Guideline-discordant care in acute myocardial infarction: predictors and outcomes

Potential subjects were consecutive patients with a principal discharge diagnosis of AMI (International classification of diseases, 9th revision, clinical modification9 [ICD-9-CM] code 414) who had been admitted to two community hospitals (a 300-bed public hospital and a 120-bed private hospital) in Ipswich, south-east Queensland, during the period 1 July 1997 to 31 December 2000. Neither institution had on-site cardiologist services or invasive coronary facilities; consultant staff comprised emergency and general physicians.

Subjects were excluded if they had been transferred to a study hospital from another site during the acute episode, or failed to satisfy two of three diagnostic criteria for AMI:

• acute chest pain lasting for 20 minutes or more;

• a rise in serial cardiac enzyme levels to more than twice the upper normal reference range; or

• diagnostic electrocardiographic (ECG) changes of infarction.

If patients had recurrent hospitalisations for AMI during the study period, the first admission only was studied.

During the study period, clinicians within both hospitals were made aware of locally developed, evidence-based clinical practice guidelines relating to inhospital AMI care. These guidelines were adapted from those issued in 1996 by the American College of Cardiology (ACC) and the American Heart Association (AHA).8 Guideline dissemination was coupled with a six-monthly performance feedback, based on clinical audits comparing routine care with guideline recommendations.10

Recommendations from the 1996 ACC/AHA guidelines8 were used in determining patient eligibility criteria for specific treatments. We chose six key treatment recommendations associated with strong evidence from randomised controlled trials of effects on short-term mortality and recurrent coronary events (Box 1). One recommendation was "negative" in that it stipulated avoidance of a drug class (calcium antagonists) in patients with left ventricular systolic dysfunction.

For each patient, a recommendation concordance score was calculated as shown in Box 1.2 Applying pre-specified thresholds, patients with a concordance score of ≥ 0.75 were assigned to the guideline-concordant care group and those with a score of < 0.75 to the guideline-discordant care group.

Of 790 patient records screened, 607 (77%) were found to satisfy the selection criteria. At least one treatment recommendation was applicable in 602 of the 607 patients (99%); the mean number of applicable recommendations per patient was 2.7 (Box 2A). The treatment recommendation which applied most frequently was administration of aspirin, while the least frequent was avoidance of calcium antagonists (Box 2B). The proportion of patients for whom an applicable treatment recommendation was not followed (discordance frequency) was low for thrombolysis administration to eligible patients (1%) and higher for timely administration of thrombolysis (36%), prescribing of ACE inhibitors (33%) and avoidance of calcium antagonists (31%) (Box 2B).

Mortality rates were inversely related to concordance scores grouped as quartiles (correlation coefficient [r] = – 0.969; P = 0.03) and were highest in those subjects with lowest concordance scores (score, < 0.25; 23%), intermediate in near-concordant-care patients (score, > 0.5– < 0.75; 9%), and lowest in those receiving concordant care (score, ≥ 0.75; 5%) (Box 3). A similar but non-significant trend was noted for reinfarction.

Of the 602 patients to whom recommendations did apply, 411 (68%) were assigned to the concordant-care group, and 191 (32%) to the discordant-care group. Patient characteristics of the two groups are compared in Box 4. Those which positively predicted discordant care at presentation were age > 65 years, silent infarction, anterior infarction, history of heart failure, chronic atrial fibrillation, tachycardia (heart rate, ≥ 100 beats/min), and previous AMI. Negative predictors included current smoking status, ST-segment-elevation infarction, and previous coronary revascularisation. After fitting regression models, age ≥ 65 years (OR, 1.7; 95% CI, 1.1–2.6; P = 0.02), anterior infarction (OR, 2.2; 95% CI, 1.4–3.4; P < 0.001), and history of heart failure (OR, 4.1; 95% CI, 2.2–7.6; P < 0.001) remained as independent positive predictors of discordant care.

Older age predicted discordant care, as also noted in other studies assessing use of thrombolysis,5,14 β-blockers4 and ACE inhibitors15 in eligible elderly patients. Pre-existing heart failure, which is prevalent in older patients (many of whom are female) and in those with previous AMI, and which signifies significant myocardial damage, may engender more conservative management.16 Equivocal ECG findings or the absence of chest pain as a presenting symptom may temper a more aggressive approach due to diagnostic uncertainty.17 Chronic atrial fibrillation and tachycardia at presentation often co-exist in older patients with advanced coronary disease.18 An explanation for the association of anterior infarction with discordant care remains elusive.

