Article Text

Download PDFPDF

Original research article
Facility-level association of preoperative stress testing and postoperative adverse cardiac events
  1. Javier A Valle1,2,
  2. Laura Graham3,
  3. Thejasvi Thiruvoipati4,
  4. Gary Grunwald1,2,
  5. Ehrin J Armstrong1,2,
  6. Thomas M Maddox5,
  7. Mary T Hawn6,
  8. Steven M Bradley7
  1. 1 Department of Cardiology, Veterans Affairs Eastern Colorado Health Care System, Denver, Colorado, USA
  2. 2 Division of Cardiovascular Medicine, University of Colorado School of Medicine, Aurora, Colorado, USA
  3. 3 Birmingham Veterans Affairs Medical Center, Birmingham, Alabama, USA
  4. 4 Division of Cardiology, University of Michigan Medical School, Ann Arbor, Michigan, USA
  5. 5 Department of Cardiovascular Medicine, Washington University School of Medicine, St Louis, Missouri, USA
  6. 6 Department of Surgery, Stanford University School of Medicine, Stanford, California, USA
  7. 7 Minneapolis Heart Institute, Minneapolis, Minnesota, USA
  1. Correspondence to Dr Javier A Valle, Veterans Affairs Eastern Colorado Health Care System, University of Colorado School of Medicine, Division of Cardiology, Denver, CO 80220, USA; javier.valle{at}Ucdenver.edu

Abstract

Background Despite limited indications, preoperative stress testing is often used prior to non-cardiac surgery. Patient-level analyses of stress testing and outcomes are limited by case mix and selection bias. Therefore, we sought to describe facility-level rates of preoperative stress testing for non-cardiac surgery, and to determine the association between facility-level preoperative stress testing and postoperative major adverse cardiac events (MACE).

Methods We identified patients undergoing non-cardiac surgery within 2 years of percutaneous coronary intervention in the Veterans Affairs (VA) Health Care System, from 2004 to 2011, facility-level rates of preoperative stress testing and postoperative MACE (death, myocardial infarction (MI) or revascularisation within 30 days). We determined risk-standardised facility-level rates of stress testing and postoperative MACE, and the relationship between facility-level preoperative stress testing and postoperative MACE.

Results Among 29 937 patients undergoing non-cardiac surgery at 131 VA facilities, the median facility rate of preoperative stress testing was 13.2% (IQR 9.7%–15.9%; range 6.0%–21.5%), and 30-day postoperative MACE was 4.0% (IQR 2.4%–5.4%). After risk standardisation, the median facility-level rate of stress testing was 12.7% (IQR 8.4%–17.4%) and postoperative MACE was 3.8% (IQR 2.3%–5.6%). There was no correlation between risk-standardised stress testing and composite MACE at the facility level (r=0.022, p=0.81), or with individual outcomes of death, MI or revascularisation.

Conclusions In a national cohort of veterans undergoing non-cardiac surgery, we observed substantial variation in facility-level rates of preoperative stress testing. Facilities with higher rates of preoperative stress testing were not associated with better postoperative outcomes. These findings suggest an opportunity to reduce variation in preoperative stress testing without sacrificing patient outcomes

  • perioperative management
  • stress testing
  • non-cardiac surgery
  • quality and outcomes

Statistics from Altmetric.com

Introduction 

Adverse cardiac events are a significant source of perioperative morbidity and mortality for patients with known coronary artery disease (CAD) undergoing non-cardiac surgery.1 2 Cardiac stress testing is often used to estimate the risk of these events,3–5 however national and societal statements suggest limited indications for preoperative stress testing.6 Moreover, prior studies have raised concerns over excess cost and surgical delays associated with routine preoperative stress testing.7–9

Current guidelines recommend consideration of preoperative stress testing in patients with both poor functional capacity and elevated risk, based on clinical and surgical factors.6 Prior studies suggest the presence of significant variability in the use of preoperative stress testing,5 10 11 but evaluations of outcomes associated with this variability at the patient level are likely confounded by selection bias and case mix. Facility level analyses may mitigate these limitations, yet there are limited data on facility-level practice patterns of preoperative stress testing and their relationship with postoperative outcomes. If more frequent stress testing is associated with improved facility-level outcomes, this would suggest an opportunity to address preprocedural processes at high-use facilities associated with better postoperative outcomes. Alternatively, if facilities that perform more stress testing do not achieve better outcomes, this would suggest an opportunity for more parsimonious use of preoperative stress testing.

