Treatment strategy and percentage of patients taking lipid-lowering agents 30 days and 6 months after hospital discharge.
Aronow HD, Novaro GM, Lauer MS, Brennan DM, Lincoff AM, Topol EJ, Kereiakes DJ, Nissen SE. In-Hospital Initiation of Lipid-Lowering Therapy After Coronary Intervention as a Predictor of Long-term UtilizationA Propensity Analysis. Arch Intern Med. 2003;163(21):2576-2582. doi:10.1001/archinte.163.21.2576
Despite multiple randomized trials demonstrating their efficacy for the secondary prevention of coronary disease, lipid-lowering agents remain underused. Few studies have examined the relationship between predischarge initiation of lipid-lowering therapy and long-term use.
Using data from patients at 69 centers from the United States and Canada enrolled in the Evaluation in PTCA to Improve Long-term Outcome With Abciximab GP IIb/IIIa Blockade (EPILOG) trial, we performed a retrospective propensity-analyzed cohort study. Patients underwent percutaneous coronary intervention for stable or recently unstable coronary disease and were older than 21 years, were not taking lipid-lowering therapy at the time of admission, and survived to hospital discharge; 175 were discharged taking lipid-lowering therapy and 1951 were not.
After 6 months, 77% of patients who started taking lipid-lowering agents before hospital discharge continued taking therapy, compared with only 25% of those discharged without these agents (relative risk, 3.17; 95% confidence interval, 2.88-3.41; P<.001). After restricting the analysis to propensity-matched patients (n = 477) and adjusting for other potential confounders, initiation of a lipid-lowering agent during hospitalization was the strongest independent predictor of use at 6 months (relative risk, 2.50; 95% confidence interval, 2.29-2.65; P<.001).
Inpatient initiation of lipid-lowering therapy is a strong and independent positive predictor of subsequent use, with patients who start taking lipid-lowering therapy before hospital discharge nearly 3 times as likely to be taking these agents 6 months later. Inpatient initiation of lipid-lowering therapy appears to be an effective strategy for bridging the gap between current medical knowledge and practice.
DESPITE AN abundance of randomized trial data demonstrating the clinical benefits of lipid-lowering agents in patients with stable coronary disease,1- 4 and preliminary data suggesting an early benefit in unstable coronary syndromes,5- 7 these agents remain underused. Low rates of use of lipid-lowering agents are evident in large clinical trials,8 in prospective observational studies,9 at academic centers,10 in the community,11 in the United States,12,13 and abroad.14- 16 Given the unequivocal reductions in cardiovascular morbidity and mortality afforded by these agents, interventions that improve their use are urgently needed.
Although lipid-lowering therapy can be initiated in the inpatient or outpatient setting, the former environment may provide a more effective milieu for risk factor modification.17 In addition, it is plausible that if a cardiologist fails to prescribe lipid-lowering therapy to an eligible patient at hospital discharge, the primary care physician may believe that the patient was a poor candidate and forgo the opportunity to initiate lipid-lowering therapy during outpatient follow-up. While an inpatient initiation strategy has been advocated by some investigators,18 few studies14,19,20 have compared the effectiveness of predischarge initiation on subsequent use. We aimed to further test the hypothesis that patients with coronary disease would be more apt to use lipid-lowering therapy 6 months after hospitalization for a percutaneous coronary intervention if treatment were initiated during hospitalization. For this analysis, we examined data from the Evaluation in PTCA to Improve Long-term Outcome With Abciximab GP IIb/IIIa Blockade (EPILOG) trial.21
The EPILOG trial methods have been published previously.21 Briefly, this trial enrolled 2792 patients older than 21 years from the United States and Canada between February 27 and December 11, 1995, who were undergoing percutaneous coronary intervention for stable or recently (>24 hours before enrollment) unstable coronary disease. Patients undergoing planned stent placement or rotational coronary atherectomy were excluded from the main trial. In addition, patients who were taking lipid-lowering therapy at the time of admission or who died during their index hospitalization were excluded from this substudy. An institutional review board approved the EPILOG trial protocol at each clinical site, and informed consent was obtained from all patients for trial participation.
