• adherence;
  • antihypertensive;
  • cardiovascular disease;
  • drug combination;
  • dyslipidemia;
  • economic analysis;
  • statin


  1. Top of page
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

Objective:  We sought to determine lifetime costs, morbidity, and mortality associated with varying adherence to antihypertensive and 3-hydroxy-3-methylglutaryl-coenzyme A reductase inhibitors (statin) therapy in a hypertensive population.

Methods:  A model was constructed to compare costs and outcomes under three adherence scenarios: no treatment, ideal adherence, and real-world adherence. Simulated patients’ characteristics matched those of participants in the Anglo-Scandinavian Cardiac Outcomes Trial–Lipid-Lowering Arm and event probabilities were calculated with Framingham Heart Study risk equations. The real-world adherence scenario employed adherence data from an observational study of a US population; risk reductions at each level of adherence were based on linear extrapolations from clinical trials. Outputs included life expectancy, frequencies of primary and secondary coronary heart disease and stroke, and direct medical costs in 2006 US$. The incremental cost per life-year gained and incremental cost per event avoided were calculated comparing the three adherence scenarios.

Results:  Mean life expectancy was 14.73 years (no-treatment scenario), 15.07 (real-world adherence), and 15.49 (ideal adherence). The average number of cardiovascular events per patients was 0.738 (no treatment), 0.610 (real-world adherence), and 0.441 (ideal adherence). The incremental cost of real-world adherence versus no treatment is $30,585 per life-year gained, and ideal adherence versus real-world adherence is $22,121 per life-year gained.

Conclusions:  Hypertensive patients taking antihypertensive and statin therapy at real-world adherence levels can be expected to receive approximately 50% of the potential benefit seen in clinical trials. Depending on its cost, the incremental benefits of an effective adherence intervention program could make it an attractive value.


  1. Top of page
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

Numerous clinical trials and meta-analyses have demonstrated the benefits of antihypertensive medications and 3-hydroxy-3-methylglutaryl-coenzyme A reductase inhibitors (statins) for reducing coronary heart disease (CHD) and stroke risk in patients at a high risk of CHD [1–4]. A substudy of the Anglo-Scandinavian Cardiac Outcomes Trial (ASCOT), ASCOT–Lipid-Lowering Arm (ASCOT–LLA), recently showed a lower cardiovascular event incidence among hypertensive patients with total cholesterol ≤6.5 mmol/L (241 mg/dl) who were treated concomitantly with antihypertensive and statin therapies compared with antihypertensive therapy alone [5]. Economic analyses consistently show antihypertensive and statin therapies to be cost-effective in a myriad of patient populations for both primary and secondary prevention [6–13]. Efficacy estimates are typically drawn from clinical trials, where close monitoring encourages optimal drug adherence (accurate timing, dosage, and frequency of medication administration) and persistence (duration of therapy) [14].

In real-world patient-care settings, however, adherence and persistence fall short of clinical trial levels. Previous studies have reported suboptimal persistence and adherence among patients taking antihypertensives [15–17] and statins [18–23] alone or concurrently. For example, one study of concurrent therapy found that after 3 months, <45% of patients were adherent to both regimens; after 6 months <36% were adherent [22]. Although most clinicians and policymakers would agree that such use is suboptimal, the clinical and economic impacts of nonadherence with these medications have not been well described.

To inform policies and practices that might affect adherence and persistence with antihypertensive and statin therapies, we assessed the clinical and economic burden of nonadherence in an adult hypertensive population with normal to mildly elevated serum cholesterol levels (modeling that of the ASCOT–LLA population [5].


  1. Top of page
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

Model Structure

Three Markov models were developed to compare costs and outcomes of primary and secondary prevention with antihypertensive and statin therapy in hypertensive patients aged 40 to 79 years with total cholesterol ≤6.5 mmol/L (241 mg/dl) to simulate the ASCOT–LLA population [5]. Costs and events were estimated using Monte Carlo microsimulation for three scenarios: no treatment, real-world adherence, and ideal adherence. The no-treatment scenario was a disease natural history model, including neither costs nor benefits of pharmacologic therapy. The ideal adherence scenario was based on the antihypertensive and statin adherence and effectiveness observed in ASCOT–LLA [5]. The real-world adherence scenario employed real-world adherence rates and transitions based on filled prescription records from the California Medicaid (Medi-Cal) system.

The base-case analysis compared event rates and costs among the three models over a lifetime horizon using 1-year Markov cycles. Events included primary and secondary myocardial infarction (MI) or CHD death, primary and secondary angina (angina pectoris or unstable angina), and primary and secondary stroke (any type). The model employed a payer perspective, including direct pharmacy and medical costs in 2006 US$ with future costs and benefits discounted by 3% annually [24]. The models were constructed and analyzed using TreeAge Pro 2006 (TreeAge Software Inc., Williamstown, MA).

