Abstract
Background
Coronary artery calcification (CAC) is a well-established risk factor for the occurrence of adverse ischemic events. However, the economic impact of the presence of CAC is unknown.
Objectives
Through an economic model analysis, we sought to estimate the incremental impact of CAC on medical care costs and patient mortality for de novo percutaneous coronary intervention (PCI) patients in the 2012 cohort of the Medicare elderly (≥ 65) population.
Methods
This aggregate burden-of-illness study is incidence-based, focusing on cost and survival outcomes for an annual Medicare cohort based on the recently introduced ICD9 code for CAC. The cost analysis uses a one-year horizon, and the survival analysis considers lost life years and their economic value.
Results
For calendar year 2012, an estimated 200,945 index ( de novo ) PCI procedures were performed in this cohort. An estimated 16,000 Medicare beneficiaries (7.9%) were projected to have had severe CAC, generating an additional cost in the first year following their PCI of $3500, on average, or $56 million in total. In terms of mortality, the model projects that an additional 397 deaths would be attributable to severe CAC in 2012, resulting in 3770 lost life years, representing an estimated loss of about $377 million, when valuing lost life years at $100,000 each.
Conclusions
These model-based CAC estimates, considering both moderate and severe CAC patients, suggest an annual burden of illness approaching $1.3 billion in this PCI cohort. The potential clinical and cost consequences of CAC warrant additional clinical and economic attention not only on PCI strategies for particular patients but also on reporting and coding to achieve better evidence-based decision-making.
Highlights
- •
The economic impact of the presence of coronary artery calcification is unknown.
- •
We estimated the incremental impact of moderate and severe coronary artery calcification on medical care costs and patient mortality for percutaneous coronary intervention patients in the 2012 cohort of the Medicare elderly (≥ 65) population.
- •
Considering both moderate and severe coronary artery calcification, the associated burden of illness is approximatively $1.3 billion per year.
1
Introduction
Coronary artery calcification (CAC) is both an established risk factor for poor cardiovascular clinical outcomes and a predictor of additional resource utilization and overall health care costs . Although most percutaneous coronary intervention (PCI) trials have excluded patients with either moderately- or severely-calcified coronary lesions, Genereux and colleagues’ recent analysis of pooled data from the HORIZONS-AMI and ACUITY trials demonstrates that patients with moderate and severe coronary calcification experience worse ischemic outcomes – stent thrombosis, target lesion revascularization (TLR), and mortality – after PCI compared to patients with no CAC .
Less is known about the population-level burden of coronary calcification. Prior to 2011, clinically significant CAC lacked a specific ICD9 diagnosis code, an important administrative data element needed for conducting a population-level, epidemiologically-based analysis. Although a new diagnostic code (ICD9 414.4 for coronary calcification) was introduced late in 2011, actual documentation of CAC via administrative coding practice has lagged. This is the first published analysis to examine the reporting and use of this ICD9 code and to assess the clinical and cost burden of calcification in de novo PCI patients in the elderly (age 65 and older) Medicare population. In addition, we estimated the economic burden of CAC in this select Medicare PCI population in 2012 by addressing this question: What is the estimated incremental impact of CAC – at both the PCI patient-level and in the aggregate – on associated medical care costs and patient mortality in the 2012 cohort of the Medicare elderly population?
2
Methods
2.1
Study design
The target study population was the Medicare elderly with atherosclerosis in calendar year 2012 experiencing a new ( de novo ) index PCI, defined as a patient receiving a coronary angiogram with no prior coronary revascularization in the preceding six months. This was an aggregate, population-level economic burden study, which is also known as a “cost-of-illness” or “burden-of-disease” study . The study design was incidence-based, focusing on cost and survival outcomes for an annual cohort of the target study population. The horizon of the cost analysis was one year because the greatest cost impact tends to occur in the first year post-procedure, and due to the limited data for these patients. The horizon for survival analysis was a patient’s lifetime, given that mortality differences at one year can be modeled as life years lost over the remaining expected lifetime.
CAC could potentially adversely affect patients’ clinical outcomes for those who underwent PCI, coronary artery bypass graft surgery (CABG), or prescribed medical therapy for severe atherosclerosis deemed not treatable via surgical intervention. Of coronary angiography patients without a revascularization in the last 6 months, 5% underwent CABG (3.1% documented calcification), 31% underwent PCI (1.9% documented calcification), and 64% underwent a medical intervention (1.5% documented calcification). However, the focus of this analysis was on patients receiving PCI, due to the limited number of CABG and medical therapy patients documented with the calcification code in the available Medicare data.
