Sulodexide for the Prevention of Recurrent Venous Thromboembolism

Supplemental Digital Content is available in the text.


Proportion of events assigning the outcome to patients lost to follow-up by propensity score
A sensitivity analysis of the outcome was also performed by assigning the outcome to the patients lost to follow-up by propensity score.
If we assume that the risk of recurrence among those who abandoned the study is determined by the factors considered putative predictors of the event -with the exclusion of treatmentwe can estimate the propensity score for recurrence from the monitored patients. From the relevant equation, we can estimate the score for those lost to follow-up; subsequently the patients lost to follow-up are assigned the status (event/no-event) of the nearest neighbor.
We estimated the propensity score for having the primary event using the data from the 586 patients who either had the event or reached the 24 months without event. As predictors, the same used for the Cox survival analysis were employed, once considering treatment and once not considering treatment.
The equations estimating the propensity score were then applied to the 29 patients lost to follow-up.

CASE: CONSIDERING TREATMENT IN THE EQUATION
The 29 patients lost to follow-up were assigned the outcome exhibited by the subject of the same treatment group, having the nearest propensity score. This assigned 1 case among placebo and none among treated to the category FAILURE. The resulting estimate of the proportion of events was 31/308 (10.1%) among controls, and 15/307 (4.9%) among treated (P=0.021, Fisher's test; incidence risk ratio: 0.49 [0.27-0.88]).
We repeated the same procedure, assigning to the 29 cases the outcome exhibited by the subject with the nearest propensity score, regardless of the treatment group. This assigned 2 cases among placebo and none among treated to the category FAILURE. The resulting 4 estimate of the proportion of events was 32/308 (10.4%) among controls, and 15/307 (4.9%) among treated (P=0.014, Fisher's test; incidence risk ratio: 0.44 [0.22-086]).

CASE: NOT CONSIDERING TREATMENT IN THE EQUATION
The 29 cases were assigned the outcome exhibited by the subject with the nearest propensity score, regardless of the treatment group. This assigned 0 cases among placebo and 1 among treated to the category FAILURE.
Regardless of the approach taken, the results consistently confirmed that the probability of having a recurrence of DVT/PE was significantly greater among controls than among treated patients.
The variations that could be seen with the different procedures to assign outcomes to the patients lost to follow-up affected the size, but not the direction, of the effect.

NNT estimates for the primary clinical endpoint (recurrence of DVT)
We estimated the NNT to avoid one event more of recurrent DVT/PE in two years with the indicated dosage scheme of sulodexide added to the standard of care, vs. the standard of care alone. Since the probability of recurrence was estimated under different assumptions and with different techniques, several different estimates of NNT were computed.

ESTIMATES FROM THE ABSOLUTE RISK REDUCTION
The most common estimate of NNT is from the absolute risk reduction that, however, in this study should be estimated under the different assumptions made about the cases lost to follow-up.
However, these estimates do not take into account neither the actual exposure to treatment, nor the effect of potential confounders that, even in a randomised study, is definitely evident (as shown by the significant effects of predictors at the Cox analysis). We therefore estimated the NNT from the Kaplan-Meyer procedure, the unadjusted NNT from the Cox regression analysis and the NNT from the adjusted Cox regression analysis (using the covariates Further studies, which will allow estimating the NNT from the summary measure of effect, would allow to better estimate the point NNT and to reduce the width of the confidence interval.

Results in the per-protocol population
The per-protocol population was composed of 521 patients, of whom 263 received sulodexide and 258 received placebo (Figure 1). Venous thromboembolism recurred in 44 patients (one patient with a primary event was excluded from this population because of a major protocol violation) and was due to deep-vein thrombosis in 36 patients and to pulmonary embolism in 8 patients (fatal in 1 patient).
The primary outcome, recurrence of venous thromboembolism, occurred in 14 of the 263 patients who received sulodexide, as compared with 30 of the 258 patients who received placebo (hazard ratio, 0.45; 95% CI, 0.24 to 0.85; P = 0.014).

Unplanned subgroup analysis of the incidence of primary events
We estimated the risk ratio of recurrence in different subgroups of potential prognostic relevance, after exclusion of the cases lost to follow-up. The analysis was performed with epiR in R. No formal comparison was performed across subgroups, since the 95% confidence intervals are already sufficient to estimate the extent of superposition across levels of subgroups, and the displacement of the individual estimate from the overall estimate of the effect.
This unplanned subgroup analysis was performed with the exclusive aim of detecting whether there was any major discrepancy across potentially important subgroups, that could suggest major modifications to protocol in future randomized controlled trials. Indeed, being the analysis unplanned, any possible difference seen by subgroups levels, could only be considered a hypothesis-generating finding.
The results are summarized in Figure S-1.

Secondary vascular events
Five patients had distal leg DVT (4 randomized to placebo vs. 1 randomized to sulodexide), 10 had superficial vein thrombosis (6 vs. 4), and 5 had arterial events considered secondary endpoints (3 vs. 2). The incidence of these events did not differ between groups, although each of these events occurred more frequently among controls. The number of patients who had any one of these secondary events was 13/308 among the patients randomized to placebo, and 7/307 among those randomized to sulodexide (4.2% vs. 2.3%), without evidence of a significant difference (P=0.26).
Some arterial events were considered secondary study endpoint (AMI, stroke, peripheral ischemia); others were not (identification of carotid stenosis or peripheral artery thrombosis).