Unmeasured Confounders in Observational Studies Comparing Bilateral Versus Single Internal Thoracic Artery for Coronary Artery Bypass Grafting: A Meta-Analysis.

BACKGROUND: Observational studies suggest a survival advantage with bilateral single internal thoracic artery (BITA) versus single internal thoracic artery grafting for coronary surgery, whereas this conclusion is not supported by randomized trials. We hypothesized that this inconsistency is attributed to unmeasured confounders intrinsic to observational studies. To test our hypothesis, we performed a meta-analysis of the observational literature comparing BITA and single internal thoracic artery, deriving incident rate ratio for mortality at end of follow-up and at 1 year. We postulated that BITA would not affect 1-year survival based on the natural history of coronary artery bypass occlusion, so that a difference between groups at 1 year could not be attributed to the intervention. METHODS AND
RESULTS: We searched MEDLINE and Pubmed to identify all observational studies comparing the outcome of BITA versus single internal thoracic artery. One-year and long-term mortality for BITA and single internal thoracic artery were compared in the propensity-score-matched (PSM) series, that is, the form of observational evidence less prone to confounders. Thirty-eight observational studies (174 205 total patients) were selected for final comparison. In the 12 propensity-score-matched series (34 019 patients), the mortality reduction for BITA was similar at 1 year and at the end of follow-up (incident rate ratio, 0.70; 95% confidence interval, 0.60-0.82 versus 0.77; 95% confidence interval, 0.70-0.85; P for subgroup difference=0.43).
CONCLUSIONS: Unmeasured confounders, rather than biological superiority, may explain the survival advantage of BITA in observational series.

Clinical Perspective

What Is New?

Our findings suggest that factors not related to the conduit patency, such as the patients’ general status or quality of the target vessels, play a role in determining the outcome of observational studies and that a selection bias is present even in propensity‐score–matched analyses.

What Are the Clinical Implications?

Our findings elicit concerns regarding the ability of the propensity‐matching process to overcome selection bias and assure comparability between groups.

The long‐term clinical outcomes data from the ART (Arterial Revascularization Trial) trial and new randomized studies are needed to clarify the effect of bilateral internal thoracic artery grafting in patients undergoing coronary bypass surgery.

A clear contradiction between observational and randomized studies exists in the literature on the effect of multiple internal thoracic artery grafts in patients undergoing coronary artery bypass surgery.

In the 1980s, it was recognized that in coronary artery bypass surgery patients long‐term survival was enhanced when the left anterior descending (LAD) was grafted with a left internal thoracic artery, rather than a saphenous vein graft (SVG).1 By extension, the use of bilateral internal thoracic arteries (BITAs) should further increase postoperative survival, compared with the use of a single internal thoracic artery (SITA).2 This difference is generally attributed to greater and more‐durable patency of the internal thoracic artery compared with the SVG, as well as increased late SVG atherosclerosis.3

In the past 25 years, a very large amount of observational data, including 6 meta‐analyses,4, 5, 6, 7, 8, 9 have supported this concept. On this basis, the use of BITA is a class IIA recommendation in patients with a long anticipated life expectancy by current guidelines and professional society position papers.10, 11, 12

The randomized studies, however, reported different results. To date, there have been 4 randomized controlled trials (RCTs) comparing BITA and SITA.13, 14, 15, 16 In these studies, survival has been similar following BITA and SITA grafting. In the largest of the RCTs, the ART (Arterial Revascularization Trial), mortality was 8.7% after BITA grafting and 8.4% following SITA at 5 years.16

There are several possible explanations for the discrepant findings between observational and RCT evidence. The RCTs may not have sufficient sample size or follow‐up to detect a mortality difference compared with observational series. In the ART trial, a relatively higher proportion of crossovers in the patients randomized to BITA, as well as the allowed use of a radial artery in the SITA group, may have diluted the treatment effect.

The other possible explanation, however, is that the benefit observed in the observational studies for BITA grafting is largely related to unmeasured confounders.

The objective of this study is to perform a meta‐analysis of the observational literature comparing survival following BITA and SITA grafting. To evaluate whether unmeasured confounders rather than biological superiority explained the BITA effect, we chose to compare both 1‐year as well as late survival in the BITA and SITA cohorts. We postulated that BITA would not affect 1‐year survival based on the natural history of SVG occlusion. The latter analysis was restricted to propensity‐score–matched studies, because PSM is considered the best method to minimize confounding in observational series.

Methods

The data, analytical methods, and study materials will not be made available to other researchers for purposes of reproducing the results or replicating the procedure.