The study design and analysis were influenced by the availability of CAC data. A new diagnostic code for calcification (ICD9 414.4) was introduced in the last quarter of 2011. Because the reporting of new codes can take time to become a part of regular medical documentation practice, there is a high likelihood of under-reporting of CAC in this initial period. This limited reporting affected the ability to do a strict epidemiologically-based comparison of patients with reported calcification versus those with non-calcified coronary lesions, given that such an approach would be subject to misclassification bias in which many patients with severe calcification would be inappropriately classified in the non-CAC group. Therefore, an aggregate estimate was constructed using an economic modeling approach that synthesizes data from multiple sources described below.
The principal data sources for cost and survival analyses were Medicare’s Standard Analytic Files (SAFs). The SAFs comprise seven data sets containing detailed claims information about health care services rendered to Medicare beneficiaries in fee-for-service (FFS) Medicare. SAFs are available for institutional (inpatient, outpatient, skilled nursing facility, hospice, or home health agency) and non-institutional (physician and durable medical equipment providers) claim types. Data are organized at the claim level and include basic beneficiary demographic information, date of service, diagnosis and procedure code, provider number, and reimbursement amount. Two SAF databases were used: the Medicare 5% random sample SAF and 100% SAF. The 5% sample of beneficiaries includes all of their claims (inpatient, outpatient, physician, durable medical equipment, etc.) except drugs, which are tracked and reported separately via Medicare Part D. The 100% SAF includes only inpatient and outpatient claims but includes all FFS beneficiaries. In addition, a special sub-sample was defined from the 100% SAF that included all hospitals (n = 17) that coded more than 10% of their PCI patients as having CAC (using code 414.4).
The Medicare SAF analyses were conducted with Limited Data Set (LDS) files, which encrypt beneficiary identifiers, and this research complies with the Centers for Medicare and Medicaid Services Data Use Agreement rules on blinding and data use. Institutional Review Board approval is not required for use of LDS administrative data under the HIPAA Privacy Rule.
To better estimate the degree of calcification and incidence of major adverse cardiovascular events among this target patient group, additional data from the HORIZONS-AMI/ACUITY pooled sample were used, though this dataset does not contain information on costs . Approval for analyses of this pooled sample was obtained from the Institutional Review Boards or Ethics Committees at each of the enrolling sites.
2.2
Economic model: cost burden of coronary calcification
A population-level economic model was constructed drawing parameter estimates from both a detailed analysis of the available Medicare claims data and from the published literature. A modeling approach also allowed for assessing the sensitivity of the results to varying assumptions about key parameters.
As depicted in Fig. 1 , the economic model had three major components: 1) the estimated annual incidence of elderly Medicare patients receiving an index PCI (typically a drug-eluting stent); 2) the incidence of CAC in these PCI patients, classified as severe, moderate, or mild/none; and 3) the estimated impact of the three levels of CAC on Medicare per-patient costs and health outcomes (viz., mortality).
2.3
Incidence of PCI in 2012
The 100% Medicare SAFs of administrative claims data were used to define the index population in 2012, which includes only patients in the Medicare FFS population. The cohort was restricted to elderly patients (i.e., age 65 or older). Although Medicare Advantage (MA) patients represented about 21.7% of the Medicare population in 2012- and rising substantially annually – these data are not included in this database. To account for MA patients, we made a simple, crude adjustment, increasing the aggregate burden estimate for the whole population upward to reflect the inclusion of this subpopulation.
2.4
Incidence of calcification in the elderly
We reported incidence of CAC in the Medicare population based on the 414.4 code. However, given the likely under-reporting of CAC in the 2012 Medicare SAF cohort, the estimated incidence of moderate and severe CAC was based on estimates from the HORIZONS-AMI/ACUITY pooled sample . The estimated means for the elderly subsample were used in the base case. The range for the assumption was based on the literature . Calcification incidence from the HORIZONS-AMI/ACUITY sample was multiplied by the Medicare population to determine population size for CAC.