Accordingly, we sought to describe the facility-level variation in cardiac stress testing among patients undergoing non-cardiac surgery in the Veterans Affairs (VA) Health Care System and its association with postoperative outcomes. Hypothesising that preoperative stress testing and adverse events would be most prevalent among patients with known CAD, we focused on a cohort of patients with prior percutaneous coronary intervention (PCI). We assessed the facility-level association between risk-standardised rates of preoperative stress testing and major adverse cardiac events (MACE) in the postoperative period. Our findings may inform opportunities to optimise the use of preoperative stress testing.

Methods

Study population

The derivation of the initial study cohort has been previously described.12–14 Briefly, we identified all patients who underwent non-cardiac surgery within 24 months of PCI in the VA Health Care System between October 2000 and September 2011 using the VA Clinical Assessment, Reporting and Tracking (CART) Program, the Corporate Data Warehouse (CDW) and the VA National Patient Care Database (NPCD). The VA CART programme is a national clinical quality programme for all VA cardiac catheterisation laboratories, using a software application embedded into the electronic health record to collect patient and procedural data for all VA cardiac catheterisation procedures.15 16 The VA CDW and NPCD collect clinical data from inpatient and ambulatory encounters,17 18 and allowed identification of patients undergoing non-cardiac surgery within 24 months of PCI, as defined by data using Current Procedure Terminology (CPT) codes and Center for Medicare and Medicaid Services (CMS) data. Patients were excluded if they underwent emergent non-cardiac surgery, as they were unable to undergo stress testing prior to their operation, or if they underwent cardiac surgery. Patients were excluded if they underwent non-operative procedures like tracheostomies, endoscopies, cast placement or injections. Finally, patients were excluded if they underwent a subsequent non-cardiac surgical procedure within 30 days of the index operation or if the index surgery occurred during the same hospital stay as a previous surgery. A total of 10 377 patients (25.7%) were excluded.

Study variables

We identified cardiac stress tests in the 3 months prior to non-cardiac surgery. Stress tests were identified using CPT and International Classification of Diseases, Ninth Revision (ICD-9) diagnosis codes from the VA National Surgery Office and CDW data (online supplementary table 1). Preoperative cardiac stress tests were defined as those tests occurring within the 3 months prior to the index operation in order to selectively identify testing used to inform perioperative risk. Finally, we also identified any cancellation of scheduled surgical procedures following stress testing, using VA CDW data.

Supplementary file 1

In the 30-day postoperative period, we identified all MACE defined as all-cause mortality, myocardial infarction (MI) and coronary revascularisation (online supplementary table 1). ICD-9 codes for MACE events were obtained from VA NPCD data, and patient deaths were identified using the VA Vital Status file.18–20

Covariates were chosen to adjust for perioperative risk, based on prior literature and clinical judgement. Patient-level covariates included demographic and clinical data for each patient, derived from the VA NPCD and VA Decision Support System. These covariates included age, race and sex, history of cerebrovascular disease (CVD), MI, hypertension, diabetes mellitus, the presence of a serum creatinine ≥2 g/dL within the past year, and revised Cardiac Risk Index (rCRI)11 and its component factors. The rCRI score, a validated metric used to estimate the risk of perioperative cardiac events, was calculated using ICD-9 codes for its component factors of congestive heart failure, CVD, MI and diabetes mellitus, CPT codes associated with high-risk surgery (intraperitoneal, intrathoracic or suprainguinal vascular) and laboratory studies with one or more values of serum creatinine greater than 2 mg/dL (176.8 µmol/L) in the year prior to surgery. PCI and operative features were similarly collected as covariates for adjustment, including indication for PCI (acute coronary syndrome or stable angina), stent type (bare metal vs drug eluting), time between PCI to surgery, surgical site (digestive, eye/ear, genitourinary, integumentary, musculoskeletal, nervous, vascular, respiratory and other) and inpatient versus outpatient status.

Statistical analyses

Patient-level characteristics were compared by the presence or absence of preoperative stress testing using Χ2 tests for categorical variables, Student’s t-tests for continuous variables and Wilcoxon signed-rank tests for ordinal variables. We determined the rate of preoperative stress testing for each hospital, and also compared facility-level patient characteristics across hospital quartiles of preoperative stress test use, using Χ2 tests for categorical variables and analysis of variance for continuous variables. A Cochran-Armitage test for trend was used to assess potential trends in the occurrence of each outcome across hospital quartiles of preoperative stress test use.