The EPILOG trial randomized patients to 1 of 3 groups: placebo and standard-dose weight-adjusted heparin sodium; abciximab and standard-dose weight-adjusted heparin; or abciximab and low-dose weight-adjusted heparin. Abciximab or placebo was given as an intravenous bolus before the percutaneous coronary intervention and continued as an infusion for 12 hours thereafter. Heparin boluses were given before and during the coronary intervention as needed. Daily aspirin therapy was protocol-mandated in all patients. The administration of other medications was at the discretion of the treating physician. Specifically, there were no policy recommendations regarding the prescription of lipid-lowering therapy. Medications used on admission, at discharge, and at 30-day and 6-month follow-up, including lipid-lowering agents, were captured prospectively on the trial case report form; however, data on the specific class of lipid-lowering therapy prescribed were not available.
We compared use of lipid-lowering therapy at 30 days and 6 months according to whether treatment with these agents had been initiated before hospital discharge. Medication use during follow-up was ascertained through telephone calls and/or clinic visits.
Continuous data are displayed as medians with interquartile ranges. Mann-Whitney rank-sum tests were used in univariate analyses of continuous data. Categorical data appear as frequencies and percentages. The χ2 or Fisher exact test was used where appropriate for univariate analyses of categorical data. Multivariable logistic regression was used to ascertain the independent association between inpatient initiation of lipid-lowering therapy and drug use at 6 months. Adjusted odds ratios generated from multivariable logistic regression models were converted to corrected risk ratios (RRs) by a validated method.22
Unlike in randomized trials, treatment decisions in observational studies are frequently biased by multiple external factors because of their relationship with particular outcomes. Therefore, before undertaking standard multivariable logistic regression techniques to adjust for the effect of confounding on drug use, we performed a propensity analysis to minimize any bias inherent in the decision to prescribe lipid-lowering agents before hospital discharge.23 Propensity analyses attempt to identify patients who are similar except for their treatment assignment and in this way approximate the conditions of randomized trials.24 The propensity analysis was conducted as follows.25 First, the probability that any patient would receive a lipid-lowering drug during hospitalization (ie, the propensity score) was estimated with a multivariable logistic regression model that incorporated available demographic, clinical, and treatment variables, irrespective of their clinical relevance. The ability of this model to discriminate between patients who did and did not have lipid-lowering therapy initiated before hospital discharge was assessed with the c statistic (ie, area under the receiver operating characteristic curve). Second, patients who did not receive lipid-lowering agents before hospital discharge were closely matched (2:1 ratio) with those who did, according to their propensity score. Nearest available matching was performed on the propensity score, initially to the fifth digit and then as needed to the first digit. Third, propensity scores were entered in a continuous fashion along with treatment strategy (ie, lipid-lowering agent administration before hospital discharge) and other potential confounders into a multivariable logistic regression model predicting lipid-lowering agent utilization at 6 months. In this model, the entire cohort served as the unit of analysis as patients were matched on treatment, not on outcome. All statistical analyses were performed with the SAS system, version 6.12 (SAS Institute Inc, Cary, NC).
Overall, there were 2126 patients in the EPILOG trial who were not taking a lipid-lowering agent at the time of admission and who survived until hospital discharge. Of these patients (median age, 60 years; 28% female), 175 (8%) had treatment with a lipid-lowering agent initiated during hospitalization. In the remainder, the opportunity to initiate lipid-lowering therapy was deferred to the outpatient setting. There were a number of notable between-group differences at baseline (Table 1). Patients discharged on a regimen of lipid-lowering therapy were younger, more obese, more likely to have a history of hypercholesterolemia, less likely to have a history of heart failure, and more likely to have undergone unplanned stent placement. Finally, there were nonsignificant trends toward less diabetes, more previous myocardial infarction, more coronary interventions performed for a recent myocardial infarction, fewer for stable ischemia, and a greater prevalence of β-blocker treatment before discharge among those in whom lipid-lowering therapy was initiated before discharge.
