FormalPara Key Points for Decision Makers

There are limited economic evaluations comparing the cost-effectiveness of coronary artery bypass grafting (CABG) and percutaneous coronary intervention (PCI) for patients with complex multi-vessel coronary artery disease, using data from routine clinical practice.

Economic evaluations using data from randomised controlled trials are constrained by reduced generalisability to a wider population, including high-risk cases, e.g. the elderly and those with chronic kidney disease.

This economic evaluation used real-world data from prospective clinical and administrative registries, and a novel matching method appropriate for small-size cohorts (propensity score bin bootstrapping).

At a mean follow-up of 2.7 years, CABG compared with PCI was found costly yet more effective in high-risk subgroups, including those with diabetes, a history of myocardial infarction, left main coronary artery disease, and heart failure. These results support current clinical guidelines available for coronary revascularisation in Australia and New Zealand.

A personalised, multidisciplinary approach to treatment of patients may enhance cost containment, as well as improving clinical outcomes following coronary revascularisation.

1 Introduction

Over the last 3 decades, treatment of coronary artery disease (CAD) through coronary revascularisation has undergone rapid advancements [1]. Improvements to coronary artery bypass grafting (CABG) are associated with a threefold reduction of in-hospital mortality [2]. The introduction of drug-eluting stents (DES) for percutaneous coronary intervention (PCI) reduced the risk of target revascularisation by 30–55% [3, 4]. CABG is traditionally recommended for patients with complex multi-vessel CAD (MVCAD) and those with diabetes mellitus, and PCI is recommended for those with single-vessel CAD and acute coronary syndrome (ACS) [5, 6]. Significant improvements in clinical safety and efficacy of PCI with DES triggered several randomised controlled trials (RCTs) comparing the clinical and cost-effectiveness of CABG versus PCI among non-traditional patients for PCI, including those with MVCAD [6,7,8,9,10,11,12].

RCTs are the gold-standard for comparative efficacy studies; however, some issues are apparent in their execution and application in the real world [13]. RCTs are typically expensive and time consuming, and final results from trials may lag several years behind, delaying any potential economic modelling using data from the in-trial period. Due to their controlled nature, the generalisability of data from RCTs of surgical interventions to routine clinical practice may also be limited. For instance, in routine clinical practice, patients may have more co-morbidities and may receive less intensive treatment and monitoring than in RCTs. These drawbacks with RCTs flow through to economic evaluations conducted within RCTs.

The use of observational data from clinical registries and administrative databases may overcome the apparent drawbacks of RCTs, although these datasets carry their own limitations [14]. For instance, administrative datasets may be limited by the aggregation of service use, censored cost data, limited timely updates, missing resource utilisation data, and the limited ability to pool data from different centres due to inconsistencies in coding and costing approaches [15, 16]. Observational datasets from clinical registries and administrative databases, however, provide quick, inexpensive, and relevant data for comparisons of treatment strategies in contemporary clinical practice [14]. The drawbacks of clinical registries, including treatment bias, finite patient follow-up and missing data, may be accounted for through use of propensity-score (PS) matching and record linkage approaches [14].

To date, several comparative effectiveness evaluations for CABG and PCI have been carried out using data from different clinical registries across various jurisdictions [5, 17,18,19,20,21]. However, only a limited number of studies utilised real-world data to evaluate the costs and cost-effectiveness [22, 23]. These evaluations were primarily carried out by collaborating researchers of large clinical and quality registries in the USA, including the Society for Thoracic Surgery (STS), and the ‘CathPCI’ registry, which were linked recently to Medicare claims data [24,25,26]. Although contemporary clinical practice is largely similar across developed Western economies, results from the USA may not be generalisable to other jurisdictions because of differences in healthcare systems [27]. Currently, there is limited evidence in relation to the cost-effectiveness of surgery versus PCI in the Australian setting. The current evaluation investigates the real-world cost-effectiveness of CABG versus PCI for MVCAD, using data from two prominent clinical registries in Australia and a local administrative dataset.

