ABSTRACT
Purpose To perform an individual patient data-level meta-analysis of randomized controlled trials comparing drug-coated balloon angioplasty (DCB) against conventional percutaneous transluminal angioplasty (PTA) in the treatment of dysfunctional hemodialysis venous access.
Methods A search was conducted from inception till 13th November 2020. Kaplan-Meier curves comparing DCB to PTA by target lesion primary patency (TLPP) and access circuit primary patency (ACPP) were graphically reconstructed to retrieve patient-level data. One-stage meta-analyses with Cox-models with random-effects gramma-frailties were conducted to determine hazard ratios (HRs). Dynamic restricted mean survival times (RMST) were conducted in view of violation of the proportional hazards assumption. Conventional two-stage meta-analyses and network meta-analyses under random-effects Frequentist models were conducted to determine overall and comparative outcomes of paclitaxel concentrations utilised. Where outliers were consistently detected through outlier and influence analyses, sensitivity analyses excluding those studies were conducted.
Results Among 10 RCTs (1,207 patients), HRs across all models favoured DCB (one-stage shared-frailty HR=0.62, 95%-CI: 0.53–0.73, P<0.001; two-stage random-effects HR=0.60, 95%-CI: 0.42–0.86, P=0.018, I2=65%) for TLPP. Evidence of time-varying effects (P=0.005) was found. TLPP RMST was +3.47 months (25.0%) longer in DCB-treated patients compared to PTA (P=0.001) at 3-years. TLPP at 6-months, 1-year and 2-years was 75.3% vs 58.0%, 51.1% vs 37.1% and 31.3% vs 26.0% for DCB and PTA respectively. P-Scores within the Frequentist network meta-analysis suggest that higher concentrations of paclitaxel were associated with better TLPP and ACPP. Among 6 RCTs (854 patients), the one-stage model favoured DCB (shared-frailty HR=0.72, 95%-CI: 0.60–0.87, P<0.001) for ACPP. Conversely, the two-stage random-effects model demonstrated no significant difference (HR=0.76, 95%-CI: 0.35–1.67, P=0.414, I2=81%). Sensitivity analysis excluding outliers significantly favoured DCB (HR=0.61, 95%-CI: 0.41–0.91, P=0.027, I2=62%).
Conclusion Overall evidence suggests that DCB is favoured over PTA in TLPP and ACPP. The increased efficacy of higher concentrations of paclitaxel may warrant further investigation.
Introduction
Venous stenosis is the commonest cause of dysfunction in arteriovenous fistulas (AVFs) and arteriovenous grafts (AVGs) used for hemodialysis [1]. The Kidney Disease Outcomes Quality Initiative (KDOQI) recommends percutaneous transluminal angioplasty (PTA) as first-line treatment to restore vessel patency [2]. While PTA procedural success is generally high, microscopic damage to the vessel wall upon inflation may cause reactive intimal hyperplasia [3-5], inducing restenosis and poor long-term patency [6].
Presently, drug-coated balloon (DCB) angioplasty has become an increasing subject of discussion as an adjunct to PTA. DCBs are commonly coated with paclitaxel, which prevents neointimal hyperplasia in blood vessels [7-9].
Paclitaxel DCB has been demonstrated to be superior to PTA in the contexts of femoropopliteal artery disease [10] and coronary instent restenosis [11]. However, findings are mixed in the setting of hemodialysis access – some meta-analyses significantly favoured DCB over PTA [12, 13], while others reported no significant benefit [14, 15]. Not long ago, two new randomized controlled trials (RCTs) on this subject were published [16, 17]; this article aims to re-evaluate the field with these new inclusions. To our knowledge, this is first meta-analysis which includes these 2 new RCTs.
In lieu of this area’s uncertain and developing nature, more precise methods are needed to quantify the comparison between DCB and PTA. Thus, we utilised graphical reconstructive algorithms to attain survival information of individual patients [18, 19] from Kaplan-Meier curves. Individual patient data (IPD) is recognised as the gold standard approach for evidence synthesis [20, 21]. The source of IPD for this study was restricted to RCTs only, as baseline characteristic differences between patients and differing indications for treatment render non-randomized studies difficult to interpret definitively.
