Customize your JAMA Network experience by selecting one or more topics from the list below.
Ko DT, Mamdani M, Alter DA. Lipid-Lowering Therapy With Statins in High-Risk Elderly Patients: The Treatment-Risk Paradox. JAMA. 2004;291(15):1864–1870. doi:10.1001/jama.291.15.1864
Author Affiliations: Division of Cardiology, Schulich Heart Centre, and Department of Medicine, Sunnybrook and Women's College Health Sciences Centre (Drs Ko and Alter), and Institute for Clinical Evaluative Sciences (Drs Ko, Mamdani, and Alter), Toronto, Ontario.
Context The benefits of cardiovascular therapies such as statins for secondary
prevention have been well documented, although they may not be optimally used
in patients most likely to benefit. Ideally, aggressiveness in the use of
these beneficial therapies should correlate with baseline cardiovascular risk.
Objective To examine the association between physicians' treatment aggressiveness
and baseline cardiovascular risk.
Design, Setting, and Patients Retrospective cohort study incorporating the use of multiple linked
health care administrative databases covering more than 1.4 million elderly
residents of Ontario. We included 396 077 patients aged 66 years or older
who had a history of cardiovascular disease or diabetes while undergoing medical
treatment and who were alive on April 1, 1998. Baseline cardiovascular risk
was derived using a risk-adjustment index in which we modeled probability
of death after 3 years of follow-up.
Main Outcome Measure Likelihood of statin use, stratified by baseline cardiovascular risk,
after adjusting for age, sex, socioeconomic status, and rural or urban residence.
Results Only 75 617 patients (19.1%) in this secondary prevention cohort
were prescribed statins. In patients 66 to 74 years old, the adjusted probabilities
of statin prescription were 37.7%, 26.7%, and 23.4% in the categories of low,
intermediate, and high baseline risk, respectively. The likelihood of statin
prescription was 6.4% lower (adjusted odds ratio, 0.94; 95% confidence interval,
0.93-0.95) for each year of increase in age and each 1% increase in predicted
3-year mortality risk. The influence of age also interacted synergistically
with baseline risk on the prescription of statins (P<.001).
Conclusions We found that prescription of statins diminished progressively as baseline
cardiovascular risk and future probability of death increased. Since the benefits
of a therapy are dependent on the baseline risk, the maximum benefits of statins
may not be fully realized until implementation of therapy includes patients
at highest risk.
Available evidence has demonstrated that the impact of cardiovascular
evidence-based therapies is predominantly dependent on patients' baseline
risk of future adverse cardiovascular events.1 If
physicians are appropriately attuned to the risk profiles of their patients,
one might reasonably assume that patients who are at highest baseline risk
should be treated most aggressively. Yet, for many cardiovascular therapies,
this is not the case.2-6 For
example, studies have consistently demonstrated an inverse relationship between
treatment propensity and age.2,7 Moreover,
patients with multiple chronic conditions are less likely to receive evidence-based
therapies than healthier patients with lower illness severity,8 an
observation that may relate to high baseline risk and/or concerns about treatment
complications. Nonetheless, the extent to which the discordant relationship
between baseline risk and treatment propensity is a phenomenon driven by age
alone, arguably the most important determinant of baseline risk in the population,
is unknown. Furthermore, the extent to which the treatment-risk paradox applies
only to extremes of illness severity or, conversely, applies incrementally
throughout the entire spectrum of risk is also unclear.
Statins for secondary prevention provide a useful test case for several
reasons. First, statins are among the most efficacious therapies in reducing
future cardiac events and mortality.9-15 Initially
shown to be beneficial in patients with substantially elevated cholesterol,9,10 the benefits of statin therapy have
currently extended to previously "normal" cholesterol levels.11-14 Second,
clinical trials have demonstrated consistent treatment effects across multiple
subgroups, including elderly persons.15 Third,
clinical guideline recommendations advocate the use of statins according to
individual baseline risk of future cardiovascular events.16 Accordingly,
the primary objective of this study was to examine the association between
physician aggressiveness in the prescription of statins among a secondary
prevention cohort of elderly patients and baseline risk throughout the entire
The Canadian health insurance system provides free universal coverage
for most hospitals and ambulatory medical services. The Ontario Drug Benefit
(ODB) program is a government-funded drug benefit program that covers outpatient
drug costs for all Ontario residents aged 65 years or older. Patients are
responsible for a dispensing fee of approximately Can $6, but this fee is
waived for patients whose annual income falls below a threshold of Can $15 500
(US $1 = Can $1.34 on March 17, 2004).