We then calculated the facility-level risk-standardised rates of stress testing, MACE, and the individual components of MACE (mortality, MI and revascularisation) using the method currently endorsed by the CMS for hospital profiling.21 22 Hierarchical logistic regression was used to model hospital-level rates of preoperative stress testing, postoperative MACE, death, MI and revascularisation. Backwards stepwise selection was used to identify the most parsimonious model. Then, we calculated the ratio of predicted events to expected events for each hospital, where predicted events were computed as the sum of the predicted probabilities of events including that hospital’s specific random effect, and expected events were computed as the sum of the predicted probabilities excluding the hospital effect; that is, for an ‘average’ other hospital within the cohort. Next, we multiplied each hospital’s predicted/expected ratio by the overall study event rate, to obtain risk-standardised event rates. We then plotted each hospital’s risk-standardised MACE and mortality rates against the risk-standardised hospital-level rates of preoperative stress testing, calculating a Spearman correlation coefficient to assess the relationship between stress testing and clinical outcome.

As patients could undergo stress testing prior to both PCI and non-cardiac surgery (ie, stress testing and non-cardiac surgery with interval PCI), we performed a sensitivity analysis excluding all patients who underwent their index PCI between stress test and surgery. Using the same methods as described above, we determined the risk-standardised facility rates of stress testing and MACE within this restricted population, and determined a Spearman correlation coefficient (r) to assess the relationship. Finally, to determine whether rates of surgical cancellation could possibly account for the primary findings, we also conducted additional exploratory analyses to identify the proportion of patients who were scheduled but never received surgery after a stress test.

All statistical analyses were conducted using R and SAS V.9.2, using only two-sided p values and a level of ≤0.05 considered statistically significant.

Results

Study cohort

The study cohort consisted of 29 937 patients who underwent non-cardiac surgery within 24 months of PCI across 131 total facilities (median 179 patients per facility; IQR 87–341). In this cohort, 4103 patients (13.7%) underwent cardiac stress testing in the 3 months prior to surgery. Median time from stress test to surgery was 50 days (IQR 24–77). Cancellation of scheduled surgery occurred in 3.0% of patients undergoing preoperative stress testing.

Patient-level characteristics associated with preoperative stress testing

Stress testing occurred more frequently prior to surgery in patients who were numerically younger, a statistically significant but not clinically meaningful difference (median age 65 vs 66, p<0.001). There was no significant difference in race or gender. Patients undergoing stress testing had higher rates of CVD, hypertension, chest pain or angina within the 6 months prior to PCI, and an acute coronary syndrome 6 months prior to PCI. Rates of renal dysfunction and diabetes were not significantly different between patients undergoing preoperative stress testing and those not. Preoperative stress testing occurred more frequently prior to elective inpatient surgeries than outpatient surgeries, and prior to surgeries defined as high risk. Stress tests were performed more often in patients with rCRI scores ≥2 (56.4% vs 52.6%, p<0.001). Preoperative stress testing occurred more frequently prior to surgeries occurring within shorter time intervals from PCI (median 255 days vs 365 days, p<0.001). There was no statistically significant difference in stress testing prior to surgeries in patients with prior bare metal coronary stents versus drug-eluting stents (table 1).

Table 1

Patient characteristics associated with preoperative stress testing

Facility-level use of preoperative stress testing

At the facility level, the rate of preoperative stress testing varied from 0.0% to 25.6%, with a median rate of 14.7% (IQR 12.3%–16.7%). By hospital quartiles of preoperative stress testing use, the median rate of stress testing was 7.0% (IQR 6.0%–9.1%) in the lowest quartile (Q1), 12.0% (IQR 10.8%–12.3%) in the second lowest quartile (Q2), 14.5% (IQR 14.1%–15.3%) in the third lowest quartile (Q3) and 18.2% (IQR 16.6%–21.5%) in the highest quartile (Q4). Facilities with higher rates of preoperative stress testing had patient cohorts with overall higher rCRI scores (rCRI 2; Q1: 48.7%, Q2: 52.9%, Q3: 54.8%, Q4: 54.6%, p<0.001), performed a higher proportion of high-risk surgeries (Q1: 9.3%, Q2: 10.6%, Q3: 11.0%, Q4: 11.5%, p=0.002) and had the lowest proportion of outpatient surgeries (Q1: 61.7%, Q2: 56.0%, Q3: 56.0%, Q4: 52.7%, p<0.001) (table 2).

