A systematic review and economic evaluation of cilostazol, naftidrofuryl oxalate, pentoxifylline and inositol nicotinate for the treatment of intermittent claudication in people with peripheral arterial disease

Cilostazol

For CRD quality assessment items, studies scored well in most cases and for most items, with some exceptions. Sequence generation and allocation concealment both scored poorly across studies, with most studies failing to report on these items. Imbalances between dropouts was poorly reported (Elam et al. ,64 Dawson et al. ,63 Money et al. ,62 Hiatt et al. ,49 Otsuka 21-95-20134 and Otsuka 21-94-30134) and may be a source of bias. There was some evidence of selective reporting in the Strandness et al. ,56 Elam et al. 64 and Dawson et al. 58 trials, as additional data were found in published systematic reviews. These were mostly AE data and have been incorporated in this review, but they highlight the possibility that there are additional unreported data. It is unclear how this may affect current estimates of the study outcomes. For EMA items, quality was largely good, although there is potential for some problems, mainly due to poor reporting. Treatment duration varied between studies, with only the Strandness et al. ,56 Beebe et al. ,61 Hiatt et al. ,64 O’Donnell et al. ,51 Dawson et al. ,58 Otsuka 21-94-30134 and Otsuka 21-98-21334 trials treating patients for at least 24 weeks (this will not affect the meta-analysis, which considers only studies that treated patients for 24 weeks). The use of concomitant treatment was poorly reported, and studies did not generally state that they had stratified for diabetes. Less than half of the studies stated that only patients with a < 25% change in baseline walking distances were selected, and this may introduce unwanted variability to the results. However, there does not appear to be clinical evidence to suggest that these patients respond differently to treatment. A placebo run-in period was reported only by Dawson et al. 63 and Hiatt et al. 49 – for between 2 and 6 weeks in both cases. All studies reporting walking distance outcomes used a standardised treadmill test. There was, however, heterogeneity in tests between protocols, which is discussed elsewhere in this report.

Naftidrofuryl oxalate

Studies of naftidrofuryl oxalate scored moderately well overall for both CRD and EMA items. Items that scored poorly were sequence generation and allocation concealment, as most studies scored unclear for these items. Baseline characteristics may influence results, as Trubestein et al. 67 and Ruckley et al. 68 did not score positively according to CRD criteria in that the former had more smokers in the intervention group and the latter did not report baseline characteristics. It is unclear what effects this would have on results. For EMA items, more specific problems with patient characteristics were identified as follows: concomitant treatment was unclear in every case; the distribution of diabetics was stratified only in the Kieffer et al. 65 trial and proportions of diabetics are unknown for the Adhoute et al. ,66 Trubestein et al. 67 and Ruckley et al. 68 trials; only Kieffer et al. 65 and Adhoute et al. 66 selected patients with a < 25% change at baseline measurements. Other EMA items were generally well addressed.

Pentoxifylline

Overall, studies were of mixed quality and some items may impact on estimates of treatment effect. Among the CRD quality items, sequence generation and allocation concealment may present problems, with two-thirds of studies scoring unclear. Items that were of mixed quality included the use of an ITT analysis, follow-up of at least 80% of participants, imbalances between dropouts and selective reporting of outcomes. EMA items were also only partially fulfilled. Although diagnosis, history of condition and treatment duration were mostly good, other items were mixed. Among items relating to patient characteristics, it was largely unclear whether or not concomitant treatment was comparable across groups; there may have been imbalances in the numbers of diabetics, and patients may not have always been selected on the basis of having a < 25% change in baseline assessments. Outcomes may also have been affected by the lack of a 2- to 6-week placebo run-in. 34,38,58

Inositol nicotinate

Overall, studies of inositol nicotinate scored well for most CRD quality assessment items but very poorly for the EMA items. This reflects the age of the studies and is likely to introduce a considerable degree of inaccuracy to the study findings. Among the CRD quality assessment items, methods of randomisation and treatment allocation were poorly reported in every case. Baseline characteristics were not similar in the study by Head. 81 All studies stated that they were double blind. An ITT analysis was provided in every case, and at least 80% of participants were followed up in the final analysis. Imbalances in dropouts were not reported or did not occur, and this seems unlikely to affect results. There is no evidence of selective reporting within the studies. Several EMA items scored poorly or were unclear. Only Kiff and Quick80 stated that IC was objectively diagnosed, and only this same study stated that patients had a 6-month history of the condition. None of the studies treated patients for 24 weeks or longer, and it was unclear in every case whether or not concomitant treatments were comparable across groups. Kiff and Quick80 and Head81 did not stratify for diabetes, and did not report how many were diabetic in each group. Although MWD and/or PFWD were reported in O’Hara et al. 78 and Head,81 neither of these studies used a treadmill test, although the alternative walking distance tests used did follow a standard protocol.

Maximal walking distance narrative summary

Details of MWD results, where reported, are shown in Tables 12–17 (from published trial reports) and Appendix 4 (which includes details from reviews and the manufacturer’s submission). Across trials, there was a tendency for all groups, including placebo groups, to show improvement with time. Of the 10 studies of cilostazol 200 mg versus placebo comparison, seven significantly favoured cilostazol over placebo,56,58,61–64,83 whereas three trials (the three unpublished Otsuka trials34) did not find any significant difference between groups. As patient populations were similar across trials, in terms of disease, diabetes, hypertension, smoking and age range, these characteristics cannot explain any significant differences between treatment groups. Other issues of trial design were similar across trials: all were blinded, randomised and presented ITT analyses, and all measured baseline walking distance with two treadmill tests. As the graded test encourages longer walking distances than the constant-load protocol, absolute mean walking distance in metres is not directly comparable between protocols (see Chapter 3, Measurement of disease). 28 The use of the treadmill protocol (see Appendix 4) may go some way to explaining the presence of heterogeneity across trials. All three trials using the graded test treadmill protocol reported a significantly greater improvement in the cilostazol group than in the placebo group: Dawson et al. 58 at 24 weeks’ follow-up (p = 0.0005), Money et al. 62 at 16 weeks’ follow-up (p < 0.05) and Elam et al. 64 at 12 weeks’ follow-up (p = 0.004). However, treadmill protocol does not explain why some trials report significant differences and others do not, as this differs between trials using the same treadmill protocol (the constant treadmill protocol). Trials with non-significant results do not have shorter follow-up or smaller sample sizes than trials with significant results.

Of the seven trials using the constant-workload treadmill protocol, four56,61,63,83 reported a significantly greater improvement in the cilostazol group than in the placebo group, although for one of these trials83 significance was only borderline; three of these had follow-up time of 24 weeks [O’Donnell et al. 83 (p = 0.048), Strandness et al. 56 (p = 0.0003), Beebe et al. 61 (p < 0.001)], and the fourth trial, Dawson et al. ,63 had a follow-up time of 12 weeks (p < 0.01). The three trials34 that did not find any significant difference between groups used the constant-workload treadmill protocol, and two of these trials had follow-up times of 24 weeks (Otsuka 21-94-301, p = 0.06;34 Otsuka 21-98-213, p = 0.9134), and the other had a follow-up time of 12 weeks (Otsuka 21-95-201, p = 0.9034). Lack of significant treatment effect cannot be explained by sample size, as these three trials did not have smaller sample sizes than the other trials (see Appendix 4).