In this cohort, only 1% of patients eligible to receive thrombolysis did not receive this treatment, which contrasts with other studies showing a much lower rate of use of thrombolysis, particularly in elderly subjects, women, and patients with silent infarction.5,19 However, the high discordance frequency for its timely administration may account for a significant fraction of preventable mortality.20

The low discordance frequency of 10% for aspirin may reflect its ease of administration and low toxicity. Higher discordance frequencies of 18% and 33% noted with β-blockers and ACE inhibitors, respectively, mirror the underuse of these drugs in unselected populations reported by others.21,22 Such discordance may reflect clinician uncertainty as to clear indications, as well as concern about possible drug-related adverse events immediately after the acute event.22 Such concerns appear to be more pronounced in older patients. In contrast, recent research confirms that patients with comorbidities traditionally regarded as contraindications to specific treatments (such as obstructive lung disease in the case of β-blockers) exhibit higher death rates than those without such comorbidities,23 and gain greater reductions in absolute risk of death if, in fact, they receive these treatments.23

The discordance seen for avoidance of calcium antagonists invites scrutiny given the increased mortality associated with inappropriate use of these drugs in patients with impaired left ventricular systolic function.24 If left ventricular function is preserved, these drugs can prevent reinfarction in patients with non-ST-segment-elevation infarction,25 and are frequently used in patients with poorly controlled atrial fibrillation and postinfarction angina, which in our study accounted for 60% of all instances of calcium antagonist administration in patients with left ventricular dysfunction. While trials published since 1996 show new-generation dihydropyridine agents such as amlodipine and felodipine do not worsen prognosis in patients with left ventricular dysfunction,26,27 this is not the case with first-generation agents such as verapamil and diltiazem,24 which were used in 25% of patients (and in three of the five who died) receiving calcium antagonists. This emphasises the importance of formulating unambiguous recommendations for high-risk patient groups.

More deaths occurred among patients receiving calcium antagonists (5 of 30, 17%) and among eligible patients who did not receive aspirin (8 of 56, 14%) or ACE inhibitors (13 of 78, 17%) than in those who did not receive timely lysis (4 of 60, 7%) or β-blockers (7 of 75, 9%) (Box 2B). Of the estimated 10% absolute-risk reduction in 30-day mortality after AMI achieved over the past 20 years, a recent analysis concluded that aspirin use accounted for 34% of the decrease, followed by thrombolysis (17%), β-blockers (7%), and ACE inhibitors (3%).28 In our cohort, all eligible patients received thrombolysis (although delayed in many), and most eligible patients received aspirin (90%), β-blockers (82%) and ACE inhibitors (67%). Nevertheless, the omission of one or more of these therapies appears to significantly increase the risk of early death.

The uncontrolled study design and risk-adjustment methods may not adequately account for unknown or unmeasured confounding factors responsible for the observed differences in outcome between the two patient-care groups. However, for the primary outcome measure of early death, a recent analysis of 2153 patients with confirmed AMI (in whom 250 demographic, clinical and treatment variables were measured) showed three admission characteristics alone (age, heart rate and systolic blood pressure) predicted 30-day mortality with a high level of accuracy (area under the receiver operating characteristics curve of 0.78).13 These same factors, as well as others, were included in our risk-adjustment models.

Specific treatment recommendations (eg, for β-blockers, ACE inhibitors and specific calcium antagonists) are, on current evidence, restrictive in terms of patient eligibility, but were considered the standard of care during the study period. β-Blockers have been shown to reduce mortality in patients with heart failure,29 and the use of ACE inhibitors could now be justified in all patients with AMI, not just those with left ventricular dysfunction.30 We aimed to define those patients in whom the benefit–risk ratio was highest, based on evidence available in 1996.

We did not attempt to ascertain all legitimate clinical reasons as to why guideline recommendations were not followed in seemingly eligible patients, and which may themselves predict a poorer prognosis. However, we adjusted outcome results at the individual patient level for age and various comorbidities likely to be associated with justifiable clinician disinclination to prescribe selected therapies. This contrasts with other studies comparing concordant with discordant care which have not subjected their outcome results to any form of risk adjustment,2 or have reported outcomes in patients simply stratified according to baseline risk (low, medium and high).3

The method for deriving concordance scores, and the choice of 0.75 as the discordant–concordant cut-point, have not been validated as accurate predictors of suboptimal care. Further, the equal weighting of scores (1 or 0) for each recommendation implies equivalent treatment effects on each outcome, which, as discussed above, is not the case.28 However, the method has face validity, has been used by others,2 and serves as a starting point for further refinement.