Exploring the Treatment-Risk Paradox in Coronary Disease FREE

It is well documented that proved, efficacious therapies are often suboptimally applied in clinical practice.¹ For example, almost half of eligible patients with coronary artery disease (CAD) do not receive statins, and older individuals, women, and patients with substantial comorbidities are less likely to be prescribed evidence-based therapies for their cardiovascular diseases than younger and/or healthier individuals.^1- 9 It has recently been suggested that there is a “treatment-risk paradox” in cardiovascular pharmacotherapy, such that clinicians preferentially initiate treatment in low-risk individuals and are less likely to prescribe new therapies in those patients at highest risk for poor outcomes from the underlying condition.^8- 13

The treatment-risk paradox has been cited by some as an indicator of poor-quality care by practicing clinicians^8- 15; in the words of one editorialist, “it is the premise of matching risk to level of care that physicians fail to accept, heed, or understand.”^14(p382) However, as clinicians, we recognize that prescribing decisions and patient adherence are influenced by far more factors than those that are coded in databases constructed and maintained for billing claims. Thus, before attributing the treatment-risk paradox to a systematic prescribing bias by clinicians, we believe it is important to explore other potential explanations in order to properly inform the design of interventions to address the treatment-risk paradox.

To explore the potential causes of the treatment-risk paradox, we designed a prospective study to collect a variety of clinical and functional variables in a population-based cohort of individuals with CAD. In particular, we were interested in the potential role depressive symptoms, functional capacity, or both may play in generating the treatment-risk paradox for 3 reasons. First, both conditions are common in patients with CAD and influence prognosis.^16- 17 Second, they are not coded in administrative databases. Third, both conditions could conceivably influence a clinician's willingness to prescribe another medication (especially preventive medications that do not relieve symptoms) to an already overburdened patient and a patient's adherence to such a prescription.

We prospectively enrolled all adult patients diagnosed as having CAD (defined as >50% stenosis in at least 1 vessel) by angiography in any of the 3 cardiac catheterization centers in Alberta between February 1, 2004, and November 30, 2005. We excluded patients who died or underwent coronary artery bypass surgery during their index hospitalization. We conducted our study within the framework of the Alberta Provincial Project for Outcome Assessment in Coronary Heart Disease (APPROACH), a population-based registry capturing all cardiac catheterizations performed in Alberta (a province of approximately 3.1 million individuals).¹⁸ The APPROACH study and this substudy were approved by the University of Alberta Health Research Ethics Board.

Information on sociodemographic factors (sex, age, and postal code as a proxy for income), indication for the cardiac catheterization, CAD-specific variables, comorbid conditions, and coronary anatomical features (using Heartview software; Duke Medical Center, Durham, NC) were collected through APPROACH (a project in which cardiologists assign diagnoses; these are double checked and entered into the database by specially trained cardiac nurses). A full description of these data sets (including the variables collected and definitions used) has been published previously.^18- 21

All patients were asked to complete a package of mailed questionnaires 1 month after their index cardiac catheterization, which included those medications they were taking at the time of the survey, the Seattle Angina Questionnaire (a 19-item health-related quality-of-life instrument specific for CAD),²² the EuroQol 5D (a 5-item generic health-related quality-of-life instrument),²³ and the 10-item Center for Epidemiological Studies Depression Scale (CES-D).²⁴ Each of these scales has been validated in patients with CAD and has been responsive and reliable in this patient population.^25- 28 For this study, the exertional capacity scale of the Seattle Angina Questionnaire was used as a measure of functional status.

Based on previous studies,^21,29 we a priori defined patients as being at “low risk,” “medium risk,” or “high risk” based on their coronary anatomical features (using the Duke Coronary Index, which classifies patients based on the number, severity, and location of their stenosed coronary arteries and has been validated as a prognostic marker in patients with CAD).^19- 21 In an examination of all 69 816 APPROACH subjects enrolled between January 1, 1995, and December 31, 2004, 1-year mortality was 3% in those with low-risk Duke Coronary Index scores, 7% in those with medium-risk scores, and 11% in those with high-risk scores, and the mortality prediction model for these 69 816 subjects incorporating the Duke Coronary Index had a C statistic of 0.82 (Colleen Norris, PhD, written communication, April 19, 2006).