The review by Pande et al. 31 included nine industry-sponsored trials, of which six trials (Otsuka trials56,57,61–64) found a significant difference between treatment groups, and three trials (the three trials without published trial reports) found no significant difference between cilostazol 200 mg and placebo groups (Otsuka trials 21–94–301, 21–98–213, 21–95–201 et al. 34). 31 The Pande et al. review31 presented a pooled analysis of these nine trials as a ratio of geometric means, and calculated an estimate of treatment effect31 of 1.15 95% [confidence interval (CI) 1.11 to 1.19], which significantly favoured cilostazol over placebo. 31 This analysis31 did not include the O’Donnell et al. trial,51 which found a borderline significant treatment effect for the whole trial population (p = 0.048) but found no significant difference between treatment groups when considering the subgroups of patients with diabetes (p = 0.09, n = 26)53 or without diabetes (p = 0.27, n = 80),83 which may reflect the small sample sizes rather than lack of actual treatment effect. The cilostazol versus placebo comparison trials that reported significant treatment effect for MWD generally also reported significant treatment effect for PFWD (see Chapter 5, Pain-free walking distance) and vice versa; however, there were a couple of exceptions in that the O’Donnell et al. 83 and Elam et al. 64 trials, found a significant treatment effect for MWD but did not find a significant treatment effect for PFWD.

Two trials for the naftidrofuryl oxalate 600 mg versus placebo comparison reported MWD; Kieffer et al. 65 reported significantly greater improvement for naftidrofuryl oxalate 600 mg versus placebo (p < 0.001), and Trubestein et al. 67 found no significant difference between groups. It may be that this difference could be explained in terms of length of follow-up, in that Kieffer et al. 65 had a follow-up of 24 weeks, whereas Trubestein et al. 67 had a follow-up of 12 weeks. These trials both used the constant-workload treadmill protocol and designs were similar in terms of having a placebo run-in, being randomised, presenting ITT analyses, measuring baseline walking distance with two tests and being blinded. There was little difference between these two trials in baseline MWD (see Appendix 4). However, both naftidrofuryl oxalate trials (Kieffer et al. ,65 Trubestein et al. 67) had higher baseline MWD than the cilostazol trials that used the constant workload treadmill protocol (Strandness et al. ,57 Beebe et al. ,61 Dawson et al. ,63 O’Donnell et al. ,51 Otuska trials 21–94–301, 21–98–213, 21–95–201) (see Appendix 4). 34,83 The Kieffer65 trial found a significant treatment effect for PFWD (see Chapter 5, Pain-free walking distance) as well as for MWD. However, the Trubestein et al. 67 trial, which had not found a significant treatment effect for MWD, did report a significant effect for PFWD favouring naftidrofuryl oxalate (see Chapter 5, Pain-free walking distance).

Of the eight trials34,58,70,71,73,76,77 comparing pentoxifylline versus placebo in terms of MWD, two trials significantly favoured pentoxifylline over placebo: Creager et al. (p = 0.039)70 and Di Perri et al. (p < 0.01). 77 Of these, the Di Perri77 trial did not use a treadmill protocol, instead measuring the distance that a patient could walk on a horizontal level at metronome controlled speed of 120 steps per minute, with a follow-up at 8 weeks. The Creager et al. 70 trial used a graded treadmill test protocol, and found a significant effect on MWD at 24 weeks. Of the six trials finding no significant difference between groups for MWD, one of these used the graded test (Dawson et al. ;58 p = 0.82) and had a follow-up of 24 weeks. The five trials using the constant-workload treadmill protocol all found no statistically significant difference between pentoxifylline and placebo groups (Gallus et al. ,76 Lindgarde et al. ,71 Porter et al. ,73 Otsuka 21-98-21334 and Otsuka 21-94-30134). Of these, the Gallus et al. 76 study had a follow-up of only 8 weeks [ratio of percentage change from baseline 1.05 (95% CI 0.81 to 1.36), which was non-significant], and the other studies had follow-up of 24 weeks: Lindgarde et al. 71 (p = 0.09), Porter et al. 73 (two-sided p = 0.32), Otsuka 21-98-21334 (p = 0.24) and Otsuka 21-94-30134 (p = 0.29). The pentoxifylline-versus-placebo comparison trials that reported significant treatment effect for MWD generally also reported significant treatment effect for PFWD (see Chapter 5, Pain-free walking distance) and vice versa; however, there were a couple of exceptions in that the Creager et al. 70 trial, which found a significant treatment effect for MWD, did not find a significant treatment effect for PFWD, and the Dawson et al. 58 trial did not find an effect for MWD but did find a treatment effect for PFWD (see Chapter 5, Pain-free walking distance).

For the comparison of inositol nicotinate 4 g versus placebo, only the Kiff and Quick trial80 reported MWD. This trial found no significant difference between inositol nicotinate and placebo groups at 12 weeks for MWD measured by patients walking at their most comfortable speed on a treadmill set at a 10% gradient. 80

Three trials reported MWD for the comparison of cilostazol versus pentoxifylline, all with 24 weeks’ follow-up (Dawson et al. ,58 Otsuka 21-98-21334 and Otsuka 21-94-30134). Two trials found no significant difference between the cilostazol and pentoxifylline groups (Otsuka 21-98-213,34 p = 0.65; Otsuka 21-94-301,34 p = 0.87), both using the constant-workload treadmill protocol. One trial (Dawson et al. 58), which used the graded test treadmill protocol, found a significantly greater improvement in MWD (p = 0.0002) in the cilostazol group than in the pentoxifylline group. 58

In the one trial (Hobbs et al. 82) comparing cilostazol (with or without supervised exercise) versus usual care (with or without supervised exercise), all treatment groups improved, but there was significantly more improvement for cilostazol added to supervised exercise or usual care (p = 0.005). 82 This trial used the constant workload treadmill protocol and measured MWD at 24 weeks.

Maximal walking distance meta-analysis

The reanalysis of the cilostazol trials included within the Cochrane review28 is presented in Table 18, in terms of change from baseline in absolute mean walking distance.

The posterior distribution of a parameter is a weighted compromise between the prior information and the sample data. In particular, if for some value of the likelihood function (expressing what is known about a parameter based on the sample data) the likelihood of that value is small, so that these data suggest that this value is implausible, then the posterior distribution will also give small probability to this value. Similarly, if for some value of the prior distribution (expressing what is known about a parameter in addition to the sample data) the prior probability of that value is small, so that the prior information suggests that this value is implausible, then, again, the posterior distribution will also give small probability to this value. In general, the posterior probability will be high for some value only when both information sources support that value. The posterior mean treatment effect for the cilostazol studies, together with the 95% credible interval, is shown in Table 19. Table 19 also shows the posterior mean of the between-study SD, together with the 95% credible interval.

The random effects meta-analysis of the change from baseline in absolute mean walking distance showed that treatment with cilostazol resulted in an increase of 57.27 m (95% credible interval 24.93 to 86.57 m) compared with placebo.

For the overall comparison of the treatment options, of the 26 studies identified by the systematic literature review, 12 studies were excluded from the meta-analysis of MWD for the reasons provided within Table 20.

The evidence base for the log of the geometric mean change from baseline in MWD and PFWD generates a network of trials comparing different pairs or triplets of treatments, as shown in Figure 2. The numbers within Figure 2 represent the number of times that specific treatment arms are compared within studies.

FIGURE 2.

Network of evidence used in the analysis of the change from baseline in log mean walking distance (log m).