2 Methods

2.1 Study Population

2.1.1 Data Sources

Data from two large Australian prospective multi-centre cardiac databases were used, including the Australian and New Zealand Society of Cardiac and Thoracic Surgeons (ANZSCTS) registry for CABG and the Melbourne Interventional Group (MIG) registry for PCI. The ANZSCTS and MIG registries have been described in detail previously [28]. The registries undergo salient audit programmes on a regular basis to ensure data quality and integrity. Data accuracy of 97% has been achieved for baseline data, which is comparable to the accuracy of other large cardiac registries [28].

The study population comprised all patients who underwent either CABG or PCI at a single centre, The Alfred Hospital, a major metropolitan quaternary teaching hospital in Melbourne, Australia, from 1 July 2009 to 31 December 2013. The data from ANZSCTS and MIG registries are routinely linked to the Australian National Death Index (NDI). The records were further linked to an administrative clinical costing dataset maintained by the Alfred Hospital Clinical Performance Unit, for this analysis. The data were censored at 31 December 2014. The study was approved by the Alfred Hospital’s Human Research Ethics Committee (Project number 142/15).

2.1.2 Inclusion and Exclusion Criteria

All patients who had MVCAD or left main CAD (LMCAD) and underwent isolated CABG or PCI using DES and/or bare metal stents (BMS) were included in this evaluation. Patients who underwent other procedures concomitant with CABG surgery (such as valve surgery), had PCI without stents, or had incomplete or missing administrative clinical costing records were excluded from the study population.

2.1.3 Subgroup Analyses

Subgroup analyses were performed for all patients with diabetes (irrespective of treatment modality), chronic kidney disease (CKD), history of myocardial infarction (MI) (defined as all patients who had a record of MI at any point in time in their medical history), LMCAD, and congestive heart failure (reffered to as ‘heart failure’) and for the elderly (age > 75 years). Further subgroups analyses were performed for device categories, including PCI with BMS and PCI using second-generation DES and third-generation DES. Different generations of DES were defined based on design similarities (see the electronic supplementary material, Online Resource 1).

2.2 Clinical Endpoints

Two primary clinical endpoints were used, including major adverse cardiac and cerebrovascular events (MACCE) and all-cause mortality. Only same-hospital readmissions were considered, obtained through the linkage of ANZSCTS and MIG records to administrative clinical costing data. Clinical endpoints were defined using the Australian-refined diagnostic related group (AR-DRG) code included in each linked subsequent hospital episode record (see Online Resource 1). Two additional clinical endpoints were considered, including other cardiovascular events (OCVE) and non-cardiovascular events (NCVE). MACCE was defined as a combination of hospital readmissions for repeat CABG, repeat PCI, stroke, cerebrovascular events, acute MI, heart failure, angina, chest pain, and arrhythmia. Subsequent hospitalisations that were not MACCE, but were cardiovascular related, were defined as OCVE (see Online Resource 1). All subsequent hospitalisations that were not MACCE or OCVE were defined as NCVE (Online Resource 1).

2.3 Costs and Resource Utilisation

Medical costs of hospitalisations were obtained from the administrative clinical costing dataset maintained by the Alfred Hospital Clinical Performance Unit (see Online Resource 1).

2.4 Statistical Analysis

PS matching analysis was performed to select the CABG and PCI groups for the economic evaluation. Two approaches were considered for PS matching, including the 1:1 nearest neighbour approach [29, 30] and the propensity score bin bootstrap (PSBB) technique [29, 31]. The PSBB technique is an emerging PS matching approach that was used by Zhang et al. [22]. It relates to sub-classification (stratification), which involves the formation of groups of individuals who are similar, as defined by quintiles (bins) of the PS distribution, and bootstrapping of results within these bins [6]. Results are aggregated within bins and averaged across the groups to obtain a final estimate on average for the cohort.

The rationale for the PSBB approach was twofold: (1) as an alternative matching method to test the sensitivity of results, and (2) to conduct subgroup analyses, as sample sizes of certain subgroups were too small to undertake matching through the 1:1 nearest neighbour matching technique.