The Geriatric Ontario Longitudinal Database (GOLD) was created by linking
several major health care administrative databases with follow-up tracking
of mortality over time, regardless of location of death. Briefly, GOLD includes
1.44 million residents aged 66 years or older who were alive in Ontario on
April 1, 1998. Unique encrypted patient identifiers were used for linkage
in the multiple databases to protect patient confidentiality. We identified
previous hospitalizations using the Canadian Institute of Health Information
hospital discharge abstracts and identified physician visits and previous
cardiac interventions using physician claims data obtained by the Ontario
Health Insurance Plan. We used the ODB to obtain information on medication
prescriptions within 1 year before cohort inception. Demographic and geographic
information was identified using the Registered Persons Database and data
from the official 1996 Census. We excluded all non-Ontario residents and those
who did not have a valid health card number. The research ethics board of
Sunnybrook and Women's College Health Science Center, University of Toronto,
approved the study and waived a requirement for informed consent.
We selected patients at high risk of future cardiovascular events and
included patients older than 66 years with a history of cardiovascular disease
or diabetes mellitus while undergoing medical therapy. We defined a statin
prescription as one in which a patient had been dispensed any prescription
for statin medication within the year before the inception of the cohort.
Statins covered in the ODB at the time of the study included atorvastatin,
cerivastatin, fluvastatin, lovastatin, pravastatin, and simvastatin. We defined
cardiovascular disease in the study sample as 1 or more of the following:
cardiovascular hospitalization within 5 years, coronary intervention (cardiac
catheterization, percutaneous coronary intervention, or coronary artery bypass
graft surgery) within 5 years, or angina (defined as concurrent use of nitrates
within the year of cohort inception). Using hospital discharge abstracts,
a previous cardiovascular hospitalization was defined as an admission with
any of the following: unstable angina, myocardial infarction, congestive heart
failure, stroke or transient ischemic attack, or peripheral vascular disease
(abdominal aortic aneurysm, peripheral vascular surgery, or carotid endarterectomy).
A diagnosis of drug-treated diabetes mellitus required receipt of insulin
or oral hypoglycemic agents within 1 year before cohort inception. The definitions
and codes that were used to identify eligible patients focused on maximizing
specificity rather than sensitivity to ensure the construction of a valid
secondary prevention cohort. The coding accuracy of acute myocardial infarction,
heart failure, and diabetes had positive predictive values of more than 90%.17-19 Although individual
cholesterol profiles were unavailable, we assumed that most of these patients
would be eligible for statin therapy based on the fact that they were all
secondary prevention patients.16 We excluded
patients with a history of cancer within 5 years because of competing risks.
Our final cohort included 396 077 patients after applying inclusion and
The total number of medications dispensed within the previous year was
used as a marker of comorbidity. The use of such measures has been validated
as a comorbidity index and is described elsewhere.20 Socioeconomic
status was defined as a binary variable (impoverished: yes or no). Any patient
whose annual individual income was Can $15 500 or less (or who had a
household income of Can $22 000 or less) and who applied for full subsidization
for prescription drug dispensing fees was classified as impoverished. Patient
residence was categorized as a binary term (urban vs rural) based on residential
To disentangle baseline risk for future adverse events from the effects
of other factors on mortality, we derived a baseline risk index using multiple
logistic regression models that adjusted for various clinical characteristics
but excluded age, sex, socioeconomic status, and location of residence. We
modeled likelihood of death after 3 years of follow-up as a function of the
following clinical variables: previous cardiovascular hospitalizations (stratified
according to diagnosis), total number of previous cardiovascular hospitalizations,
number of dispensed cardiovascular medications (excluding statins), and the
comorbidity index. Similar risk-adjustment indexes have been previously used
to characterize illness severity and validated in disease-specific cohorts
(eg, acute myocardial infarction).21 The area
under the receiver operating characteristic curve (AUROC) of our baseline
risk index was 0.71. When age and sex were added into the risk index, the
AUROC increased to 0.79, suggesting good discriminating characteristics.
We first compared demographic and clinical characteristics in patients
with and without statin prescriptions. In univariate fashion, categorical
variables were compared using χ2 tests, and continuous variables
were compared using either a t test or another nonparametric
test as appropriate.
We examined the relationship among baseline risk index, age, and statin
prescription using multiple logistic regression techniques, while adjusting
for sex, socioeconomic status, and rural vs urban status. We tested for multiple
statistical interactions, including the interaction between age and baseline
risk, in each of our models. Adjusted probability curves (ie, the probability
of receiving vs not receiving statins) were constructed according to age and
the baseline risk index by imputing average covariate patterns for sex, income,
and geographic residence. When examining the probability of statin prescription
by age, we stratified the baseline risk index into 3 groups (25th, 50th, and
75th percentiles of death). Similarly, when examining the probability of statin
prescription by baseline risk index, we categorized age into 3 subgroups (a
typical 71-year-old, a typical 75-year-old, and a typical 81-year-old).