Health States

Each model included eight health states: baseline (no CHD history), angina, MI, stroke, postangina, post-MI, poststroke, and dead (Fig. 1). Patients remained in the baseline state until a CHD or stroke event or death due to noncardiovascular causes (noncardiovascular death). Patients surviving the year in which an event occurred transitioned to the corresponding postevent state, where they remained until incurring a subsequent event of any type or noncardiovascular death.


Figure 1. Model flow diagram. MI, myocardial infarction.

Download figure to PowerPoint

Event Risks

Natural history event rates were determined annually using sex-specific equations from the Framingham Heart Study for primary and secondary CHD [25] and primary stroke [26]. The equations’ end points of MI, sudden CHD death, and nonsudden CHD death were classified as MI or CHD death in the models; angina pectoris and coronary insufficiency were classified as angina [25]. Because no Framingham equation estimates secondary stroke risk, the annual risk was calculated as the primary stroke risk multiplied by the relative risk for recurrent stroke [27].

The Framingham equations’ risk-factor inputs were randomly drawn for each subject in the microsimulation from probability distributions. All distributions were based on ASCOT–LLA characteristics except atrial fibrillation, drawn from the Framingham stroke population [28], and menopause. Binomial distributions were created for the prevalence of male sex, cigarette smoking, diabetes mellitus, left ventricular hypertrophy, stroke history, and previous cardiovascular disease (CVD) (excluding stroke and CHD); a normal distribution for alcohol consumption was created based on the reported mean and SD from ASCOT–LLA data. Correlations among age, systolic blood pressure (SBP), total cholesterol and high-density lipoprotein (HDL) cholesterol were estimated based on a patient sample from the National Health and Nutritional Examination Survey 2003 to 2004 [29] that approximated ASCOT–LLA inclusion criteria, except the requirement of ≥3 cardiovascular risk factors, which could not be reproduced [5]. Sex-specific hypothetical sets of the four variables were created based on age distributions from ASCOT–LLA and were randomly assigned to patients entering the model.

Risk-factor values were held constant over Markov cycles with the exception of age, which increased with time; menopause, which was assumed to affect all female patients at the age of 51 years [30], and history of CVD, which was adjusted with event occurrences.


All-cause mortality rates were based on age- and sex-specific US life tables [31]. Rates for the baseline state were adjusted by removing the portion attributable to CHD and stroke for that age [32].

In-hospital mortality following MI was age-specific, based on age distributions and mortality probabilities [33] coupled with age-specific hazard ratios for in-hospital mortality [34]. The mortality rate among patients surviving after hospitalization was equal to the general rate, adjusted for the relative risk of mortality following MI. Relative risks, conditional on age and years since MI, were extracted from post-MI survival curves [34].

Poststroke mortality rates were calculated similarly using published survival curves [35]. Because of limited data being available, no differentiation was made between in-hospital mortality and mortality for 1 year following stroke. Poststroke mortality rates were functions of age and years since stroke (<5 years; year 5 rate was applied subsequently).

Mortalities occurring in the year following MI or stroke were attributed to that event; all other mortalities were classified as noncardiovascular deaths. Mortality rates for patients in the postangina state were assumed to follow general population rates.


The models incorporated emergency, hospitalization, physician, and follow-up direct medical costs associated with included events (Table 1). Emergency costs for angina and for MI were based on a study by Russell et al. [36]; separate costs reported for unstable angina and angina pectoris were averaged using relative frequency weights [25]. Emergency costs for stroke were not available from the literature and were conservatively assumed to be $0.

Table 1.  Cost inputs
ComponentCost inputs (2006 US$)*Source reference
Base-caseLower boundUpper bound
  • *

    All costs were inflated to 2006 US$ using the Medical Care Consumer Price Index for all urban consumers [44].

  • Bounds for the annual cost of statin and antihypertensive therapy are base-case values ±20%. Bounds for all other costs are the 25th and 75th percentiles from the distribution of Medicare hospital reimbursement for 2006.

  • Follow-up costs in year 1 apply only to patients surviving hospitalization.

  • §

    Institutional care costs for stroke are age- and sex-specific; the reported costs are mean values.

  • CHD, coronary heart disease; MI, myocardial infarction.

Nonfatal MI19,58518,17122,152 
 Follow-up in year 19,1178,45910,312[36]
Fatal MI28,76226,68632,533 
 Follow-up in year 117,82416,53820,161[36]
Primary angina5,1554,7385,831 
 Follow-up in year 11,3851,2851,567[36]
Secondary angina7,8817,3128,914 
 Follow-up in year 14,0483,7564,578[36]
Primary stroke41,38238,39546,807 
 Annual institutional care (years 1 and 2)§34,25531,78338,746[42]
Secondary stroke49,76146,16956,284 
 Annual institutional care (years 1 and 2)§42,63439,55748,223[42]
Post-CHD per year1,2171,1291,377[36]
Annual statin therapy832666999[43]
Annual antihypertensive therapy8596871031[43]
Annual discount3%[23]

Hospitalization costs were the national median 2006 Medicare reimbursement for the appropriate diagnosis-related group [37–39], and associated physician reimbursement estimates were based on Miller and Welch [40], updated to 2006 units [41]. Hospitalization and physician costs for nonfatal MI were an average of corresponding diagnosis-related groups, weighted by the relative frequency of Medicare claims [39]. Costs for stroke subtypes were averaged using literature-based relative frequency weights [28,42].