As shown in Figure 1, after 30 days of follow-up, 387 patients (18%) were taking lipid-lowering therapy: 141 (81%) of those in whom therapy was initiated in the inpatient setting (inpatient group) and 246 (13%) of those in whom it was not (outpatient group) (P<.001). Similarly, lipid-lowering therapy was used in 628 patients (29%) at the 6-month follow-up, 134 (77%) of those who began taking lipid-lowering therapy in the inpatient setting and 494 (25%) in the remainder (RR, 3.05; 95% confidence interval [CI], 2.76-3.29; P<.001). Besides inpatient initiation, other significant positive univariate predictors of lipid-lowering agent use at 6-month follow-up included the following (shown as using vs not using lipid-lowering therapy at 6 months, respectively): younger median age (57 vs 61 years; P<.001), white race or ethnicity (93% vs 89%; P = .01), history of hypercholesterolemia (67% vs 42%; P<.001), absence of diabetes (84% vs 76%; P = .001), current smoking (40% vs 35%; P = .04), no history of congestive heart failure (96% vs 91%; P<.001), percutaneous coronary intervention for an indication other than chronic stable angina (82% vs 76%; P = .002), discharge with a β-blocker (60% vs 55%; P = .04), and discharge without a nitrate (14% vs 11%; P = .04). Absence of hypertension (45% vs 41%; P = .10), history of myocardial infarction (52% vs 48%; P = .12), and absence of a prescription for a calcium channel blocker at discharge (51% vs 48%; P = .13) were also positively related to the use of lipid-lowering therapy at 6 months, although these associations were of marginal statistical significance.
The final multivariable logistic regression model of treatment with a lipid-lowering agent before discharge included the following significant positive predictors in order of their importance in the model: history of hypercholesterolemia, discharge while taking a β-blocker, greater body mass index, younger age, placement of a coronary stent, absence of diabetes, and later randomization date. The c statistic was 0.73, suggesting that the propensity model discriminated acceptably between patients who were prescribed lipid-lowering drugs before discharge and those who were not. The final (ie, "propensity-matched") study cohort comprised 477 well-matched patients, 159 who began lipid-lowering therapy in the inpatient setting and 318 who did not. Baseline characteristics, including median propensity score, were similar in each group (Table 2).
In the overall cohort, after adjusting for propensity score and other potential confounders, inpatient initiation of lipid-lowering therapy was a strong, significant positive predictor of use at 6 months (RR, 3.17; 95% CI, 2.88-3.41; χ2 = 141; P<.001). Other significant and independent predictors in order of their importance in the model included initiation of lipid-lowering therapy within 30 days of hospital discharge (RR, 3.43; 95% CI, 3.21-3.60; χ2 = 208; P<.001), history of hypercholesterolemia (RR, 1.56; 95% CI, 1.28-1.87; χ2 = 17; P<.001), percutaneous coronary intervention for an indication other than chronic stable angina (RR, 0.74; 95% CI, 0.58-0.93; χ2 = 7; P = .01), previous congestive heart failure (RR, 0.74; 95% CI, 0.58-0.93; χ2 = 5; P = .02), increasing age (per year) (RR, 0.99; 95% CI, 0.98-0.99; χ2 = 5; P = .03), and diabetes mellitus (RR, 0.78; 95% CI, 0.60-0.99; χ2 = 4; P = .04).
After restricting the analysis to propensity-matched patients (n = 477) and adjusting for the propensity score and other potential confounders, initiation of a lipid-lowering agent during hospitalization was the strongest predictor of use at 6 months (RR, 2.50; 95% CI, 2.29-2.65; χ2 = 97; P<.001) (Table 3). In this model, the only additional significant and independent predictors of lipid-lowering agent use at 6 months were initiation of lipid-lowering therapy within 30 days of hospital discharge (RR, 2.69; 95% CI, 2.41-2.83; χ2 = 48; P<.001) and percutaneous coronary intervention for an indication other than chronic stable angina (RR, 0.57; 95% CI, 0.36-0.86; χ2 = 7; P = .007).
When we examined 6-month persistence rates (100% use at discharge minus the discontinuation rate) among patients discharged taking lipid-lowering agents (134/175 [77%]) and compared them with 5-month persistence rates (ie, those measured 6 months after hospital discharge) among patients who began lipid-lowering agents 1 month after discharge (179/219 [82%]), the rates were statistically similar (P = .28). A propensity model predicting initiation of lipid-lowering therapy at discharge vs 1 month after discharge was only somewhat discriminative given the small sample size (n
= 394; c statistic, 0.64). In a propensity-adjusted multivariable model of 6-month use of lipid-lowering agents, both the discharge and 1-month strategies remained statistically identical (P = .19).
