Rivaroxaban versus warfarin to treat patients with thrombotic antiphospholipid syndrome, with or without systemic lupus erythematosus (RAPS): a randomised, controlled, open-label, phase 2/3, non-inferiority trial

Summary Background Rivaroxaban is established for the treatment and secondary prevention of venous thromboembolism, but whether it is useful in patients with antiphospholipid syndrome is uncertain. Methods This randomised, controlled, open-label, phase 2/3, non-inferiority trial, done in two UK hospitals, included patients with antiphospholipid syndrome who were taking warfarin for previous venous thromboembolism, with a target international normalised ratio of 2·5. Patients were randomly assigned 1:1 to continue with warfarin or receive 20 mg oral rivaroxaban daily. Randomisation was done centrally, stratified by centre and patient type (with vs without systemic lupus erythematosus). The primary outcome was percentage change in endogenous thrombin potential (ETP) from randomisation to day 42, with non-inferiority set at less than 20% difference from warfarin in mean percentage change. Analysis was by modified intention to treat. Other thrombin generation parameters, thrombosis, and bleeding were also assessed. Treatment effect was measured as the ratio of rivaroxaban to warfarin for thrombin generation. This trial is registered with the ISRCTN registry, number ISRCTN68222801. Findings Of 116 patients randomised between June 5, 2013, and Nov 11, 2014, 54 who received rivaroxaban and 56 who received warfarin were assessed. At day 42, ETP was higher in the rivaroxaban than in the warfarin group (geometric mean 1086 nmol/L per min, 95% CI 957–1233 vs 548, 484–621, treatment effect 2·0, 95% CI 1·7–2·4, p<0·0001). Peak thrombin generation was lower in the rivaroxaban group (56 nmol/L, 95% CI 47–66 vs 86 nmol/L, 72–102, treatment effect 0·6, 95% CI 0·5–0·8, p=0·0006). No thrombosis or major bleeding were seen. Serious adverse events occurred in four patients in each group. Interpretation ETP for rivaroxaban did not reach the non-inferiority threshold, but as there was no increase in thrombotic risk compared with standard-intensity warfarin, this drug could be an effective and safe alternative in patients with antiphospholipid syndrome and previous venous thromboembolism. Funding Arthritis Research UK, Comprehensive Clinical Trials Unit at UCL, LUPUS UK, Bayer, National Institute for Health Research Biomedical Research Centre.


Contents Page
Adequate contraception is defined as hormonal contraception or barrier method contraception.

Subtherapeutic anticoagulant therapy
Patients who develop a new venous thromboembolic event while on warfarin with an INR (International Normalised ratio) of <2.0 (target INR 2.5, i.e. range 2.0-3.0) will be considered to have had the event when subtherapeutically anticoagulated. The same would apply if the patient was on a dose of another anticoagulant (e.g. low molecular weight heparin) administered at less than the manufacturer's recommended therapeutic dose for VTE at the time of the event.

Thrombotic APS
Thrombotic APS is defined as venous thromboembolism (VTE) associated with persistent positivity of one or more antiphospholipid antibodies (aPL; i.e. lupus anticoagulant, IgG and/or IgM anticardiolipin and/or anti beta 2 glycoprotein 1 antibodies > the 99th centile. Persistent aPL positivity is defined as aPL present on at least two consecutive occasions at least 12 weeks apart as defined in the International (Sydney) Consensus Statement criteria.

Clinical outcome events
All thrombotic events, defined below, will be classified by the PI at the trial site.

Venous thromboembolism
The diagnosis of symptomatic recurrent venous thromboembolism (VTE), i.e. deep-vein thrombosis (DVT) or non-fatal or fatal pulmonary embolism (PE) is based on the following: DVT is diagnosed objectively on venous doppler or duplex scanning or venography.
Fatal PE: the diagnosis of fatal PE is based on objective diagnostic testing, autopsy, or death which cannot be attributed to a documented cause and for which PE cannot be ruled out (unexplained death).

