The American Journal of Managed Care August 2008
A Meta-analysis Update: Percutaneous Coronary Interventions
Objective: To update the most recent meta-analysis comparing percutaneous coronary interventions (PCIs) with medical therapy (MT) in patients having stable coronary artery disease (CAD) by including 2 new large trials that double the total number of patients.
Study Design: Meta-analysis was used to update previous meta-analyses of PCIs in stable CAD. Eleven previously analyzed randomized controlled trials (RCTs) and 2 new RCTs were included.
Methods: Summary estimates of relative risk (RR) are obtained by applying fixed-effects and random-effects models. Statistical tests for assessing between-study heterogeneity and biases are performed. Cumulative estimates and results from influence analysis are reported.
Results: No difference between PCIs and MT alone was found for risk of mortality. There was a 12% increase in the RR of cardiac death or myocardial infarction (MI) associated with PCIs, as well as a 22% increase in the RR of nonfatal MI associated with PCIs. Cumulative analysis favored MT over PCIs as early as 1997, but recent study results have increased confidence in this finding. Because of heterogeneity between studies, no certain conclusions are drawn for the use of PCIs in preventing follow-up PCI or coronary artery bypass graft surgery.
Conclusion: Recent RCTs comparing PCIs with conservative MT in stable CAD increase confidence in previous findings that the use of PCIs does not offer marginal benefit over that of the use of MT alone for mortality risk, cardiac death or MI, and nonfatal MI.
(Am J Manag Care. 2008;14(8):521-528)
The weight of evidence has not shown percutaneous coronary interventions (PCIs) to have marginal benefit in the treatment of stable coronary artery disease over that of medical therapy alone; recent large trials increase our confidence in this finding.
Original favorable coverage decisions by Medicare were not based on whether PCIs provided marginal benefit over that of medical therapy; no randomized controlled trials were available for evaluation at that time.
Early diffusion of PCIs was likely influenced by Medicare‚Äôs decision to pay the same rate as open heart surgery and by early evaluation suggesting that PCIs could save money.It has been 31 years since the 1977 development of percutaneous transluminal angioplasty by Grüntzig1,2 for use in coronary arteries. The early diffusion of percutaneous coronary interventions (PCIs, previously percutaneous transluminal coronary angioplasty3) benefited from favorable Medicare pricing.4,5 Early Medicare payment for coronary angioplasty was based on coding to International Classification of Diseases, Ninth Revision, Clinical Modification procedure code 36.0 (“removal of a coronary obstruction”), which was reimbursed through the inpatient hospital diagnosis related group system as open heart surgery (diagnosis related group 108), despite the lower costs of PCIs.6 An analysis by Sawi7 cites 1982 costs for coronary artery bypass graft (CABG) surgery as $20,000 and costs for PCIs as $3000. Zweifel8 notes that, while new technology lowers the unit cost of production among industry in healthcare, it drives up the total cost. The use of PCIs in the United States grew from 1800 cumulative procedures from June 1979 through May 1981, to 32,300 procedures in 1983, to more than 200,000 procedures in 1988. From 2000 through 2005, more than 7 million coronary artery angioplasties, arthrectomies, or stent insertions have been performed in the United States.9-14 Using the mean national Medicare hospital payment rate for fiscal 2006 of $13,793, the procedures could represent hospital cost ranging from $54 billion to $97.4 billion. Estimates of national commercial insurance mean costs for PCIs, including physician fees, approximate $24,400 per case, giving a total cost estimate of $96 billion to $170 billion for the 6-year period from 2000 to 2005. (The $24,400 value is based on the BlueCross BlueShield of Tennessee treatment cost estimator. The treatment cost estimator is a collaborative project of Consortium Health Plans, Blue- Cross BlueShield Association, and Milliman USA. Based on the ClaimsQuest cost modeling system, the treatment cost estimates are episode treatment group–based. For the $96-billion to $170-billion estimate, there is considerable uncertainty surrounding the values. The source of the uncertainty is accounting for the number of procedures versus the number of patients; that is, a single patient could have both angioplasty and stent procedures performed within a single stay, for which there could be a single-case rate payment, a per diem rate–based payment, or a payment based on the fees for individual procedures and implanted stents.) Despite the high level of investment and experience with this technology, the marginal benefit of use in stable coronary artery disease (CAD) remains controversial, with study results often in conflict.15-18 Two previous meta-analyses19,20 of randomized controlled trials (RCTs) of PCIs versus medical therapy (MT) have shown no marginal benefit for PCIs over that of MT alone in death rates, myocardial infarction (MI), or need for subsequent revascularization in patients with nonacute CAD. Since the publication of the most recent meta-analysis,20 2 large RCTs have been added to the literature,21,22 more than doubling the total number of patients studied. The results of these new studies potentially have an important role in the development of policy regarding PCIs. The new studies have been portrayed in the lay press as “landmark,”23 implying that there is a historical change of course from the findings of previous studies. We update the meta-analysis of Katritsis and Ioannidis by including the 2 new studies 21,22 and by addressing the consistency, heterogeneity, and bias of previous findings.
We employed eligibility criteria, studied outcomes data, and meta-analytic methods of Katritsis and Ioannidis,20 whose principal outcome measure was relative risk (RR). In addition to summary outcome measures and heterogeneity test results, we present cumulative estimates and influence analysis, and test for the existence of biases.
