Adjustment for Atherosclerosis Diagnosis Distorts the Effects of Percutaneous Coronary Intervention and the Ranking of Hospital Performance.

BACKGROUND: Coronary atherosclerosis raises the risk of acute myocardial infarction (AMI), and is usually included in AMI risk-adjustment models. Percutaneous coronary intervention (PCI) does not cause atherosclerosis, but may contribute to the notation of atherosclerosis in administrative claims. We investigated how adjustment for atherosclerosis affects rankings of hospitals that perform PCI. METHODS AND
RESULTS: This was a retrospective cohort study of 414 715 Medicare beneficiaries hospitalized for AMI between 2009 and 2011. The outcome was 30-day mortality. Regression models determined the association between patient characteristics and mortality. Rankings of the 100 largest PCI and non-PCI hospitals were assessed with and without atherosclerosis adjustment. Patients admitted to PCI hospitals or receiving interventional cardiology more frequently had an atherosclerosis diagnosis. In adjustment models, atherosclerosis was associated, implausibly, with a 42% reduction in odds of mortality (odds ratio=0.58, P<0.0001). Without adjustment for atherosclerosis, the number of expected lives saved by PCI hospitals increased by 62% (P<0.001). Hospital rankings also changed: 72 of the 100 largest PCI hospitals had better ranks without atherosclerosis adjustment, while 77 of the largest non-PCI hospitals had worse ranks (P<0.001).
CONCLUSIONS: Atherosclerosis is almost always noted in patients with AMI who undergo interventional cardiology but less often in medically managed patients, so adjustment for its notation likely removes part of the effect of interventional treatment. Therefore, hospitals performing more extensive imaging and more PCIs have higher atherosclerosis diagnosis rates, making their patients appear healthier and artificially reducing the expected mortality rate against which they are benchmarked. Thus, atherosclerosis adjustment is detrimental to hospitals providing more thorough AMI care.

Clinical Perspective

What Is New?

While coronary atherosclerosis is a risk factor for acute myocardial infarction (AMI), its notation on an administrative claim may be affected by hospital practice patterns.

Although biologically implausible, adjustment models used by the Centers for Medicare & Medicaid Services suggest that a diagnosis of atherosclerosis is associated with an ≈40% reduction in the odds of 30‐day mortality in older Medicare beneficiaries hospitalized for AMI.

Adjusted comparisons of hospital AMI outcomes that include atherosclerosis in the risk‐adjustment model may underestimate the quality of percutaneous coronary intervention hospitals and overestimate the quality of non–percutaneous coronary intervention hospitals.

What Are the Clinical Implications?

Hospitals that more frequently use interventional cardiology for treatment of AMI may be adversely affected by AMI mortality models that adjust for atherosclerosis.

The notation of atherosclerosis on an administrative claim in the context of performing a percutaneous coronary intervention could introduce bias to comparisons of hospital quality of AMI care.

Coronary heart disease affects >15 million adults and is a leading cause of mortality, responsible for ≈1 in 7 US deaths in 2013.1 Coronary atherosclerosis is the most common cause of myocardial ischemia,2 which can lead to acute myocardial infarction (AMI). In 2011, MI ($11.5 billion) and coronary atherosclerosis ($10.4 billion) were 2 of the 10 most expensive US hospital principal discharge diagnoses.1, 3 As a dominant source of morbidity and cost in the healthcare system, AMI has long been a focus of hospital quality measurement.

Percutaneous coronary interventions (PCI), such as percutaneous transluminal coronary angioplasty, are commonly used in the elective workup4 and urgent management5 of coronary artery disease, including severe types of AMI such as ST‐elevation MI. Percutaneous transluminal coronary angioplasty has been described as among the definitive medical advances of modern cardiology.6 Current guidelines note coronary angioplasty to be the treatment of choice for management of ST‐elevation MI if performed within certain parameters.5

Atherosclerosis can begin early in life and exist asymptomatically for decades. Atherosclerosis of the coronary arteries has been described in autopsy studies of young individuals, as well as in US soldiers killed in various conflicts.7, 8, 9, 10, 11 While coronary atherosclerosis may frequently be present in older patients with coronary heart disease, its presence might not be noted as frequently in the absence of coronary arteriography or PCI.

In adjustment models used by the Centers for Medicare & Medicaid Services (CMS), a diagnosis of coronary atherosclerosis or angina is implausibly associated with a ≈40% reduction in the odds of 30‐day mortality in older Medicare beneficiaries.12 Is it conceivable for atherosclerosis to cut risk of death from AMI by almost half? If not, what is the mechanism by which such an association is produced?

