who developed inhibitory antibodies during the registry-specified treatment period. The endpoint is therefore fundamentally different from a continuous laboratory measurement: the primary outcome classifies whether a participant crossed a prespecified inhibitor threshold.
5. Results and Statistical Reporting
The ClinicalTrials.gov record identifies results as posted, but no formal statistical analyses are posted for the primary endpoints. The record therefore supports reporting the endpoint definitions and the available safety counts, while the appropriate inferential framework for the inhibitor endpoint is described below.
Primary Endpoint: Prophylaxis
Incidence of inhibitory antibodies against factor IX
Time frame: 52 weeks after treatment start for patients on prophylaxis
Endpoint reported as the number of subjects who developed inhibitory antibodies.
The primary prophylaxis endpoint is a binary participant-level outcome: a subject either develops an inhibitory antibody meeting the registry threshold or does not during the specified observation period. The clinically relevant quantity would therefore be the number and proportion of affected subjects within each prophylaxis arm.
A proportion describes the frequency of the observed event; it does not establish that treatment caused or prevented the event without considering the randomized design, follow-up, and analysis population. A confidence interval around a proportion would quantify statistical precision, particularly important when the number of events is small.
A p-value, if used for an inferential comparison, would address compatibility with a specified null hypothesis. It would not measure the size or clinical importance of the observed inhibitor incidence. With rare events, confidence intervals and the actual event counts are especially important because a small change in the number of affected participants can materially change the estimated proportion.
Primary Endpoint: On-Demand Treatment
Incidence of inhibitory antibodies against factor IX
Time frame: 28 weeks after treatment start on on-demand treatment
Endpoint reported as the number of subjects who developed inhibitory antibodies.
The on-demand endpoint uses the same inhibitor definition but a different prespecified observation period. The 28-week time frame should therefore be kept distinct from the 52-week prophylaxis time frame rather than treating the two periods as interchangeable.
For this type of endpoint, a natural descriptive analysis is the number of subjects with inhibitors divided by the number of subjects evaluated within the relevant treatment arm. An exact binomial confidence interval can be useful when event counts are small because normal-approximation intervals may perform poorly near the boundaries of the probability scale.
The endpoint does not measure the magnitude of an antibody titre beyond the threshold definition. A participant with a titre meeting the threshold and a participant with a much higher titre would both contribute an event to this particular binary endpoint.
6. Safety Results
The registry reports serious adverse events by treatment arm. These figures are presented as affected participants divided by participants at risk.
| Treatment arm | Time frame | Serious adverse events |
|---|---|---|
| Prophylaxis, Low Dose 10 U/kg | 52 Weeks | 1/30 |
| Prophylaxis, High Dose 40 U/kg | 52 Weeks | 3/29 |
| On-Demand | 28 Weeks | 0/15 |
The numerators and denominators are more informative than a simple ordering of the arms. The observation periods also differ between prophylaxis and on-demand treatment, so direct comparisons of crude event frequencies should account for the different follow-up windows.
7. Statistical Methodology
Binary incidence analysis
The primary endpoints are binary incidence outcomes. For each arm, the basic descriptive quantity is the proportion of participants who developed an inhibitory antibody meeting the threshold:
where x is the number of subjects who developed inhibitory antibodies and n is the number of subjects evaluated for the endpoint.
The registry specifically defines the event using a titre equal to or above 0.6 BU. That threshold converts the underlying assay result into a binary endpoint for the primary incidence analysis.
Confidence intervals for rare events
When the number of inhibitor events is small, an exact binomial confidence interval is generally preferable to relying automatically on a large-sample normal approximation. The exact interval respects the discrete nature of the event count and remains applicable when the observed event count is zero.
The numerator and denominator should always accompany the estimated incidence because they reveal how much information supports the observed proportion.
Between-arm comparisons
If the objective were to formally compare inhibitor incidence between randomized arms, the analysis would ordinarily use a method appropriate for binary outcomes and the small event counts. Depending on the prespecified estimand and sample size, this could include an exact test or a model for binary response. The choice should be specified before looking at the results because different methods target somewhat different inferential questions.