For base-case economic evaluation, the 1:1 nearest neighbour approach was performed. A PS was generated for each patient, using the nearest neighbour 1:1 matching algorithm (with no replacement) with a calliper width of 0.25 times the standard deviation (SD) of the PS distribution. Multiple logistic regression was utilised for both the generation of the PS using the type of treatment (CABG or PCI) as the dependent variable and the identification of matching variables based on significant predictors of long-term all-cause mortality. Baseline variables selected for matching included age, diabetes, diabetes treated with insulin, dialysis user, heart failure, cerebrovascular disease, cardiogenic shock, intra-operative balloon pump, admission status, number of diseased vessels, LMCAD and the AusSCORE II (see Online Resource 1 for further details).

Statistical analyses were conducted using R Version 3.3.2 (Windows) and Microsoft Excel 2013 (Windows). A program for the PSBB technique in R was developed for the purpose of this cost-effectiveness analysis.

2.5 Economic Evaluation

2.5.1 Type of Analysis

The real-world cost-effectiveness of CABG compared with PCI was estimated using a purely database approach and cohorts defined by 1:1 PS matching. Costs and benefits were not projected beyond the observation period—similar to a within-trial analysis.

2.5.2 Treatment

To be consistent with prior economic evaluations, CABG was defined as the ‘treatment’ under consideration, despite being standard of care for patients with MVCAD and/or LMCAD.

2.5.3 Comparator

The comparator was defined as PCI with BMS or DES. All generations of DES deployed in routine clinical care were included. Different generations of DES were defined on the basis of design similarities (see Online Resource 1).

2.5.4 Perspective

The evaluation took the perspective of the Australian public hospital payer.

2.5.5 Costs

Only medical costs related to hospital episodes were considered. Out-of-pocket costs and other costs related to loss of productivity of patients or carers, travel time, and lost income were not included. Medical costs associated with pharmaceuticals, and rehabilitation beyond hospital episode were not included due to absence of data. The base year for costs was 2013. All costs were measured in Australian dollars (AU$) (AU$1 = US$1.37, January 06, 2017).

To adjust for censoring, costs were weighted using the Lin 1997 method [32], a time interval-based approach for adjusting mean costs according to the survival probability of patients at various time periods for which they were completely observed.

2.5.6 Economic Measure

The economic measure under consideration was the incremental cost-effectiveness ratio (ICER). Three primary ICERs were calculated, including the incremental costs per deaths avoided, per life years gained (LYG), and per MACCE avoided. For the matched cohort, the incremental costs per OCVE avoided and per NCVE avoided were also evaluated. Due to the absence of quality-of-life data from this population, an incremental cost per quality-adjusted life year (QALY) could not be generated. To compare ICERs, willingness-to-pay thresholds of $45,000, $50,000, $60,000, and $75,000 were used, based on current considerations by the Australian Medical Services Advisory Committee [33].

2.5.7 Time Horizon

A maximum period of observation ranging from 6 months (0.5 years) to 60 months (5 years) was estimated for the study population. Outcomes were censored at 1 July 2014.

2.5.8 Inflation Adjustment

All costs were adjusted for inflation using the Australian Total Health Price Index [34]

2.5.9 Discounting

Net overall costs and benefits beyond the first year were discounted by 5%, the standard rate as determined by the Australian Medicare Services Advisory Committee [35].

2.5.10 Sensitivity Analyses

The structural uncertainty of base-case results (primary ICERs) was evaluated using the PSBB technique. As only same-hospital readmissions were considered in this analysis, the sensitivity of ICERs to different MACCE rates was evaluated through two-way deterministic sensitivity analysis (assumed + 5%, 10–95% greater than observed rate). Probabilistic sensitivity analysis was carried out by way of PSBB for this evaluation, which utilised bootstrapping of results from propensity-matched cohorts.

3 Results

3.1 Study Population

3.1.1 Baseline Characteristics

At baseline, there were 2000 patients who underwent either isolated CABG (n = 1022) or PCI (n = 978) and met the inclusion criteria. Prior to matching, the two cohorts were similar (p > 0.05) in several characteristics, but were significantly different (p < 0.05) in relation to multi-vessel disease status, urgency of procedure, and history of MI (see the electronic supplementary material, Online Resource 1). The size of CABG and PCI cohorts were reduced to 1094 patients in total (547 in each group) following PS matching. Post-matching, the majority of baseline characteristics were similar between the CABG and PCI cohorts (Online Resource 2).