The 10 studies (leading to 16 comparisons) included within the meta-analysis of MWD, represented in Figure 2, are the seven two-arm and three three-arm 24-week studies that are described in Table 21. Three 12-week studies34,63,64 and one 16-week study62 in which there were data on MWD available, as described in Table 21, were excluded from this analysis as the outcomes from these studies with a shorter follow-up period are not directly comparable.

Table 22 also shows the estimated treatment effect of cilostazol relative to placebo in the study by Money et al. ,62 for which individual arm data were not available. This study was excluded from the meta-analysis because of the 16-week follow-up.

Goodness-of-fit was assessed by calculating the arm-specific and total residual deviance. The total residual deviance was 23.03, which compares favourably with the 23 data points being analysed. The arm-specific deviance terms were not indicative of any particular sample mean, being poorly represented by the model. The posterior mean treatment effect for these studies, together with the 95% credible interval, is shown in Table 23. Table 23 also shows the posterior mean of the between-study SD, together with the 95% credible interval.

The random effects meta-analysis of the change from baseline in log walking distance showed that treatment with naftidrofuryl oxalate had the greatest effect [60.3% = 1 – exp(0.472)] relative to placebo, followed by cilostazol (24.6%) and pentoxifylline (10.6%).

The 95% credible intervals suggest that treatment with naftidrofuryl oxalate and cilostazol produces real increases in the percentage change from baseline walking distance relative to placebo, although there was some uncertainty as to the true effect.

Table 24 gives a matrix of results from the network meta-analysis (see Table 23), where the upper right cells are the pair-wise posterior means (i.e. the direct effects) and the lower left cells are the network meta-analysis posterior means. Only four direct effects were estimable, and in three cases the mean results were essentially the same as the results from the network meta-analysis but with greater uncertainty, as expected. There was a greater difference in the results for the comparison between cilostazol and pentoxifylline, with the direct effect giving approximately 4% improvement in MWD in favour of cilostazol and the network meta-analysis result giving an approximately 11% improvement in favour of cilostazol, although there was considerable uncertainty as to the true effect based on the direct evidence alone.

There was moderate between-study variation, which suggests that the treatment effect varied depending on the characteristics of the study. The trial by Strandness et al. 56 had the largest observed effect of cilostazol compared with placebo (0.394) and the Otsuka 21-98-21334 trial had the smallest observed cilostazol effect compared with placebo (0.016). The trial by Lindgarde et al. 71 had the largest observed pentoxifylline effect compared with placebo (0.190) and the trial by Dawson et al. 58 had the smallest observed pentoxifylline effect compared with placebo (–0.031).

Forest plots are shown in Figure 3.

FIGURE 3.

Posterior distribution for the change from baseline in log mean MWD for cilostazol, naftidrofuryl oxalate and pentoxifylline vs placebo.

Figure 3 suggests that the relative treatment effects are unrelated to the date of the study. Furthermore, of the trials used within the meta-analysis it is only some of the pentoxifylline trials that pre-date the last 12 years. All trials of cilostazol and naftidrofuryl oxalate were published within the last 12 years.

The percentage change from baseline to 24 weeks in MWD, estimated by projecting the treatment effects from the network meta-analysis on to an estimated placebo response, is shown in Table 25.

Tables 26–28 give the results of five sensitivity analyses, which allow for more informative prior distributions for the between-study SD and treatment effects. Tables 26 and 27 provide the results of the sensitivity analyses for the placebo mean and the treatment effects relative to placebo, respectively, whereas Table 28 provides the results for the between-study standard SD.

Table 26 shows that the placebo mean is insensitive to relatively informative prior distributions [i.e. sensitivity analysis 5 (SA5)] compared with the results based on weak prior distributions (i.e. SA1).

Table 27 shows that the treatment effects are insensitive to relatively informative prior distributions compared with the results based on weak prior distributions.

Table 28 shows that the between-study SD is relatively insensitive to more informative prior distributions than the results based on weak prior distributions, although the mean from SA5 is smaller than the results of the other analysis. This is a consequence of the prior distribution in SA5 excluding estimates considered plausible in the other sensitivity analyses, i.e. posterior estimates > 0.2.

Pain-free walking distance narrative summary

Tables 29–33 show PFWD results as reported by the trials. This may be reported as the difference in mean PFWD between baseline and final measurement, the change from baseline as a percentage or as an effect size. Details of treadmill protocols are shown in Appendix 4. As the graded test encourages longer walking distances than the constant-load protocol, absolute mean walking distance in metres is not directly comparable between protocols. 28

For the 10 studies comparing cilostazol 200 mg with placebo, five did not find any significant difference between groups (O’Donnell et al. ,83 Otsuka trials 21–94–301, 21–98–213, 21–95–201,34 Elam et al. 64), and five significantly favoured cilostazol over placebo (Strandness et al. ,57 Dawson et al. ,58,63 Beebe et al. ,61 Money et al. 62). Table 29 shows the PFWD data from the published trial reports. The review by Pande et al. 31 included nine industry-sponsored trials, of which five trials (Strandness et al. ,57 Dawson et al. ,58,63 Beebe et al. ,61 Money et al. 62) found a significant difference between treatment groups, and four trials (including the three trials without published trial reports) found no significant difference between cilostazol 200 mg and placebo groups (Otsuka trials 21–94–301, 21–98–213, 21–95–201,34 Elam et al. 64). The five trials finding a significant difference between treatment groups reported this in published trial reports (see Table 29). The Pande et al. review31 presented a pooled analysis of these nine trials as a ratio of geometric means, and calculated an estimate of treatment effect of 1.15 (95% CI 1.10 to 1.20), which significantly favoured cilostazol over placebo. 31 This analysis31 did not include the O’Donnell et al. trial. 51 The O’Donnell et al. trial51 did not find any significant treatment effect, with both the cilostazol and placebo groups showing improvement in PFWD (see Table 29). O’Donnell et al. 83 also found no significant difference between treatment groups when considering the subgroups of patients with diabetes (p = 0.14, n = 26)53 or without diabetes (p = 0.63, n = 80), although, as described above (see Maximum walking distance), this may be due to the small sample sizes.

As patient populations were similar across trials, in terms of disease, diabetes, hypertension, smoking and age range, these characteristics do not appear to explain whether or not, significant differences between treatment groups were found and nor does sample size. Of the trials using the constant-workload treadmill protocol, three out of seven favoured cilostazol over placebo (Strandness et al. ,56 Beebe et al. ,61 Dawson et al. 63), whereas four out of seven were non-significant (O’Donnell et al. ,83 Otsuka trials 21–94–301, 21–98–213, 21–95–20134). Of the trials using the graded test treadmill protocol, two out of three favoured cilostazol over placebo (Dawson et al. ,58 Money et al. 62), whereas one was non-significant (Elam et al. 64). For the graded test protocol trials, the trial with the non-significant result (Elam et al. 64) was the trial with the shortest follow-up, at 12 weeks, whereas the trials with significant results had follow-up periods of 16 weeks (Money et al. 62) and 24 weeks (Dawson et al. 58). However, length of follow-up cannot explain the difference between significant and non-significant results in the constant-workload protocol trials. For the constant-workload protocol trials with 24 weeks’ follow-up, three were non-significant in terms of PFWD comparing cilostazol versus placebo (O’Donnell et al. ,83 Otsuka trials 21–94–301, 21–98–21334), whereas two were significant (Strandness et al. ,56 Beebe et al. 61). Two of the constant-workload protocol trials had follow-up of 12 weeks, and, of these, one produced significant results (Dawson et al. 63), whereas the other one (Otsuka 21-95-20134) did not find any difference between treatment groups. Only Dawson et al. 63 specified administration of placebo during the run-in period of the study. These trials had similar designs: all were blinded, randomised and presented ITT analyses, and all measured baseline walking distance with two tests.