Risk difference and risk ratio
For two treatment groups, two useful effect measures are the risk difference and risk ratio. The risk difference describes the absolute difference in event probability, whereas the risk ratio describes the relative event frequency.
Neither measure should be interpreted without its confidence interval and the underlying event counts.
8. Statistical Methods Explained
Why is the inhibitor endpoint binary?
The registry defines the primary endpoint as the incidence of inhibitory antibodies using a prespecified titre threshold of 0.6 BU. The statistical endpoint therefore records whether the participant developed an inhibitor meeting that definition rather than analysing the titre as a continuous measurement.
Why should the numerator and denominator always be reported?
An incidence proportion without its underlying count can obscure the amount of information available. Reporting affected subjects together with subjects at risk allows the reader to distinguish a result based on many participants from one based on a small denominator.
Why are exact confidence intervals useful here?
Rare safety or immunogenicity events can produce very small numerators, including zero. Exact binomial methods avoid relying on approximations that can become unreliable with sparse data.
Why should the 28-week and 52-week endpoints remain separate?
The registry specifies different observation periods for prophylaxis and on-demand treatment. Incidence depends on the period during which participants are observed, so combining the periods would change the definition of the endpoint.
Does a zero-event arm prove that inhibitors cannot occur?
No. An observed zero count means that no qualifying event was observed among the participants evaluated in that arm during its specified follow-up. Statistical uncertainty remains, which is precisely why a confidence interval is useful even when the observed event count is zero.
Why does a p-value not measure safety?
A p-value evaluates a statistical hypothesis under a specified testing framework. It does not describe the severity of an adverse event, the magnitude of an incidence difference, or the clinical importance of a finding. Event counts, effect estimates, confidence intervals, and clinical context provide different information.
9. Analysis Populations and Denominators
For an incidence endpoint, the denominator is fundamental. The registry's safety reporting gives explicit denominators for serious adverse events: 30 for low-dose prophylaxis, 29 for high-dose prophylaxis, and 15 for on-demand treatment. The primary endpoint descriptions specify the relevant treatment periods but do not provide corresponding inhibitor event counts in the posted statistical-analysis fields.
| Question | Statistical importance |
|---|---|
| Who was randomized? | Defines the population created by random allocation and protects the principal treatment comparison. |
| Who was evaluable for the inhibitor endpoint? | Defines the denominator for incidence estimation. |
| Who was at risk for serious adverse events? | Defines the denominator shown in the registry safety results. |
| How long was each group observed? | Determines the exposure period underlying the incidence measure. |
A careful analysis should not silently substitute a safety denominator for an efficacy or immunogenicity denominator. Each endpoint has its own analysis population and observation rules.
10. Randomization and Parallel-Group Interpretation
The registry identifies PARADIGM-2 as randomized with a parallel design. Randomization is statistically important because it creates treatment groups whose differences, in expectation, are attributable to the randomized allocation rather than systematic assignment of participants to treatment strategies.
What randomization provides
It provides the foundation for comparing outcomes between the randomized treatment groups without relying solely on adjustment for measured baseline variables.
What randomization does not provide
It does not guarantee identical groups in every finite sample, nor does it eliminate uncertainty arising from the number of participants and number of observed events.
Why parallel design matters
Participants remain associated with their randomized treatment strategy rather than creating a crossover comparison between treatment periods.
Why masking matters
The registry identifies the study as single-masked. Masking can reduce certain sources of bias, although its statistical value depends on which participants or study personnel were masked.
11. Safety Interpretation
The serious-adverse-event results provide a straightforward example of why absolute event counts should accompany percentages. The registry reports 1/30 for low-dose prophylaxis, 3/29 for high-dose prophylaxis, and 0/15 for on-demand treatment.
The observed event counts differ across the three arms, but the denominators are also different and the on-demand group has a shorter specified treatment period. With only the reported counts, a descriptive comparison is possible, while a formal comparative safety inference would require an explicitly defined analysis population, estimand, and statistical test.
In particular, an observed difference in crude incidence should not automatically be interpreted as evidence of a treatment-dose effect. Randomization supports causal comparison, but sparse events and limited sample size can make estimates imprecise.