3.1.2 Clinical and Procedural Features

Among patients who underwent isolated CABG, 98.4% (n = 1006) received an arterial graft, while the remainder received vein grafts only. In the PCI group, DES, BMS, or both were deployed in 59% (n = 579), 39% (n = 379), and 2% (n = 20) of procedures, respectively. On average, 1.1 ± 0.37 stents were used per PCI procedure. Sixty per cent of PCI procedures (n = 351) used a third-generation DES, while 37% (n = 213) used a second-generation DES. A first-generation DES was used only in 3% of PCI procedures. Following PS matching, similar distributions of clinical and procedural characteristics were apparent in the matched PCI and CABG cohorts.

3.1.3 Follow-Up

All patients were followed up for a period between 6 months and 5 years. The average follow-up period was 2.7 years, which was not significantly different (p > 0.05) between the CABG and PCI groups (2.69 years vs. 2.72 years).

3.2 Effectiveness

The clinical outcomes of the PS-matched cohorts and the discounted benefits are presented in Table 1. The probability of survival for each event was illustrated in Kaplan-Meier curves (see Online Resource 2).

Table 1 Clinical outcomes in propensity-matched cohorts at mean 32 months (2.7 years) post-intervention
Full size table

3.2.1 MACCE

At the end of the observation period, those treated with PCI had a significantly higher number of MACCE, than CABG (287 vs. 54). The mean number of MACCE avoided from treatment with CABG compared with PCI was 0.43 (95% confidence interval [CI] 0.33–0.52) per patient treated at mean follow-up. With the 5% discount rate applied, the benefit was reduced to 0.38 (0.29–0.46).

The clinical outcomes related to individual components of MACCE are presented in Online Resource 2.

3.2.2 OCVE

There were fewer OCVE following CABG than PCI at the end of the observation period (122 vs. 173). However, at mean follow-up, the mean number of OCVE prevented by CABG was not significant.

3.2.3 NCVE

A greater number of NCVE were observed following CABG compared with PCI at the end of the observation period (784 vs. 632). However, the mean difference of NCVE between CABG and PCI groups did not reach significance.

3.2.4 All-Cause Mortality

The total number of deaths (any) at the end of the observation period was greater in the CABG group than PCI group (33 vs. 7). At discharge and 30-day follow-up, this figure was five versus seven. The mean number of deaths avoided by CABG was − 0.04 (95% CI − 0.06 to − 0.02).

3.2.5 Life Years

The total life years lived by both CABG and PCI groups were similar at the end of the period (1473 vs. 1487); as such, the mean difference (i.e. LYG) did not reach significance.

3.3 Costs

The costs within PS-matched groups related to the initial procedure, readmissions, and total costs, before and after adjustments for censoring and discounting, are reported in Table 2.

Table 2 Medical costs related to the initial procedure and readmissions in propensity-matched cohorts at mean 32 months (2.7 years) post-intervention
Full size table

3.3.1 Initial Costs

The total initial procedural cost of CABG was around 3–4 times greater than PCI. The mean incremental initial cost of CABG was $26,728 (95% CI 25,029–30,466).

3.3.2 Readmissions Costs

The total unadjusted cost of MACCE-related readmissions was approximately 13-fold greater for the PCI group than the CABG group at the end of the observation period. After adjusting for censoring through the Lin 1997 method (Table 2 and Online Resource 2) and discounting, the mean difference was − $3164 (95% CI − 3967 to − 2360). The costs related to individual components of MACCE are available in Online Resource 2.

Similar to MACCE, the total unadjusted costs for OCVE- and NCVE-related readmissions were greater for the PCI than the CABG group, by one- to twofold. The mean difference in costs for OCVE and NCVE readmissions did not reach significance.

The total unadjusted costs related to all readmissions were two- to threefold greater for the PCI group than the CABG group. The adjusted incremental (CABG-PCI) costs of all readmissions was − $6140 (95% CI − 8945 to − 3335).