Annual pharmacy costs for antihypertensive and statin therapy were calculated as the daily 2006 wholesale acquisition cost (WAC) [43] of ASCOT drugs multiplied by 360 days. ASCOT–Blood Pressure Lowering Arm (ASCOT–BPLA) randomized patients to two antihypertensive regimens: amlodipine followed by perindopril and doxazosin, or atenolol followed by bendroflumethiazide and doxazosin; all agents were available in two strengths [45]. Antihypertensive cost was an average of the two regimens, assuming patients took higher-strength amlodipine or atenolol, and that patients were evenly distributed between doses of the other drugs. Statin costs were based on ASCOT–LLA regimen of 10 mg atorvastatin [5].

All costs were defined as distributions and sampled every cycle. For pharmacy costs, triangular distributions were created as WAC ± 20%, chosen for upper costs corresponding to the average wholesale price [43]. For Medicare-based costs, triangular distributions were created around the means with bounds at 25th and 75th reimbursement percentiles. Distributions were defined for all other event-related costs in the same proportion.

Medication Adherence and Persistence

The approach to modeling use of antihypertensive and statin medications differed across the no treatment, ideal adherence, and real-world adherence scenarios. The no-treatment model projected the natural history of CHD and stroke in the absence of treatment and formed the foundation for the other models.

The ideal adherence scenario extended the no-treatment scenario by adding antihypertensive and statin therapy and the associated relative event risk reductions and costs. It was assumed that adherence rates observed in clinical trials represented the gold-standard (ideal) adherence for a real-world practice setting. To model adherence benefits, predicted event risks were multiplied by relative risk reductions expected with full adherence to antihypertensive and statin therapy. In ASCOT–LLA, there was no placebo group; the relative risk for patients taking antihypertensive and lipid-lowering therapy compared with those taking only antihypertensives was 0.64 for MI or CHD death and 0.73 for stroke over the 3.5-year follow-up [5]. To determine the relative risk under ideal adherence to antihypertensives and statins compared with patients on no treatment, we multiplied the ASCOT–LLA effect estimates by relative risks from a meta-analysis of placebo-controlled trials of antihypertensive therapy [1], resulting in relative risks of 0.56 for MI or CHD death, and 0.50 for stroke. Although ASCOT–LLA reported risk reductions for angina alone, these risk reductions were unavailable from meta-analyses. Thus, risk reductions with therapy for angina were assumed to follow those for MI or CHD death.

Pharmacy costs in the ideal adherence scenario were equal to 1 year's therapy multiplied by the proportion of days covered (PDC) observed in the BPLA of ASCOT, ASCOT–BPLA [45] (for antihypertensives), and in ASCOT–LLA [5,6] (for statins).

The real-world adherence scenario was similar to the ideal adherence scenario but added real-world adherence data and costs and benefits associated with each level of adherence. For simplicity, we defined three possible levels of adherence and allowed patients to transition among them over time: adherent (PDC ≥ 80%); partially adherent (20% ≤ PDC < 80%); and nonadherent (PDC < 20%) [18]. Adherence status was treated independently for antihypertensives and statin therapy, resulting in nine possible adherence states. All nondead health states from the natural history model were subdivided into these nine states to account for adherence status.

Initial adherence status and transitions among categories of adherence were drawn from an analysis of prescription and medical claims data from Medi-Cal. A sample of 8277 patients initiating antihypertensive and statin therapies within a 180-day period were extracted, using first prescription claim date as the index date. Patients were aged ≥40 years, with a diagnosis of hypertension, and no history of CHD, stroke, or diabetes in the 6 months prior to the index date. Patients were followed for up to 6 years or until death or data stream termination. An adherence transition matrix was determined empirically for each year of therapy. The initial and final adherence status distributions and transition probabilities for years 1 to 2 are reported, as an example, in Table 2. The transition probabilities for year 6 and beyond equaled those for years 5 to 6, as adherence had stabilized at this point.

Table 2.  Adherence status transition probabilities for years 1 to 2
  Year 2 adherence*
  • *

    Adherence status defined as antihypertensive adherence/lipid-lowering adherence.

Year 1 adherence*Proportion of patients0.

Annual event risks predicted by Framingham equations were multiplied by the relative risk for each adherence category (Table 3). Relative risk reductions for patients adherent to both antihypertensive and statin therapy equaled those used in the ideal adherence scenario. For patients fully adherent to one medication and nonadherent to the other, relative risk reductions were based on meta-analyses of placebo-controlled trials of antihypertensive [1] and statin [2–4] therapies.