12. What an Inhibitor Incidence Estimate Means
An incidence estimate for the primary endpoint would represent the proportion of participants who developed inhibitory antibodies meeting the registry's threshold during the specified observation period.
It would not mean that the treatment caused an inhibitor in that proportion of every comparable patient population. The estimate applies to the study population and observation framework, with uncertainty around the estimate.
A confidence interval around an incidence proportion communicates how precisely the observed event frequency has been estimated. When event counts are small, the interval can remain wide even when the point estimate appears straightforward.
The primary endpoint is not simply "any antibody." It is specifically an inhibitory antibody against factor IX defined by a titre equal to or above 0.6 BU, with testing using either the Nijmegen modified factor IX Bethesda assay or a heat/cold Nijmegen modified factor IX Bethesda assay.
13. Endpoint-Specific Interpretation
| Endpoint feature | Statistical implication |
|---|---|
| Binary event | Incidence is naturally summarized as a proportion. |
| Prespecified threshold | The 0.6 BU threshold determines which assay results count as endpoint events. |
| Fixed observation window | Prophylaxis and on-demand treatment have distinct registry-specified time frames. |
| Rare-event setting | Exact methods and confidence intervals can be particularly informative. |
| Three treatment arms | Multiple pairwise comparisons can create multiplicity concerns if formal hypothesis tests are performed. |
14. Multiplicity Considerations
The three-arm structure creates a potential multiplicity issue if investigators conduct multiple formal pairwise tests of the same primary endpoint. A confirmatory statistical analysis should specify in advance which comparisons are primary and how the familywise type I error rate is controlled.
Multiplicity is distinct from the endpoint definition itself. The registry identifies the inhibitor endpoint at two treatment-period specifications, but the information available here does not establish a hierarchy of pairwise hypotheses or an alpha-allocation strategy.
15. Missing Data and Observation Windows
Incidence endpoints depend on whether a participant is observed sufficiently to contribute to the relevant endpoint. A participant who does not complete the specified observation period may require a prespecified rule determining whether and how that participant contributes to the analysis.
Complete follow-up
Participants observed through the relevant endpoint period provide the clearest contribution to a fixed-period incidence estimate.
Early discontinuation
Participants who leave treatment early can create ambiguity unless the statistical analysis plan specifies how their endpoint status is handled.
Imputation
Imputing an inhibitor status can materially affect a rare-event endpoint and should therefore be prespecified rather than selected after seeing the data.
Time-to-event alternative
If the scientific question concerns the timing of inhibitor development rather than simply incidence during a fixed window, a time-to-event framework can preserve more temporal information.
The registry endpoint is explicitly framed as incidence during a specified treatment period. That makes a fixed-period binary analysis a natural descriptive framework, while a time-to-event analysis would represent a different estimand and would require corresponding event-time and censoring information.
16. Bayesian Methods and Non-Inferiority
The registry information does not identify a Bayesian statistical analysis or a non-inferiority margin for the primary endpoints. The primary endpoint is an incidence measure rather than a registry-defined non-inferiority comparison, so non-inferiority logic should not be imposed on the trial without an explicitly specified margin and hypothesis.
Likewise, Bayesian posterior probabilities should not be substituted for frequentist confidence intervals unless a Bayesian model and prior specification are explicitly defined. The choice between these frameworks is part of the statistical design, not something that can be determined from the endpoint name alone.
17. Planned Analysis
For the inhibitor endpoints, the registry measures the number of subjects who develop inhibitory antibodies against factor IX, using the threshold of 0.6 BU and the specified Bethesda assay approaches. The natural statistical summary is the incidence proportion within each relevant treatment arm and observation period.
| Analysis component | Typical statistical approach |
|---|---|
| Primary endpoint | Number and proportion of subjects developing inhibitory antibodies |
| Precision | Confidence interval for a binomial proportion, with exact methods useful for sparse events |
| Arm comparison | An appropriate exact or binary-outcome method, prespecified according to the intended estimand |
| Effect measure | Risk difference and/or risk ratio can provide absolute and relative comparisons |
| Observation period | 52 weeks for prophylaxis; 28 weeks for on-demand treatment |
| Safety | Descriptive serious-adverse-event counts and proportions by treatment arm |
18. Trial Timeline
Study start
The registry lists April 27, 2011 as the start date for PARADIGM-2.