A series of sensitivity analyses was performed to examine the robustness
of our results. First, analyses were repeated by evaluating the prescription
of all lipid-lowering therapies (eg, fibric acid derivatives) rather than
the prescription of statins alone. Second, all data were reanalyzed when confining
the cohort to patients with a history of preexisting cardiovascular disease
(ie, excluding those with diabetes alone). Third, due to potential concerns
arising from the confounding "protective" survival effects of statins themselves,
baseline risk was derived with and without the inclusion of statins in our
risk-adjustment models. Fourth, we undertook additional modifications to the
derivation of our risk-adjustment index in which baseline risk was modeled
as a function of the composite of death or myocardial infarction rather than
as a function of death alone. Fifth, because of concerns that we might have
included patients with substantial comorbid conditions for which statins might
not be appropriate, we repeated the analysis using 2 different cohorts. One
cohort excluded patients in the 75th percentile of the comorbidity index,
and the other excluded patients who died within 1 year of cohort inception.
Finally, we examined a different risk index that excluded the comorbidity
index and adjusted only for cardiovascular risks in both our original cohort
and lower risk cohorts. In all of these sensitivity analyses, our overall
results did not materially change. All statistical analyses were performed
using SAS statistical software, version 8.2 (SAS Institute Inc, Cary, NC). P<.05 was considered statistically significant for all
In our study sample, 271 504 patients (68.6%) had a history of
cardiovascular disease alone, 70 535 (17.8%) had diabetes mellitus alone,
and 54 038 (13.6%) had both preexisting cardiovascular disease and diabetes
mellitus. The median age of the overall cohort was 75 years; 216 089
(54.6%) were women, 143 790 (36.3%) had low socioeconomic status, and
69 210 (17.5%) lived in rural areas (Table 1). The correlation between age and the baseline risk index
was modest (r = 0.31). In this cohort, 75 617
patients (19.1%) were prescribed statin therapy. Patients prescribed statins
were younger; were more likely to be men; had a history of angina, acute myocardial
infarction, or prior cardiac invasive procedures; and had more visits to a
cardiologist within the past year. Conversely, patients not prescribed statins
were more likely to have diabetes, congestive heart failure, or stroke; to
have lower socioeconomic status; and to live in rural areas (P<.001 for all) (Table 1). Table 2 illustrates the observed rates
of death and the expected probabilities of death at 3 years according to different
categories of age and different categories of the baseline risk index.
Table 3 illustrates the
independent determinants of statin prescription according to multivariable
analysis. Both age and baseline risk inversely correlated with the likelihood
of receiving statins, after adjusting for sex, socioeconomic status, and rural
vs urban residence. Moreover, age interacted with baseline risk in determining
statin prescription (P<.001 for the interaction
term). For each year of increase in age and each 1% increase in the baseline
risk index, there was a 6.4% lower odds of statin prescription (adjusted odds
ratio, 0.94; 95% confidence interval, 0.93-0.95; P<.001)
after adjusting for all other factors.
Progressively lower use of statins in patients with higher cardiovascular
risk existed across the full spectrum of cardiovascular risk (Figure 1, A). Similarly, we also observed lower use of statins in
elderly patients across the entire spectrum of age (Figure 1, B). The 95% confidence intervals were very small in the
probability estimations of statin use across the whole spectrum of baseline
risk and age (data not shown).
Despite convincing evidence demonstrating substantial mortality reductions
from the use of statins in secondary prevention, we found that only a small
proportion of elderly patients with preexisting cardiovascular disease were
prescribed statin therapy in Ontario. Moreover, the likelihood that physicians
prescribed statin therapy was inversely correlated with baseline risk and
did so across the entire spectrum of illness severity, independent of age.
Finally, the effects of age and baseline risk had a synergistic effect on
the prescribing behaviors of physicians.
Our observed low prescription rate of statin therapy adds to a growing
body of literature demonstrating that statin therapy is substantially underused.22-25 This
is concordant with recent data collected by the National Registry of Myocardial
Infarction, in which only 31.7% of all patients discharged from the hospital
with acute myocardial infarction were prescribed lipid-lowering therapy.22 Furthermore, when examining treatment targets, a
recent survey demonstrated that only 5.4% of patients known to have hyperlipidemia
had achieved a target total cholesterol level in the United States,26 representing a substantial opportunity for improvement.
Although the overall prescription of statins in the population was low,
we found that the decision to withhold therapy was not only observed at the
extremes of illness severity but increased progressively and incrementally
in patients with advancing age and advancing baseline risk. Furthermore, we
observed a significant interaction effect between age and risk in the prescription
of statin therapy. Thus, age and risk acted in concert to further reduce the
propensity of physicians to prescribe statins in secondary prevention.
We do not believe that our findings are explainable by variations in
affordability and/or accessibility of medications. Although patients in Ontario
have to pay out of pocket for dispensing fees, the acquisition costs for statins
are provided to all elderly patients free of charge. Moreover, our analyses
also, to some extent, adjusted for baseline socioeconomic and geographic differences.