Leveraging the IPD of 11 RCTs, this meta-analysis aims to compare the relative efficacy of DCB to PTA in maintaining target lesion primary patency (TLPP) and access circuit primary patency (ACPP) in hemodialysis patients with stenosed venous access.
Methodology
Literature Search
The electronic literature search was conducted on EMBASE, Scopus, PubMed and Web of Science for RCTs from inception to 13th November 2020 using a search string [Supplementary Table 1], in line with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses and Cochrane Guidelines (PRISMA) for IPD and network meta-analysis (NMA) [22, 23]. Only full-text articles were included. If the same publication appeared more than once, the most recent publication was used for the analysis.
We included RCTs reporting TLPP and ACPP between DCB and PTA that provided Kaplan-Meier curves. The exclusion criteria for this study were: (a) Studies only comparing differences within a single approach. (b) Studies with combination approaches including multiple intra-arterial approaches.
For studies with multiple treatment arms, we only analysed the treatment arms in our inclusion criteria. The abstracts were reviewed by 3 investigators J.J.Z., K.Y.F. and E.T. and conflicts resolved by C.W.T. The data was extracted by K.Y.F., J.J.Z. and E.T. using predefined data fields including study characteristics, patient demographics and primary endpoints.
Risk of Bias Assessment
The RCTs were assessed for risk of bias by K.Y.F. and J.J.Z. using the Cochrane Risk of bias (RoB 2) tool for RCTs [24]. Studies were also assessed for trustworthiness and relevance with the Joanna Briggs Institute (JBI) critical appraisal tool for RCTs [25].
Statistical Analysis
Reconstruction of Individual Patient Data
Prior to meta-analyses, we reconstructed IPD from published survival curves using a graphical reconstructive algorithm by Guyot et al [18]. Images of Kaplan-Meier curves from included studies were digitised to obtain the step function values and timings of the steps. Survival information of individual patients were then recovered based on the numerical solutions to the inverted Kaplan-Meier product-limit equations [18]. The IPD dataset was reconstructed by J.J.Z and K.Y.F. and were approved by E.T. by visual comparisons against original curves, and comparing reconstructed raw hazard ratio (HR) and P-values to reported values [Supplementary Table 2, Supplementary Comparisons 1, 2].
One-Stage Meta-Analysis
The Kaplan-Meier method was used to determine TLPP and ACPP for both groups. To account for between-study heterogeneity, Cox-models with random-effects gamma-frailties and stratification were conducted to determine HRs of patients undergoing DCB versus PTA. The primary analysis was based on a shared-frailty approach which assumes that individual patients are at an equal risk of death as other individuals within the same study. Then, we stratified the Cox model by study subgroups, modelling inter-study heterogeneity by assuming a baseline hazard amongst patients in each unique study.
For Cox-based models, we verified the proportional hazards (PH) assumption by examining scaled Schoenfeld residuals [26] and through the quantitative Grambsch-Therneau test [27]. Where the PH assumption was violated, we analysed restricted mean survival times (RMST) and modelled patency outcomes at prespecified epochs up to 1-, 2- and 3-years [28]. RMST is an alternative treatment outcome measure that can be estimated as the area under the survival curve up to a prespecified time horizon and hence can account for all survival information before that time horizon [28, 29]. To fully capture the dynamic changes of RMST differences and ratios, we computed RMST values over a range of values to the restriction time (tau) to trace out an evolving treatment effect profile over time [30].
Patency rates were calculated from the one-stage meta-analysis along with Greenwood 95% confidence intervals to account for censorship status. To account for the group of patients who no longer contribute to excess hazard (as seen by long plateaus on Kaplan-Meier curves), we fitted flexible parametric cure models to estimate “long term patency rates” using an identity link [31].
Two-Stage Meta-Analysis
We computed summary HRs for individual studies based on the reconstructed individual patient dataset and pooled them under a conventional two-stage Frequentist meta-analysis. The random-effects model was chosen in light of the high degree of heterogeneity found (I2). I2 can be interpreted as the disparity in study results due to between-study variability. In light of the potentially high degree of clinical heterogeneity of included studies (as evidenced by differences in inclusion criteria, paclitaxel dose concentrations, proportions of AVFs versus AVGs and devices utilised), a random-effects model was used. Quantitative analysis of funnel plot asymmetry was done using Egger’s regression test [32]. Heterogeneity was considered low, moderate, or considerable for I2 values <40%, 40-75%, and >75% respectively [33, 34]. Where considerable heterogeneity was found, we searched for extreme effect sizes (outliers) using outlier and influence analysis [35, 36]. Where outliers were consistently detected across these analyses, sensitivity analyses excluding those studies were conducted.