Baseline characteristics of the study patients and their medications were compared across risk strata by analysis of variance. We compared clinical characteristics in those patients taking and not taking statins using χ² tests for dichotomous variables and t tests for continuous variables.

We calculated the crude odds ratio (OR) for taking a statin in patients with high-risk Duke Coronary Index scores vs other patients and then explored factors that might modify the observed association between baseline risk and statin use rates by sequentially adjusting for progressively more clinical detail in a series of “partially adjusted” models (Figure). For example, in one model, we examined the OR for statin use in high-risk patients using multiple logistic regression, incorporating as covariates sociodemographic factors plus nonstatin prescriptions and clinical comorbidities (identified with the backward stepwise selection technique selecting all clinically important variables and other prespecified factors, with P<.25 on bivariate analyses and a prevalence of at least 1%, and accepting statistical significance at P<.05) (Table 1 provides a full list of covariates). To illustrate the impact of incorporating the influence of a patient's mental outlook and functional status on treatment decisions, we incorporated scores from the CES-D questionnaire and the exertional capacity scale of the Seattle Angina Questionnaire into 2 separate logistic regression models that included all of the covariates in the partially adjusted model previously described plus either the CES-D questionnaire or the exertional capacity scale of the Seattle Angina Questionnaire. Finally, in our fully adjusted model, we included all covariates in Table 1, except the EuroQol 5D scores (we did not include the EuroQol 5D scores in our multivariate models because they demonstrated substantial colinearity with the CES-D score and the exertional capacity scale score of the Seattle Angina Questionnaire). For all models, the OR for statin use in high-risk patients compared with other patients was the variable of interest and we tested for first-order interactions. Because the OR may exaggerate a risk association when the outcome of interest is common, we also calculated corrected risk ratios (RRs) using the method of Zhang and Yu.³⁰

In sensitivity analyses to examine the robustness of our findings, we explored the following: (1) the use gradient for statins in only those patients who would have fulfilled the eligibility criteria for the published statin trials, and adjusted for the covariates previously described; (2) the use gradient across baseline risk for angiotensin-converting enzyme (ACE) inhibitors and for aspirin (non–symptom-relieving medications); and (3) the use gradient for symptom-relieving medications (long-acting nitrates, calcium channel blockers, and β-blockers).

Of 5015 consecutive patients approached for this study, 716 either declined participation or did not return the questionnaires, 416 were excluded because of death or urgent coronary artery bypass surgery during the index hospitalization, and 12 were excluded because they did not have coronary disease on angiography. Thus, our study sample consisted of 3871 patients who had CAD proved on angiography, were discharged alive, and completed all 3 questionnaires assessing functional capacity, quality of life, and depressive symptoms. Compared with study participants, nonparticipants were older (mean age, 70 years), less likely to be male (68.3%), and more likely to have hypertension (73.6%), peripheral vascular disease (8.9%), cerebrovascular disease (11.3%), malignancy (5.7%), or heart failure (11.0%) (P<.001 for all).

The characteristics of our cohort (Table 2) closely mirror those for the full APPROACH cohort.²¹ Statins were being taken by 2436 (62.9%) of the 3871 patients 1 month after their cardiac catheterization, and high-risk patients were less likely to be taking statins than lower-risk patients (55.8% vs 63.5%; crude OR, 0.72 [95% confidence interval {CI}, 0.57-0.92]; RR, 0.88 [95% CI, 0.79-0.97]) (Table 3). Fasting lipid profiles did not differ significantly across risk strata (or indeed between statin users and nonusers) (Tables 1 and 2). Compared with those not taking statins, patients taking statins were younger and more likely to be male, had fewer comorbidities, were less likely to report depressive symptoms or functional limitations, and reported better health-related quality of life (Table 1). While patients with chronic kidney disease (adjusted OR, 0.69 [95% CI, 0.53-0.89]; RR, 0.73 [95% CI, 0.58-0.91]) or peripheral vascular disease (adjusted OR, 0.72 [95% CI, 0.49-0.98]; RR, 0.74 [95% CI, 0.51-0.98]) were less likely to be taking statins, patients with acute coronary syndromes (adjusted OR, 1.43 [95% CI, 1.19-1.70]; RR, 1.21 [95% CI, 1.10-1.31]) and patients prescribed other evidence-based medications (adjusted OR, 4.89 [95% CI, 4.07-5.88] for β-blocker users, 3.22 [95% CI, 2.69-3.85] for ACE inhibitor users, and 2.03 [95% CI, 1.70-2.42] for aspirin users; RR, 2.00 [95% CI, 1.91-2.10], 1.99 [95% CI, 1.83-2.14], and 1.46 [95% CI, 1.34-1.57], respectively) were more likely to be taking statins.