Search Strategy and Study Selection

This systematic review was conducted in accord to the Preferred Reporting Items for Systematic Reviews and Meta‐Analyses guidelines.17

Pubmed and OVID's version of MEDLINE was searched from January 1972 to August 2017 for publications comparing BITA versus SITA grafting on all‐cause mortality. The following keywords were combined with the Boolean operator “or”: “bilateral internal mammary,” “bilateral internal thoracic,” “total arterial revascularization,” and “multiple arterial revascularization.” The full search strategy can be found in Data S1. All citations were screened for study inclusion independently by 2 investigators (A.D.F. and M.G.). Any disagreements were discussed and resolved by consensus. In addition, the bibliography of all studies and meta‐analyses was searched to identify further publications.

Inclusion criteria for analysis were:

Observational study (unadjusted and adjusted studies were eligible).

Sample size of at least 100 patients in each group.

Follow‐up duration longer than 30 days.

Written in English language.

We excluded studies that were: RCTs, not performed in humans, review articles, case reports, editorials, and expert opinions. To ensure that the analysis was strictly limited to a comparison of BITA versus SITA, we excluded studies where an additional arterial graft was used in 1 of the 2 groups and it was not possible to abstract the exact information for the isolated BITA and SITA series. In case of overlapping between studies or multiple publications from the same center, only the publication with the largest sample size was considered.

The quality of included studies was assessed using the Newcastle–Ottawa Scale for observational studies by 2 investigators independently (A.D.F. and M.G.).18 The highest possible score is 9 stars; <6 stars was considered low quality whereas ≥6 stars was considered high quality.

Data Abstraction

Two investigators (A.D.F. and M.G.) independently abstracted the following: study demographics (study period, country, and centers involved, sample size), study design methods, completeness of follow‐up, and follow‐up duration. In addition, the following patient characteristics in the unmatched and matched groups were also obtained: age, female sex, diabetes mellitus, left ventricular ejection fraction, and chronic obstructive pulmonary disease. Continuous variables were expressed as median (25th, 75th percentile) or as mean±SD. Categorical variables are reported as frequency (%).

For all‐cause mortality, crude event rates, unadjusted and adjusted hazard ratios (HRs) for BITA versus SITA grafting, and their respective 95% confidence intervals (CIs) and log P‐rank values were abstracted.

Outcome Analyses

The primary outcome was all‐cause mortality. Long‐term all‐cause mortality for BITA and SITA patients was compared in all the studies.

Subgroup analyses for the primary outcome were performed as follows:

Studies in the general population versus studies in specific subgroups of patients (ie, diabetics, elderly patients as defined by the individual studies, patients with renal failure, urgent/emergent cases, and patients with low ejection fraction).

Unadjusted versus adjusted studies (including regression‐adjusted and PSM) in the general population.

Regression‐adjusted versus PSM studies in the general population.

To assess for possible treatment allocation bias in the observational studies, we chose to compare 1‐year mortality between matched treatment groups. The 1‐year interval was chosen because the patency rate of SVGs at 1 year remains high and a survival difference related to difference in patency between arterial and venous conduits is unlikely.19 PSM is a robust method used to balance against confounding by indication in observational studies20; for this reason, we compared all‐cause mortality for BITA and SITA at 1 year in the PSM studies only.

Analytical Plan

Long‐term all‐cause mortality between BITA and SITA patients was compared in all studies initially. Comparisons were then performed in the general population studies after exclusion of studies restricted to specific patient subgroups (diabetes mellitus, elderly as defined in the individual studies, renal failure, urgent/emergent, and reduced left ventricular ejection fraction) and in the individual specific patient subgroups. Next, separate comparisons were made between BITA and SITA in the unadjusted and adjusted series (covariate adjusted and PSM combined). Last, comparisons were performed in the covariate adjusted and PSM series separately.

One‐year mortality between BITA and SITA patients was compared in PSM studies only.

Statistical Analysis

The generic inverse variance method21 was used to pool the natural logarithm of the incident rate ratio (IRR) across studies to account for potentially different follow‐up durations between the groups. We estimated the IRR through several means depending on the available study data. When HRs were provided, we took the natural logarithm of the HR; the SE was derived from the 95% CI or log rank P value.22 When Kaplan–Meier curves were present, we estimated the number of events from the curves to calculate the IRR, as previously described.23 The SE was estimated from the number of events in each arm.22 When event rates were not readily available, they were extracted from Kaplan–Meier curves using GetData Graph Digitizer software (version 2.26; http://getdata-graph-digitizer.com/) according to a previously described method.24

A random‐effects model was used for statistical survival pooling, computing risk estimates with 95% CIs. Funnel plots were used to assess publication bias by graphical inspection.25 Hypothesis testing for equivalence was set at the 2‐tailed 0.01 level. Hypothesis testing for statistical heterogeneity was set at the 2‐tailed 0.10 level and was based on the Cochran Q test, with I2 values of 0% to 25%, 26% to 50%, and 51% to 100% representing low, moderate, and high heterogeneity, respectively.26

Metaregression analysis examining the following variables—age, sex, diabetes mellitus, and left ventricular ejection fraction—was performed. In addition, a “leave‐one‐out” analysis and a cumulative meta‐analysis were performed in all studies ordered by year of publication.