2.5
Medical care cost impact
Estimation of the first-year impact of calcification following index PCI on costs was based on a generalized estimating equation (GEE) regression analysis of data from the Medicare 5% Standard SAFs, which were used because they contain the most complete cost data including physician claims—adjusting for age, gender, and the Charlson comorbidity index . The Charlson index was used rather than individual comorbidities to decrease the effect of correlation and increase power of the models. As previously noted, an adjustment for the MA patients was applied in the aggregate population model. In the aforementioned available data, approximately 20% of patients had one-year of follow-up data. Techniques developed by Lin et al. and Basu et al. were applied to address censoring in order to fully utilize patient cost data up to the point of loss-to-follow-up or death.
2.6
Mortality impact
We reported the impact of CAC on mortality using the 100% Medicare SAF. However, given the likely under-reporting of CAC in the 2012 Medicare SAF cohort, the estimated mortality impact of moderate and severe CAC was based on new tabulations from the HORIZONS-AMI/ACUITY pooled sample, an in-depth analysis of cardiac death in the sample studied by Genereux et al. . Cardiac death was used, as all causes of mortality were not available and cardiac death was assumed to be most relevant. A Cox proportional hazards model was used to account for censoring with adjustment for age, gender, the Charlson comorbidity index, and the comorbidities end-stage renal disease and diabetes, which were included explicitly in order to evaluate their specific effects due to their potential key role in the causal framework. The impact of the mortality differential on the health outcomes for the 2012 cohort was assessed both in terms of lost life-years and the associated monetized value of those lost life-years. For the life expectancy calculation, the estimate of lost life years was based on the median age and gender of an elderly PCI patient (i.e., a 74-year-old male) who would normally have an expected lifetime of 11.5 years: this was then adjusted down to 9.5 years after standard discounting at 3% per annum. A sensitivity analysis scenario used a mortality differential 25% higher than the base case as a ‘high case’ for mortality, and a ‘low case’ sensitivity analysis conservatively assumed no survival difference between CAC and non-CAC PCI patients.
2.7
Aggregate burden-of-illness model
In order to assess the overall burden of CAC, we aimed to include both true economic costs as well as the effects on health. Fig. 1 illustrates how the economic model of the burden of illness integrated each of the components above to produce an aggregate estimate of the economic burden. The first component was the incidence of PCI by age and gender. The second component represented the degree of CAC within these age and gender groups.
The third component was patient outcomes with two dimensions: cost and mortality impacts, which were included in the model in two capacities. First, the attributed medical cost burden was the product of the number of patients with CAC and the expected average attributable annual cost for patients in the 2012 cohort receiving an index PCI procedure. Second, to combine health and economic effects, we converted health impact into a monetary value. We utilized a method valuing life-years according to willingness-to-pay thresholds, as discussed in the recently published ACC/AHA Task Force on value and cost in guideline development . For example, if the societal willingness-to-pay for one year of healthy life is $100,000, we valued a year of healthy survival in the model at that amount. In this way we could report a singular economic impact rather than separate economic and health impacts. The willingness-to-pay threshold has the limitations both of being subject to societal acceptance, as well as being difficult to estimate with a high degree of certainty. Accordingly, we varied the willingness-to-pay threshold at three levels in order to address societal uncertainties in the threshold. Estimates of societal willingness to pay for life year gains were based on the Andersen et al. thresholds for “levels of value”—high, medium, and low : the base case assumed $100,000 per life year gained, while the lower bound was $50,000 and the upper bound was $150,000.
For all inputs, the model was evaluated at a base case, based on the results of the analyses, and at low and high values in the scenario sensitivity analyses based on the expert judgment of the co-authors as to the degree of uncertainty and plausible variation.
2
Methods
2.1
Study design
The target study population was the Medicare elderly with atherosclerosis in calendar year 2012 experiencing a new ( de novo ) index PCI, defined as a patient receiving a coronary angiogram with no prior coronary revascularization in the preceding six months. This was an aggregate, population-level economic burden study, which is also known as a “cost-of-illness” or “burden-of-disease” study . The study design was incidence-based, focusing on cost and survival outcomes for an annual cohort of the target study population. The horizon of the cost analysis was one year because the greatest cost impact tends to occur in the first year post-procedure, and due to the limited data for these patients. The horizon for survival analysis was a patient’s lifetime, given that mortality differences at one year can be modeled as life years lost over the remaining expected lifetime.