Frequentist Network Meta-Analysis
Where concentration of paclitaxel was provided, natural log-transformed hazard ratios were estimated for each IPD study and were pooled together in an NMA within a Frequentist setting. Treatment strategies were ranked using P-Scores, with higher P-Scores corresponding to greater efficacy [37].
Meta-Regression
To identify potential sources of heterogeneity, we performed a meta-regression with aggregate level data. Prior to meta-regression, we imputed missing means and standard deviations (SDs) from medians, ranges (minimum to maximum), and interquartile ranges (IQRs) using the methods proposed by Hozo et al [38], Wan et al [39] and Furukawa et al [40]. Conventional mixed effects meta-regression of variables (with more than 10 studies) were performed against logarithmic transformed hazard ratios.
All analyses were conducted in R-4.0.0 (with packages ‘digitize’, ‘survival’, ‘flexsurvcure’, ‘metafor’. ‘dmetar’ and ‘netmeta’). P < 0.05 were regarded to indicate statistical significance.
Results
Study Selection
The search strategy retrieved 503 studies; after de-duplication and screening, 11 RCTs [16, 17, 41-48] comprising 1,243 patients were eligible and included in the meta-analysis [Figure 1]. One RCT was excluded as no curve was provided for graphical reconstruction to retrieve patient-level patency information [49].
PRISMA indicates preferred reporting items for systematic reviews and meta-analyses. N=, number of studies; n=, number of patients; TLPP, target lesion primary patency; ACPP, access circuit primary patency; IPD, individual patient data; RCT, randomized controlled trials.
Study Characteristics
Breakdown of studies and extracted outcomes are reported in Table 1 and in Supplementary Table 2 respectively. Of note: 1) 5 of the 11 studies had 20 or less patients per arm; 2) all studies had a higher percentage of males except Liao et al [47]; 3) a significant proportion of patients in most studies received antiplatelet treatment, either preoperatively or postoperatively; 4) 6 studies reported a significant benefit with DCB based on primary and secondary endpoints, while the other 5 found no significant difference compared with PTA; 5) 4 studies involved AVGs, of which 1 study (Liao et al [47]) studied only AVGs; the other 7 involved only AVFs.
Reconstructed Curves
All included studies had IPD reconstructed from published Kaplan-Meier curves [Figure 2]. Comparisons against original Kaplan-Meier curves are shown in Supplementary Comparisons 1 & 2. The reconstructed dataset is available on request.
DCB, drug coated balloon; PTA, percutaneous transluminal angioplasty.
Quality Assessment of Trials
Our risk-of-bias analysis [Table 2] yielded overall ‘some concerns’ for all 11 studies. As all studies were single-blinded or non-blinded, they ran the risk of observer bias, wherein knowledge of which patients were assigned DCB or PTA might have influenced the judging of outcomes. Particularly, decisions on the need for reintervention in all studies may have been influenced by knowledge of assignment to treatment or control arm, since the reinterventions in most studies involved clinical evaluation of graft dysfunction. Our JBI analysis [Supplementary Table 3] noted that Björkman et al [44] discontinued their trial after only 39 patients were recruited, despite requiring a sample size of 140 in their initial power calculations.
Survival Outcomes
Target Lesion Primary Patency
Among 10 RCTs comprising 1,207 patients, the stratified and gamma-frailty models yielded HRs in favour of DCB (shared-frailty HR=0.62, 95%-CI: 0.53–0.73, P<0.001; stratified HR=0.62, 95%-CI: 0.53–0.73, P<0.001) [Figure 3]. Patency rates at intervals of 6-months are reported in Figure 3.
HR, hazard ratio; 95CI, 95% confidence intervals; DCB, drug coated balloon; PTA, percutaneous transluminal angioplasty.
While time-varying effects were noted (Grambsch-Therneau test, P=0.005) [Supplementary Figure 1A], dynamic RMST curves demonstrate that DCB is consistently favoured throughout the follow-up period [Figure 4A, B].