All analyses were performed using CMA software (version 3; Biostat, Englewood, NJ).

Results

Selected Studies

From 2921 titles, 149 pertinent studies were included for full‐text review. We excluded 111 studies that did not meet inclusion criteria. Further details of the study flow are shown in Figure S1. A total of 38 observational studies were selected for the quantitative analysis. Eight nonadjusted, 9 covariate‐adjusted, and 21 PSM studies were included (see Table 1).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, 60, 61, 62, 63, 64 Twenty‐eight studies (162 989 patients) were performed in the general population, whereas 10 (11 216 patients) were performed in specific subgroups of patients (diabetics: 3 studies [1533 patients]; elderly: 4 studies [6033 patients]; renal failure patients: 1 study [1203 patients]; urgent/emergent cases: 1 study [652 patients]; and patients with low ejection fraction: 1 study [1795 patients]). An overview of the studies is summarized in Tables 1 and 2, 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, 60, 61, 62, 63, 64 (variables included for PSM are summarized in Table S1).

Table 1

Characteristics of the Studies Included in the Primary Analysis

Study	Year	Center	Study Period	Setting	Type of Study	Adjustment Performed
Ashraf27	1994	Manchester Royal Infirmary, Manchester, UK	1989–1992	Isolated primary CABG	Retrospective	NS
Benedetto28	2014	Harefield Hospital, London, UK	2001–2013	Isolated primary CABG	Retrospective	PSM
Berreklouw29	2001	Catharina Hospital, Eindhoven, The Netherlands	1985–1990	Isolated primary CABG	Retrospective	MCPHR
Bonacchi30	2006	University of Florence, Italy	1997–2003	Non‐elective CABG in unstable angina patients	Retrospective	MCPHR
Buxton31	1998	Austin and Repatriation Medical Center, University of Melbourne, Victoria, Australia	1985–1995	Isolated primary CABG	Retrospective	MCPHR
Calafiore32	2004	University Hospital, Torino, Italy and “G D'Annunzio” University, Chieti, Italy	1986–1999	Isolated primary CABG in patients <75 years old	Retrospective	PSM
Carrier33	2009	Montreal Heart Institute, Montreal, Quebec, Canada	1995–2007	Isolated primary CABG	Retrospective	MCPHR
Dalén34	2014	Nationwide population‐based cohort study (Sweden)	1997–2008	Isolated primary CABG	Retrospective	PSM
Danzer35	2001	University Hospital, Geneva, Switzerland	1983–1989	Isolated primary CABG	Retrospective	NA
Dewar36	1995	Vancouver Hospital and Health Sciences Centre, University of British Columbia, Vancouver, Canada	1984–1992	Isolated primary CABG	Retrospective	Univariate regression
Endo37	2001	Tokyo Women's Medical University, Tokyo, Japan	1985–1998	Isolated primary CABG	Retrospective	MCPHR
Gansera 200438	2004	Klinikum Bogenhausen, Munich, Germany	1997–1999	Isolated primary CABG	Retrospective	NA
Gansera 201739	2017	Klinikum Bogenhausen, Munich, Germany	2000–2011	Isolated CABG in diabetic patients, <65 years old	Retrospective	PSM
Grau40	2015	The Valley Columbia Heart Center, Columbia University College of Physicians and Surgeons, Ridgewood, NJ, USA	1994–2013	Isolated CABG	Retrospective	PSM
Hirotani41	2003	Tokyo Saiseikai Central Hospital, Minato‐Ku, Tokyo, Japan	1991–2003	Isolated primary CABG in diabetic patients	Retrospective	NA