CAC could potentially adversely affect patients’ clinical outcomes for those who underwent PCI, coronary artery bypass graft surgery (CABG), or prescribed medical therapy for severe atherosclerosis deemed not treatable via surgical intervention. Of coronary angiography patients without a revascularization in the last 6 months, 5% underwent CABG (3.1% documented calcification), 31% underwent PCI (1.9% documented calcification), and 64% underwent a medical intervention (1.5% documented calcification). However, the focus of this analysis was on patients receiving PCI, due to the limited number of CABG and medical therapy patients documented with the calcification code in the available Medicare data.
The study design and analysis were influenced by the availability of CAC data. A new diagnostic code for calcification (ICD9 414.4) was introduced in the last quarter of 2011. Because the reporting of new codes can take time to become a part of regular medical documentation practice, there is a high likelihood of under-reporting of CAC in this initial period. This limited reporting affected the ability to do a strict epidemiologically-based comparison of patients with reported calcification versus those with non-calcified coronary lesions, given that such an approach would be subject to misclassification bias in which many patients with severe calcification would be inappropriately classified in the non-CAC group. Therefore, an aggregate estimate was constructed using an economic modeling approach that synthesizes data from multiple sources described below.
The principal data sources for cost and survival analyses were Medicare’s Standard Analytic Files (SAFs). The SAFs comprise seven data sets containing detailed claims information about health care services rendered to Medicare beneficiaries in fee-for-service (FFS) Medicare. SAFs are available for institutional (inpatient, outpatient, skilled nursing facility, hospice, or home health agency) and non-institutional (physician and durable medical equipment providers) claim types. Data are organized at the claim level and include basic beneficiary demographic information, date of service, diagnosis and procedure code, provider number, and reimbursement amount. Two SAF databases were used: the Medicare 5% random sample SAF and 100% SAF. The 5% sample of beneficiaries includes all of their claims (inpatient, outpatient, physician, durable medical equipment, etc.) except drugs, which are tracked and reported separately via Medicare Part D. The 100% SAF includes only inpatient and outpatient claims but includes all FFS beneficiaries. In addition, a special sub-sample was defined from the 100% SAF that included all hospitals (n = 17) that coded more than 10% of their PCI patients as having CAC (using code 414.4).
The Medicare SAF analyses were conducted with Limited Data Set (LDS) files, which encrypt beneficiary identifiers, and this research complies with the Centers for Medicare and Medicaid Services Data Use Agreement rules on blinding and data use. Institutional Review Board approval is not required for use of LDS administrative data under the HIPAA Privacy Rule.
To better estimate the degree of calcification and incidence of major adverse cardiovascular events among this target patient group, additional data from the HORIZONS-AMI/ACUITY pooled sample were used, though this dataset does not contain information on costs . Approval for analyses of this pooled sample was obtained from the Institutional Review Boards or Ethics Committees at each of the enrolling sites.
2.2
Economic model: cost burden of coronary calcification
A population-level economic model was constructed drawing parameter estimates from both a detailed analysis of the available Medicare claims data and from the published literature. A modeling approach also allowed for assessing the sensitivity of the results to varying assumptions about key parameters.
As depicted in Fig. 1 , the economic model had three major components: 1) the estimated annual incidence of elderly Medicare patients receiving an index PCI (typically a drug-eluting stent); 2) the incidence of CAC in these PCI patients, classified as severe, moderate, or mild/none; and 3) the estimated impact of the three levels of CAC on Medicare per-patient costs and health outcomes (viz., mortality).
2.3
Incidence of PCI in 2012
The 100% Medicare SAFs of administrative claims data were used to define the index population in 2012, which includes only patients in the Medicare FFS population. The cohort was restricted to elderly patients (i.e., age 65 or older). Although Medicare Advantage (MA) patients represented about 21.7% of the Medicare population in 2012- and rising substantially annually – these data are not included in this database. To account for MA patients, we made a simple, crude adjustment, increasing the aggregate burden estimate for the whole population upward to reflect the inclusion of this subpopulation.
2.4
Incidence of calcification in the elderly
We reported incidence of CAC in the Medicare population based on the 414.4 code. However, given the likely under-reporting of CAC in the 2012 Medicare SAF cohort, the estimated incidence of moderate and severe CAC was based on estimates from the HORIZONS-AMI/ACUITY pooled sample . The estimated means for the elderly subsample were used in the base case. The range for the assumption was based on the literature . Calcification incidence from the HORIZONS-AMI/ACUITY sample was multiplied by the Medicare population to determine population size for CAC.