We found that initiation of lipid-lowering agents before discharge was the most important independent predictor of their use at follow-up. In fact, patients in whom lipid-lowering therapy was initiated before discharge were nearly 3 times as likely to be taking these agents 6 months later. This cohort of patients, with angiographically documented coronary disease, represents a particularly high-risk subset in secondary prevention,1- 3,26 and lipid-lowering therapy is especially cost-effective in this setting27- 29 (as few as 20 patients must be treated to prevent an event30). Accordingly, it is striking that, if treatment is not initiated during hospitalization, only 25% of patients not previously taking lipid-lowering agents were receiving therapy 6 months after discharge. By comparison, 77% of patients in whom lipid-lowering therapy was initiated during hospitalization were using these agents 6 months after discharge.
The question of when to initiate lipid-lowering therapy has broad implications for the treatment of patients with vascular disease. Multiple secondary prevention trials have demonstrated that lipid lowering, particularly with statins, can reduce the long-term risk of death, myocardial infarction, stroke, and coronary revascularization when begun within 3 to 6 months of an acute coronary syndrome1- 3; the recent Heart Protection Study4 has extended these benefits to individuals with stable coronary disease and cholesterol levels well below the treatment initiation threshold recommended in current National Cholesterol Education Program–Adult Treatment Panel III guidelines.31 Recent studies suggest that early initiation of lipid-lowering therapy after an acute coronary syndrome may reduce short-term complications as well.5- 7 Therefore, it may not be prudent to wait for "normalization" of lipid levels in the months following an acute coronary syndrome before determining patient eligibility for lipid-lowering therapy.32,33 Accordingly, any strategy that increases early use of therapy has the potential to dramatically reduce early and late cardiovascular morbidity and mortality.
Although national consensus guidelines now recommend simultaneous initiation of therapeutic lifestyle changes and lipid-lowering therapy in patients hospitalized for a coronary event or procedure,31,34,35 no randomized studies have examined the effect of an in-hospital initiation strategy on subsequent use of lipid-lowering agents,14,19,20 and existing observational studies have not used multivariable analyses. The 1998 French PREVENIR survey14 observed that among 1456 patients with unstable angina or acute myocardial infarction, 94% of those discharged taking statins continued taking these agents 6 months later. Conversely, only 15% of those not discharged taking a statin were taking one of these agents at 6 months. In the Cardiac Hospitalization Atherosclerosis Management Program (CHAMP), a program in which in-hospital secondary preventive measures were implemented among patients admitted with an acute myocardial infarction,20 investigators compared 256 patients who were admitted to a single university hospital in 1992 to 1993 with 302 similar patients admitted in 1994 to 1995 (CHAMP's inception was in January 1994). Prescription of lipid-lowering agents at discharge increased dramatically after initiation of the program (6% vs 86%; P<.01). At 1 year, only 10% of patients discharged in the pre-CHAMP era were taking a statin, whereas 91% of those discharged after CHAMP were prescribed these agents (P<.01). Finally, Muhlestein and others19 observed a cohort of 600 patients from 1994 through 1997 who had angiographically documented coronary disease and met National Cholesterol Education Program–Adult Treatment Panel II criteria36 for lipid-lowering therapy at the time of their diagnostic cardiac catheterization. This single-center study found that, among the 64% of surviving patients with available follow-up, discharge taking a lipid-lowering agent was a significant univariate predictor of use during the next 3 years. Of the 278 patients discharged without lipid-lowering therapy, only 40% began taking these agents during follow-up. In contrast, 77% of the 65 patients discharged taking lipid-lowering therapy continued taking treatment at follow-up (P<.001).
Collectively, these unadjusted analyses suggest that patients with coronary disease discharged taking lipid-lowering agents are more likely to use them at follow-up than those discharged without these agents. Nevertheless, the relatively small size and reach of these studies, inclusion of patients previously treated with lipid-lowering therapy, and limited statistical analyses make it difficult to ascertain the true independent effect of in-hospital initiation on subsequent use. Accordingly, we sought to address each of these important limitations. First, our study was performed in a cohort of patients from 69 medical centers across the United States and Canada. Second, we excluded patients in whom such therapy was started before admission. Finally, we performed both propensity and standard multivariable logistic regression analyses to minimize the influence of bias and confounding on our findings.