Itoh42	2016	Saitama Medical Center, Jichi Medical University, Saitama, Japan	1990–2014	Isolated CABG in patients ≥75 years old	Retrospective	PSM
Johnson43	1989	Milwaukee Heart Surgery Associates, S.C., and St. Mary's Hospital, Milwaukee, WI, USA	1972–1986	Isolated CABG	Retrospective	NA (patients matched with the general US population)
Jones44	2000	Baylor College of Medicine and Veterans Affairs Medical center, Houston, TX, USA	1986–1996	Isolated primary CABG in patients >65 years old	Retrospective	NA
Joo45	2012	Yonsei Cardiovascular Hospital, Seoul, Republic of Korea	2000–2009	Isolated OPCAB	Retrospective	PSM
Kelly46	2012	Queen Elizabeth II Health Sciences Center, Halifax, Nova Scotia, Canada	1995–2007	Isolated primary CABG	Retrospective	Non‐parsimonious MCPHR including PS quintiles
Kieser47	2011	The Province of Alberta, Canada	1995–2008	Isolated primary CABG	Retrospective	MCPHR
Kinoshita48	2015	Shiga University of Medical Science, Otsu, Japan	2002–2014	Isolated CABG‐patients stratified by GFR	Retrospective	PSM
Kurlansky49	2010	Florida Heart Research Institute, Miami, FL, USA	1972–1994	Isolated CABG	Retrospective	PSM
Locker50	2012	Mayo Clinic, Rochester, MN, USA	1993–2009	Isolated primary CABG	Retrospective	PSM and MCPHR
Lytle51	2004	The Cleveland Clinic Foundation, Cleveland, OH, USA	1971–1989	Isolated primary CABG	Retrospective	PSM
Medalion52	2015	Tel Aviv Sourasky Medical Center, Tel Aviv, Israel	1996–2008	isolated CABG in patients ≥70 years old	Retrospective	PSM
Mohammadi53	2014	Quebec Heart and Lung Institute, Quebec City, Canada	1991–2011	Isolated primary CABG in patients with EF ≤40%	Retrospective	PSM
Nasso54	2012	Multicenter	2003–2008	Isolated primary CABG	Retrospective	PSM
Naunheim55	1992	St. Louis University Medical Center, St. Louis, MS, USA	1972–1975	Isolated CABG	Retrospective	NA
Navia56	2016	Instituto Cardiovascular de Buenos Aires, Buenos Aires, Argentina	1996–2014	Isolated CABG	Retrospective	PSM
Parsa57	2013	Duke University Medical Center, Durham, NC, USA	1984–2009	Isolated CABG	Prospective	MCPHR
Pettinari58	2015	Ziekenhuis Oost Limburg, Genk, Belgium and University Hospitals Leuven, Leuven, Belgium	1972–2006	Isolated CABG in patients ≥70 years old	Retrospective	PSM
Pick59	1997	Mayo Clinic, Rochester, MN, USA	1983–1986	Isolated CABG	Retrospective	MCPHR
Rosenblum60	2016	Emory University School of Medicine, Atlanta, GA, USA	2003–2013	Isolated primary CABG	Retrospective	PSM
Schwann61	2016	Multicenter	1987–2011	Isolated CABG	Retrospective	PSM
Stevens62	2004	Montreal Heart Institute, Montreal, Quebec, Canada	1985–1995	Isolated primary CABG	Retrospective	MCPHR including PS
Tarelli63	2001	Varese Hospital, Varese, Italy	1988–1990	Isolated CABG	Retrospective	NA
Toumpoulis64	2006	St. Luke's–Roosevelt Hospital Center at Columbia University, NY, USA	1992–2002	Isolated CABG in diabetic patients	Retrospective	MCPHR in PS‐matched patients

CABG indicates coronary artery bypass grafting; MCPHR, multivariable Cox proportional hazards regression; NA, not applicable; NR, not reported; NS, not specified; OPCAB, off‐pump coronary artery bypass; PS, propensity score; PSM, propensity‐score matching.