3.3.3 Total Costs

Compared with PCI, the total unadjusted cost, inclusive of initial and all readmissions costs, was 1–2 times greater for the CABG group. The adjusted incremental total cost of CABG compared with PCI at mean follow-up was $20,997 (95% CI 16,897–25,098).

3.4 Cost-Effectiveness

The ICERs corresponding to primary and secondary outcomes at mean follow-up are presented in Table 3. CABG was associated with greater total costs, but also greater total benefits in terms of MACCE, with an ICER of $55,255 per MACCE avoided.

Table 3 The ICERs corresponding to primary and secondary outcomes at mean 32 months (2.7 years) post-intervention
Full size table

3.5 Sensitivity Analyses

The results from the PSBB approach are reported in Table 4 and Table 5. For the overall PS-matched cohort, an ICER of $56,153 per MACCE avoided was reported (Table 4), similar to the corresponding base-case ICER (Table 3). For the overall PS-matched cohort, probabilities of cost-effectiveness in terms of MACCE avoided of 30.6, 35.1, 43.3, and 56.5% were reported at thresholds of $45,000, $50,000, $60,000, and $75,000, respectively (Table 5). Two-way deterministic analyses for the evaluation of sensitivity of MACCE rates demonstrated an improvement in ICER for CABG with uniform increase of observed MACCE rates for both groups (Online Resource 2).

Table 4 Results from matching using the PSBB approach: incremental costs and effectiveness and ICERs at mean 32 months (2.7 years) as observed for patient subgroups from the original cohort
Full size table
Table 5 Probabilities of cost-effectiveness for CABG: incremental costs per MACCE avoided at mean 32 months (2.7 years) for a range of willing-to-pay thresholds (data for CEACs)
Full size table

3.6 Subgroup Analyses

The cost-effectiveness results for small groups of high-risk MVCAD patients including diabetics, the elderly (age > 75 years), and those with prior MI, CKD, LMCAD, and heart failure, as well as device-specific subgroups are reported in Tables 4 and 5. There were not enough patients treated with first-generation DES (3% of original PCI cohort) to generate cost-effectiveness results.

3.6.1 High-Risk Patients

In terms of MACCE avoided, ICERs for CABG were lowest (< $50,000) among patients with diabetes, ACS, LMCAD and heart failure (Table 4), for whom probabilities of cost-effectiveness of CABG were high (> 50%) (Table 5). CABG resulted in significant incremental all-cause mortality when compared with PCI across all high-risk patient subgroups, deeming CABG less effective and more costly, in terms of ICER per death avoided at mean follow-up. Elderly patients and those with CKD were among those that incurred the highest incremental costs (Table 4).

3.6.2 Device-Specific Subgroups

When compared against BMS, DES (overall), second-generation DES, and third-generation DES, CABG was associated with greater incremental costs and lower mean MACCE at mean follow-up (Table 4 and Fig. 1). The ICER for CABG varied across comparisons against BMS (ICER $25,815/MACCE avoided), all DES ($56,861), second-generation DES ($42,925), and third-generation of DES ($88,535). With each comparative technological advancement of PCI (from BMS to DES), the ICERs for CABG increased. In terms of mean all-cause mortality, CABG resulted in significantly higher deaths (any) at mean follow-up, when compared with all PCI groups, except second-generation DES, where the difference did not reach significance (Table 4).

Fig. 1
figure 1

Scatter plots generated from evaluations of cost-effectiveness in terms of incremental cost per major adverse cardiac and cerebrovascular events (MACCE) avoided using the propensity score bin bootstrap approach. AUD$ Australian dollars, BMS bare-metal stents, CABG coronary artery bypass grafting, DES drug-eluting stents, LC less costly, LE less effective, MC more costly, ME more effective

Full size image

4 Discussion

4.1 Summary of Findings

In contemporary clinical practice, CABG compared with PCI was associated with significantly lower MACCE rates, but significantly higher all-cause mortality, and significant incremental costs for patients with MVCAD at mean follow-up of 2.7 years. These results were expected given the high procedural risk associated with CABG, where early mortality following surgery is typically high when compared with PCI. In the short-term, clinical benefit of CABG is also overshadowed by its high initial procedures cost. An ICER (for CABG) of $55,255 per MACCE avoided was reported for the overall cohort, from the perspective of the Australian public hospital payer.