Other thrombotic events Arterial thrombosis/thromboembolism
Stroke is defined as a sudden focal neurologic deficit of presumed cerebrovascular aetiology that persists beyond 24 hours and is not due to another identifiable cause. An event matching this definition but lasting less than 24 hours is considered to be a transient ischemic attack. The diagnosis of stroke is based on brain imaging (computed tomography or magnetic resonance imaging).
Systemic embolism is defined as abrupt vascular insufficiency associated with clinical or radiological evidence of arterial occlusion in the absence of another likely mechanism (e.g. atherosclerosis, instrumentation, or trauma).
Myocardial infarction (MI) is defined by typical symptoms, cardiac biomarker elevation and electrocardiogram changes, or confirmation at autopsy.
Deaths are classified as either vascular (e.g. due to stroke, embolism, myocardial infarction, or arrhythmia) or non-vascular (e.g. malignancy or infection).

Microvascular thrombosis
Thrombosis in the microvasculature, generally associated with evidence of organ dysfunction in patients with thrombotic APS, is diagnosed on histological examination of a tissue biopsy.

Lack of Drug Effect (LODE)
Non-serious adverse events (AE) will be considered due to LODE if the investigator explicitly states that the non-serious adverse event occurred due to the drug not working. A non-serious AE would not be considered to be due to LODE if e.g. there has been a medication error (insufficient dose was used, prescribed dosing regimen was not followed), handling error (instructions for use not followed, product stored incorrectly or used after expiry date) or there was a misuse or abuse of the product.

Lupus anticoagulant (LA) testing in patients on VKA
At least one of the following should be positive: 1) Taipan snake venom time (TVT) with a platelet neutralisation procedure or ecarin time as confirmation.
If the screening step on the mixture is abnormal, this may be taken as grounds for considering that an inhibitor is present and the confirmatory step will demonstrate phospholipid dependence. The calculation of results should be according to the laboratory standard operating procedure (SOP)/manufacturers' recommendations.
Points to note: • The DRVVT and TVT are appropriate, but other methods may also be used (e.g. Acticlot dilute PT). 2. The protocol states: "Regimen for patients converting from warfarin to rivaroxaban: • For patients randomised to rivaroxaban, the warfarin should be stopped on the day of conversion from warfarin to rivaroxaban. This means that patients randomised to rivaroxaban should not take the next due dose of warfarin after randomisation, which will generally be either that evening or the following morning.
• If the INR is ≤2.5, the rivaroxaban should be started the following morning. This means that if the INR is ≤2.5 on the baseline sample taken on the day of randomisation, then the rivaroxaban should be started the next morning (the patient having not taken the warfarin the evening of randomisation/the following morning -and of course from then onwards). Therefore the rivaroxaban will be started around 24 hours or more after the last warfarin dose.
• If the INR is 2.6 to ≤3, the rivaroxaban should be started 2 days later in the morning. This means that if the INR is 2.6 to ≤3, the rivaroxaban will be started around 48 hours or more after the last dose.

F. Blinded review of bleeding events
Bleeding events across all sites were psuedoanonymised and reviewed by a single investigator, removing the potential for bias and inter-operator variation. Bleeding events were categorised as 'minor,' 'clinically relevant non-major,' and 'major' using the following criteria as defined in the RAPS Protocol Glossary (p 7-8). There was no major bleeding, so the bleeds were categorised as minor or clinically relevant non-major.

Major bleeding
Major bleeding is defined as clinically overt bleeding associated with any of the following: fatal outcome, involvement of a critical anatomic site (intracranial, spinal, ocular, pericardial, articular, retroperitoneal, or intramuscular with compartment syndrome), fall in hemoglobin concentration of at least 20 g/L, transfusion of 2 or more units of red blood cells, or permanent disability.

Clinically relevant non-major bleeding
Clinically relevant non-major bleeding is defined as overt bleeding not meeting the criteria for major bleeding but associated with medical intervention, unscheduled contact (visit or telephone) with a physician, temporary interruption of study drug (i.e. delayed dosing), or associated with any other discomfort such as pain or impairment of activities of daily life.

Minor bleeding
Bleeding events that do not meet the criteria of major or clinically relevant non-major bleeding are defined as minor.  The primary objective is to demonstrate that the intensity of anticoagulation achieved with rivaroxaban is not inferior to that with warfarin, by measurement of the dynamics of ex vivo thrombin generation using the thrombin generation test (TG) with the endogenous thrombin potential (ETP) as the key parameter.