Table 3.  Relative risks of coronary heart disease and stroke by adherence status
Antihypertensive adherenceStatin adherenceRelative risks (95% CI)Calculation
  1. Calculation A: Relative risk of (Full–Full) to (Full–Non) [5] times the mean of the relative risk of (Full–Non) to (Non–Non) [1] and the relative risk of (Non–Full) to (Non–Non) [2,3]. Confidence interval calculated as proportionate to wider of two source confidence intervals [5].

  2. Calculation B: Mean of (Full–Full) and (Full–Non).

  3. Calculation C: Relative risks and confidence intervals as reported in antihypertensive meta-analysis [1].

  4. Calculation D: Mean of (Full–Full) and (Non–Full).

  5. Calculation E: Mean of (Full–Full), (Full–Non), (Non–Full), and (Non–Non).

  6. Calculation F: Mean of (Full–Non) and (Non–Non).

  7. Calculation G: Relative risks and confidence intervals as reported in meta-analyses of statins in CHD prevention [2] and stroke prevention [3].

  8. Calculation H: Mean of (Non–Full) and (Non–Non).

  9. Calculation I: Referent.

  10. CHD, coronary heart disease; CI, confidence interval.

FullFull0.50 (0.39–0.65)0.54 (0.41–0.71)A
FullNon0.87 (0.80–0.94)0.68 (0.61–0.76)C
NonFull0.69 (0.64–0.74)0.79 (0.73–0.85)G

As data were lacking for the effectiveness achieved by partially adherent patients, a linear association between adherence and efficacy was assumed in the base-case analysis. The efficacy of partial adherence was thus derived as the midpoint of known relative risks. For example, the relative risk of stroke for patients fully adherent to antihypertensive therapy, and partially adherent to statin therapy (fully adherent/partially adherent) equaled the mean relative risk for fully adherent/nonadherent (0.68) and fully adherent/fully adherent (0.54), or 0.61) (see Table 3 for explanatory calculations). Pharmacy costs in this scenario equaled 1-year therapy costs multiplied by the actual PDC. As parameters in the model, PDCs were defined as β-distributions derived from observed means and SDs. For antihypertensives, mean ± SD PDC for: full adherence was 94.1 ± 6.0%; partial adherence, 52.9 ± 17.2%; and nonadherence, 12.0 ± 4.8%. For statins, PDC for: full adherence was 92.5 ± 6.2%; partial adherence, 50.3 ± 17.4%; and nonadherence, 11.9 ± 4.7%.


Monte Carlo microsimulations were performed for each scenario, and events, costs, and patient-years were recorded. The model considered a lifetime horizon, such that each simulation ran until death. Each model ran 10 times with 50,000 trials (patients) to enable variability analyses.

As a primary analysis, the incremental cost-effectiveness ratio (ICER) per life-year gained was calculated between the ideal and real-world adherence scenarios. Additional analyses included the ICER per life-year gained between the real-world adherence and no-treatment scenarios, and calculations of ICERs per event avoided. Mean ICER and the 95% confidence interval (CI) were calculated for each analysis.

Additional simulations assessed the significance of risk reduction associated with partial adherence. The base-case analysis assumed that partial adherence yielded 50% of the effectiveness of full adherence. The percentage of full adherence effectiveness seen with partial adherence was varied from 0% (no effectiveness, equivalent to no treatment) to 100% (full effectiveness, equivalent to full adherence). Two-way sensitivity analyses were conducted to examine the influence of partial adherence effectiveness when segmenting by starting age.


  1. Top of page
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

Base-Case Analysis

Baseline model patient characteristics approximated those of the ASCOT–LLA participants, except mean age (58.7 years vs. ASCOT–LLA: 63.1 years), SBP (150.7 mm Hg vs. ASCOT–LLA:164.2 mm Hg), total cholesterol (199.8 mg/dl vs. ASCOT–LLA 5.5 mmol/L [211.5 mg/dl]), and HDL cholesterol (48.8 mg/dl vs. ASCOT–LLA 1.3 mmol/L [50.0 mg/dl]), all of which were lower in the model than in ASCOT–LLA by <10% (Table 4). This resulted in slightly lower population-level event risks.

Table 4.  Patient characteristics at baseline
Patient characteristicMean (across all treatment scenarios)*
  • *

    Includes no treatment, ideal adherence, and real-world adherence.

  • Excludes coronary heart disease and stroke.

  • Data are shown as % or mean (standard deviation).

  • CVD, cardiovascular disease; HDL, high-density lipoprotein; LVH-ECG, left ventricular hypertrophy by electrocardiogram.