Subgroup analyses revealed that ICERs were lower among MVCAD patients with diabetes (ICER 42,819 per MACCE avoided), ACS ($30,431), LMCAD ($38,864) and heart failure ($36,966). The incremental costs of CABG compared with PCI were the highest among elderly patients (age > 75 years) and those with CKD. Device-specific subgroup analyses revealed that the ICER for CABG (in terms of MACCE avoided) increased with each comparative technological advancement of PCI, including BMS (ICER $25,815), DES ($56,861), second-generation DES ($42,925), and third-generation of DES ($88,535).

4.2 How Current Findings Compare to Prior Evidence

The current paper presents the first economic evaluation of CABG and PCI for MVCAD patients in Australia, in the setting of a large, quaternary hospital that typically treats complex cases. The evaluation, however, only assessed costs and benefits over a short-to-medium- term time horizon (mean 2.7 years). Our recent systematic review of cost-effectiveness revealed that in the long term, CABG was more cost-effective than PCI, whereas PCI may be more favourable in the short term for patients with MVCAD [27].

The current evaluation revealed that the total cost of CABG was significantly higher than that of PCI at mean follow-up of 2.7 years. By comparison, CABG incurred significantly lower total readmissions costs than PCI (− $6140) at mean follow-up. The total incremental cost of CABG ($20,997) was largely driven by its mean procedural cost, which was more expensive than PCI by $26,728 initially, suggesting that the clinical benefit of CABG is overshadowed by its high initial procedural cost in the short and medium term. This is a notion that is consistent across published economic literature for CABG [27].

Only four prior economic evaluations comparing CABG and PCI reported ICERs in terms of MACCE (‘events’) avoided by CABG compared with PCI [27, 36,37,38,39]. In an economic model that was developed from the perspective of the Canadian healthcare payer, traditional CABG was found less effective and more costly in terms of MACCE avoided, when compared with BMS and DES at 1-year follow-up [36]. In a 3-year analysis of costs and benefits yielded from the Arterial Revascularization Therapy Study, CABG was found more costly, but also more effective, when compared with BMS in terms of event-free survival, driven significantly by a lower repeat revascularisation rate [38]. This evaluation was carried out from the Netherlands healthcare payer perspective [38]. The same group also reported an ICER of $35,809 per MACCE avoided for patients with unstable angina at 1-year follow-up [39]. The latter findings are consistent with our results for the ACS subgroup, which includes patients with unstable angina. In an economic evaluation conducted from the patient perspective in Armenia, CABG compared with DES was found more effective and less costly in terms of MACCE avoided at 5-year follow-up [37].

More recent economic evaluations from large RCTs comparing CABG versus first-generation DES suggest that CABG is more cost-effective for high-risk MVCAD patients, within the trial periods and over a life-time time-horizon, from the US healthcare payer perspective [10, 11]. Direct comparison of our results with these findings was not possible as these studies reported effectiveness in terms of QALYs gained, using the EuroQOL (EQ-5D) health status instrument [10, 11].

An economic evaluation from the ASCERT study, which used observational data from the STS, CathPCI, Medicare Claims registries and sourced health-related utility estimates from a local longitudinal study, also reported costs per QALY gained from the US healthcare payer perspective [22, 26, 40]. Nevertheless, ICERs reported by the ASCERT study were higher than those reported using data from RCTs [27]. Perhaps, this may be as a result of inclusion of more complex cases treated in routine care, which may have had greater resource utilisation needs, and therefore greater costs, particularly following CABG rather than PCI.

4.3 Strengths and Limitations

Our economic evaluation is strengthened by the use of real-world data from high-risk patient subgroups with MVCAD who underwent revascularisation at a contemporary clinical practice. Our results provide further evidence that ICERs obtained from real-world analyses are typically higher than those from RCTs, which exclude complex cases.