G. Statistical Analysis
Population studied: 116 patients with thrombotic APS, with or without SLE.
Trial design: RAPS is a two arm unblinded phase II/III prospective randomised controlled non-inferiority clinical trial. Patients on warfarin will be randomly allocated to either continue with their standard care or to change to rivaroxaban. Each patient will have a six month treatment period and a final visit 30 days after the end of trial treatment.
Sample size: Using a one-sided 2.5% significance level and 80% power it is calculated that 51 subjects per group are required for the study. Allowing for a 12% drop-out rate, a total of 116 patients will need to be randomised.
Randomisation: Random permuted blocks of varying length will be used to equally allocate participants to either remain on warfarin or switch to rivaroxaban, stratified by site and patient type.
This trial will not be blinded or masked as potentially this could be unsafe for the participants.

Primary outcome
The primary outcome measure is the percentage change in ETP from randomisation to day 42. ETP is measured in nmol/L per minute (nM/min). 3. Percentage time in therapeutic range (TTR) between baseline and day 180 in patients on warfarin.

EQ-5D-5L
The EQ-5D-5L consists of a self-reported matrix comprising 5 items or dimensions (i.e., mobility, self-care, usual activities, pain-discomfort and anxiety-depression) rated on 5-point scales ranging from 0 to 4 and a selfrated health state 100mm visual analogue scale (VAS). 4 Respondents' ratings can be combined into a single health utility score.
Scoring: Value sets based on preferences directly elicited from representative general population samples to derive the EQ 5D 5L health utility score are not yet available. If they are not disseminated by the time follow-up CRF forms are completed, we shall use the UK value sets of the "EQ-5D-5L Crosswalk Index Value Calculator (CIVC), 5 available for download at the EuroQol Group website, to obtain the EQ 5D 5L health utility score. As its name implies, the CIVC is a "crosswalk" between the EQ-5D-3L value sets and the new EQ-5D-5L descriptive system resulting in crosswalk value sets for the EQ-5D-5L.
Missing items: If the instrument has no more than one item missing (i.e. 20%), then we shall impute the missing item by the weighted mean of the completed items and score it as described above. If the instrument has more than 1 missing item (i.e.>20%), then the score will be set to missing.

TTR
The time in therapeutic range (TTR) is a method of summarising INR control over time. Copies of CRFs are included in the Trial Master File (TMF).

Management of datasets
At the time of analysis:  A copy of each dataset will be prepared by the Trial Statistician (frozen dataset, downloaded from the database) and saved in section 3 'Annotated output from statistical analysis programme' of the Statistical Master File (eSMF).
 If necessary, data can be added to or amended in the MACRO database after data download.
 If any outstanding queries are resolved during the analysis that relate to data in the frozen dataset (e.g., problems that are found during analysis or amended CRFs that are returned to the University College London Comprehensive Clinical Trials Unit (CCTU), the MACRO database and frozen dataset should both be updated.
 If any outstanding data queries are resolved while the analysis files are being prepared (when only a practice dataset has so far been copied), the changes need only be made to the MACRO database and an updated frozen copy created in section 3 of the eSMF.

Data completion and schedule
The last patient for the RAPS trial was randomised on the 11 th November 2014. All forms for 210-day follow-up should therefore be available by mid June 2015, with leeway of 14 days for study visits.

Data verification
Data verification, consistency and range checks are performed during data entry, as well as checks for missing data (copies of these checks can be found in the TMF). Additional range, consistency and missing data checks will be performed when the datasets for analysis are constructed, before the statistical analysis is performed. All variables will be examined for unusual, outlying, unlabelled or inconsistent values.
Any problems with trial data will be queried with the Trial Manager. If possible, data queries will be resolved; although it is accepted that due to administrative reasons and data availability a small number of problems will continue to exist. These will be minimised.

Data coding
Details of the variables, including variable coding lists are included in the metadata which forms part of the TMF.

Type of Comparison and Hypothesis
RAPS is designed as a non-inferiority trial, to demonstrate that the intensity of anticoagulation in patients on Rivaroxaban is not inferior to that obtained with warfarin as assessed by the ETP. Rivaroxaban would be regarded as non-inferior to warfarin if the percentage change in ETP is not more than 20% higher (i.e. less anticoagulant effect) than that for warfarin. The non-inferiority limit of 20% is based on the inter site assay variability of test performance 7 and on clinical relevance.
The null hypothesis (H0) is that Rivaroxaban is inferior to warfarin with respect to the mean percentage change in ETP at six weeks: H0: |μR -μW|≥ 20% The alternative hypothesis (H1) is that Rivaroxaban is not inferior to warfarin with respect to the mean percentage change in ETP at six weeks: H1: |μR -μW| < 20% Where: μR = mean percentage change in ETP in the rivaroxaban group μW = mean percentage change in ETP in the warfarin group In order to conclude non-inferiority, we need to reject the null hypothesis.

