The optimal coronary revascularization strategy (coronary artery bypass grafting [CABG] or percutaneous coronary intervention [PCI]) in patients with end-stage renal disease (ESRD) remains uncertain. We performed an updated systematic review and meta-analysis of observational studies comparing CABG and PCI in patients with ESRD using a random-effects model for the primary outcome of long-term all-cause mortality. Our review registered through PROSPERO included observational studies published after 2011 to ensure overlap with previous studies and identified 7 new studies for a total of 23. We found that the median sample size in the selected studies was 125 patients (25 to 15,784) with a large variation in the covariate risk adjustment and only 3 studies reporting the indications for the revascularization strategy. CABG was associated with a small reduction in mortality (relative risk 0.92, 95% CI 0.89 to 0.96) with significant heterogeneity demonstrated (p = 0.005, I 2 = 48.6%). Subgroup analysis by categorized “year of study initiation” (<1990, 1991 to 2003, >2004) further confirmed the summary estimate trending toward survival benefit of CABG along with a substantial decrease in heterogeneity after 2004 (p = 0.64, I 2 = 0%). In conclusion, our updated systematic review and meta-analysis demonstrated that in patients with ESRD referred for coronary revascularization, CABG was associated with a small decrease in the relative risk of long-term mortality compared with PCI. The generalizability of the finding to all patients with ESRD referred for coronary revascularization is limited because of a lack of known indications for coronary revascularization, substantial variation in covariate risk adjustment, and lack of randomized clinical trial data.
The current evidence of the optimal coronary revascularization strategy (coronary artery bypass grafting [CABG] or percutaneous coronary intervention [PCI]) to improve long-term survival in patients with end-stage renal disease (ESRD) has been predominantly through the conduct of observational studies and not by randomized clinical trials. A recent meta-analysis of 17 observational studies in patients with ESRD noted that CABG was associated with an improved long-term survival (odds ratio 0.86, 95%CI 0.83 to 0.89) over PCI. However, significant heterogeneity was demonstrated (p <0.0001, I 2 = 83%). Variation in study sample sizes has been suggested as a possible origin for the heterogeneity. Other possibilities are disparities in risk adjustment, the use of analytic methods such as propensity score matching, inclusion of angiographic variables, and temporal changes resulting in improvements of technique and equipment. The aim of the current report is to incorporate recently published studies on the comparative effectiveness between CABG and PCI on long-term mortality in patients with ESRD along with an improvement in our understanding of the origin of heterogeneity.
Materials
This review was conducted using a prespecified protocol. The protocol was registered at PROSPERO (“an international database of prospectively registered systematic reviews in health and social care”, http://www.crd.york.ac.uk/PROSPERO/ ), PROSPERO number: CRD42015030148. Studies were eligible if they included at least 20 patients older than 18 years with ESRD. The primary intervention was CABG with PCI being the comparator. PCI included all devices and techniques used for percutaneous coronary artery revascularization including bare-metal and drug-eluting stents. CABG included the use of vein grafts and arterial grafts. The inclusion of updated studies was considered eligible if published from 2011 to March 2015. Studies were excluded if they included only patients with advanced chronic kidney disease (non-ESRD) or included patients who received another cardiac surgical procedure along with CABG (e.g., concomitant valve surgery). The primary outcome was all-cause mortality more than one year after coronary revascularization. We searched PubMED and EMBASE and the COCHRANE database for studies using the search query ([‘chronic kidney failure’/exp or ‘albumin dialysis’/exp] and ([‘angioplasty’/exp or ‘transluminal coronary angioplasty’/exp or ‘coronary stent’/exp or ‘coronary artery bypass graft’/exp] or ‘coronary artery disease’/exp/dm_su) and [humans]/lim and [english]/lim) and (2011:py or 2012:py or 2013:py or 2014:py or 2015:py). The reference list of relevant studies and reviews was searched. A single reviewer independently screened each citation and those considered potentially applicable were retrieved for full text review. The number of events was estimated from Kaplan–Meier survival graphs when no other information was provided in the manuscript. This may overestimate the number of events but not influence the relative risk (RR). The 2 primary reviewers independently abstracted data from the full texts. There was no attempt to contact primary investigators of relevant studies. No formal assessment of the quality of observational studies was performed.