Table 2

Patient characteristics by quartile of facility stress testing rates

Facility-level postoperative outcomes

Overall, there were 1409 (4.7%) MACE events, composed of 445 (1.5%) deaths, 895 (3.0%) MIs and 312 (1.1%) revascularisation events. In unadjusted analyses, there were 194 MACE (3.9% of patients within quartile) in the facilities within the lowest quartile of preoperative stress testing, 383 (4.8%) in the second lowest quartile, 493 (4.8%) in the third lowest quartile and 447 (5.2%) in the highest quartile (p=0.01 for trend). Unadjusted differences in individual outcomes of mortality, MI and repeated revascularisation did not reach statistical significance across quartiles of stress testing (p≥0.05 for trend; table 3).

Table 3

Thirty-day postoperative outcomes by quartile of facility stress testing rates

Risk-standardised stress testing and patient outcomes

The risk-standardised facility median rate of stress testing was 12.7% (IQR 8.5%–17.4%; online supplementary table 2). The risk-standardised facility median rate of MACE was 3.8% (IQR 2.3%–5.6%). Risk-standardised facility rates of the individual MACE components of MI, repeated revascularisation and death are seen in online supplementary table 2 (distribution plots for risk-standardised stress testing and outcomes in online supplementary figure 1A–D). There was no correlation between risk-standardised facility-level rates of stress testing and risk-standardised rates of 30-day MACE (correlation coefficient r=0.022, p=0.81; figure 1A), death (r=0.062, p=0.49; figure 1B), MI (r=−0.026, p=0.77; figure 1C) or repeat revascularisation (r=−0.038, p=0.66; figure 1D). In a sensitivity analysis that excluded patients who underwent stress testing prior to both PCI and surgery (n=1074 patients), there remained no relationship between risk-standardised rates of facility-level stress testing and 30-day MACE (r=−0.017, p=0.85) or individual rates of death, MI, or revascularisation (table 4).

Table 4

Spearman correlation coefficients for facility risk-standardised rates of stress testing versus outcomes, excluding patients with stress tests prior to PCI and surgery

Figure 1

Spearman correlation between risk-standardised stress testing and 30-day postoperative outcomes. (A) Risk-standardised stress testing versus major adverse cardiac events (MACE). (B) Risk-standardised stress testing versus death. (C) Risk-standardised stress testing versus myocardial infarction (MI). (D) Risk-standardised stress testing versus repeat revascularisation.

Discussion

In this large national cohort of patients undergoing non-cardiac surgery following PCI, we sought to describe facility-level use of preoperative stress testing and its association with postoperative MACE. Risk-standardised rates of preoperative stress testing varied across facilities, with a greater than twofold difference in rates of use across centres. Furthermore, we found that facilities performing more preoperative stress tests did not achieve lower rates of postoperative MACE, and that facilities with lower rates of preoperative stress testing did not demonstrate higher rates of adverse events. These findings suggest an opportunity to refine selection for preoperative stress testing and reduce variation while maintaining patient outcomes.

Current guidelines for the perioperative management of cardiac patients undergoing non-cardiac surgery do not endorse stress testing outside of patients at elevated risk for perioperative MACE.6 These recommendations were made on the basis of prior studies demonstrating reduction in postoperative cardiac events when selecting for high-risk patients.3 23–26 In lower risk populations, preoperative stress testing has offered little additional information above clinical evaluation25 and has even been associated with increased harm.3 Despite these recommendations, a recent analysis of Medicare patients suggests that patient characteristics may have limited influence on the use of stress testing, finding variability in the use of preoperative stress testing independent of patient or operative risk, and significant influence from clinician and practice factors.5 Our facility-level analysis of preoperative stress testing adds to these prior data, again demonstrating wide variability in practice patterns for the use of preoperative stress testing in non-cardiac surgery, independent of patient risk. It is possible that variation in preoperative evaluation, across medical, surgical and anaesthesia specialties may contribute to the demonstrated variation. Alternatively, the use of dedicated ‘preoperative clinics’ with facility-specific algorithms for testing may similarly contribute to the development of variable practice patterns. This is an important area for further study to assess the main factors driving variability in the use of preoperative stress testing.

Prior studies have also raised concerns over the costs of routine preoperative stress testing,7 as well as the potential to result in delays to surgery.8 9 Without a demonstrable benefit in postoperative outcomes, routine preoperative stress testing represents unnecessary and low value care, delaying therapy and increasing medical cost. The present study demonstrates a lack of association between higher rates of preoperative stress testing and postoperative outcomes. While stress testing is beneficial in preoperative risk stratification for some patients, these findings suggest an opportunity to reduce variability in preoperative stress testing without diminishing surgical outcomes.