The treatment-risk paradox for statin prescriptions in high-risk patients was still present in our data after adjusting for sociodemographic factors (OR, 0.78 [95% CI, 0.61-0.99]; RR, 0.91 [95% CI, 0.81-0.996]), but it was progressively attenuated as we adjusted for additional baseline variables (Figure). In our fully adjusted model, high-risk patients were not less likely to be taking statins than lower-risk patients (OR, 1.07 [95% CI, 0.78-1.47] [P = .68]; RR, 1.02 [95% CI, 0.91-1.13]). Indeed, adjusting for only self-reported exertional capacity on the Seattle Angina Questionnaire and depressive symptoms on the CES-D completely removed the treatment-risk paradox (OR, 0.98 [95% CI, 0.75-1.28]; RR, 0.99 [95% CI, 0.89-1.09]) for statin use in high-risk patients.

Statin use exhibited a similar gradient across baseline risk in the subset of 3127 patients in our cohort who met the eligibility criteria for the major statin trials³¹: 56.5% in high-risk patients, 63.1% in medium-risk patients, and 66.8% in low-risk patients. The unadjusted OR for statin use in trial-eligible high-risk patients was 0.69 (95% CI, 0.52-0.90) and the OR after adjustment for sociodemographics was 0.70 (95% CI, 0.53-0.92), with an RR of 0.87 (95% CI, 0.76-0.97). However, this apparent treatment-risk paradox was also completely attenuated in the fully adjusted model (OR, 1.12 [95% CI, 0.76-1.64] [P = .57]; RR, 1.04 [95% CI, 0.90-1.16]).

Our findings with statins also extended to other secondary prevention therapies. High-risk patients were less likely to be taking ACE inhibitors (44.5% vs 55.6%; crude OR, 0.64 [95% CI, 0.51-0.81]; RR, 0.80 [95% CI, 0.70-0.91]) (Table 3), and although this treatment-risk paradox was still present after adjusting for sociodemographic factors (OR, 0.66 [95% CI, 0.52-0.84]; RR, 0.82 [95% CI, 0.71-0.92]), it was attenuated as additional clinical variables were included in the analyses (OR, 0.89 [95% CI, 0.62-1.26] [P = .50]; RR, 0.95 [95% CI, 0.79-1.10] in the fully adjusted model). Although aspirin prescribing did differ across baseline risk, it was not as pronounced as the gradient for statins or ACE inhibitors and, as a result, the unadjusted OR for aspirin use in high-risk patients was only 0.82 (95% CI, 0.65-1.04), and the RR was 0.92 (95% CI, 0.82-1.02); this apparent treatment-risk paradox was also completely absent in our fully adjusted model (OR, 1.01 [95% CI, 0.76-1.34]; RR, 1.00 [95% CI, 0.88-1.12]). Finally, high-risk patients were less likely to be taking all 3 agents than lower-risk patients (25.8% vs 32.3%; crude OR, 0.73 [95% CI, 0.56-0.95]; RR, 0.80 [95% CI, 0.65-0.97]), an association that did not persist in the fully adjusted model (OR, 0.98 [95% CI, 0.72-1.33] [P = .87]; RR, 0.99 [95% CI, 0.79-1.20]). On the other hand, there was no treatment-risk paradox in the use of antianginal medications for symptom relief: high-risk patients were more likely than low-risk patients to be taking long-acting nitrate prescriptions (OR, 1.67 [95% CI, 1.14-2.45]; RR, 1.60 [95% CI, 1.13-2.22]) and equally likely to be taking calcium channel blockers (OR, 1.22 [95% CI, 0.87-1.71]; RR, 1.19 [95% CI, 0.88-1.58]) or β-blockers (OR, 0.86 [95% CI, 0.68-1.10]; RR, 0.95 [95% CI, 0.87-1.03]) (Table 3).