Relative risks across studies were combined using the Mantel- Haenszel fixed-effects model and the DerSimonian and Laird random-effects model.24-27 In general, both methods give greater weight to studies with smaller variances. Weighting is shown graphically in forest plots, in which the size of the box is proportional to study weight (which is also shown in the plots) and in which the center of the box represents the size of the treatment effect estimated from that study (point estimate). The confidence interval (CI) for the treatment effect from each study is also shown. The pooled treatment effect is indicated by the middle of the diamond. An estimate less than 1.0 favors PCIs, while an estimate of 1.0 or greater favors MT alone. Influence analysis was performed by omitting 1 study at a time from the Mantel-Haenszel or DerSimonian and Laird model. A study is influential if, when omitted, the combined meta-analytic estimate differs significantly from that achieved when the study is included. Cumulative effects are shown in the cumulativeeffects plot, in which the first circle represents the results of the first study only, while subsequent circles represent the weighted results of all previous studies in addition to the current study. Between- study heterogeneity was tested using χ2-based Q statistic. Publication and other biases were investigated using funnel plots28 and formally using Egger’s regression asymmetry test.29-32 We account for any biases detected by applying the “trim and fill” method developed by Duval and Tweedie.33-35
Problems in the computation of RR and standard errors of RR arise when a study contains no events (the meta-analysis studies21,22,36-46 are summarized in the Table). For example, Sievers et al45 report no deaths in the intervention group, so that the estimated RR ratio is zero and its standard error cannot be estimated. To overcome this problem, we add 0.5 to each observed frequency (ie, to each cell of the 2 × 2 contingency table for the trial). When the treated and the control groups of the study contain zero events, presenting effect size estimates as RR ratios is meaningless. We have one such case (by Hambrecht et al40) for death due to all causes, and we discard it from that portion of the meta-analysis.
The primary end points are the following clinical outcomes: death (due to all causes), cardiac death or MI, nonfatal MI, and follow-up procedures that include PCIs and CABG surgery.
There were 509 deaths (250 in the PCI group and 259 in the MT group), 855 patients had cardiac death or MI (452 in the PCI group and 403 in the MT group), 520 patients had nonfatal MI (287 in the PCI group and 233 in the MT group), and 1628 patients underwent follow-up PCIs or CABG surgery (726 in the PCI group and 902 in the MT group). Pooled estimates for the 13 RCTs 21,22,36-46 are given in the Table.
No difference between PCIs and MT alone was found for risk of mortality. There was a 12% increase in the RR of cardiac death or MI associated with PCIs, as well as a 22% increase in the RR of nonfatal MI associated with PCIs.
One trial, the Arbeitsgemeinschaft Leitende Kardiologische Krankenhausärzte (ALKK),46 reports an RR and a 95% CI that are below 1.0 for mortality (RR, 0.36; 95% CI, 0.15-0.88), while another trial, the Medicine, Angioplasty, or Surgery Study (MASS-II),43 reports an RR and a 95% CI that are above 1.0 for mortality (RR, 3.63; 95% CI, 1.03-12.82). The remaining trials report RRs and CIs for mortality that include 1.0 (Figure 1). The pooled RR for mortality is 0.97 (95% CI, 0.82-1.14). For cardiac death or MI, the ALKK46 reports an RR and a 95% CI that terminate at 1.0 from below, while the MASS-II43 reports an RR and a 95% CI that terminate at 1.0 from above (Figure 2). The pooled RR for cardiac death or MI is 1.12 (95% CI, 0.99-1.27). No individual trial reports an RR and a CI that do not include 1.0 for nonfatal MI; the pooled RR is 1.22 (95% CI, 1.04-1.44) (Figure 3). Evidence from individual RR ratios is mixed for follow-up PCI or CABG surgery, with a pooled RR estimate of 1.00 (95% CI, 0.80-1.26) (Figure 4). Additional results are given in the online appendix (eAppendix; available at: www.ajmc.org).
This study adds to the current body of literature on PCIs by considering between-study heterogeneity, bias, consistency of outcomes among studies, inclusion of 2 recent large studies, and historical aspects of PCI acceptance by payers and healthcare providers. Consideration of these features affirms and increases confidence in the previous findings. We also find that contrary to perception, except for the first RCT,45 cumulative effects support the finding that PCIs offer no marginal benefit over MT for the 4 outcomes studied herein. Consideration of how PCIs became covered by Medicare and were accepted by healthcare providers lends credence to the conclusion that PCIs provide no marginal benefit over MT alone. Policy may be further informed by these findings because they may prompt additional service review by payers, employers, and consumers in addition to providers.
A meta-analysis by Katritsis and Ioannidis20 of 11 RCTs comparing PCIs with conservative treatment in patients with stable CAD was updated to include 2 recent large trials.21,22 The following 4 outcomes were considered: (1) For death, the combined estimate of RR is 0.97, implying no difference between PCIs and MT. (2 and 3) For cardiac death or MI and for nonfatal MI, the combined estimates are greater than 1.0 (1.12 and 1.22, respectively), favoring MT. (4) When the clinical outcome is follow-up PCI or CABG surgery, randomeffects estimation accounting for between-study heterogeneity demonstrates a wide 95% CI for the pooled RR, which prevents inferences about the superiority of PCIs versus MT. No evidence of between-study heterogeneity is found for death, cardiac death or MI, and nonfatal MI (Q statistics, P = .48, P = .38, and P = .86, respectively). Because of the small number of studies, the test of homogeneity has low power to detect excess variation (variation beyond sampling error) and cannot be interpreted as evidence of homogeneity. However, fixed-effects and random-effects estimates are remarkably close for these outcomes, implying the absence of between-study heterogeneity. The null hypothesis for unitary RR is rejected for nonfatal MI (P < .05).