This analysis is also strengthened by the use of standard and emerging PS matching techniques for adjusting baseline differences between CABG and PCI groups. Only two prior economic evaluations of CABG and PCI utilised such approaches [22, 41]. Reynolds et al. [41] used a simple 2:1 matching approach based on several patient risk factors. The ASCERT study [22] used the PSBB approach, which was introduced by Faries et al. [31]. The current evaluation also used the PSBB approach in addition to the nearest neighbour 1:1 PS matching technique, which was used to derive base-case results [30]. Several advantages were apparent with the use of the PSBB approach, including the retention of all observations found in the original CABG and PCI cohorts, and overcoming issues of truncated cohorts resulting from standard PS matching techniques, thus enabling valid subgroup analyses [30]. The PSBB approach was also used to estimate willingness-to-pay probabilities through bootstrapping results. The PSBB approach may therefore be an appropriate tool to assess structural sensitivity of results yielded from standard PS matching methods [30].

Our evaluation also yielded cost-effectiveness estimates for high-risk MVCAD patients, including those with CKD, HF, and ACS, who have been typically excluded from RCTs comparing CABG and PCI [27]. In particular, the CKD group remains an under-examined group in relation to costs and benefits of different revascularisation strategies, despite evidence of an incremental risk and resource burden from renal failure [42,43,44]. Further evidence is therefore warranted for this patient subset from similar economic evaluations, but over a longer time horizon.

The current analysis was impacted by limitations inherent to the data sources used. We could not estimate the QALY gained from CABG compared with PCI because of the absence of local evidence of health-related quality of life following revascularisation. Patient follow-up beyond 30 days and 12 months (MIG only) were not available through ANZSCTS and MIG registries, thus enforcing the use of record linkage to collect readmissions data, albeit only up to 5 years and from a single centre. Multiple administrative datasets from different centres could not be used for this evaluation, as technical difficulties were anticipated merging such datasets in addition to obtaining permission from different sites. Difficulties with inter-hospital merging of admissions records in Australia are further amplified by issues related to data linkage capacity, its dependencies, timeliness of data supply, and different linkage maturities across states, territories, and public-private institutions, and privacy legislation [45]. Issues related to the use of censored medical costs for this evaluation was accounted for by the Lin 1997 method [32]. To overcome issues pertaining to analyses of only same-hospital readmissions, a sensitivity analysis was carried out, which suggested an improvement of ICER (for CABG) with uniform increase of MACCE rates.

The perspective of the analysis was limited to that of the Australian public hospital payer, as data on post-intervention outpatient physician visits and pharmaceutical use were not available. Prior evidence suggests that post-intervention costs for PCI in high-risk patients are primarily driven by hospitalisations rather than medication use [43]. Moreover, treatment guidelines for long-term medical management of cardiovascular disease tend to be similar despite initial revascularisation approach. The Alfred Hospital, which is a large, quaternary teaching hospital in Australia, is representative of other similar quaternary teaching hospitals in developed countries that provide modern healthcare facilities for patients, but with similar challenges inherent to public hospitals (e.g. limited funding and scarce resources). Future economic analyses may consider broader perspectives such as that of the society, including patient perspectives.

Finally, although the use of PS matching analysis led to some removal of selection bias, we cannot eliminate unobservable bias, including the impact of unknown confounders that were not part of the ANZSCTS or MIG data collection. For instance, circumstances beyond clinician control, e.g. resource constraints at hospitals, urgency of procedures, anticipated adherence to medication by the patient, etc., may force clinical decisions outside of guideline-driven revascularisation with CABG or PCI with BMS or DES.

5 Conclusions

In terms of MACCE avoided, ICERs for CABG were lowest among high-risk MVCAD patients, including those with diabetes, history of MI, LMCAD, and heart failure. These results support a personalised, multidisciplinary approach to treatment of high-risk MVCAD patients that may enhance cost containment and clinical outcomes following intervention. Despite the single-centre nature of this economic evaluation, policy makers, clinicians, and researchers may benefit from the reported methods and, more importantly, the findings of this evaluation, which are in line with clinical guidelines recommended for coronary revascularisation of MVCAD patients in Australia and New Zealand.