RMST-R, restricted mean survival time ratio; RMST-D, restricted mean survival time difference; HR, hazard ratio; 95CI, 95% confidence intervals; DCB, drug coated balloon; PTA, percutaneous transluminal angioplasty.
This was consistent within a two-stage Frequentist random-effects model (HR=0.60, 95%-CI: 0.42–0.86, P=0.018, I2=65%) [Figure 4C]. The funnel plot was visually symmetrical and did not suggest publication bias (Egger’s test = 0.507) [Supplementary Figure 4A].
Access Circuit Primary Patency
Among 6 RCTs comprising 854 patients, the stratified and gamma-frailty models yielded HRs in favour of DCB (shared-frailty HR=0.72, 95%-CI: 0.60–0.87, P<0.001; stratified HR=0.72, 95%-CI: 0.60–0.86, P<0.001) [Figure 5A]. Time-varying effects were not found (Grambsch-Therneau test, P=0.300) [Supplementary Figure 1B]. Patency rates at intervals of 6-months are reported in Figure 5A.
HR, hazard ratio; 95CI, 95% confidence intervals; DCB, drug coated balloon; PTA, percutaneous transluminal angioplasty.
This was inconsistent within the overall two-stage Frequentist random-effects model (HR=0.76, 95%-CI: 0.35–1.67, P=0.414, I2=81%) [Figure 5B]. In view of the considerable heterogeneity found, we undertook efforts to identify potential outliers. The Baujat plot suggested that Björkman et al [44] contributed to the most between-study heterogeneity [Supplementary Figure 5A]. At cut-offs proposed by Viechtbauer and Cheung [35], Björkman et al was likewise detected as an influential case [Supplementary Figure 5B]. Leave-one-out analyses showed that omission of Björkman et al yielded the least between-study heterogeneity (I2=62%) [Supplementary Figure 5C].
Sensitivity analysis excluding Björkman et al significantly favoured DCB (two-stage Frequentist random-effects model HR=0.61, 95%-CI: 0.41–0.91, P=0.027, I2=62%) [Figure 5C]. The one-stage model excluding this study yielded congruent results with one-stage meta-analysis of the overall cohort [Supplementary Figure 6]. The funnel plot was visually asymmetrical and suggestive of publication bias [Supplementary Figure 4B].
Frequentist Network Meta-Analysis of Paclitaxel Concentrations
6 RCTs comprising 1,050 patients were included in a network meta-analysis to evaluate TLPP within a random-effects Frequentist setting (I2=56.1%). Among all comparisons against PTA, only DCB 3.5μg/mm2 (HR=0.37, 95%-CI: 0.17–0.80, P=0.012) and DCB 3.0μg/mm2 (HR=0.52, 95%-CI: 0.34–0.80, P=0.002) had a statistically significant TLPP advantage [Table 3]. Indirect comparisons between DCB 3.5μg/mm2 and DCB 3.0μg/mm2 showed no differences (HR=0.71, 95%-CI: 0.29–1.70, P=0.437) [Table 3].
3 RCTs comprising 736 patients were included in a network meta-analysis to evaluate ACPP within a random-effects Frequentist setting. Heterogeneity was not estimable. Björkman et al was excluded from this analysis in view of the upstream influence analysis. Among all comparisons against PTA, only DCB 3.5μg/mm2 (HR=2.34, 95%-CI: 1.59–3.45, P<0.001) had a statistically significant ACPP advantage [Table 3]. Indirect comparisons between DCB 3.5μg/mm2 and DCB 2.0μg/mm2 was significantly in favour of DCB 3.5μg/mm2 (HR=2.09, 95%-CI: 1.30–3.36, P=0.002) [Table 3].
The comparison adjusted funnel plots were symmetrical [Supplementary Figure 7]. Within the network meta-analysis, we surmise that the transitivity assumption was likely to be met, based on the observations that the common treatment (PTA) was reasonably consistent across trials, effect modifiers were equally distributed across studies, and participants may, in principle, be randomized to any of the treatments being compared in the network. Given the star shaped nature of our network of treatment arms, statistical inconsistency could not be evaluated [Supplementary Figure 8].