To be valid, a risk factor or covariate must describe the condition of the patient before treatment, and its value must not be changed by the treatment a patient subsequently receives.13 As a biological condition, atherosclerosis upon admission is a valid risk factor, but if its notation in a medical chart or in administrative claims is affected by PCI, then its notation is not a risk factor but rather a consequence of treatment. If models adjust for consequences of a treatment, they remove or distort the effects of that treatment. This study explores the implications of adjustment for atherosclerosis when auditing hospitals for quality of AMI care.

Methods

Data use agreements with the CMS do not permit data sharing, but the corresponding author may be contacted for additional details on analytic methods.

Data Set

This research protocol was approved by the institutional review board of Children's Hospital of Philadelphia; informed consent was not required because the institutional review board determined this was not human subjects research. We studied older Medicare beneficiaries admitted to short‐term acute‐care hospitals nationwide with a principal diagnosis of AMI between July 1, 2009 and November 30, 2011 in the Medicare Provider Analysis and Review file, with additional data drawn from the Outpatient, Carrier/Part B, and Hospice files. We followed methods established by Krumholz et al for application to Medicare administrative claims data.14, 15, 16, 17 Using the master beneficiary summary file, we excluded patients <65.5 years of age at admission, who had missing sex, who were admitted from hospice, whose date of death preceded the admission date, or who lacked Part B coverage or were in a Health Maintenance Organization at any point in the 6 months before admission. As is done for CMS AMI quality assessment, outcomes for transferred patients were assigned to the first admitting hospital14, 15, 16, 17 (Table S1). If a patient had multiple qualifying admissions, we chose a random one.

Patient Characteristics

We defined each patient's age at admission, sex, category of AMI principal diagnosis, and comorbidities validated for AMI risk‐adjustment using the inpatient record, any bills from other files in the 6 months before admission, or both, as indicated in established methods in use by Medicare.14, 15, 16, 17 International Classification of Diseases Ninth Revision (ICD‐9) diagnosis codes in the 412 to 414 groups or ICD‐9 code 74685 indicated a diagnosis of coronary atherosclerosis or angina.14, 15, 16, 17 We determined history of percutaneous transluminal coronary angioplasty or coronary artery bypass grafting (CABG) via the same process as was used for ascertaining comorbidities. Using the index inpatient bill, we also established whether each patient underwent PCI or CABG during the admission.

Hospital Characteristics

We defined a hospital as providing PCI services if they had a minimum of 10 inpatient bills per year noting a PCI in the procedure code fields, using ICD‐9 procedure codes 00.66, 17.55, 36.01, 36.02, and 36.05 to 36.07. We also defined other hospital characteristics using the Medicare Provider of Services file18: teaching status, size in beds, nurse‐to‐bed ratio, nurse‐mix, and the availability of comprehensive cardiac technology (the presence of a coronary care unit and catheterization laboratory, and provision of cardiothoracic surgery services).

Outcomes

We examined all‐location mortality within 30 days of admission.

Statistical Analysis

Mortality was modeled using LOGISTIC and GLIMMIX procedures in SAS Version 9.3 for UNIX.19 Models adjusted for age, sex, AMI principal diagnosis type, history of percutaneous transluminal coronary angioplasty or CABG, and comorbidities, following as closely as possible the established methods in use by Medicare for hospital quality assessment.12, 14, 15, 16 Hierarchical models added hospitals as random effects. The 2 types of models were each fit with and without atherosclerosis in the comorbidity set.

In each logit model, we calculated the risk‐adjusted number of expected deaths had all patients been treated at PCI hospitals versus non‐PCI hospitals, a form of direct standardization.20, 21

In analyses that ranked hospitals, rather than use the logit models to calculate observed‐to‐expected (O/E) mortality rates for each hospital, we used the CMS hierarchical model to calculate predicted‐to‐expected (P/E) mortality ratios. The P/E mortality rate differs from the O/E mortality rate in that, unlike the O/E ratio, which uses a hospital's own actual death rate as the numerator, the CMS hierarchical model yields a predicted mortality rate (“P”) that is based partially on a hospital's own death rate and partially on the national death rate. In the CMS hierarchical model, the weighting of P/E toward the hospital's own outcome rate versus the national outcome rate is contingent, in part, upon the hospital's volume, with lower‐volume hospitals’ P/E weighted more toward the national outcome rate. The CMS model has been criticized for shrinking to a national mean rather than a mean more appropriate to a hospital's specific characteristics22, 23, 24; regardless, P/E is believed to yield a more stabilized hospital estimate for smaller hospitals, and is the measure used by CMS. We likewise calculated and reported hospital ranks using P/E mortality ratios from CMS hierarchical models generated by the GLIMMIX procedure in SAS.19