For the naftidrofuryl oxalate 600 mg versus placebo comparison (see Table 30), the three trials using constant-workload treadmill protocol – Kieffer et al. 65, Adhoute et al. 66 and Trubestein et al. 67 – all reported significantly greater improvement in PFWD in the naftidrofuryl oxalate group than in the placebo group. These trials had similar designs: all had placebo run-in, were randomised, presented ITT analyses, measured baseline walking distance with two tests, and were blinded with the exception that the clinicians in the Adhoute et al. trial66 were not blinded to treatment group. The three naftidrofuryl oxalate 600 mg trials using constant-workload treadmill protocol – Kieffer et al. ,65 Adhoute et al. 66 and Trubestein et al. 67 – had some variation across trials in baseline PFWD; however, all had higher baseline PFWD (see Appendix 4) than the constant-workload cilostazol trials (Strandness et al. ,56 Beebe et al. ,61 Dawson 1998 et al. ,63 O’Donnell et al. ,83 Otsuka trials 21–94–301, 21–98–213, 21–95–20134). The Spengel et al. trial47 reported significantly greater improvement in claudication distance in the naftidrofuryl oxalate 600 mg group than in the placebo group; however, this was based on patient estimates of PFWD at baseline and 24 weeks, not treadmill testing. The Ruckley et al. trial68 of naftidrofuryl oxalate 300 mg versus placebo reported that there was no significant difference between groups for PFWD at 12 weeks’ follow-up,68 as measured by patients’ normal walking pace on a level.

Of the seven trials comparing pentoxifylline 1200 mg versus placebo in terms of PFWD, five found no significant difference between treatment groups (Creager et al. ,70 Lindgarde et al. ,71 Gallus et al. 76 and Otsuka 21-98-213, 21–94–30134), whereas two significantly favoured pentoxifylline over placebo for PFWD (Dawson et al. 58 and Porter et al. 73). The Gallus et al. 76 study had a follow-up of only 8 weeks, whereas the other studies had a follow-up of 24 weeks. Three published trials (see Table 31) comparing pentoxifylline with placebo – Creager et al. ,70 Lindgarde et al. 71 and Gallus et al. 76 – reported no significant difference between treatment groups in PFWD. Two trials without published trial reports – Otsuka 21-98-21334 and Otsuka 21-94-30134 – also found no significant difference between the pentoxifylline and placebo groups in PFWD. Two trials (reported in Table 31) – Dawson et al. 58 and Porter et al. 73 – reported significantly greater improvement in PFWD for the pentoxifylline group than for the placebo group. Five of the trials comparing pentoxifylline with placebo used constant workload treadmill protocols, of which four found no significant treatment effect (Lindgarde et al. ,71 Gallus et al. ,76 Otsuka 21-98-21334 and Otsuka 21-94-30134), and one favoured pentoxifylline (Porter et al. 73). Of the two trials using a graded test protocol, one found a significant treatment effect (Dawson et al. 58), whereas the other did not (Creager et al. 70).

For the comparison of inositol nicotinate 4 g versus placebo, none of the trials reported PFWD. However, O’Hara et al. 78 measured pain-free walking paces and claudication time, and reported that there was no significant difference between groups in claudication time, and that pain-free walking paces improved significantly in both groups, with inositol nicotinate showing significantly greater improvement (p < 0.05) than the placebo group at 12 weeks. 78 The Head81 trial reported improved claudication times for both treatment groups at 12 weeks, but there was a significant difference between treatment groups (p < 0.001) only for patients with moderate disease, among whom the inositol nicotinate group (n = 24) had a significantly greater improvement than the placebo group (n = 28).

For the comparison of cilostazol 200 mg versus pentoxifylline 1200 mg (see Table 32), Dawson et al. 58 found a significantly greater improvement in PFWD for the cilostazol group than for the pentoxifylline group. Two trials without published trial reports – Otsuka 21-98-21334 and Otsuka 21-94-30134 – found no significant difference between cilostazol and pentoxifylline groups in PFWD. The Dettori et al. 69 trial did not report PFWD, but did report pain-free walking time, which was statistically more improved for patients taking pentoxifylline (p < 0.05) in an analysis including all four trial arms of the study (see Table 3 for summary table of included studies) at 1-year follow-up.

For the trial comparing cilostazol 200 mg (with or without supervised exercise) versus usual care (with or without supervised exercise) (see Table 33), all treatment groups improved but there was no significant effect of cilostazol added to supervised exercise or usual care. 82

Pain-free walking distance meta-analysis

The 10 studies included within this analysis are the same as for those used in the meta-analysis of MWD, shown in Table 21. Table 34 shows the change from baseline in log mean PFWD, including the three studies that are excluded from the meta-analysis because they had follow-up times of < 24 weeks.

Goodness-of-fit was assessed by calculating the arm-specific and total residual deviance. The total residual deviance was 23.07, which compares favourably with the 23 data points being analysed. The arm-specific deviance terms showed that the data from the Beebe et al. 61 and Strandness et al. 56 studies had the largest deviances.

The posterior mean treatment effects for these studies, together with the 95% credible intervals, are given in Table 35. Table 35 also gives the posterior mean of the between-study SD, together with the 95% credible interval.

The random effects meta-analysis of the change from baseline in log walking distance showed that treatment with naftidrofuryl oxalate had the greatest effect [64.2% = 1 – exp(0.496)] relative to placebo, followed by cilostazol (13.4%) and pentoxifylline (9.2%).

The 95% credible intervals suggest that treatment with naftidrofuryl oxalate and cilostazol produces real increases in the percentage change from baseline PFWD relative to placebo, although there was some uncertainty as to the true effect.

There was moderate between-study variation, which suggests that the treatment effect varied depending on the characteristics of the study. The trial by Strandness et al. 56 had the largest observed effect of cilostazol effect compared with placebo (0.291) and the Otsuka 21-94-30134 trial had the smallest observed cilostazol effect compared with placebo (0.003). The trial by Porter et al. 72 had the largest observed pentoxifylline effect compared with placebo (0.219) and the trial by Dawson et al. 58 had the smallest observed pentoxifylline effect compared with placebo (–0.034).

Forest plots are shown in Figure 4.

FIGURE 4.

Posterior distribution for the difference from placebo in change from baseline log mean PFWD for cilostazol, naftidrofuryl oxalate and pentoxifylline vs placebo.

The percentage change from baseline to 24 weeks in PFWD, estimated by projecting the treatment effects from the network meta-analysis on to a estimated placebo response, is shown in Table 36.