2.5
Medical care cost impact
Estimation of the first-year impact of calcification following index PCI on costs was based on a generalized estimating equation (GEE) regression analysis of data from the Medicare 5% Standard SAFs, which were used because they contain the most complete cost data including physician claims—adjusting for age, gender, and the Charlson comorbidity index . The Charlson index was used rather than individual comorbidities to decrease the effect of correlation and increase power of the models. As previously noted, an adjustment for the MA patients was applied in the aggregate population model. In the aforementioned available data, approximately 20% of patients had one-year of follow-up data. Techniques developed by Lin et al. and Basu et al. were applied to address censoring in order to fully utilize patient cost data up to the point of loss-to-follow-up or death.
2.6
Mortality impact
We reported the impact of CAC on mortality using the 100% Medicare SAF. However, given the likely under-reporting of CAC in the 2012 Medicare SAF cohort, the estimated mortality impact of moderate and severe CAC was based on new tabulations from the HORIZONS-AMI/ACUITY pooled sample, an in-depth analysis of cardiac death in the sample studied by Genereux et al. . Cardiac death was used, as all causes of mortality were not available and cardiac death was assumed to be most relevant. A Cox proportional hazards model was used to account for censoring with adjustment for age, gender, the Charlson comorbidity index, and the comorbidities end-stage renal disease and diabetes, which were included explicitly in order to evaluate their specific effects due to their potential key role in the causal framework. The impact of the mortality differential on the health outcomes for the 2012 cohort was assessed both in terms of lost life-years and the associated monetized value of those lost life-years. For the life expectancy calculation, the estimate of lost life years was based on the median age and gender of an elderly PCI patient (i.e., a 74-year-old male) who would normally have an expected lifetime of 11.5 years: this was then adjusted down to 9.5 years after standard discounting at 3% per annum. A sensitivity analysis scenario used a mortality differential 25% higher than the base case as a ‘high case’ for mortality, and a ‘low case’ sensitivity analysis conservatively assumed no survival difference between CAC and non-CAC PCI patients.
2.7
Aggregate burden-of-illness model
In order to assess the overall burden of CAC, we aimed to include both true economic costs as well as the effects on health. Fig. 1 illustrates how the economic model of the burden of illness integrated each of the components above to produce an aggregate estimate of the economic burden. The first component was the incidence of PCI by age and gender. The second component represented the degree of CAC within these age and gender groups.
The third component was patient outcomes with two dimensions: cost and mortality impacts, which were included in the model in two capacities. First, the attributed medical cost burden was the product of the number of patients with CAC and the expected average attributable annual cost for patients in the 2012 cohort receiving an index PCI procedure. Second, to combine health and economic effects, we converted health impact into a monetary value. We utilized a method valuing life-years according to willingness-to-pay thresholds, as discussed in the recently published ACC/AHA Task Force on value and cost in guideline development . For example, if the societal willingness-to-pay for one year of healthy life is $100,000, we valued a year of healthy survival in the model at that amount. In this way we could report a singular economic impact rather than separate economic and health impacts. The willingness-to-pay threshold has the limitations both of being subject to societal acceptance, as well as being difficult to estimate with a high degree of certainty. Accordingly, we varied the willingness-to-pay threshold at three levels in order to address societal uncertainties in the threshold. Estimates of societal willingness to pay for life year gains were based on the Andersen et al. thresholds for “levels of value”—high, medium, and low : the base case assumed $100,000 per life year gained, while the lower bound was $50,000 and the upper bound was $150,000.
For all inputs, the model was evaluated at a base case, based on the results of the analyses, and at low and high values in the scenario sensitivity analyses based on the expert judgment of the co-authors as to the degree of uncertainty and plausible variation.
3
Results
3.1
Incidence of index PCI—Medicare cohort
Based on the analysis for the Medicare 100% SAF for calendar year 2012, there were 157,340 index PCI procedures performed, which amounted to an estimated 200,945 when adjusted for the MA patients. Table 1 shows mean Charlson comorbidity scores and ages by gender. The mean age was 73.4 years, 61.8% were males, and 66.4% were between ages 65 and 74.