Adjusted 30‐day mortality rates for PCI hospitals and non‐PCI hospitals were calculated by summing the predicted mortality and expected mortality from the hierarchical models for all patients at PCI hospitals and for all patients at non‐PCI hospitals, then dividing them to form a P/E mortality ratio for each type of hospital. We then multiplied the national mortality rate by these P/E mortality ratios to obtain adjusted mortality rates for PCI hospitals and non‐PCI hospitals.

In each hierarchical model, we calculated P/E mortality for the 100 largest PCI hospitals and 100 largest non‐PCI hospitals, and ranked them from smallest to largest, with a rank of 1 signifying the smallest (best) P/E ratio. We then determined how their ranks among all hospitals were affected by removing atherosclerosis by subtracting each hospital's rank in the first model from its rank in the second. Inferences that compare pairs of 2 models, 1 with and 1 without atherosclerosis, used the bootstrap.25

A 2‐tailed P value was significant if ≤0.05.

Role of the Funding Source

The Agency for Healthcare Research and Quality had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; and preparation, review, or approval of the article.

Results

Patient Characteristics by Hospital PCI Status

Table 1 shows patient and hospital characteristics split by whether patients were admitted to PCI hospitals or non‐PCI hospitals. Important differences can be observed between patients at PCI hospitals and non‐PCI hospitals. Patients at PCI hospitals were younger on admission. Also, patients at PCI hospitals more frequently had ST‐elevation MIs amenable to treatment with PCI than patients at non‐PCI hospitals (just >25% in PCI hospitals, compared with <10% at non‐PCI hospitals).

Table 1

Differences in Patient Demographics, MI Types, and Hospital Characteristics Between PCI and Non‐PCI Hospitals

Covariate	All Patients	Patients at PCI Hospitals	Patients at Non‐PCI Hospitals
Number of patients	414 715	359 685	55 030
Age at admission, y (mean)	78.2	77.9	80.6
Sex (% male)	52.0	53.1	44.9
Anterior or anterolateral MI (principal diagnoses 410.00–410.11)	9.6	10.5	4.0
Other ST‐elevation MI of specified sites (principal diagnoses 410.20–410.61)	13.7	15.2	4.6
History of PTCA	5.6	5.8	4.2
History of CABG	7.2	7.2	6.7
Hospital characteristics
% at teaching hospitals	32.6	35.1	16.9
% at large hospitals (size >250 beds)	71.1	77.9	26.5
% at hospitals with comprehensive cardiac technologya	33.9	38.7	3.2
Nurse‐to‐bed ratio (mean)	1.34	1.36	1.20
Nurse mix (% RNs, mean)	0.90	0.90	0.86

CABG indicates coronary artery bypass graft; MI, myocardial infarction; PCI, percutaneous coronary intervention; PTCA, percutaneous transluminal coronary angioplasty; RN, registered nurse.

Presence of a cardiac catheterization laboratory and a coronary care unit, and provision of cardiothoracic surgery services.

In comparison to non‐PCI hospitals, PCI hospitals were more often teaching hospitals, with larger bed size and better technology and nursing characteristics. For example, 35% of patients in the PCI hospital group were treated at teaching hospitals, compared with 16.9% of patients in the non‐PCI hospital group. Similarly, 38.7% of PCI hospital patients were seen at hospitals with comprehensive cardiac technology, compared with just 3.2% of patients at non‐PCI hospitals.

Comorbidities by Hospital PCI Status

Table 2 shows comorbidity rates split by hospital PCI status. Patients at PCI hospitals had lower rates of all comorbidities except for coronary atherosclerosis or angina, which was diagnosed in 86.5% of patients at PCI hospitals, compared with 70.1% of patients at non‐PCI hospitals. When directly comparing patients by treatment status rather than hospital type, patients who received PCI or underwent CABG had a rate of atherosclerosis diagnosis of 93.9%, compared with 75.4% among those who were managed medically (Table S2).