Randomized treatment study
The study used randomized allocation, a parallel design, single masking, and treatment as its primary purpose.
Primary completion
The registry lists March 31, 2013 as the primary completion date.
Registry status
The ClinicalTrials.gov record identifies the study as completed.
19. Why This Trial Matters Statistically
PARADIGM-2 is a useful teaching example because it illustrates a statistical situation in which event definition, denominator, follow-up period, and sparse-event uncertainty matter more than a single headline effect estimate.
| Concept | How it appears in PARADIGM-2 |
|---|---|
| Randomization | Participants were allocated randomly among three parallel treatment arms. |
| Binary endpoint | The primary endpoint records whether a participant developed an inhibitor meeting the 0.6 BU threshold. |
| Assay-based definition | Factor IX inhibitors were analysed using specified Nijmegen modified Bethesda assay approaches. |
| Fixed follow-up | Prophylaxis and on-demand treatment have different registry-specified observation periods. |
| Rare events | Small event counts make confidence-interval interpretation particularly important. |
| Three-arm design | Multiple possible comparisons require attention to multiplicity if formal testing is undertaken. |
| Safety denominators | Serious adverse events are reported as affected subjects divided by subjects at risk. |
| Endpoint-specific populations | The denominator for one endpoint should not automatically be substituted for another endpoint. |
20. Important Limitations and Interpretation Issues
- No formal statistical analyses for the primary endpoints: the registry reports the endpoint definitions but does not provide formal inferential analyses for the inhibitor outcomes.
- Sparse-event uncertainty: inhibitor and serious-adverse-event endpoints can be driven by small event counts, making point estimates alone potentially misleading.
- Different observation periods: prophylaxis is assessed over 52 weeks, while on-demand treatment is assessed over 28 weeks.
- Three-arm multiplicity: multiple pairwise comparisons can require type I error control if formal hypothesis testing is performed.
- Denominator dependence: incidence proportions are meaningful only when the population contributing to the denominator is clearly defined.
- Binary threshold: the primary endpoint captures whether the inhibitor threshold is reached, not the full continuous distribution of factor IX inhibitor titres.
- Generalizability: the study population, treatment structure, and registry-defined eligibility determine the population to which the findings can be generalized.
- Safety versus efficacy: serious adverse events and inhibitory antibodies answer different clinical questions and should not be combined into a single statistical outcome.
21. Statistical Interpretation vs Clinical Interpretation
Statistical interpretation
The primary outcomes are binary incidence endpoints with a prespecified factor IX inhibitor threshold. Their natural analysis is based on event counts and binomial proportions, with confidence intervals providing essential information about precision.
Clinical interpretation
The clinical meaning of an inhibitor event depends on the treatment context, the assay-defined threshold, the duration of observation, and the consequences of developing an inhibitory antibody. Safety findings should be considered separately.
22. Sources
- ClinicalTrials.gov: NCT01333111 — PARADIGM-2.
Continue through the Clinical Biostats clinical-trial library
Explore additional clinical-trial statistical analyses and educational material covering trial design, endpoints, statistical methods, and interpretation.
23. Record Summary
PARADIGM-2 was a completed randomized phase 3, parallel, single-masked treatment study enrolling 74 participants with congenital bleeding disorder and haemophilia B. All three treatment arms involved nonacog beta pegol, with the registry identifying low-dose prophylaxis at 10 U/kg for 52 weeks, high-dose prophylaxis at 40 U/kg for 52 weeks, and on-demand treatment for 28 weeks. The primary endpoints focused on inhibitory antibodies against factor IX defined by a titre equal to or above 0.6 BU. The registry reports serious adverse events of 1/30, 3/29, and 0/15 across the three arms, respectively.
Statistically, the trial is most useful as an example of rare binary clinical-trial endpoints. The key analytical issues are defining the event precisely, preserving the appropriate observation period, reporting the underlying numerator and denominator, using confidence intervals suited to sparse data, and distinguishing descriptive incidence from formal between-arm inference. The three-arm randomized design also makes prespecification of comparative hypotheses and multiplicity control important whenever confirmatory statistical testing is undertaken.