Meta-Regression
Meta-regression of proportions of males, diabetics, arteriovenous fistulas and publication year against logarithmic transformed HRs for TLPP did not demonstrate any significant associations [Table 4]. Meta-regression for variables against ACPP between DCB and PTA was not conducted as there were insufficient studies.
Discussion
In view of cost [41, 50], potential morbidity [51] and uncertain theoretical benefit in prolonging patency rates, DCB angioplasty has yet to earn its place as the “holy grail” for dysfunctional hemodialysis venous access [52]. In 2019, in a meta-analysis of RCTs and retrospective studies by Wee et al [12], DCB was concluded to be superior to PTA (6-month TLPP RR=0.57, 95%-CI: 0.44-0.74, P<0.0001). In contrast, in a 2020 meta-analysis by Liao et al [14], DCB did not demonstrate significant patency benefit (6-month TLPP RR=0.75, 95%-CI: 0.56-1.01, P=0.06), even in the AVF-only subgroup.
Between the randomized trials conducted, significant heterogeneity was quantitatively, and qualitatively noted with regard to patient recruitment and access details – rendering the interpretation of its authentic efficacy challenging. Devices utilized (from IN.PACT Admiral, Passeo-Lux and Lutonix) varied in a multitude of aspects; such as its excipient, a chemical added to locally retain paclitaxel at the intended site of action.
Amidst this unresolved backdrop, this patient-level meta-analysis of 11 RCTs comprising 1,243 patients presents a statistically robust and up-to-date pool of evidence demonstrating a consistent benefit of using DCB over PTA in prolonging TLPP.
Despite incorporating random-effects shared-frailties and stratification to account for clinical heterogeneity within the one-stage model, patients treated with DCB were at a lower hazard rate of restenosis. Sensitivity analyses with aggregate data within a random-effects Frequentist models likewise re-affirm the upstream findings. In view of the violation of the proportionality assumption, dynamic restricted mean survival times consistently favoured DCB across the follow-up period. Collectively, these strongly suggest that the benefit of paclitaxel is likely a reflection of a true biological phenomenon, attesting to its superiority to plain old balloon angioplasty.
Albeit the promising outcomes in prolonging TLPP, comparisons of ACPP in the overall two-stage Frequentist random-effects model showed no significant differences as opposed to the significant differences demonstrated within one-stage models. Nonetheless, sensitivity analysis excluding outliers significantly favoured DCB in both one & two-stage models.
Other factors which are not directly related to the target lesion may cause eventual dysfunction elsewhere in the circuit. Perhaps, the puncture of AVGs and AVFs during the procedure could cause platelet thrombi and cytokine release; even though in most instances antiplatelets are given, gradually causing vessel stenosis [3]. Additional lesions in the dialysis circuit may also have formed over time, resulting in reduced circuit patency [53].
In particular, Björkman et al [44] heavily favoured PTA over DCB in the setting of ACPP. Björkman et al postulated that the younger and more immature AVFs (<1 year compared to the other studies) were more venous than arterial in nature; and with thinner venous walls, local paclitaxel overdose and potential toxicity could be more pronounced, accounting for the poor performance in the DCB arm. The authors stated that the study was discontinued due to slow recruitment; it was also limited by a smaller sample size than necessary for statistical power calculations, slow randomization and possibility of type I error. Insufficient studies provided information on age of AVFs to elucidate an association between age of AVFs and DCB efficacy in our study.
Looking forward, it is critical to determine the optimal concentration of paclitaxel where benefits are best observed. As suggested by the wide intervals between ranking P-Scores, there seems to be some evidence that higher concentrations of paclitaxel may be associated with longer TLPP and ACPP. Of note, among all comparisons against PTA, only DCBs with paclitaxel concentrations of 3.5μg/mm2 and 3.0μg/mm2 had a statistically significant TLPP advantage. However, upon closer inspection of indirect treatment comparisons generated through the Frequentist network meta-analysis, we did not find a significant difference in TLPP between concentrations of 2.0, 3.0 and 3.5μg/mm2.
While specific and potentially influential lesion covariates (such as length, location, type/location of access point and balloon diameter) were not consistently accounted for amongst the RCTs, this may suggest that the true benefit of adding paclitaxel may lie beyond concentrations beyond 3.0μg/mm2 – and may be an area of inquiry for further investigations.