Statin therapy is a useful medication to examine because of the substantial
reductions in cardiovascular mortality and the lower adverse effect profiles
compared with thrombolytic therapy or invasive cardiac procedures. The relationship
between baseline risk and statin avoidance in secondary prevention patients
likely reflects a systematic bias, which may be generalizable to other evidence-based
therapies. Studies demonstrating the diminishing use of thrombolytics, cardiac
catheterizations, and β-blockers in elderly patients in the setting of
acute myocardial infarction2,27,28 may
represent a similar phenomenon—an aversion to treat patients at high
risk of future adverse events. The inverse relationship between baseline risk
and treatment aggressiveness implies suboptimal benefits of evidence-based
therapies when applied to real-world settings. To our knowledge, the treatment-risk
paradox has not been demonstrated in a similar fashion. Other studies are
needed to confirm our findings in different patient subsets using different
Several factors may explain the treatment-risk paradox for statins.
First, physicians may have misconceptions about the benefit-harm tradeoffs.
For example, physicians may feel reluctant to generalize clinical trial results
to elderly patients with comorbidities on the grounds that such patients may
experience fewer benefits and greater harm from the adverse effects of therapy.
However, the relative survival benefits associated with statins appear to
be consistent across multiple subgroups, including elderly patients.9-15 Furthermore,
the impact of any therapy in the population depends on baseline risk more
than relative efficacy.1 Although the absolute
rate of serious harmful adverse effects may be increased for patients at highest
baseline risk compared with their healthier counterparts, the rate of life-threatening
complications required to negate potential survival benefits from treatment
rarely approaches the incidence encountered in real-world settings, especially
for statins, where the rate of severe complications is extremely low.1,29,30 This overemphasis
of harm combined with an underappreciation of benefits may favor a more conservative
hands-off approach to treatment.
Second, physicians may prejudge the compliance of their patients and
be less inclined to prescribe therapies to patients thought unlikely to adhere
to treatment. Patients at higher baseline risk may not perceive the benefits
of additional therapy and have an increasing unwillingness to be receptive
to physicians' recommendations. Indeed, factors that contribute to poor compliance
in elderly patients include cognitive, functional, and social decline31—all factors associated with higher baseline
risk of adverse cardiac events. However, preconceived attitudes toward compliance
may too be misguided, given the evidence suggesting improved compliance rates
of statins among elderly patients with higher severity of cardiac illness.32 Therefore, it is important for physicians to address
the diverse needs of elderly patients and emphasize therapies that would derive
substantial benefits to promote adherence.
Finally, the treatment-risk paradox may be explained by physician inattentiveness
to cardiovascular prevention, especially when multiple conditions coexist.
For example, available evidence suggests that clinicians who care for patients
with chronic diseases become less attentive when managing the necessities
of other concurrent conditions due to constraints in time, expertise, and
Several limitations of our study merit consideration. First, we did
not have access to individual cholesterol levels, and it is not possible to
determine the appropriateness of statin prescription on an individual basis.
Since all patients in our cohort were eligible for secondary prevention, we
believe that most of our cohort would have qualified for and derived substantial
benefits from statin therapy. Furthermore, patients eligible for statins are
likely equally distributed across the risk spectrum, and thus, our observed
treatment pattern across the age-risk spectrum is unlikely to be affected.
Second, patients' risk profiles characterized by administrative data may be
subject to undercoding of comorbidity. Therefore, we designed our cohort to
include patients with a prior history of cardiovascular disease or diabetes
to maximize specificity. Third, we used all-cause mortality as our main determinant
of baseline risk. Although admittedly, the benefits of statins are largely
mediated through their effects on cardiovascular outcomes, randomized trials
have demonstrated that statins exert consistent reductions in all-cause mortality.
Moreover, studies33,34 have found
that the determination of cardiac death may be inaccurate and could potentially
lead to misinterpretation of data. Finally, our data reflected prescribing
patterns approximately 5 years ago, and overall utilization rates of statins
have likely increased in the interim.35,36 However,
given the magnitude of discordance between baseline risk and treatment propensity
observed in this study, it is unlikely that the treatment-risk paradox will
cease to exist.
In conclusion, we demonstrate that physician aggressiveness in the prescription
of statin therapy to elderly patients for secondary prevention in Ontario
was inversely correlated with baseline cardiovascular risk independent of
age. The treatment-risk paradox phenomenon was not only applicable at the
extremes of illness severity but was also observed throughout the entire spectrum
of illness severity. Given the importance of baseline risk in determining
the impact of therapy in the population, the treatment-risk paradox implies
that the survival benefits of statin therapy may never be fully realized until
physicians appropriately attune their prescribing behaviors to the risk profiles
of their patients.