Sample size: Primary outcome
Using information on the ETP in patients on warfarin at a comparable INR range the standard deviation of % change in ETP was 36% six weeks after starting warfarin treatment. 8 Using a one-sided 2.5% significance level and 80% power it is calculated that 51 patients per group are required for the study. Based on our experience with the first 84 patients recruited, 2 patients have been lost to follow-up or withdrawn before the first trial visit post randomisation, and 5 patients have not provided paired ETP values. If we conservatively allow for 12% non-evaluable patients, it is anticipated that a total of 116 patients will need to be randomised. The sample size has not been adjusted for non-compliance since, unlike in a superiority trial, it does not necessarily reduce the power for a non-inferiority design. 9

Secondary outcomes: efficacy and safety
The trial is not powered to detect differences between the two randomised groups for secondary efficacy (i.e. recurrent thrombosis) or safety endpoints (i.e. SAEs and bleeding events).

Intention-to-treat (ITT) or per-protocol?
To retain the validity of the randomisation process all analyses will be by intention-to-treat, and will include all consented patients for whom outcomes are available.
In addition, a per-protocol analysis will be performed for the primary outcome (see section 7.4.2)

Significance level of tests
All confidence intervals will be 95% and two-sided. Statistical hypothesis tests will use a two-sided p value of 0.05, unless otherwise specified. There will be no formal adjustment of p values for any interim analyses performed.

Baseline comparability
Baseline characteristics will be summarised by randomised group.

Adjustment for design factors
Since randomisation was stratified by site (i.e. UCLH and Guy's and St. Thomas') and type of patient (i.e. SLE, non-SLE), analyses of continuous outcomes will involve adjustment for these two factors as recommended in ICH E9. 10 Treatment effects are then estimated conditional on site and type of patient. The model for the primary outcome will be adjusted for baseline ETP values. Similar adjustments will be made for other continuous secondary outcomes if a baseline value is available.
Adjustment for design factors will not be made for binary secondary outcomes since there are likely to be too few events to fit logistic regression models.

Follow-up and losses to follow-up: handling missing data
The sample size estimation assumed 12% of patients would not provide evaluable ETP measurements. We have examined our actual retention rate to the trial and at the time of writing this report two participants have dropped out from the 84 randomised to date. However, the primary outcome is not available for five patients as the TG test failed to give a result. This might be due to the blood samples being haemolysed. Haemolysis occurred in vitro, this means it is caused by the way in which the blood samples are drawn and treated and not related to any medical condition. Hence, patients are not systematically different from those providing viable samples and therefore it is reasonable to assume that these results are missing completely at random (MCAR).
Allowing for a conservative 12% drop out rate, data will be assumed to be MCAR and we will perform a complete case analysis. This approach is consistent with current practice; in a randomised controlled trial setting with missing data only in the outcome measure, multiple imputation offers no advantage over complete case analysis using analysis of covariance (ANCOVA) or mixed models, where the fraction of missing data is small and MAR can reasonably be assumed. 11

Summarising models
Where possible, analysis of outcomes will involve a parametric model. Treatment effect estimates will be presented as regression coefficients and 95% confidence intervals.
If necessary, ETP and other continuous outcomes which are not expected to be normally distributed will be log transformed for the statistical analysis. Estimates and 95% confidence intervals on the log scale will be back transformed to percentage changes for presentation as the difference between two logarithms is the log of their ratio.

ANALYSIS DETAILS
The results of the analyses will be reported following the principle of the ICH E3 guidelines 12 on the Structure and Content of Clinical Study Reports.

Recruitment and follow-up patterns
Patients screened but not enrolled in the Trial and reasons for exclusions will be reported by site.
Recruitment will be presented by year and site.
The number of outcomes completed -excluding patients who have been withdrawn from therapy and were unwilling to continue follow-up, will be reported by treatment group.
The number of patients who have been withdrawn from therapy, were unwilling to continue follow-up or died while on study, will be reported by treatment group.