Table 2

Overview of the Studies Included in the Primary Analysis

Study	Overall Population, n	UNM BITA, n	UNM SITA, n	PSM BITA, n	PSM SITA, n	Mean/Median Follow‐up (Y)	Completeness of Follow‐up
Ashraf27	300	150	150	NA	NA	Median (IQR) BITA: 1.9 (1.3–2.6) Median (IQR) SITA: 2.3 (1.7–3)	NR
Benedetto28	4195	750	3445	750	750	4.8±3.2 (PSM sample)	100%
Berreklouw29	482	NA	NA	249	233	BITA: 9.7±2.7 SITA: 10.1±2.4	94%
Bonacchi30	652	NA	NA	320	332	5.6±1.4	99.7%
Buxton31	2853	1296	1557	NA	NA	4.3	95.9%
Calafiore32	1602	1026	576	570	570	Overall: 7.3±4.8 BITA: 7.1±5.0 SITA: 7.5±4.7	100%
Carrier33	6655	Statin+: 1166 Statin−: 69	Statin+: 4835 Statin−: 585	NA	NA	10	99%
Dalén34	49 702	559	49 143	558	558	7.5	100%
Danzer35	521	382	139	NA	NA	10	97.5%
Dewar36	1142	377	765	NA	NA	4	NR
Endo37	1131	443	688	NA	NA	6.2	99.3%
Gansera 200438	1378	716	662	NA	NA	5.3	NR
Gansera 201739	250	NA	NA	125	125	9.3±3.5	100%
Grau40	6666	1544	5122	1006	1006	Overall: 10.5±5 BITA: 10.9±5 SITA: 10.1±5	100%
Hirotani41	303	179	124	NA	NA	NR	95%
Itoh42	400	107	293	98	196	9.0±5.8	95.6%
Johnson43	2014	576	1438	NA	NA	NR	100%
Jones44	510	172	338	NA	NA	5.0±3.1	100%
Joo45	1749	392	1357	366	366	Overall: 7.0±2.0 BITA: 6.9±2.1 SITA: 7.1±2.7	98.1%
Kelly46	7633	1079	6554	NA	NA	BITA: 5.4 SITA: 4.6	NR
Kieser47	5067	1038	4029	NA	NA	Overall: 7 BITA: 6.4±3.2 SITA: 7.1±3.4	NR
Kinoshita48	1203	750	453	412	412	PSM BITA: 5.6±3.3 PSM SITA: 4.9±3.2	99%
Kurlansky49	4584	2215	2369	Quintiles	Quintiles	Overall: 11.5 BITA: 12.7 SITA: 11.1	BITA=96.7% SITA=98.3%
Locker50	8295	BITA only: 271 BITA/SVG: 589	7435	NR	NR	7.6±4.6	100%
Lytle51	10 124	2001	8123	1152	1152	BITA: 16.2±2.4 SITA: 16.3±2.5	100%
Medalion52	1627	1045	582	NA	NA	8.2±4.5	98%
Mohammadi53	1795	129	1666	111	111	Overall PSM: 8.0±5.3 PSM BITA: 8.6±5.1 PSM SITA: 7.7±5.5	92.7%
Nasso54	8054	4088	3966	3584	3584	3.1	98%
Naunheim55	365	100	265	100	100	NR	96.5%
Navia56	2486	2098	388	485	NR	Median: 5.5 (IQR: 2.6–8.8)	95%
Parsa57	17 609	728	16 881	NA	NA	NR	100%
Pettinari58	3496	1328	2168	892	892	3.1	100%
Pick59	321	NA	NA	160	161	9.8±2.8	100%
Rosenblum60	8254	873	7381	306	306	Median: 2.8 (IQR: 1.1–4.9)	100%
Schwann61	5125	641	4484	551	551	NR	100%
Stevens62	4382	1835	2547	NA	NA	Overall: 11±3 BITA: 8±2 SITA: 12±3	98%
Tarelli63	300	150	150	NA	NA	Overall: 9.2 BITA: 9.2±2.8 SITA: 9.1±2.5	100%
Toumpoulis64	980	NA	NA	490	490	4.7±3.0	99.1%

BITA indicates bilateral internal thoracic arteries; IQR, interquartile range; NA, not applicable; NR, not reported; PSM, propensity‐score matched; SITA, single internal thoracic artery; SVG, saphenous vein graft.

The selected studies reported on 174 205 patients (BITA: 32 206; SITA: 141 999) for final comparisons.

Overall, the BITA and SITA groups presented different preoperative risk‐factor distribution (mean age, BITA versus SITA: 60 versus 64.1 years; female sex, BITA versus SITA: 16% versus 20.8%; diabetes mellitus, BITA versus SITA: 32.2% versus 40.5%; chronic obstructive pulmonary disease, BITA versus SITA: 9.6% versus 11.8%; Table S2).

Long‐Term All‐Cause Mortality

Mean follow‐up time across the 38 studies was 7.25 years (range, 2.1–16.3). The overall mortality rate at the end of follow‐up was 28.03±18.4% in the BITA versus 39.96±23.5% in the SITA series.

Use of BITA was associated with a statistically significant reduction of mortality at the end of follow‐up when compared with SITA (IRR, 0.74; 95% CI, 0.69–0.80; P<0.001; I2=71%; Figure 1A27, 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, 60, 61, 62, 63, 64 and Figure S2). This finding was consistent across the general population and all the specific patient subgroups and all the study designs (Figures S3 through S5) and was not influenced by age, sex, diabetes mellitus, and ejection fraction (Figure 2).

Figure 1

A, Forest plot comparing the effect of the use of BITA vs SITA on end of follow‐up mortality across all the included studies (38 studies; 174 205 patients). B, Cumulative analysis of all the included studies using random‐effect model (38 studies; 174 205 patients). BITA indicates bilateral internal thoracic artery; CI, confidence interval; SITA, single internal thoracic artery. Incident rate ratio (IRR) is used.