Age (years)58.7(10.3)
Alcohol consumption (oz/week)12.2(7.6)
Systolic blood pressure (mm Hg)150.7(8.8)
Total cholesterol (mg/dl)199.8(29.8)
HDL cholesterol (mg/dl)48.8(13.5)
Male (%)81.1 
Current smoker (%)33.2 
Diabetes (%)24.3 
LVH-ECG (%)14.5 
Atrial fibrillation (%)2.8 
History of CVD (%)8.6 

Mean life expectancy, discounted by 3% per year, was 14.73 years (95% CI 14.71–14.75) for the no-treatment scenario, 15.07 years (95% CI 15.06–15.08) for real-world adherence, and 15.49 years (95% CI 15.47–15.51) for ideal adherence. Mean frequency of events per patient were 0.738 (95% CI 0.737–0.740) for the no-treatment scenario, 0.610 (95% CI 0.608–0.612) for real-world adherence, and 0.441 (95% CI 0.440–0.442) for ideal adherence.

Effectiveness and cost-effectiveness results are provided in Table 5. Lifetime discounted direct medical costs per patient were $12,831 (95% CI 12,785–12,877) for the no-treatment scenario, $23,295 (95% CI 23,270–23,321) for real-world adherence, and $32,492 (95% CI 32,462–32,522) for ideal adherence. Compared with real-world adherence, ideal adherence was associated with an additional $9197 and 0.42 patient-years, giving an ICER of $22,121 per life-year gained (95% CI 21,067–23,176). In the ideal adherence scenario, there were 0.17 fewer events per patient than in the real-world adherence scenario, yielding an ICER per event avoided of $54,364 (95% CI 53,493–55,234).

Table 5.  Incremental cost-effectiveness ratios over lifetime
Adherence scenarioAverageIncrementalICER (95% CI)
  • *

    Discounted by 3% annually.

  • All costs 2006 US$.

  • CI, confidence interval; ICER, incremental cost-effectiveness ratio.

No treatment12,83114.73
No treatment12,8310.74

Compared with no treatment, real-world adherence was associated with an additional $10,464 and 0.35 patient-years, giving an ICER of $30,586 per life-year gained (95% CI 28,742–32,431). Patients in the real-world adherence scenario incurred 0.13 fewer events than those in no treatment, giving an ICER of $81,591 per event avoided (95% CI 80,353–82,830).

Risk Reduction with Real-World Adherence

The ICER per life-year gained for ideal versus real-world adherence was modestly sensitive to assumptions about the drugs’ effectiveness under partial adherence. When partial adherence was assumed to yield 0% of full effectiveness, the ICER decreased from $22,121 (base-case) to $16,153 (95% CI 15,664–16,642). Conversely, when partial adherence was assumed to yield 100% of full effectiveness, the ICER increased to $32,529 (95% CI 29,854–35,205). The ICER per event avoided between the ideal and real-world adherence scenarios showed similar sensitivity, varying between $41,005 (95% CI 40,000–42,011) and $76,763 (95% CI 75,272–78,254).

Figure 2 shows ICER per life-year gained between the ideal and real-world adherence scenarios as a function of effectiveness in the partial-adherence state and patients’ age at baseline. ICERs were lower for older patients, from $11,214 (95% CI 10,530–11,897) to $24,934 (95% CI 23,346–26,522) among patients aged ≥65 years, and $19,144 (95% CI 18,210–20,079) to $35,532 (95% CI 32,333–38,731) among patients aged 45 to 54 years. Figure 3 shows the same analysis of ICER per event avoided between ideal and real-world adherence scenarios. As with life-years gained, ICERs per event avoided were lower with increasing age, from $34,636 (95% CI 32,665–36,607) to $70,915 (95% CI 67,614–74,215) among patients aged ≥65 years, and $43,875 (95% CI 42,550–45,200) to $79,635 (95% CI 75,205–84,065) among patients aged 45 to 54 years.


Figure 2. Incremental cost-effectiveness ratio per life-year gained for ideal versus real-world adherence for varying partial adherence effectiveness. USD, United States Dollars.

Download figure to PowerPoint


Figure 3. Incremental cost-effectiveness ratio per event avoided for ideal versus real-world adherence scenarios with varying partial adherence effectiveness. USD, United States Dollars.

Download figure to PowerPoint


  1. Top of page
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

Nonadherence to concomitant antihypertensive and statin therapy undermines the drugs’ potential effectiveness in preventing cardiovascular events. We estimated the population-level events and lost life-years attributable to nonadherence and, consequently, the reduction in events and increase in life expectancy that could be realized through improved adherence.

Among a hypertensive population with multiple cardiovascular risk factors, real-world adherence to antihypertensive and statin therapy added 4.2 months to life expectancy compared with no treatment. Assuming ideal adherence as seen in clinical trials, the life expectancy gain versus no treatment was 9.2 months. Accordingly, real-world adherence foregoes ∼54% of life expectancy gained from ideal adherence. For a population of 10,000 patients, 6100 lifetime CHD and stroke events would be expected at real-world adherence. At ideal levels of adherence, 1700 of those events (28%) could be avoided.