Strengths of our study include the use of a large, national, complete cohort of patients with prior PCI undergoing non-cardiac surgery with detailed procedural and outcome data. Performing this study among a group of patients who have undergone PCI ensures prior interaction with cardiovascular care, thereby suggesting that the behaviours and patterns demonstrated may be more representative of management of patients with known cardiovascular disease as opposed to a generalised cohort. Additionally, prior observational analyses of the use of preoperative stress testing at the patient level have been limited in the ability to appropriately assess the relationship between stress testing and outcomes due to the significant selection bias of which patients undergo preoperative testing and subsequent surgery. Performing this evaluation at the facility level allows for evaluation of aggregated practice patterns and outcomes, minimising the effects of selection bias, and the use of risk standardisation to account for case mix. Finally, our cohort was limited to those patients with known coronary disease and prior PCI, enriching the population for the likelihood of both undergoing stress testing and suffering postoperative events.

However, the study should be interpreted in light of several limitations. First, this is an observational analysis with the possibility of residual confounding beyond our robust model. However, we included a large number of covariates and used the advanced statistical methods currently endorsed by CMS to risk-standardise for case mix across facilities.21 Further, with reliance on administrative data, there is the possibility of missing data for stress testing and outcomes. We used all available data sets within the VA to minimise this possibility. Second, we were unable to assess appropriateness of stress testing due to lack of information on patient functional status. Given this inability to directly assess appropriateness, we instead assessed the relationship between overall variation in use at the facility level and outcomes while accounting for case mix at the facility level. Third, the facility-level relationships between testing and outcomes may not reflect individual patient-level relationships between testing and outcomes. However, any such relationships are likely to be heavily influenced by patient selection. By performing our analysis at the facility level and performing risk standardisation, the unmeasured effects of patient selection should be minimised, but any conclusions from our analysis should be interpreted as facility-level associations and should not be extended to the individual patient to avoid ecologic fallacy. Fourth, we were unable to ascertain the results of stress tests. It is possible that a relationship exists between rates of positive stress tests and postoperative outcomes, however in an exploratory analysis of surgery cancellation rates, we found that only 3.0% of patients scheduled for an operation had their surgery cancelled after completion of a stress test. Even under the unlikely assumption that each cancellation can be attributed solely to the stress testing results, the overall low rates of surgical cancellation after stress testing suggest that the results of these tests are unlikely to impact the decision to pursue an operation, and argue strongly against selection bias significantly influencing our results. It is also possible that stress test results would alter management of these patients, outside of surgical cancellation, through systematic intensification and optimisation of medical therapy for those with positive results. This would likely improve the outcomes for patients undergoing stress testing, and bias our results away from the null. As we found no correlation between stress testing and outcome, this is unlikely to play a large role in our findings. Additionally, our analysed data set includes patients through the year 2011. It is possible that the results from our analysis do not reflect more recent changes in preoperative testing and perioperative management. Finally, our findings in this cohort may not be generalisable to populations not well represented in the VA.

Conclusions

In patients undergoing non-cardiac surgery at VA hospitals, there is significant facility-level variation in the use of preoperative stress testing. In addition, facilities with higher rates of stress testing were not associated with lower rates of postoperative MACE. These findings suggest an opportunity to reduce variation in preoperative stress testing practice without diminishing postoperative outcomes.

Key messages

What is already known on this subject?

  • Despite limited indications, preoperative stress testing is often used prior to non-cardiac surgery, but its efficacy in mitigating perioperative events remains unclear, and prior patient-level analyses of the relationship between stress testing and outcomes are limited by case mix and selection bias.

  • Preoperative stress testing can be costly and can introduce delays in surgical care.

What might this study add?

  • This observational facility level analysis of preoperative stress testing in a national cohort of patients with coronary artery disease undergoing noncardiac surgery found substantial variation in the use of preoperative stress testing, and did not find a relationship between increased stress testing and reduction in major adverse events.

How might this impact on clinical practice?

  • These suggest an opportunity to reduce variation in preoperative stress testing, without sacrificing patient outcomes.

References

Footnotes

  • Contributors All authors listed have contributed as warranted to merit authorship for this manuscript—from study design, data collection, analysis, interpretation and to manuscript review.

  • Funding Works further supported by NIH/NCATS Colorado CTSA (Grant No UL1 TR001082).

  • Disclaimer Contents are the authors’ sole responsibility and do not necessarily represent official NIH views.

  • Competing interests JAV was supported by an NIH T32 training grant (HL0782) at the University of Colorado, Aurora, CO, while work was completed.

  • Patient consent Not required.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.