Data were analyzed using STATA statistical software, version 13 (StataCorp LP, College Station, Texas). Study characteristics were described using percentages and qualitative descriptors. The RR of death ≥1 year for CABG compared to PCI from each study was then pooled using the random-effects model. Heterogeneity in the final selected studies was assessed using the Q test and I-statistic. Subgroup analysis was performed by “year of study initiation” (≤1990, 1991 to 2002, and ≥2003), and removal of consecutive large studies. Meta-regression was performed with the independent variable “year of study initiation.” Publication bias was reported as noted by the Meta-analysis of Observational Studies in Epidemiology specifications.
Results
We excluded one study as there was no direct comparison between CABG and PCI, for a final total of 23 studies ( Figure 1 ). We initially identified 7 studies published from 2011 to 2015 and added 17 others that met our inclusion criteria from 3 previous meta-analysis. A Cochrane review search found no published randomized clinical trials for this specific comparison. Relevant details of each study that was included in this meta-analysis are provided in Table 1 . Large variations were noted between the individual studies in the sample size, year of initiation, study method, and extent of covariate collection for risk adjustment. For example, only 3 studies documented indications for coronary revascularization; whereas liver disease as a covariate was included in only 13% of the studies. Although age was available in all studies, it was often categorized and not given as a mean age for the cohort; gender however was available in 95% of the studies ( Table 2 ).
| Author | Year Published | Location | Start year | End year | Total | CABG | PCI |
|---|---|---|---|---|---|---|---|
| Takeshita | 1992 | Japan | 1983 | 1989 | 25 | 10 | 15 |
| Rinehart | 1995 | USA | 1977 | 1991 | 84 | 60 | 24 |
| Koyanagi | 1996 | Japan | 1984 | 1995 | 43 | 23 | 20 |
| Simsir | 1998 | USA | 1992 | 1196 | 41 | 22 | 19 |
| Herzog | 1999 | USA | 1978 | 1995 | 14,306 | 7419 | 6887 |
| Ohmoto | 1999 | Japan | 1983 | 1997 | 139 | 47 | 92 |
| Agirbasli | 2000 | USA | 1987 | 1997 | 252 | 130 | 122 |
| Ivens | 2001 | Germany | 1982 | 1994 | 105 | 65 | 40 |
| Szczech | 2001 | USA | 1993 | 1995 | 407 | 244 | 163 |
| Baldovinos | 2002 | Uruguay | 1994 | 1999 | 51 | 23 | 28 |
| Chertow | 2000 | USA | 1994 | 1995 | 75 | 29 | 46 |
| Herzog | 2002 | USA | 1995 | 1998 | 15,784 | 6668 | 9116 |
| Aoki | 2003 | Japan | 1997 | 2001 | 125 | 55 | 70 |
| Hemmelgarn | 2004 | Canada | 1995 | 2001 | 300 | 153 | 147 |
| Fujimoto | 2007 | Japan | 1984 | 2002 | 145 | 64 | 81 |
| Manabe | 2009 | Japan | 2004 | 2007 | 46 | 28 | 18 |
| Ashrith | 2010 | USA | 2003 | 2006 | 87 | 54 | 33 |
| Sunagawa | 2010 | Japan | 2002 | 2006 | 104 | 29 | 75 |
| Chang | 2012 | USA | 1997 | 2009 | 14,098 | 7049 | 7049 |
| Baek | 2011 | Korea | 2003 | 2006 | 87 | 43 | 44 |
| Chou | 2014 | China | 1997 | 2008 | 1287 | 166 | 1121 |
| Marui | 2014 | Japan | 2005 | 2007 | 388 | 130 | 258 |
| Krishnaswami | 2015 | USA | 1996 | 2008 | 1015 | 446 | 569 |
| Variable | Number of studies that reported the particular variable N total = 23 | (%) | Average value of variable (if applicable) |