Cilostazol

Table 51 summarises the HRQoL data for cilostazol, and Table 52 further summarises SF-36 scales and subscales. Strandness et al. ,56,57 Beebe et al. ,61 Money et al. ,62 O’Donnell et al. ,53,83 (diabetics and non-diabetics), Dawson et al. ,58,63 and Otsuka 21-98-21334 (diabetic and non-diabetic participants were reported separately by one author but are drawn from one study)53,83 assessed the quality of life of study participants in trials of cilostazol using the SF-36. However, not all studies reported significance values for all summary measures and subscales, as can be seen from Table 51. It is likely that only scales that were significant were reported. Assuming this to be the case, no summary measure or subscale shows a consistent positive outcome for physical or mental health. The subscale physical function improved significantly in Strandness et al. ,56 Beebe et al. ,61 Money et al. 62 and O’Donnell et al. 83 (non-diabetics), although, of these, the magnitude of the change is only reported in Beebe et al. 61 and the effect does not seem to be strong enough to lead to significant changes in the summary physical function score. No study reported significant differences for between-group comparisons for any mental health component of the SF-36. COM was reported only by Beebe et al. ,61 and the results of this corresponded with the SF-36 results reported for the same trial.

VascuQoL was used in one study that reported diabetic (O’Donnell et al. 53) and non-diabetic (O’Donnell et al. 83) patients separately. VascuQoL scores did not correspond with SF-36 scores in either diabetic or non-diabetic patients. VascuQoL is a disease-specific measure designed for use with critical as well as chronic limb ischaemia,87 and as such may have different psychometric properties. Across the five studies that used WIQ, results were conflicting, with some significant results reported in Beebe et al. 61 and Money et al. ,62 significant trends reported in Strandness et al. 56 but no significant changes reported for the remaining two studies, O’Donnell et al. 83 (non-diabetics) and Dawson et al. 58

Naftidrofuryl oxalate

• Only Spengel et al. 47 reported HRQoL for naftidrofuryl oxalate versus placebo. The outcome measure used was CLAU-S, and results were significant across four domains (daily living, pain, social life and mood) but not for the disease-specific anxiety domain.

Pentoxifylline

• Only Otsuka unpublished trial 21–98–21334 and Creager et al. 70 reported HRQoL using the SF-36 for pentoxifylline versus placebo. Creager et al. 70 also reported WIQ. 70 No significant differences were reported.

Inositol nicotinate

• No studies reported HRQol data.

Summary

There is some evidence that cilostazol affects the physical function subscale of the SF-36, which suggests that there are some tangible improvements in physical function for the patient. This is somewhat supported by mixed evidence from WIQ, which suggests that patients perceive improvements in walking speed and distance in some cases. These health status improvements do not appear to translate into an overall improvement in HRQoL, with no changes in the mental health components, such as social functioning and the role–emotional subscale. The very limited evidence for naftidrofuryl oxalate suggests that improvements in pain are associated with improvements in daily living, social life and mood, but not with improvements in anxiety. The very limited evidence for pentoxifylline suggests that it does not improve HRQoL.

Patient population

A Markov model was developed in Excel^® (Microsoft Corporation, Redmond, WA, USA) to determine the cost-effectiveness of each drug compared with no vasoactive drugs for PAD and with the remaining vasoactive drugs. The population considered was patients who have stable (at least for the past 3 months, which is the inclusion criterion for most of the RCTs identified) and symptomatic IC, secondary to PAD. Furthermore, only patients whose symptoms continue despite a period of conservative management, such as advice to cease smoking or do more exercise, were considered by the model. The model did not distinguish between patients who are in primary care and secondary care, because no published evidence was identified to support this classification. The model also did not distinguish between patients with different severity of the disease, because the considered patient population was already narrowly defined (i.e. patients with stable IC who fail conservative management) and no published evidence was identified to support a subgroup analysis. However, an exploratory subgroup analysis was undertaken around patients with more severe IC who might receive angioplasty following drug discontinuation.

Interventions and comparators

The four drugs considered within their licensed indications for IC were cilostazol (200 mg per day), naftidrofuryl oxalate (600 mg per day), pentoxifylline (1200 mg per day) and inositol nicotinate (4 g per day). These were compared with each other and no vasoactive drugs for PAD. As it was not possible to include inositol nicotinate within the meta-analysis of MWD or PFWD, this drug has not been included within the main analysis. However, inositol nicotinate has been included within a threshold analysis to assess how many QALYs would be required for it to have a cost per QALY gained below £20,000 and £30,000 compared with no vasoactive drug.

Outcomes

The model outcome is the incremental cost per QALY gained. Owing to lack of evidence around the utilities of patients having naftidrofuryl oxalate, pentoxifylline and inositol nicotinate, change in MWD was used as a surrogate measure to estimate the change in utilities based on a regression model. MWD was selected as the surrogate because it is the primary outcome of most identified RCTs. More detail about this regression is provided below (see Estimate of model parameters). The life-years gained outcome is not presented within this analysis because the drugs are not expected to have an impact on life-years, only on patient quality of life.

Model structure

The structure of the decision model is presented in Figure 6. The model includes three main health states: vasoactive drug treatment (where patients receive one of the four drugs under evaluation); no vasoactive drug treatment (where patients receive none of the four drugs or have discontinued); and death. Patients begin in the vasoactive drug treatment state and have a weekly probability of moving to the no vasoactive drug treatment state by discontinuing. Patients also have a weekly probability of dying of any cause from both the vasoactive drug treatment state and the no vasoactive drug treatment state. Patients who do not receive any of the four drugs have zero time in the vasoactive drug treatment state. All patients are in Fontaine stage II and have had persistent IC symptoms despite a period of conservative management to be eligible to receive the vasoactive drugs for PAD. The health states are classified according to whether or not the patients are receiving vasoactive drugs for PAD rather than by progression through different disease stages (i.e. Fontaine stages II–IV), as the drugs are for symptom relief and it is assumed that they do not have an impact on disease progression.

FIGURE 6.

Diagram of the structure of the decision model.

Patients in the vasoactive drug treatment state could improve their quality of life as a result of the treatment effect. In the base-case analyses, the only extra cost for patients receiving the vasoactive drugs for PAD compared with no vasoactive drugs is the drug acquisition cost.

Patients may discontinue the drug because of AEs, death or other reasons of non-compliance. Switching to another vasoactive drug after discontinuation and returning to the same vasoactive drug after discontinuation were not considered within the model owing to lack of published evidence. In addition, expert clinical opinion suggested that this would not be standard practice within England and Wales (Steven Thomas, Jonathan Michaels and Gerard Stansby, personal communication).

Patients who have discontinued the drug therapy were assumed to incur no extra costs compared with those patients initially having no treatment. It was also assumed that the drugs are only effective while they are being given, as suggested by the study by Keiffer et al. ,65 which recorded MWD for 2 months beyond treatment discontinuation. 65 Therefore, patients discontinuing a vasoactive drug will have no extra health gains (regarding utility, walking distance or disease progression) compared with no vasoactive drug.

Time horizon

The time horizon of the model was 100 years to ensure that all differences in costs and benefits are captured within the model, and a starting age of 66 years was used to represent the average age of patients with IC. The starting age was based on the average age of patients within the CASTLE study,49 which has the longest follow-up period and the largest sample size of all RCTs (Hiatt et al. 49). A time cycle of 1 week was chosen as being sufficiently short to capture the effect of treatment, and the time period was in line with that used in the trials for measurement of walking distance and quality of life.

Discounting

All costs and QALYs were discounted at a rate of 3.5% per year.