Figure 2

Results of the metaregression analyses. Univariate metaregression analysis showed that the effect of BITA was not influenced by age (slope P value=0.625; intercept P value=0.941), sex (slope P value=0.160; intercept P value=0.0002), diabetes mellitus (slope P value=0.730; intercept P value=0.0001), and ejection fraction (slope P value=0.674; intercept P value=0.482). Similarly, multivariate metaregression analysis showed that the effect of BITA was not influenced by age (slope P value=0.270), sex (slope P value=0.412), diabetes mellitus (slope P value=0.848), and ejection fraction (slope P value=0.644) with intercept P value=0.487 (plot not shown). BITA indicates bilateral internal thoracic artery; DM, diabetes mellitus; EF, ejection fraction.

One‐Year All‐Cause Mortality in the PSM Populations

Mean follow‐up time of the 12 PSM studies was 7.41±4.4 years, and the number of patients included was 34 019. Use of BITA was associated with a similar reduction of mortality at 1‐year and at the end of follow‐up (IRR, 0.70; 95% CI, 0.60–0.82 at 1 year versus IRR, 0.77; 95% CI, 0.70–0.85 at the end of follow‐up; P for subgroup differences=0.43; Figure 3)1 (details of the statistical analysis for the PSM studies included in this analysis are summarized in Table S3). These findings were robust in a leave‐one‐out analysis (Figure 4).2

Figure 3

Forest plot comparing the effect of the use of BITA vs SITA on end of follow‐up (top) and 1‐year (bottom) mortality in PSM studies in the general population (12 studies; 34 019 patients). BITA indicates bilateral internal thoracic artery; CI, confidence interval; PSM, propensity‐score matched; SITA, single internal thoracic artery. Incident rate ratio (IRR) is used.

Figure 4

Leave‐one‐out analyisis for 1‐year mortality among PSM studies (12 studies). BITA indicates bilateral internal thoracic artery; CI, confidence interval; PSM, propensity‐score matched; SITA, single internal thoracic artery. Incident rate ratio (IRR) is used.

Publication Bias and Internal Validity Appraisal

Study quality was high across all studies included in the primary analysis (Table S4). Overall heterogeneity was high both at 1‐year analysis in the PSM studies (I2=51%) and at end of follow‐up in the overall studies analysis (I2=71%). Publication bias was low, as assessed by funnel plots, for all‐cause mortality in the primary analysis (Figure 5).

Figure 5

Publication bias as assessed by funnel plots for all‐cause mortality in the primary analysis. A, All included studies. B, Studies performed in the general population vs studies performed in specific subpopulations. C, Unadjusted studies vs adjusted studies. D, PSM studies vs adjusted non‐PSM studies. E, PSM studies at 1‐year follow‐up vs PSM studies at end of follow‐up. PSM indicates propensity‐score matched.

An overview of the results of all the analyses is provided in Table S5.

Discussion

For almost 25 years, the concept that the use of BITA is associated with improved survival after coronary artery bypass surgery has been accepted in the cardiovascular community. This concept is almost completely based on observational studies.

To date, at least 60 English‐language observational studies comparing the clinical outcome of BITA and SITA patients have been published (Figure S1). The overwhelming majority of these have shown better outcomes in the BITA treatment group. Several reports have also suggested that the advantages of BITA grafting could be extended to females,65 diabetics,66 and patients with chronic renal insufficiency.48 Over time, this evidence has been summarized in 6 meta‐analyses.4, 5, 6, 7, 8, 9 All of them showed a significant and similar survival advantage, as measured by the HR, for the use of BITA (see Table 3).4, 5, 6, 7, 8, 9

Table 3

Published Meta‐Analyses of the Observational Evidence on the BITA vs SITA Comparison

First Author, Year	Studies Included in Survival Analysis, n	Patients Included in Survival Analysis, n	Type of Observational Studies Included	Patient Populations Excluded by Inclusion Criteria	HR in Favor of BITA
Taggart, 20015	7	15 962	All	None	0.81 [95% CI 0.70–0.94]
Rizzoli, 20024	7	15 299	All	High‐risk patients, emergencies, diabetics	0.79 [95% CI 0.66–0.91]
Weiss, 20137	27	79 063	All	None	0.78 [95% CI 0.72–0.84]
Takagi, 20146	20	70 897	Adjusted	None	0.80 [95% CI 0.77–0.84]
Yi, 20148	9	15 583	Adjusted	None	0.79 [95% CI 0.75–0.84]
Buttar, 20179	29	89 399	All	None	0.78 [95% CI 0.72–0.84]

BITA indicates bilateral internal thoracic artery; HR, hazard ratio; PSM, propensity‐score matched; UNM, unmatched.