Table 2

Excess Rate of Atherosclerosis Diagnosis Among Patients at PCI Hospitals

Comorbidity	Patients at PCI Hospitals	Patients at Non‐PCI Hospitals	Difference
Number of patients	359 685	55 030
Congestive heart failure	26.9	37.7	−10.8
Past myocardial infarction	15.0	20.4	−5.4
Unstable angina	14.4	18.7	−4.4
Coronary atherosclerosis or angina	86.5	70.1	+16.4
Cardiorespiratory failure or shock	9.3	12.4	−3.1
Valvular or rheumatic heart disease	29.7	31.4	−1.7
Hypertension	84.9	85.5	−0.6
Stroke	5.8	7.9	−2.1
Cerebrovascular disease	14.2	16.7	−2.4
Renal failure	19.5	24.9	−5.4
Chronic obstructive pulmonary disease	27.2	32.5	−5.3
Pneumonia	23.2	30.5	−7.3
Diabetes mellitus	42.7	45.7	−3.0
Malnutrition	5.2	6.4	−1.3
Dementia	16.2	25.0	−8.8
Paraplegia	4.7	6.2	−1.5
Peripheral vascular disease	21.0	25.0	−4.0
Cancer	3.5	4.0	−0.5
Trauma	21.9	26.4	−4.5
Major psychiatric disorders	6.5	9.1	−2.6
Chronic liver disease	1.1	1.3	−0.2
Outcomes
Unadjusted 30‐d mortality, %	13.7	18.1	−4.4
Adjusted 30‐d mortalitya including atherosclerosis, %	14.1	14.4	−0.3
Adjusted 30‐d mortalitya omitting atherosclerosis, %	14.0	14.6	−0.6

PCI indicates percutaneous coronary intervention; P/E, predicted‐to‐expected mortality ratios.

Adjusted 30‐day mortality rates are computed by multiplying the national mortality rate by the P/E mortality ratio for each type of hospital given by the 2 models with and without atherosclerosis.

Atherosclerosis in Mortality Models

Although biologically implausible, the hierarchical model suggested that a diagnosis of atherosclerosis was associated with a reduction in the risk of death, the odds of death being reduced by a factor of 0.58 (95% confidence interval [CI], 0.55–0.61, P<0.0001) when compared with patients without a diagnosis of atherosclerosis. The logit model reached similar conclusions. See Tables S3 and S4 for hierarchical and logistic model results, respectively.

Atherosclerosis and Hospital Quality Assessment

The logit model with atherosclerosis estimated 4311 lives saved had all patients been treated at PCI hospitals (95% CI, 3111–5566). Removing atherosclerosis changed that estimate to 7296 lives saved had all patients been treated at PCI hospitals (95% CI, 6048–8553), an increase of 2985, or 62% (95% CI, 2781–3173, P<0.001) (Table S5).

The bottom of Table 2 reports unadjusted mortality rates as well as adjusted mortality rates given by the hierarchical models. Unadjusted mortality rates at PCI hospitals and non‐PCI hospitals were 13.7% and 18.1%, respectively, a difference of 4.4% between PCI hospitals and non‐PCI hospitals. Adjusted mortality with atherosclerosis in the model was 14.1% at PCI hospitals and 14.4% at non‐PCI hospitals, a difference of 0.3% between PCI hospitals and non‐PCI hospitals. After removing atherosclerosis from the model, adjusted mortality at PCI hospitals declined to 14.0%, while adjusted mortality at non‐PCI hospitals increased to 14.6%, a difference of 0.6% between PCI hospitals and non‐PCI hospitals (Table S6).

Does adjustment for atherosclerosis affect the ranking of hospitals? The Figure shows that it does. After removing atherosclerosis from the hierarchical model, 72 of the 100 largest PCI hospitals had better rankings (95% CI, 70–81), while 77 of the 100 largest non‐PCI hospitals had worse rankings (95% CI, 69–80). The alteration of the model had a significantly different effect on the 100 largest PCI hospitals relative to the 100 largest non‐PCI hospitals (P<0.001). See Table S7 for details of hospital rank changes and bootstrap results.

Figure 1

Boxplots of changes in P/E mortality rank for the 100 largest PCI hospitals and 100 largest non‐PCI hospitals after removing atherosclerosis from the model. After ranking each hospital by P/E in the first model with atherosclerosis and the second model without atherosclerosis (with rank 1 assigned to the smallest (best) P/E), we subtracted each hospital's rank in the second model from its rank in the first; thus, a negative difference implies an improved ranking in the second model. The thick horizontal line represents the median, while the box represents the interquartile ranges. Whiskers extend to the 5th and 95th percentiles, with dots beyond the whiskers representing outlier hospitals. PCI indicates percutaneous coronary intervention; P/E, predicted‐to‐expected mortality ratios.