Maximal walking distance and utilities

The majority of studies reported the change in MWD for the vasoactive drug and control arms, but only some stated that quality-of-life data were collected and only two RCTs (both for cilostazol)61,83 reported quantitative data for SF-36 quality-of-life outcomes that can be converted to utilities using published algorithms. 96 No studies of naftidrofuryl oxalate, pentoxifylline or inositol nicotinate provided sufficient quality-of-life evidence to estimate utility outcomes associated with these drugs. However, in order to compare the cost-effectiveness of these vasoactive drugs with other interventions routinely funded by the NHS, quality-of-life estimates are required for all drugs being assessed so that an incremental cost per QALY can be calculated. Given the limited published data around quality-of-life outcomes, the authors of the identified RCTs were contacted to ask for the patient-level or summary SF-36 data if the paper mentioned that the SF-36 questionnaire was used within the RCT. The aim of this was to attempt to determine a relationship between the change in MWD and the change in utility scores that could be used to estimate the utility gains associated with the drugs being assessed for which there were only MWD data and no utility data. MWD was chosen to be linked to utility because MWD is the primary outcome of most identified RCTs. Most of the authors responded (80%) but could not provide these data. One author (O’Donnell) provided a complete set of patient-level data (N = 106) for MWD and SF-36 scores based on a recent RCT in the UK comparing cilostazol and no vasoactive drug for PAD. 83

The SF-36 conversion algorithm, as defined by Ara and Brazier,96 was applied to the patient-level data from O’Donnell et al. 83 to calculate the utilities of each patient at week 0 and week 24 (the period of the RCT). These patient-level data were then used to test for a correlation between change in MWD and change in utilities from week 0 to week 24, which is the trial period of most identified RCTs. The correlation coefficient of the absolute difference in MWD on the log scale (log of MWD in week 24 minus log of MWD in week 0) and absolute difference in utilities (utility in week 24 minus utility in week 0) was 0.39 and the scatterplot is presented in Figure 7. A linear regression model was fitted to these data to predict the absolute change in utilities from the absolute change in MWD on the log scale during the RCT period. A linear relationship was assumed owing to lack of evidence of other types of underlying relationship. One regression model was fitted to the placebo and cilostazol data combined and this maximised the sample size for the regression analysis. The underlying assumption was that the relationship between MWD and utility is independent of treatment, such that the relationship can be applied to all of the vasoactive drugs being evaluated and the no vasoactive drug comparator. This was tested by including an additional term in the regression analysis representing the treatment effect which was not significant. The fitted regression based on the O’Donnell et al. patient-level data83 and used in the economic model is:

FIGURE 7.

The relationship between the absolute change in utilities and the absolute change in MWD on log scale based on patient-level data (O’Donnell et al. 83).

It is reassuring that the constant within the regression model is very close to zero, which means that when there is no change in MWD, there is also no change in the utility score. The variance–covariance matrix of the slope and the intercept of the regression model is presented in Table 57. To represent the uncertainty of the regression model, the matrix was used to sample the two coefficients of the regression model in the PSA. By using the variance–covariance matrix within the PSA, the uncertainty around the relationship between MWD and utility is propagated through the model and represented within the model results. However, this uncertainty does not account for any differences in this relationship between the vasoactive drugs.

The estimate of treatment effect that is generally reported in the RCTs of the vasoactive drugs for PAD is the percentage change between treatments based on the geometric mean change from baseline. The reason for this is because the raw data are analysed on the log scale and anti-logging the sample means on the log scale produces sample geometric means. The motivation for transforming these data to the log scale is to produce a scale on which the treatment effects can be assumed to be linear. We use a similar rationale when relating the log of the difference in MWD to the absolute difference in utilities, with treatment effects in terms of utilities assumed to be linear on the absolute scale.

The regression model was applied to all four drugs and to no vasoactive drug treatment to estimate the absolute change in utilities given a certain change in MWD from week 0 to week 24 on the log scale. The baseline utilities, i.e. the utilities for patients at week 0, were also estimated from the patient-level data. 83 All estimated absolute changes in utilities were applied to the baseline utilities. The estimated mean of the baseline utilities was 0.4838 and the estimated SD was 0.1001. The patient-level data used for this analysis were collected within the latest reported clinical trial on cilostazol and were based on patients in the UK. 83 Therefore, the mean baseline utilities should reflect the quality of life of patients with stable IC in the UK NHS context. SA was performed to test alternative baseline utilities.

The SF-36 data were also available at week 6 for each patient; and the mean utility change over time (at weeks 0, 6 and 24) for all patients is presented in Figure 8. Several RCTs of cilostazol, naftidrofuryl oxalate and pentoxifylline reported the change in MWD over time, which also suggested a linear increase. 58,61,65,66 In the absence of any additional evidence, this suggests that a linear model may be appropriate when representing the increase in utilities over the first 24 weeks. For patients who receive a vasoactive drug beyond 24 weeks, it was assumed the utility remains constant from week 24 onwards, due to the lack of published evidence beyond this time point. For patients who discontinue treatment, it was assumed the utility returns to the level of the no vasoactive drug group at the time of discontinuation.

FIGURE 8.

Mean utility change over time (at weeks 0, 6 and 24) based on patient-level data (O’Donnell et al. 83).

Given that the HRQoL of the general population is dependent upon age, it is important to take this into account in the model. General population utility estimates from Ara and Brazier97 were applied using a regression analysis of utility versus age. The age-related utility was calculated by the following formula:

where A = –0.0001728, B = –0.000034 and C = 0.9584588.

The ratio between the utility at age 66 years using this formula for the general population and the utility at age 66 years for IC patients estimated using the regression above was calculated. The age-related utility within the general population was then adjusted to account for the lower average utility associated with IC patients by multiplying it by this ratio for each age within the model.

Given the limited evidence in terms of utilities, a threshold analysis has also been undertaken to assess the QALY gain required for each of the drugs to be considered to be cost-effective compared with no vasoactive drug at willingness-to-pay thresholds of £20,000 and £30,000 per QALY gained. The threshold analysis is the only way in which the cost-effectiveness of inositol nicotinate is assessed owing to the lack of effectiveness data available for this drug.

Table 58 presents the predicted mean utility values for each vasoactive drug and no vasoactive drug treatment at week 24.

Adverse events

Within the trials identified within the systematic review, rates of SAEs were similar between the treatment groups and the placebo groups, and rates of minor AEs were similar between naftidrofuryl oxalate and placebo and inositol nicotinate and placebo. The trials of cilostazol and pentoxifylline reported higher rates of minor AEs within the treatment groups than the placebo groups, which were mainly headaches, diarrhoea, peripheral oedema and palpitations (see Chapter 3, Adverse events and serious adverse events, above, for further details). Clinical expert advice suggests that these patients are unlikely to require additional treatment as they would discontinue the vasoactive drugs, as suggested by the trials that demonstrate higher discontinuation rates for cilostazol and pentoxifylline. This means that there are unlikely to be any additional costs incurred as a result of these AEs. As these minor AEs would generally be experienced for a short time period, and, given that these patients already have a lower utility on average than that experienced by the general population, the impact of these minor AEs upon utilities is expected to be minimal (i.e. unlikely to affect total QALYs to fewer than three decimal places). The quality of life of patients will therefore be affected within the model only by the patients discontinuing and hence having a lower quality of life than if they remain on treatment.

Discontinuation of treatment

The rates of discontinuation for the first 24 weeks were based on meta-analyses of all identified RCTs. The long-term discontinuation rate (i.e. beyond 24 weeks) was reported in only one study of cilostazol (Hiatt et al. 49), which reported that 68% of patients in the cilostazol arm discontinued with the drug by 36 months. Expert clinical opinion suggests that many discontinuations beyond 24 weeks are likely to be due to the patients’ condition improving or mortality and hence the patients no longer require the drug, rather than discontinuations being because of any AEs associated with the drugs. Therefore, given the lack of published evidence, the long-term discontinuation rates of the remaining three drugs were assumed to be the same as for cilostazol.