In our prospective cohort study of 3871 patients with CAD, we found a negative correlation between baseline risk and use of statins, ACE inhibitors, and aspirin, which was similar in magnitude to previous reports^8- 13 of treatment-risk paradoxes in the literature. However, we have expanded on these earlier reports by demonstrating that the treatment-risk paradox was also present in that subset of patients in our cohort who would have been eligible for the relevant statin trials (thus refuting claims that any differential underuse of these medications in high-risk patients is attributable to the paucity of trial evidence in such patients, who are usually deemed high risk because of comorbidities that would have precluded their participation in the trials).

Similar to earlier reports, we found that the treatment-risk paradox was not appreciably altered by adjusting for sociodemographics, in either the entire cohort or the trial-eligible subset. However, in our study, we collected more than 200 variables on each patient and, thus, were able to explore the impact of far more clinical detail than is possible in administrative databases. In doing so, we found that the treatment-risk paradox for statins (and for ACE inhibitors and aspirin) in CAD was attributable to confounding from clinical and functional variables that the clinician faces during the patient-physician encounter, but that are not captured in administrative databases; in particular, the functional status and depressive symptom burden of patients seemed to be key drivers of the treatment-risk paradox for preventive therapies. This raises 2 possible explanations for the treatment-risk paradox: either clinicians preferentially avoid treating patients with depression or poor functional status or patients who are depressed, have poor functional status, or both are less likely to fill and/or adhere to prescribed therapies. The fact that we did not find a treatment-risk paradox for symptom-relieving antianginal medications (long-acting nitrates, β-blockers, or calcium channel blockers) argues for the latter explanation.

Because cardiovascular preventive therapies proved efficacious in trials also confer benefits in subgroups excluded from the trials,^{6- 7,32- 33} the underuse of statins, ACE inhibitors, and aspirin in patients with poor functional capacity or depression is an important care gap that needs to be recognized and addressed. As has been pointed out by others, the numbers needed to treat for preventive therapies are smaller when they are used by patients at higher baseline risk, and even modest increases in prescribing rates for preventive therapies in high-risk patients would confer substantial reductions in morbidity and mortality from CAD.⁸ Thus, educational interventions to address the treatment-risk paradox need to emphasize to physicians and patients that the benefits of preventive CAD therapies are applicable even to patients with poor functional capacity or depression. Indeed, our results would suggest that patients with CAD who have depression or poor functional capacity should be targeted for more education at discharge and closer follow-up to ensure they are taking optimal secondary preventive therapy.

Although our study reports on a large, well-categorized, consecutively enrolled, and population-based cohort of patients with CAD, there are some limitations. First, we used patient self-report 1 month after their cardiac catheterization to define medication use; however, an earlier study³⁴ in Alberta documented a high degree of agreement between patient self-report for statin and/or ACE inhibitor use and prescribing data from dispensing pharmacies (κ, 0.78-0.93). Regardless, we are unable to say definitively whether the underuse of these secondary preventive therapies in our high-risk patients occurred because (1) their physicians did not prescribe the drugs or (2) the patients did not fill the prescription or adhere to the medication after filling the prescription. Second, some may argue that our study was unable to find a systematic bias underlying the treatment-risk paradox because the patients in our study exhibited a narrower spectrum of risk than previous studies of this phenomenon (our high-risk patients were only 3-4 times more likely to die than our low-risk patients, while the high-risk patients in the study by Ko et al¹¹ were nearly 10 times more likely to die than their low-risk peers); however, the unadjusted OR for statin use in our study (0.72) was nearly identical to the 0.75 reported by Ko and colleagues. Third, while some may question our use of the Duke Coronary Index to stratify patients into risk categories, we think this is, in fact, a strength of our study. While previous examinations of the treatment-risk paradox have relied on multivariate prediction models to stratify patients into high, medium, or low risk, multivariate models are rarely used by clinicians in their day-to-day practice. Indeed, clinicians tend to make treatment decisions based on single pieces of information, and coronary anatomical features are a key driver of CAD therapeutic decisions made by cardiologists; in this context, all of our cohort patients were seen and treated by the cardiologists who performed their angiography, and we examined medication use 1 month after the angiograms were obtained. Thus, using the Duke Coronary Index to classify our cohort patients by baseline risk in fact mimics practice to a greater degree than using multivariate risk prediction models.