Baseline Characteristics
Baseline characteristics will be reported for each treatment arm. Summary measures for the baseline characteristics of each group will be presented as mean and standard deviation for continuous (approximate) normally distributed variables, medians and interquartile ranges for non-normally distributed variables, and frequencies and percentages for categorical variables.

Trial treatment
Adherence to treatment will be summarised by treatment group.

Primary analysis
We shall use a linear regression model to estimate the difference in ETP between the two treatments (rivaroxaban -warfarin) at day 42 together with a two-sided 95% confidence interval, adjusting for the stratification variables (i.e. site and patient type). Baseline ETP will be included as a covariate. ETP will be log transformed for the statistical analysis. Results will be back transformed and presented as geometric means (GM) and 95% confidence intervals (CI), or ratios and 95% CI. If the upper end of the 95% CI does not cross the non-inferiority limit of 20%, then rivaroxaban will be regarded as non-inferior (see B and C in the figure below). If the mean ratio is less than one and the 95% CI does not include a ratio of one then we can conclude that rivaroxaban is superior to warfarin (see A in the figure below). 1 1 7.4.2 Sensitivity analyses for the primary outcome A sensitivity analysis will be performed by fitting a tobit regression model which can handle censored values, below the limit of detection of the assay. In addition, a per-protocol analysis will be performed to assess the treatment effect in adherent patients.

Secondary analyses
As stated in section 5 the trial is not powered to detect differences between the two randomised groups for secondary efficacy or safety endpoints. Therefore, for secondary outcomes the differences between the two groups will be summarised using estimates and confidence intervals.

a) Efficacy
Continuous secondary outcome measures of the parameters derived from the thrombin generation curve (i.e. lag-time, time to peak, peak) and markers of in vivo coagulation activation (i.e. prothrombin fragment 1.2, thrombin-antithrombin complex and D-dimer) will be analysed using a linear regression model to estimate differences between the two treatments (rivaroxaban -warfarin) at Day 42 together with a two-sided 95% confidence interval. Stratification variables (i.e. site and patient type) and baseline values will be included as covariates in the models. Estimates of treatment effects with 95% CI will be presented.
Differences between the two arms for binary efficacy outcomes 1 and 2 described in section 3.2 will be analysed using Fisher's exact tests. Summary measures will be the number (%) of patients with an event in each group.
Treatment effects will be estimated by the difference in event rates and 95% CI for the differences.

b) Safety
Summary measures will be the number (%) of patients with a SAE in each group. Treatment effects will be estimated by the difference in event rates and 95% CI for the differences.

c) Compliance
The rivaroxaban anti-Xa assays in the patients on rivaroxaban (at Day 42) and the INRs and factor X amidolytic assays in patients on warfarin (at baseline and Day 42) will be correlated, using Pearson's correlation coefficient or Spearman's rank correlation coefficient if data is clearly not Normally distributed (following a log transformation, if that improves Normality), with log transformed ETP to assess whether there are any relationships. In addition, ETP and anti-Xa assays in patients on rivaroxaban will be correlated with the time elapsed since rivaroxaban ingestion on the day of blood sample collection.
Time in Therapeutic Range (TTR) for patients on warfarin will be summarised using means and standard deviations.

7.4.4
Regression diagnostics Diagnostic checks to assess the appropriateness of the regression models fitted will be through the use of residual plots. We will plot:  Histograms and probability plots to assess normality  Scatterplots of residuals against fitted values to assess constant variance and linearity, and to identify potential outliers

Subgroup analyses
There are no planned subgroup analyses.

Quality of Life
Quality of life as measured by the EQ-5D-5L will be analysed using a linear regression model to estimate the differences between the two treatments (rivaroxaban -warfarin) at Day 180. Stratification variables (i.e. site and patient type) and baseline measurements will be included as covariates in the models. Estimates of treatment difference with 95% CI will be presented.