Mortality

It was assumed that all drugs are for symptomatic relief rather than having an impact on the progression of the disease (Steven Thomas, Jonathan Michaels and Gerard Stansby, personal communication). Therefore, all patients within the model have the same overall mortality rates. General population mortalities were based on the latest life tables of the general population in England and Wales [Office for National Statistics (ONS)]. 98 The mortalities of the patient population in the model were calculated by multiplying the general population mortality by the relative risk of mortality of IC patients, which was assumed to be 1.6 based upon a study of the risk of mortality and cardiovascular disease associated with ABPI by Heald et al. 99

Resource use and costs

The cost of each drug was based on the latest drug tariff updated in October 2010. 26 Where there is more than one licensed dose available, the cost of the drug was based upon the doses used within the RCTs identified within the clinical effectiveness review, which are also current practice in the UK. Inositol nicotinate is supplied in two packs: 100 tablets of 500 mg at a price of £30.76 (6.152p per 100 mg) and 112 tablets of 750 mg at a price of £51.03 (6.075p per 100 mg). The former pack was used for estimating costs as it has a higher unit price (in terms of 100 mg) and the identified inositol nicotinate RCTs used a dose of 4 g per day (which can not be divided by 750-mg tablets). Naftidrofuryl oxalate is available both as a generic drug at a price of 5.38p per 100 mg and produced by the manufacturer that held the original patent at a higher price of 9.83p per 100 mg. The former cost was used in the base-case model because this is expected to be the acquisition cost in practice. Drug costs included within the model are presented in Table 59.

Sensitivity analysis 1: utility remains the same as when on the drug if discontinuation occurs after 24 weeks

Clinicians suggest that a proportion of patients discontinue with the drug after 24 weeks because their condition has improved. The SA assumes that if the patients discontinue the drug after 24 weeks, the utility remains the same over this subgroup of patients’ remaining lifetime as when on the drug at the time of discontinuation.

Sensitivity analysis 2: alternative baseline utility

Two RCTs of cilostazol were identified within the clinical effectiveness review, which presented SF-36 data that could be converted to utilities. 61,83 The SF-36 data were converted into utilities for both studies and found to be very different from each other: a utility of 0.4838 versus 0.7562. The former utility was used within the base-case analysis, and the latter was tested within this SA. Utility would also be variable between patients in practice. The relationship between baseline utility and utility at 24 weeks is assumed to remain the same within this analysis.

Sensitivity analysis 3: alternative cost for naftidrofuryl oxalate

In the base-case model, the cost of generic naftidrofuryl oxalate is used. The drug is also produced by the manufacturer that held the patent (with a brand name of Praxilene), but at a higher cost of £4.13 per week compared with £2.26 per week in the base case. This SA was performed to test the impact on cost-effectiveness results using the alternative drug cost.

Sensitivity analysis 4: shorter time horizon

Most of the evidence on change in MWD, change in utility and discontinuation rates is based on RCTs that have a follow-up period of < 24 weeks. Beyond 24 weeks, a number of assumptions have been made within the model around the change in utilities and the drug discontinuation rates owing to a lack of published evidence. This SA tests a shorter time horizon of 24 weeks where data are most robust.

Sensitivity analysis 5: alternative starting age

The base-case model assumes that the cohort of patients within the model begin treatment at age 66 years. This SA assesses whether or not the age at which patients begin treatment affects the cost-effectiveness of the drugs. A starting age of 55 years (the age at which the disease begins being prevalent in the population) was applied to test the robustness of the results to starting age.

Sensitivity analysis 6: alternative long-term discontinuation rates

No evidence on the long-term discontinuation rates of pentoxifylline, naftidrofuryl oxalate or inositol nicotinate was identified. The base case assumes that the discontinuation rates of these vasoactive drugs beyond 24 weeks are the same as the discontinuation rates of cilostazol, as clinicians suggest that the reasons for discontinuation beyond 24 weeks would be more likely related to improvements in disease than to AEs, as for the first 24 weeks. However, in order to test the impact of alternative discontinuation rates beyond 24 weeks, the SA assumes that the long-term discontinuation rates of pentoxifylline and naftidrofuryl oxalate maintain the same relative ratios compared with cilostazol within the 24 weeks (i.e. as the discontinuation rate of pentoxifylline and naftidrofuryl oxalate are, respectively, 5% more and 60% less than that of cilostazol within the first 24 weeks. It is assumed that beyond 24 weeks the discontinuation rates of pentoxifylline and naftidrofuryl oxalate are also, respectively, 5% more and 60% less than the long-term discontinuation rate of cilostazol).

Sensitivity analysis 7: angioplasty procedure for patients discontinuing within 24 weeks

Clinical practice regarding prescribing of vasoactive drugs for IC patients whose symptoms continue despite a period of conservative management varies among clinicians. Some clinicians will assess whether angioplasty is appropriate within this patient group and if so undertake this immediately. If angioplasty either is not appropriate or fails then those patients may receive vasoactive drugs. Alternative practice is for IC patients to be offered vasoactive drugs whether or not they may be considered for angioplasty. If the drugs are unsuccessful, patients may then be considered for angioplasty if this is an appropriate option, but, if successful, these vasoactive drugs may negate or delay the need for angioplasty. The former clinical practice is considered within the base-case analysis. This SA concerns the latter of these two alternative clinical practices.

The subgroup of patients who would be potentially offered angioplasty may be prespecified as they tend to have a worse prognosis. It may be that the cost-effectiveness of the assessed drugs is different within this subgroup of patients and hence an exploratory analysis has been undertaken around this subgroup. The analysis is considered to be exploratory, as there is no published evidence reporting the costs and outcomes associated with this subgroup and hence it is mainly based upon personal communication with the team of clinical advisors (Steven Thomas, Jonathan Michaels and Gerard Stansby, personal communication). Those patients with a worse prognosis in whom angioplasty is potentially appropriate were estimated to represent around 15% of the overall patient group included within this assessment (Steven Thomas, Jonathan Michaels and Gerard Stansby, personal communication).

A set of simplified assumptions were made for this SA:

• Patients who discontinue the vasoactive drugs within 24 weeks will have angioplasty.

• Patients in the comparator group with no vasoactive drug treatment will have angioplasty at week 0.

• The costs of angioplasty include two hospital visits (£99.03 per visit),90 one imaging (£189.90)101 and the angioplasty procedure (£925.58). 90 All costs were adjusted to 2009–10 prices. 102

• Owing to lack of comparative evidence around the utility associated with angioplasty, it will be varied over the largest plausible range within this analysis and will be related to the utility associated with naftidrofuryl oxalate (the drug associated with the highest utility). The lower bound of the utility increase due to angioplasty is assumed to be zero. The upper bound is assumed to be the same as the utility of the general population used in the model. The utility associated with angioplasty is therefore assumed to be:

  1. – equivalent to the utility associated with naftidrofuryl oxalate (the drug associated with the highest utility)

  2. – 20% higher than the utility associated with naftidrofuryl oxalate

  3. – 40% higher than the utility associated with naftidrofuryl oxalate

  4. – equivalent to the utility of the general population used in the model, which is around 60% higher than the utility associated with naftidrofuryl oxalate.