|---|---|---|---|
| Study start year | 23 | 100.0 | |
| Study end year | 23 | 100.0 | |
| Total Number of Patients | 23 | 100.0 | |
| Number of CABG patients | 23 | 100.0 | |
| Number of PCI patients | 23 | 100.0 | |
| Indications for patient entry into study | 3 | 12.5 | |
| EF-CABG | 9 | 39.1 | 48% |
| EF-PCI | 9 | 39.1 | 52% |
| Mean age- CABG | 16 | 69.6 | 62.2 years |
| Mean age- PCI | 16 | 69.6 | 61.3 years |
| Men- CABG | 22 | 95.7 | |
| Men- PCI | 22 | 95.7 | |
| DM-CABG | 20 | 86.9 | |
| DM- PCI | 20 | 86.9 | |
| HTN- CABG | 17 | 73.9 | |
| HTN- PCI | 17 | 73.9 | |
| CHF-CABG | 9 | 39.1 | |
| CHF-PCI | 9 | 39.1 | |
| Tobacco- CABG | 13 | 56.5 | |
| Tobacco- PCI | 13 | 56.56 | |
| COPD-CAPG | 7 | 30.4 | |
| COPD-PCI | 7 | 30.4 | |
| CVA-CABG | 11 | 47.8 | |
| CVA-PCI | 11 | 47.8 | |
| Cancer-CABG | 6 | 26.1 | |
| Cancer-PCI | 6 | 26.1 | |
| Liver disease-CABG | 3 | 13.0 | |
| Liver disease-PCI | 3 | 13.0 | |
| PVD-CABG | 9 | 39.1 | |
| PVD-PCI | 9 | 39.1 | |
| Prior MI- CABG | 18 | 78.2 | |
| Prior MI- PCI | 18 | 78.2 | |
| Atrial fibrillation-CABG | 3 | 13.0 | |
| Atrial fibrillation- PCI | 3 | 13.0 | |
| Prior Intervention-CABG | 5 | 21.7 | |
| Prior Intervention- PCI | 5 | 21.7 | |
| Duration of Dialysis – CABG | 11 | 47.8 | 36.8 months |
| Duration of Dialysis – PCI | 11 | 47.8 | 39.9 months |
| LM or 2 or 3 vessel intervention – CABG | 12 | 52.2 | |
| LM or 2 or 3 vessel intervention – PCI | 12 | 52.2 | |
| Number of vessels bypassed | 6 | 26.1 | 3.1 vessels bypassed |
| Number of vessels intervened | 4 | 17.4 | 1.6 vessels intervened |
| LIMA to LAD | 4 | 17.4 | |
| Average follow up | 23 | 100.0 | 38.5 months |
Among 23 studies, the total sample size was 48,994 patients (CABG: n = 22,957; PCI: n = 26,037) with an average follow-up of 37.1 ± 22.6 months. There were 29,738 total deaths, for an overall mortality risk of 60.1%. Patients who underwent CABG had a lower mortality risk (13,143 of 22,957, 57.2%) than those who underwent PCI (16,595 of 26,037, 63.8%). Figure 2 , is the forest plot with the RRs and 95% CIs using the random-effects model of individual studies along with the overall summary estimate of 0.92 (95% CI 0.89 to 0.96) favoring CABG.
Significant heterogeneity was demonstrated (I 2 = 48.6%, p = 0.005) for the main estimate. Figure 3 shows the result of subgroup analyses categorized by year of study initiation along with the results for heterogeneity for each category. The summary estimate of RR of death comparing CABG and PCI decreased from 1.02 (95% CI 0.75 to 1.38) in studies initiated before 1990 to 0.91 (95% CI 0.87 to 0.95) in studies started from 1991 to 2002. The RR summary estimate for studies initiated on or after 2003 was 0.93 (95% CI 0.83 to 1.06). The corresponding heterogeneity decreased from studies initiated before 1990 (I 2 = 63.0%, p = 0.009) to studies initiated on or after 2003 (I 2 = 0.0%, p = 0.64). The RR of removal of consecutive large studies did not substantially change the summary estimate but did affect precision (Removal of Herzog 1999 study resulted in a RR of 0.92 [0.88 to 0.97]; removal of Herzog 1999 and Herzog 2002 resulted in a RR of 0.94 [0.87 to 1.02]; and finally, removal of Herzog 1999, Herzog 2002, and Chang 2012 resulted in a RR of 0.95 [0.85 to 1.06]).