A comparison of RRs for the association between lipid-lowering agent use before discharge and use at 6 months in unadjusted, covariate-adjusted, and propensity-adjusted models suggests that the association between initiation strategy of lipid-lowering agents and 6-month use in our study was influenced by bias and confounding (Table 3). In the overall cohort, the addition of covariates to the unadjusted model increased the RR for use of lipid-lowering therapy at 6 months from 3.05 to 3.18. The addition of the propensity score to the covariate-adjusted model did not change the RR materially (3.18 to 3.17). This suggests that much of the measurable confounding present in the overall cohort could be addressed through standard multivariable modeling and that the propensity score alone added little to this analysis. Further restricting the analysis to propensity-matched patients resulted in a decrease in the unadjusted RR (3.05 to 2.26). In this smaller cohort, RRs in the covariate-adjusted and covariate- and propensity-adjusted models were similarly increased (2.26 to 2.50). Collectively, these observations suggest that restricting the analysis to propensity-matched patients addressed an element of bias not accounted for when the propensity score was simply entered into the model; standard multivariable modeling techniques then adjusted for most of the inherent and measurable confounding.
The primary aim of our analysis was to compare the effect of 2 different discharge strategies on subsequent use of lipid-lowering therapy. More specifically, we compared the persistence rate of lipid-lowering medications at 1 and 6 months among patients discharged taking these agents with the new use rate (outpatient initiation rate minus the discontinuation rate) at 1 and 6 months among those discharged without these agents. We also compared the persistence rates between patients discharged taking lipid-lowering therapy and those discharged without lipid-lowering therapy but in whom it was initiated within the first month after hospital discharge. Unfortunately, this comparison was limited by information available on medication use during follow-up. For patients who were discharged taking lipid-lowering therapy, information was available on use at 1 and 6 months after initiation. In contrast, for patients who began therapy during the first month after discharge, information on use was available only at 5 months (ie, 6 months after discharge). When we compared 6-month persistence in the first group with 5-month persistence in the second group of patients, we found these rates to be statistically identical. When considered alongside our primary study findings, this observation suggests that outpatient initiation of lipid-lowering therapy soon after a recent hospitalization may also be effective vis-à-vis drug use for a given individual, but that, as an overall population strategy, in-hospital initiation results in a much greater proportion of patients taking lipid-lowering therapy at 6 months (77% vs 25%).
First, our study was observational. Because it would be unethical to perform a placebo-controlled trial in which lipid-lowering therapy is knowingly withheld from eligible patients with coronary disease, we must look to observational data for evidence of an association between in-hospital initiation of lipid-lowering therapy and its subsequent utilization. Nevertheless, as in all observational studies, it was not possible to adjust for all potential confounders, known (eg, influence of individual physicians or medical centers) or unknown, nor to completely account for any selection bias that may have influenced who received treatment. Our propensity analysis and subsequent multivariable logistic regression modeling should have minimized the influence of these factors, although residual confounding remains a possibility.
Second, since completion of the EPILOG trial, the evidence base for lipid-lowering therapy for secondary prevention has broadened considerably and now includes patients with coronary disease who have normal or only moderately elevated cholesterol levels.2,3,26 As such, consensus guidelines now recommend initiation of lipid-lowering therapy for an even broader range of patients31 and will likely lower the threshold for treatment even further with time. Accordingly, the absolute percentages of lipid-lowering agent use may not be relevant today.
Finally, it was not possible to determine the true appropriateness of treatment on a per-patient basis, since hypercholesterolemia was not strictly defined, baseline lipid levels were not available, and no information was available on contraindications to lipid-lowering therapy in EPILOG. In addition, information regarding secondary causes of hypercholesterolemia (eg, hypothyroidism) was not available from the trial case report form. Nevertheless, most patients with coronary disease ultimately require both pharmacologic and nonpharmacologic lipid-lowering interventions to reach consensus cholesterol targets.8,9,37 It is likely that the majority of patients in the EPILOG trial would have been candidates for initiation of lipid-lowering therapy at the time of discharge or by 6 months of follow-up.
Our findings suggest that inpatient initiation of lipid-lowering therapy for the secondary prevention of coronary disease is an effective strategy to enhance subsequent use. Other modifiable factors that influence the long-term use of these agents must be identified if we are to bridge the gap between the current evidence base and practice of preventive medicine.
Corresponding author: Steven E. Nissen, MD, Department of Cardiovascular Medicine, Desk F15, Cleveland Clinic Foundation, 9500 Euclid Ave, Cleveland, OH 44195 (e-mail: firstname.lastname@example.org).
Accepted for publication January 17, 2003.
The EPILOG trial was funded by Centocor, Malvern, Pa, and Eli Lilly and Co, Indianapolis, Ind.
This study was presented at the XXIII European Society of Cardiology Annual Congress; September 3, 2001; Stockholm, Sweden.