• Patients who have no vasoactive drugs and have angioplasty will have the utility described above for 1 year. The utility will then decrease to that associated with placebo.

• Patients who have angioplasty after discontinuation of the vasoactive drugs will have the utility described above until the end of the first year; it will then decrease to the level of utility associated with placebo.

• The baseline utility for these patients in practice will be lower, as they have a worse prognosis by definition within this subgroup analysis. However, the impact of baseline utility is tested within SA3 and hence is not altered here.

Sensitivity analysis 8: informative prior distributions within network meta-analysis

The impact of more informative prior distributions for the between-study SD and treatment effects within the network meta-analysis upon the results of the meta-analysis was tested and reported (see Chapter 3, Maximal walking distance meta-analysis). The results of the network meta-analysis when an informative prior distribution from SA5 was used [τ ∼ U(0, 0.2), μ ∼ N(100)] and tested within the model.

Sensitivity analysis 1: utility remains the same as when on the drug if discontinuation occurs after 24 weeks

The SA assumes the effectiveness of the vasoactive drug continues when patients discontinue the drug after 24 weeks. The incremental cost-effectiveness results of the SA are presented in Table 62. The results show that the effectiveness of all drugs increase significantly compared with no vasoactive drug. For example, the total QALYs of naftidrofuryl oxalate increase from 5.024 in the base case to 5.174. The base-case cost-effectiveness conclusions are not changed. Both pentoxifylline and cilostazol are dominated by naftidrofuryl oxalate. The ICER of naftidrofuryl oxalate compared with no vasoactive drug decreases from £6070 in the base case to £1538, which is more favourable to the intervention.

Sensitivity analysis 2: alternative baseline utility

This SA applies an increased baseline utility of 0.7562 compared with 0.4838. The incremental cost-effectiveness results of the SA are presented in Table 63. The base-case cost-effectiveness conclusions and the ICER of naftidrofuryl oxalate compared with no vasoactive drug are similar to the base-case results, which demonstrates that the model is not sensitive to different baseline utilities.

Sensitivity analysis 3: alternative cost for naftidrofuryl oxalate

This SA applies a higher cost for naftidrofuryl oxalate, which is £4.13 per week (using the cost of the brand name Praxilene), compared with £2.26 per week (generic) in the base case. The incremental cost-effectiveness results of the SA are presented in Table 64. The results show that cilostazol is dominated by naftidrofuryl oxalate, which has lower costs and higher total QALYs. Pentoxifylline is extendedly dominated by naftidrofuryl oxalate because the ICER associated with pentoxifylline is higher than that associated with naftidrofuryl oxalate. This is due to the drug acquisition cost of naftidrofuryl oxalate becoming substantially higher than that of pentoxifylline. The ICER of naftidrofuryl oxalate compared with no vasoactive drug is £11,058 per QALY gained, which is higher than the base case of £6070, because of the cost increase of naftidrofuryl oxalate.

Sensitivity analysis 4: shorter time horizon

This SA considers a time horizon of 24 weeks. The incremental cost-effectiveness results of the SA are presented in Table 65. The base-case cost-effectiveness conclusions are not changed. The ICER of naftidrofuryl oxalate compared with no vasoactive drug increases from £6070 in the base case to £10,733.

Sensitivity analysis 5: alternative starting age

This SA assumes that patients begin treatment with these drugs at age 55 years. The incremental cost-effectiveness results of the SA are presented in Table 66. The base-case cost-effectiveness conclusions and the ICER of naftidrofuryl oxalate compared with no vasoactive drug are similar to the base-case results, which demonstrates that the model is not sensitive to the starting age of patients.

Sensitivity analysis 6: alternative long-term discontinuation rates

This SA assumes that the long-term discontinuation rates of pentoxifylline and naftidrofuryl oxalate maintain the same relative ratios compared with cilostazol within the 24 weeks. The incremental cost-effectiveness results of the SA are presented in Table 67. The results show that cilostazol is dominated by naftidrofuryl oxalate, which has lower costs and higher total QALYs. Pentoxifylline is extendedly dominated by naftidrofuryl oxalate because the drug acquisition cost of naftidrofuryl oxalate becomes substantially higher than that of pentoxifylline, owing to the lower long-term discontinuation rate of naftidrofuryl oxalate compared with the base case. The ICER of naftidrofuryl oxalate compared with no vasoactive drug is £5899 per QALY gained, which is similar to the base case.

Sensitivity analysis 7: angioplasty procedure for patients discontinuing within 24 weeks

This subgroup analysis assumes that patients who have more severe IC and discontinue with the drugs within 24 weeks will receive an angioplasty procedure that will improve HRQoL of these patients on average. These patients who receive no vasoactive drug for PAD are assumed to have an angioplasty procedure at the start of the model and experience the improved HRQoL immediately. Owing to lack of comparative evidence around the utility associated with angioplasty, four scenarios were tested: the utility associated with angioplasty is equivalent to the utility associated with naftidrofuryl oxalate, 20% and 40% higher than the utility associated with naftidrofuryl oxalate, and equivalent to the utility of general population used in the model.

The incremental cost-effectiveness results of the SA are presented in Tables 68–71. Unlike the base case and other sensitivity analyses, in which the cost of no vasoactive drug is zero, the comparator of no vasoactive drug is associated with a significant cost which is £1313, which represents the costs of angioplasty.

When it is assumed that the utility associated with angioplasty is equivalent to the utility associated with naftidrofuryl oxalate, naftidrofuryl oxalate dominates pentoxifylline, cilostazol and no vasoactive drug. When it is assumed that the utility associated with angioplasty is 20% and 40% higher than the utility associated with naftidrofuryl oxalate, no vasoactive drug (all patients receive angioplasty) is associated with the highest total QALYs. In both scenarios, pentoxifylline and cilostazol are dominated by naftidrofuryl oxalate and the ICERs of no vasoactive drug (all patients receive angioplasty) compared with naftidrofuryl oxalate are £17,992 and £6545 per QALY gained, respectively.

When it is assumed that the utility associated with angioplasty is equivalent to the utility of the general population within the model (around 60% higher than the utility associated with naftidrofuryl oxalate), no vasoactive drug (all patients receive angioplasty) is associated with the highest total QALYs. Naftidrofuryl oxalate is associated with fewer total QALYs than cilostazol because more patients discontinue with cilostazol and could therefore benefit from the angioplasty procedure. Pentoxifylline is dominated by naftidrofuryl oxalate and cilostazol is dominated by no vasoactive drug (all patients receive angioplasty). The ICER of no vasoactive drug (all patients receive angioplasty) compared with naftidrofuryl oxalate is £4094 per QALY gained.

Therefore, this exploratory analysis suggests that if angioplasty is associated with an increase in quality of life compared with the vasoactive drugs, vasoactive drugs are unlikely to be considered to be economically attractive at a willingness-to-pay threshold of £20,000 per QALY gained for this small subgroup of patients. However, this subgroup analysis is largely based upon clinical advice owing to lack of evidence. This is therefore an exploratory and highly uncertain analysis, and hence these results should be treated with caution.

Sensitivity analysis 8: informative prior distributions within network meta-analysis

This SA assesses the impact of allowing more informative prior distributions for the between-study SD and treatment effects within the network meta-analysis. The incremental cost-effectiveness results of the SA are presented in Table 72. The results suggest that changes to the prior distributions for the between-study SD and treatment effects have very little impact upon the model results.