H. RAPS exploratory analyses
Adjusting TG parameters for LA status We estimated linear regression models to estimate the difference in TG parameters between the two treatments (rivaroxaban -warfarin) at day 42 together with a two-sided 95% confidence interval, adjusting for the stratification variables (i.e. site and patient type) and LA status. Baseline values for each respective outcome were included as covariates in the models.
We also performed a subgroup analysis on LA status (i.e. LA positive vs. not detected) by testing for the interaction between treatment and LA status within each regression model to answer the question of whether any observed differences between treatment groups vary depending on LA status.
As for the primary and secondary analyses, outcomes were log transformed for the exploratory analyses. Results were back transformed and presented as geometric means ratios (GMR) and 95% confidence intervals (CI).
The exploratory subgroup analysis did not demonstrate any significant interactions between the effects of rivaroxaban and LA positivity at baseline for any of the four TG parameters. Results are presented in the tables below.

I. Transforming data
Primary and continuous secondary efficacy outcomes were log transformed for the statistical analyses to achieve normality. Results were back-transformed by anti-logging at the end of the analysis to be meaningfully interpreted. It is important for the non-statistical audience to understand that when estimating the difference between two means in log-transformed data, the back-transformation produces a ratio. For instance, to estimate the mean difference in ETP between the two treatments (rivaroxaban -warfarin) at day 42 we fitted a linear regression model adjusted for stratification variables (i.e. site and patient type) and baseline values in the log transformed data. The results yield a difference in log means (rivaroxaban -warfarin) of 0·7 (95% CI 0·5 to 0·9, p < ·0001). When back-transforming the results by anti-logging the geometric mean ratio (GMR) rivaroxaban: warfarin of the changes in ETP from baseline to day 42 in the two groups was 2·0 (95% CI 1·7 to 2·4, p < ·0001) which indicates that the mean % change in ETP from baseline to day 42 significantly doubles in patients on rivaroxaban in comparison to patients treated with warfarin. This can be expressed in terms of the percentage change, we can say that for those patients who switch their treatment from warfarin to rivaroxaban, we can expect to see a 100% (95% CI 70 to 140, p < ·0001) significant increase in the mean ETP.

J. Withdrawals, losses to follow-up, and missing outcome data
We randomised 116 patients; 57 to receive rivaroxaban and 59 to continue to receive warfarin.
In the rivaroxaban group: Patient T014 did not attend (DNA) the day 180 trial visit after experiencing an intestinal perforation.
In the warfarin group:  Patient T042 withdrew before the first trial visit (day 42) due to travel distance and time related issues, and because she did not get rivaroxaban. No follow-up data was obtained.
 Patient T047 withdrew after the first trial visit (day 42) as the general practitioner (GP) offered to start the patient on rivaroxaban outside the trial. This patient, therefore, did not continue on the allocated treatment. The patient did not attend the day 90 visit, returned to the clinic for the day 180 assessment and did not attend the day 210 visit.
 Patient T051 was lost to follow-up after the first trial visit (day 42). The site tried to contact this patient on several occasions without success.
 Patient U024 died after the third trial visit (day 180) due to high grade B-cell non Hodgkin lymphoma stage IV B and chronic obstructive pulmonary disease. The overall proportion of censored and missing data for this trial was small. Table S4 shows the proportion of cases excluded from the analyses for each outcome measure by treatment arm. Note that as explained in the Statistical Analysis Plan (above) and the methods section of the RAPS Trial manuscript, values below the lower limits of detection (LLOD) of the assays for the thrombin generation parameters and rivaroxaban levels (i.e. censored values) were excluded from the statistical analysis. Peak ( Rivaroxaban anti-factor Xa (µg/L) -day 42  Patient T047 allocated to warfarin withdrew from trial treatment after the patient's first trial visit (day 42) as his general practitioner (GP) offered to start him on rivaroxaban outside the trial. The patient did not attend the day 90 trial visit.
However, he returned to the clinic for the day 180 assessment. The remaining 58 patients remained in their allocated warfarin treatment with 6/58 (10%) of them switching to rivaroxaban after trial completion.  The proportion of patients that interrupted their treatment at some stage during the trial was similar in both treatment arms. The most common reason for temporarily interrupting the treatment was the patients missing at least one dose of their medication.
The 3 patients who did not make it to Day 180 also remained in their allocated treatment for the time they were in the study, and only one of them temporarily interrupted her treatment (i.e. warfarin) due to an increased INR.
Note that reasons for interruptions have been collected as free text variables in the CRFs. Of the 33 patients that had their treatment interrupted at some stage during the trial, 27 (14 allocated to rivaroxaban and 13 to warfarin) answered "other" to the reason for interrupting treatment.