In conclusion, we demonstrated a treatment-risk paradox for statins, ACE inhibitors, and aspirin in patients with CAD, similar to previous reports, but extended this earlier work by showing that (1) these treatment-risk paradoxes are seen even in the subset of patients with CAD who meet trial eligibility criteria and (2) such paradoxes do not exist for antianginal medications (refuting claims that physicians are less willing to prescribe any medications in high-risk patients). Our multivariate analyses suggest that these treatment-risk paradoxes for preventive therapies are attributable to baseline imbalances in functional status and depressive symptoms across risk strata, which administrative databases cannot capture. Thus, our study has shed light on the cause of the treatment-risk paradox. The main message arising from our study is that future efforts (in the reporting of randomized trials and the educating of physicians and their patients) should be directed toward confirming and clarifying that the benefits of non–symptom-relieving (but prognosis-modifying) evidence-based therapies in cardiology do extend even to those patients with poor functional capacity, depression, or both. Good medical practice should result in a direct correlation between risk and treatment such that those patients most likely to benefit from therapy are also most likely to receive that therapy (ie, be prescribed it, and be adherent to it), irrespective of functional status.

Correspondence: Finlay A. McAlister, MSc, MD, 2E3.24 Walter MacKenzie Center, 8440 112th St, Edmonton, Alberta, Canada T6G 2R7 (Finlay.McAlister@ualberta.ca).

Accepted for Publication: December 8, 2006.

Author Contributions:Study concept and design: McAlister, Graham, and Tsuyuki. Acquisition of data: McAlister, Oreopoulos, Norris, Tsuyuki, Knudtson, and Ghali. Analysis and interpretation of data: McAlister, Oreopoulos, Graham, Tsuyuki, and Ghali. Drafting of the manuscript: McAlister. Critical revision of the manuscript for important intellectual content: McAlister, Oreopoulos, Norris, Graham, Tsuyuki, Knudtson, and Ghali. Statistical analysis: Norris and Ghali. Obtained funding: Knudtson. Administrative, technical, and material support: Tsuyuki, Knudtson, and Ghali.

Financial Disclosure: None reported.

Funding/Support: This study was supported by career salary awards from the Alberta Heritage Foundation for Medical Research (Drs McAlister, Norris, and Ghali); New Investigator Awards from the Canadian Institutes of Health Research (Drs McAlister and Norris); a Canadian Institutes of Health Research Strategic Training Fellowship in Tomorrow's Research Cardiovascular Health Professionals (Ms Oreopoulos); and a Canada Research Chair (Dr Ghali). Drs McAlister and Tsuyuki hold the University of Alberta/Merck Frosst/Aventis Chair in Patient Health Management. The APPROACH initiative was initially funded in 1995 by a grant from the Weston Foundation; the ongoing operation of the APPROACH initiative is supported by Merck Frosst Canada Inc, Guidant Corporation, Boston Scientific Ltd, Hoffman–La Roche Ltd, and Johnson & Johnson Inc–Cordis.

Role of the Sponsor: The funding bodies had no role in data extraction and analyses, in the writing of the manuscript, or in the decision to submit the manuscript for publication.

Acknowledgment: We thank Danielle Southern for her assistance in creating the figure; the Calgary Regional Health Authority and the Capital Health Authority for supporting online data entry by cardiac catheterization laboratory personnel; the physicians, nurses, and technicians in the cardiac catheterization laboratories at the Foothills Medical Centre, Calgary, Alberta, and the Royal Alexandra Hospital and University of Alberta Hospital, Edmonton, for their support in data collection for APPROACH.