Our analysis pools data from 38 of these studies and 174 205 patients and confirms the previous findings (Table 3).4, 5, 6, 7, 8, 9 We used IRRs instead of HR or relative risk ratio to account for potential differences in follow‐up duration within studies and between studies. We confirmed better long‐term survival for BITA compared with SITA (IRR, 0.74; 95% CI, 0.69–0.80; P<0.001; Figure 1A).27–64 This difference was evident independently from the patient population included and the methodology used (Figures S3 through S5). The benefit was uncertain from 1989 to 2000, was consistently significant at the 0.05 level starting in 2001, and crossed the 0.01 and 0.001 levels in 2004 (Figure 1B).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, 60, 61, 62, 63, 64

Basing on these data, the current US10 and European11 Guidelines encourage the use of a second arterial graft in patients with a long life expectancy, and last year the Society of Thoracic Surgeons published a position paper strongly encouraging a wider use of arterial grafts.12

It must, however, be noted that the results of the observational studies have not been confirmed in the randomized comparisons. The 4 RCTs that have compared BITA and SITA to date have all failed to show a survival difference between the 2 revascularization strategies.13, 14, 15, 16 Two of the RCTs were small, with less than 100 patients in each arm, and had limited follow‐up, so that they were probably underpowered to detect moderate differences.13, 14 Another study was moderate in size (Stand‐in‐Y,15 800 patients) and the most recent, the ART trial,16 included more than 3000 patients.

The Stand‐in‐Y Mammary study compared the outcomes of 800 patients randomized to receive BITA using 2 different configurations: SITA and radial artery or SITA and saphenous vein.15 At a mean follow‐up of 24.1±9.8 months, no difference in survival was found between the BITA and SITA groups (P=0.62; odds ratio, 0.63; 95% CI, 0.27–1.47), although patients with arterial grafts had better cardiac event–free survival (Wilcoxon test, P<0.0001).

The ART trial randomized 3102 patients to receive 1 or 2 internal thoracic arteries.16 The primary end point is overall survival, and the study was designed to be able to detect a 20% reduction in the primary end point at 10 years. At a planned 5‐year interim analysis, no difference in survival (91.3% in the BITA group and 91.6% in the SITA group; HR, 1.04; CI, 0.81–1.32) or in the composite of mortality, myocardial infarction, and/or stroke (12.2% BITA versus 12.7% SITA; HR, 0.96; CI, 0.79–1.17) was found between groups.

Several methodological flaws in the design of the RCTs can partially explain the variance between the results of the randomized and observational studies. All the RCTs were limited to mid‐term follow‐up, and it is known the attrition rate of saphenous grafts remains low at 5 years19; it is possible that a difference between the groups would have become apparent with further follow‐up. There are additional considerations specifically regarding the ART study that may explain a negative result. A sizeable proportion (23%) of patients randomized to SITA also received a radial artery as an additional arterial graft. There was a high rate of crossover in the group allocated to BITA (16.4%). There was very high compliance with optimal medical therapy in both groups (90% of patients on aspirin, beta‐blockers, and statins). Finally, there was a treatment age interaction that approached statistical significance, favoring BITA in patients aged <70 years whereas BITA appeared harmful in patients aged >70.

There are, however, biological reasons in support of the results of the RCTs. A second arterial conduit to a non‐LAD target has less potential to impact on overall survival than the single left internal thoracic artery to the LAD. Solid evidence suggests, in fact, that in coronary artery bypass surgery, patient survival is mainly determined by the status of the LAD and that grafts to non‐LAD vessels are more likely to affect other cardiac end points (myocardial infarction, angina recurrence, and need for revascularization), but not overall survival.27, 28, 29 The LAD also can provide collaterals to other coronaries (commonly the right coronary); a persistently patent internal thoracic artery graft to the LAD can therefore supply not only the anterior wall, but, through collaterals, viable myocardium in other territories. Last, patency of grafts to the LAD generally exceed the patency of grafts to non‐LAD vessels.3

Our hypothesis, however, is that the difference in results between the RCTs and the observational evidence is attributed to unmeasured confounders and not to the difference in revascularization strategy.

In order to test this hypothesis, we repeated the BITA versus SITA comparison at 1 year, when the attrition rate of the SVGs is still low and a survival difference attributable to a difference in graft patency is unlikely.

Because PSM studies are considered the observational studies less prone to confounders, we decided to limit the 1‐year analysis to PSM studies only.