Limitations
This study was not without limitations. In the course of our analyses, we were unable to account for competing risks to restenosis, such as patient mortality, for which previous studies yielded mixed findings. A DCB trial in femoropopliteal stenosis [10] showed increased mortality compared to PTA (24-month fixed-effect RR=1.74, 95%-CI: 1.08-2.81, P=0.02), and a meta-analysis of paclitaxel-coated devices demonstrated an increased mortality by 4.6% at 5-years [51]. That being said, a meta-analysis of paclitaxel-coated devices in dialysis access [54] found no difference in short to midterm mortality among patients compared to PTA (RR=1.26, 95%-CI: 0.85-1.89, P=0.25).
Additionally, although IPD was recovered graphically, the reconstruction algorithm is unable to retrieve patient-level covariate information, which may offer beneficial insight if adjusted for in models utilised. Potentially prognostic covariates for restenosis include vascular access age (as mentioned above), type of access point (AVF or AVG – where the absence of vascular smooth muscle in AVGs may prevent paclitaxel from exerting its desired action), diabetes mellitus [55], lesion length [56], blood flow volume [57], location of the AVF or AVG (forearm versus arm) [58], and type or duration of antiplatelet treatment [59]. Unfortunately, not all studies reported these parameters; there was hence inadequate data for us to perform a meta-regression.
Finally, clinical judgement to proceed with either DCB or PTA should be considered alongside the side effects of paclitaxel, procedure-related morbidity and cost-benefit analysis which are beyond the scope of this forum.
Conclusion
Hereto, we present the first patient-level, RCT-only meta-analysis comparing DCB to PTA in hemodialysis access restenosis. Pooled-analyses of 11 RCTs comprising 1,243 patients – including 2 recent RCTs published in 2020 – demonstrate that the overall evidence suggests that DCB is favoured over PTA in prolonging TLPP and ACPP.
Intriguingly, we also found some evidence that higher paclitaxel concentrations may be associated with longer TLPP. Future investigations may benefit from exploring more dose-efficacy relationships and may benefit from adjusting for more lesion specific covariates such as age, length, location, type and balloon diameter.
Data Availability
This manuscript makes use of publicly-available data from published studies, therefore no original or additional data is available for sharing.
DISCLOSURE
Funding
None.
Conflict of interest
J.J.Z. is supported by the SingHealth SMSTDA Talent Development Award administered by SingHealth, Singapore and NUS Enterprise Innovation and Entrepreneurship Practicum Award awarded by National University of Singapore, Singapore.
Contributions
Study design: C.W.T., J.J.Z., K.Y.F.
Data collection: K.Y.F., J.J.Z., E.T.
Statistical analysis: J.J.Z., R.S., N.S., E.T.
Manuscript writing: K.Y.F., J.J.Z.
Critical revision of manuscript: all authors.
Study supervision: C.W.T.
Data Sharing Policy
This manuscript makes use of publicly-available data from published studies, therefore no original or additional data is available for sharing.
Ethics
Not applicable.
PRISMA-IPD Checklist of items to include when reporting a systematic review and meta-analysis of individual participant data (IPD)
PRISMA NMA Checklist of Items to Include When Reporting A Systematic Review Involving a Network Meta-analysis
REFERENCES
- [1].↵
- [2].↵
- [3].↵
- [4].
- [5].↵
- [6].↵
- [7].↵
- [8].
- [9].↵
- [10].↵
- [11].↵
- [12].↵
- [13].↵
- [14].↵
- [15].↵
- [16].↵
- [17].↵
- [18].↵
- [19].↵
- [20].↵
- [21].↵
- [22].↵
- [23].↵
- [24].↵
- [25].↵
- [26].↵
- [27].↵
- [28].↵
- [29].↵
- [30].↵
- [31].↵
- [32].↵
- [33].↵
- [34].↵
- [35].↵
- [36].↵
- [37].↵
- [38].↵
- [39].↵
- [40].↵
- [41].↵
- [42].
- [43].
- [44].↵
- [45].
- [46].
- [47].↵
- [48].↵
- [49].↵
- [50].↵
- [51].↵
- [52].↵
- [53].↵
- [54].↵
- [55].↵
- [56].↵
- [57].↵
- [58].↵
- [59].↵