Because lower adherence also reduced overall treatment costs, it is important to translate these findings into terms that consider both costs and treatment benefits. At $30,586 and $22,121 per life-year gained, we found treatment at both real-world and ideal adherence to be cost-effective by conventional standards [46]. Moreover, our analytic framework can be used to evaluate the cost-effectiveness of adherence interventions: from a payer's perspective, increasing adherence to ideal levels with a program costing up to ∼$8400 per patient would be as cost-effective as initiating this number of patients on dual-therapy at real-world adherence, while also conferring absolute benefits in life expectancy and event reduction.

Medication effectiveness at partial adherence levels was the greatest source of uncertainty in the models. Our sensitivity analyses showed, however, that even in the case where partial adherence yields the same benefits as full adherence, there exist cost-effective, population-level gains to be made through increased adherence. These potential gains increase with decreasing levels of partial adherence effectiveness, as the benefit gap between the ideal and real-world adherence scenarios widens. The modeling methodology employed in the present study is an important innovation because of the biases that are inherent in post hoc analyses when estimating the effects of adherence in clinical trials [47]. In the present article, we have tried to avoid such biases by constructing upper and lower bounds of effectiveness based on the treatment and placebo groups of trials, then adjusted for nonadherence using an “interpolation” between these groups based on real-world levels of adherence.

Our analysis must be interpreted in light of its limitations, including assumptions on adherence rates after a CHD event, the short-term interaction of adherence and effectiveness, and extrapolations of long-term effectiveness of antihypertensive and statin therapy. Adherence data were derived from the California Medicaid (Medi-Cal) system, which may not represent other segments of the population. We were unable to estimate the emergency cost of stroke events in the United States, which caused us to underestimate one aspect of the cost of nonadherence and thus to underestimate the cost-effectiveness of improving adherence to ideal levels. Conversely, amlodipine lost patent protection in the United States in 2007 and can now be acquired at a lower cost than the 2006 figure in our model. Amlodipine, however, was one of only five antihypertensives used by patients in the model, so the impact of this price change would not have affected the average daily WAC of antihypertensive therapy significantly. To the extent that we overestimated the cost of antihypertensive therapy, our results are conservative in that the cost per event prevented is actually lower and the cost-effectiveness of improving adherence is improved.

Assessing adherence based on prescription refill rates and the PDC may overestimate actual dosing accuracy because it assumes that patients take all of the medications for prescriptions that are filled. Prescription refill ratesl, however, have been shown to be closely associated with other measures of adherence, including serum drug levels and physiologic drug effects [48]. A given day was assumed to be covered if any drug for the indication of interest was available. Such an approach is likely to be accurate for lipid-lowering therapy, which in most patients consists of statins alone. For the treatment of hypertension, however, use of multiple drug regimens is common and we may have overestimated adherence with this method.

Our model did not consider the cost of an intervention that would be required to improve adherence; this cost would directly increase the ICER of improved adherence and thus, future studies are needed to consider the cost-effectiveness of specific adherence interventions. The model also assumed that a change in adherence resulted in a change in event risk within 1 year, as patients were assigned relative event risks based on current adherence status. ASCOT–LLA reported that the risk-reduction benefit of dual therapy, compared with antihypertensive monotherapy, emerged within 1 year [5], suggesting that the gains from ideal adherence may be recognized within the models’ 1-year cycles. The duration of effects following discontinuation of or reduced adherence to medication is unknown, however, and patients may continue to see benefits after stopping therapy. In this case, the real-world adherence scenario may have underestimated population risk reduction because most patients’ adherence deteriorates over successive model cycles. A final limitation is the coupling of shorter-term data with a lifetime horizon. Simulated model patients survived 15 years on average, longer than the duration of any trial from which effectiveness estimates were drawn. A clinical trial of statins supported this assumption by demonstrating that risk reductions persisted in older patients [49].

In summary, we found that in patients with hypertension and multiple cardiovascular risk factors, more than half of the potential benefits from antihypertensive and statin therapies are lost to poor adherence. Antihypertensive and statin therapies are cost-effective at adherence levels seen in real-world settings but are more cost-effective when taken at levels seen in clinical trials. Depending on its cost, an intervention that improved adherence from typical to trial levels has the potential to be an attractive use of health-care resources.


  1. Top of page
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References

This study was sponsored by Pfizer Inc. Editorial support was provided by Karen Burrows of Envision Pharma and funded by Pfizer Inc.

Source of financial support: This study was sponsored by Pfizer Inc.