In fact, PSM series constitute a large amount of the current evidence in the surgical fields.20 The PSM process is thought to be able to minimize differences in the preoperative risk profile of the patients, and PSM studies are often quoted as the best level of evidence after RCTs.20

We found that the relative survival advantage attributed to the BITA group at 1 year was similar to that observed at late follow‐up (Figure 3).3

This finding suggests that factors not related to the conduit patency, such as the patients’ general status or quality of the target vessels, played a role in determining the outcome and that unmatched biases are present even in PSM studies.

The use of the BITA increases the complexity and invasiveness of the procedure. It is likely that surgeons tend to reserve this operation for the patients perceived as healthier and with longer life expectancy from a cardiac and a general health perspective. A bias may also exist in terms of the graftability and location of the target vessels. This type of “eye‐balling” or clinical acumen based on the individual surgeon's experience is very difficult to quantify; the statistics can only be adjusted for the measured, and not for the unmeasured, confounders. Our findings elicit concerns on the ability of the propensity‐matching process to overcome treatment allocation biases in observational studies and assure comparability between groups.

Limitations

This analysis shares the common limitations of meta‐analysis of observational data, although the funnel plots do not indicate important publication bias.

In addition, the different studies included different surgical techniques (on‐ versus off‐pump) and grafting strategies (single versus composite grafts) as well as different definitions and matching algorithms, so that the homogeneity of the included population cannot be regarded as optimal.

In most of the series, the 1‐year IRR was not specified in the original study and had to be derived using the described statistical methods.

Upon careful review of the methods of the PSM studies, we could not confer that the original studies adjusted the variance estimates appropriately for the matched nature of the data in the original studies (Table S3). That said, the HRs would still be correct, and the leave‐one‐out analysis was consistent with the overall findings.

Finally, given that we included only articles in English, a language bias cannot be excluded, although there are no plausible biological reasons to support it.

Conclusions

In conclusion, the present meta‐analysis challenges the benefit traditionally attributed to BITA grafting. The fact that, even in the PSM series, BITA patients exhibit a significant survival advantage at 1‐year follow‐up suggests that unmeasured confounders may account for the reported survival benefit of BITA in the observational series.

In addition, our results suggest that even our best statistical methods to minimize baseline demographic differences in observational studies have major limitations.

Later reporting of the clinical outcomes of ART and new randomized studies are needed to clarify the effect of BITA grafting in patients undergoing CABG.

Sources of Funding

Prof Fremes is partially supported by the Bernard S. Goldman Chair in Cardiovascular Surgery.

Disclosures

Prof Fremes is supported, in part, by the Bernard S. Goldman Chair in Cardiovascular Surgery. The remaining authors have no disclosures to report.

Data S1. Full search strategy.

Table S1. Pretreatment Variables Included for Propensity‐Score Matching

Table S2. Risk Factor Distribution in the Populations of the Studies Included in the Primary Analysis

Table S3. Details of Statistical Analysis for the Propensity‐Score–Matched Studies Included in the 1‐Year Analysis

Table S4. Newcastle–Ottawa Scale for the Studies Included in the Primary Analysis

Table S5. Overview of the Results

Figure S1. Flow chart for study selection.

Figure S2. Leave‐one‐out analysis for the end of follow‐up mortality among all the studies included in the primary analyisis (38 studies). Incident rate ratio (IRR) is used.

Figure S3. Forest plots comparing the effect of the use of BITA vs SITA on end of follow‐up mortality after the exclusion of studies performed in specific subpopulations (28 studies; 162 989 patients, top) and in those studies performed in specific subpopulations (10 studies; 11 216 patients, bottom). BITA indicates bilateral internal thoracic artery; CI, confidence interval; SITA, single internal thoracic artery. Incident rate ratio (IRR) is used.

Figure S4. Forest plots comparing the effect of the use of BITA vs SITA on end of follow‐up mortality in adjusted (22 studies; 155, 925 patients, top) and unadjusted (6 studies; 7064 patients, bottom) studies in the general population. BITA indicates bilateral internal thoracic artery; CI, confidence interval; SITA, single internal thoracic artery. Incident rate ratio (IRR) is used.

Figure S5. Forest plots comparing the effect of the use of BITA vs SITA on end of follow‐up mortality in adjusted non‐PSM studies (10 studies; 43 855 patients, top) and PSM studies (12 studies; 34 019 patients, bottom) in the general population. BITA indicates bilateral internal thoracic artery; CI, confidence interval; PSM, propensity‐score matched; SITA, single internal thoracic artery. Incident rate ratio (IRR) is used.

Click here for additional data file.