  1. Top of page
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  • 1
    Psaty BM, Lumley T, Furberg CD, et al. Health outcomes associated with various antihypertensive therapies used as first-line agents: a network meta-analysis. JAMA 2003;289:253444.
  • 2
    LaRosa JC, He J, Vupputuri S. Effect of statins on risk of coronary disease: a meta-analysis of randomized controlled trials. JAMA 1999;282:23406.
  • 3
    Amarenco P, Labreuche J, Lavallée P, Touboul PJ. Statins in stroke prevention and carotid atherosclerosis: systematic review and up-to-date meta-analysis. Stroke 2004;35:29029.
  • 4
    Baigent C, Keech A, Kearney PM, et al. Efficacy and safety of cholesterol-lowering treatment: prospective meta-analysis of data from 90,056 participants in 14 randomised trials of statins. Lancet 2005;366:126778.
  • 5
    Sever PS, Dahlöf B, Poulter NR, et al. Prevention of coronary and stroke events with atorvastatin in hypertensive patients who have average of lower-than-average cholesterol concentrations, in the Anglo-Scandinavian Cardiac Outcomes Trial–Lipid-Lowering Arm (ASCOT–LLA): a multicentre randomized controlled trial. Lancet 2003;161:114958.
  • 6
    Lindgren P, Buxton M, Kahan T, et al. Cost-effectiveness of atorvastatin in the prevention of coronary and stroke events: an economic analysis of the Anglo-Scandinavian Cardiac Outcomes Trial–Lipid-Lowering Arm (ASCOT–LLA). Eur J Cardiovasc Prev Rehabil 2005;12:2936.
  • 7
    Shepherd J. Economics of lipid lowering in primary prevention: lessons from the West of Scotland Coronary Prevention Study. Am J Cardiol 2001;87:1922B.
  • 8
    Prosser LA, Stinnett AA, Goldman PA, et al. Cost-effectiveness of cholesterol-lowering therapies according to selected patient characteristics. Ann Intern Med 2000;132:76979.
  • 9
    Tsevat J, Kuntz KM, Orav EJ, et al. Cost-effectiveness of pravastatin therapy for survivors of myocardial infarction with average cholesterol levels. Am Heart J 2001;141:72734.
  • 10
    Hay JW, Yu WM, Ashraf T. Pharmacoeconomics of lipid-lowering agents for primary and secondary prevention of coronary artery disease. Pharmacoeconomics 1999;15:4774.
  • 11
    Grover SA, Coupal L, Paquet S, Zowall H. Cost-effectiveness of 3-hydroxy-3-methylglutaryl coenzyme A reductase inhibitors in the secondary prevention of cardiovascular disease: forecasting the incremental benefits of preventing coronary and cerebrovascular events. Arch Intern Med 1999;159:593600.
  • 12
    Ganz DA, Kuntz KM, Jacobson GA, Avorn J. Cost-effectiveness of 3-hydroxy-3-methylglutaryl coenzyme A reductase inhibitor therapy in older patients with myocardial infarction. Ann Intern Med 2000;132: 7807.
  • 13
    Franco OH, Peeters A, Looman CW, Bonneux L. Cost-effectiveness of statins in coronary heart disease. J Epidemiol Community Health 2005;59:92733.
  • 14
    Cramer JA, Roy A, Burrell A, et al. Medication compliance and persistence: terminology and definitions. Value Health 2008;11:447.
  • 15
    Kramer JM, Hammill B, Anstrom KJ, et al. National evaluation of adherence to beta-blocker therapy for 1-year after acute myocardial infarction in patients with commercial health insurance. Am Heart J 2006; 152:4548.
  • 16
    Cardinal H, Monfared AA, Dorais M, LeLorier J. A comparison between persistence to therapy in ALLHAT and in everyday clinical practice: a generalizability issue. Can J Cardiol 2004;20:41721.
  • 17
    Van Wijk BL, Klungel OH, Heerdink ER, De Boer A. Rate and determinants of 10-year persistence with antihypertensive drugs. J Hypertens 2005;23:21017.
  • 18
    Benner JS, Glynn RJ, Mogun H, et al. Long-term persistence in use of statin therapy in elderly patients. JAMA 2002;288:45661.
  • 19
    Caspard H, Chan AK, Walker AM. Compliance with a statin treatment in a usual-care setting: retrospective database analysis over 3 years after treatment initiation in health maintenance organization enrollees with dyslipidemia. Clin Ther 2005;27:163946.
  • 20
    Perreault S, Blais L, Lamarre D, et al. Persistence and determinants of statin therapy among middle-aged patients for primary and secondary prevention. Br J Clin Pharmacol 2005;59:56473.
  • 21
    Jackevicius CA, Mamdani M, Tu JV. Adherence with statin therapy in elderly patients with and without acute coronary syndromes. JAMA 2002;288:4627.
  • 22
    Chapman RH, Benner JS, Petrilla AA, et al. Predictors of adherence with antihypertensive and lipid-lowering therapy. Arch Intern Med 2005;161:114752.
  • 23
    Blackburn DF, Dobson RT, Blackburn JL, et al. Adherence to statins, beta-blockers and angiotensin-converting enzyme inhibitors following a first cardiovascular event: a retrospective cohort study. Can J Cardiol 2005;21:4858.
  • 24
    Gold MR, Slegel JE, Russell LB, Weinstein MC. Cost-Effectiveness in Health and Medicine. New York: Oxford University Press, 1996.
  • 25
    D’Agostino RB, Russell MW, Huse DM, et al. Primary and subsequent coronary risk appraisal: new results from the Framingham Study. Am Heart J 2000;139:27281.
  • 26
    D’Agostino RB, Wolf PA, Belanger AJ, Kannel WB. Stroke risk profile: adjustment for antihypertensive medication. Stroke 1994;25:403.
  • 27
    Hardie K, Hankey GJ, Jamrozik K, et al. Ten-year risk of first recurrent stroke and disability after first-ever stroke in the Perth Community Stroke Study. Stroke 2004;35:7315.
  • 28
    Wolf PA, D’Agostino RB, Belanger AK, Kannel WB. Probability of stroke: a risk profile from the Framingham Study. Stroke 1991;22:3128.
  • 29
    Centers for Disease Control and Prevention, National Center for Health Statistics. National health and nutrition examination survey data, 2003–2004. Available from:[Accessed October 4, 2007.
  • 30
    The North American Menopause Association (NMAS). About menopause, supplement. Available from:[Accessed November 12, 2007.
  • 31
    Arias E. United States Life Tables, 2002. National Vital Statistics Reports. Vol 53, no. 6. Hyattsville, MD: National Center for Health Statistics, 2004.
  • 32
    Hoyert DL, Kung HC, Smith BL. Deaths: Preliminary Data for 2003. National Vital Statistics Reports. Vol 53, no. 15. Hyattsville, MD: National Center for Health Statistics, 2005.
  • 33
    Furman MI, Dauerman HL, Goldberg RJ, et al. Twenty-two year (1975 to 1997) trends in the incidence, in-hospital and long-term case fatality rates from initial Q-wave and non-Q-wave myocardial infarction: a multi-hospital, community-wide perspective. J Am Coll Cardiol 2001;37:157180.
  • 34
    Goldberg RJ, McCormick D, Gurwitz JH, et al. Age-related trends in short- and long-term survival after acute myocardial infarction: a 20-year population-based perspective (1975–1995). Am J Cardiol 1998; 82:13117.
  • 35
    Dennis MS, Burn JP, Sandercock PA, et al. Long-term survival after first-ever stroke: the Oxfordshire Community Stroke Project. Stroke 1993;24:796800.
  • 36
    Russell MW, Huse DM, Drowns S, et al. Direct medical costs of coronary artery disease in the United States. Am J Cardiol 1998;81:11105.
  • 37
    US Department of Health and Human Services, Centers for Medicare and Medicaid Services. Diagnosis related group weights (2005). Available from:[Accessed January 4, 2008.
  • 38
    US Department of Health and Human Services, Centers for Medicare and Medicaid Services. Impact file for prospective payment system. Available from:[Accessed October 4, 2007.
  • 39
    US Department of Health and Human Services, Centers for Medicare and Medicaid Services. CMS-1428-F(2) published in Federal Register August 11, 2004. Available from:[Accessed January 20, 2008.
  • 40
    Miller ME, Welch WP. Analysis of Hospital Medical Staff Volume Performance Standards: Technical Report. Washington, DC: The Urban Institute, 1993.
  • 41
    US Department of Health and Human Services, Centers for Medicare and Medicaid Services. National physician fee schedule. Available from:[Accessed January 4, 2008.
  • 42
    Samsa GP, Bian J, Lipscomb J, Matchar DB. Epidemiology of recurrent cerebral infarction: a medicare claims-based comparison of first and recurrent strokes on 2-year survival and cost. Stroke 1999;30:33849.
  • 43
    Redbook 2006 August Update. Montvale, NJ: Thomson Micromedex, 2006.
  • 44
    US Department of Labor, Bureau of Labor and Statistics. US medical care consumer price index, all urban consumers. CUUR0000SAM. Available from:[Accessed February 4, 2008.
  • 45
    Dahlöf B, Sever PS, Poulter NR, et al. Prevention of cardiovascular events with an antihypertensive regimen of amlodipine adding perindopril as required versus atenolol adding bendroflumethiazide as required, in the Anglo-Scandinavian Cardiac Outcomes Trial–Blood Pressure Lowering Arm (ASCOT–BPLA): a multicentre randomised controlled trial. Lancet 2005;366:895906.
  • 46
    Tengs TO, Adams ME, Pliskin JS, et al. Five-hundred life-saving interventions and their cost-effectiveness. Risk Anal 1995;15:36990.
  • 47
    Coronary Drug Project Research Group. Influence of adherence to treatment and response of cholesterol on mortality in the Coronary Drug Project. N Engl J Med 1980;303:103841.
  • 48
    Steiner JF, Prochazka AV. The assessment of refill compliance using pharmacy records: methods, validity, and applications. J Clin Epidemiol 1997;50:10516.
  • 49
    Shepherd J, Blauw GJ, Murphy MB, et al. Pravastatin in elderly individuals at risk of vascular disease (PROSPER): a randomised controlled trial. Lancet 2002;360:162330.