Discussion

The biological condition of atherosclerosis is a known risk factor for cardiovascular disease and AMI. Its presence in a patient with AMI should not lower the patient's risk of death, but adjustment models presently in use suggest that it does. We found that the diagnosis of atherosclerosis was more prevalent in patients who were admitted to PCI hospitals or who underwent PCI. Possibly, the mere notation of atherosclerosis in a hospital chart or administrative claim may be a consequence of PCI or admission to hospitals that provide more thorough cardiac workup. While another possibility is that patients with atherosclerosis are more likely to be admitted to PCI hospitals, it must be remembered that all these patients were admitted for AMI, so it would be unlikely that the chronic condition of atherosclerosis would determine which hospital a patient should be brought to. Rather than the presence of atherosclerosis driving the type of hospital a patient is admitted to in the context of an AMI, it is far more plausible that diagnosis of this common medical condition is more frequently detected via interventions that can discover it. Adjusting for a consequence of treatment can, and typically does, distort the estimate of the effect caused by the treatment: it may, and typically does, remove part of the actual effect.13 In this case, some of the effect of interventional treatment for AMI may instead be misleadingly transferred to the diagnosis of atherosclerosis. Thus, adjusting for atherosclerosis can potentially be harmful to the ranking of hospitals that treat AMI patients more thoroughly.

How does this distortion occur? Table 3 shows 1 hypothetical patient who is 75 years old, male, and at the onset of a MI, may be admitted to 2 nearby hospitals A and B. Hospital B has a catheterization laboratory, while hospital A does not. By applying coefficients reported in the Medicare AMI model, we will examine what would follow should this same patient with a specific medical history have gone to either hospital.

Table 3

Admission, Treatment, and Diagnosis Process for the Same Hypothetical Patient Admitted to a PCI Hospital or a Non‐PCI Hospital

Patient History	Same Hypothetical 75‐Year‐Old Male Patient With an Inferior AMI
Admitting Hospital	Hospital A	Hospital B
Treatment status	Hospital A does not have catheterization laboratory; patient does not receive angiography or PCI	Hospital B has a catheterization laboratory; patient receives angiography and PCI
Atherosclerosis diagnosis	History of atherosclerosis is not diagnosed or noted on inpatient bill at Hospital A	History of atherosclerosis is diagnosed and noted on inpatient bill at Hospital B
Expected risk of death at each hospital	Based on medical history and AMI type, patient is expected to have a probability of death of 19.3% on admission to Hospital A	Based on the same medical history and AMI type, but adding the diagnosis of atherosclerosis, the same patient is expected to have a probability of death of 11.2% on admission to Hospital B
Effect on quality assessment	Hospital A's P/E mortality ratio is spuriously lower because of an artificially elevated “E” because of the absence of an atherosclerosis diagnosis. Its adjusted rank is better than it would have been, had there not been adjustment for atherosclerosis	Hospital B's P/E mortality ratio is spuriously higher because of a depressed “E” that is reduced because of a diagnosis noted via treatment. Its adjusted rank is worse than it would have been, had there not been adjustment for atherosclerosis

AMI indicates acute myocardial infarction; P/E, predicted‐to‐expected mortality ratios; PCI, percutaneous coronary intervention.

Upon presentation to hospital A, the patient is given aspirin at arrival, diagnosed with an inferior MI, and managed medically. In the absence of angiography or PCI, coronary atherosclerosis is not identified or noted on the claim. The model would assign the patient an expected probability of 30‐day mortality of 19.3%.

Now suppose the same patient instead went to Hospital B. Hospital B sends the patient to its catheterization laboratory, and performs coronary angiography and a PCI. The patient's coronary atherosclerosis is noted on the chart and the administrative claim. Crucially, this 1 patient's baseline risk factors seem, misleadingly, to have changed, not because the patient changed, but because the diagnostic workup is more complete. Models do not understand cardiology; they understand what predicts what. The risk adjustment model does not see the PCI—it is not a baseline risk factor, so it does not go into risk adjustment—and so the model cannot use the PCI to adjust the patient's risk, but the model does see the notation of coronary atherosclerosis, so it misleadingly attributes a reduced risk to patients with a diagnosis of coronary atherosclerosis. From a cardiologist's perspective, that is silly, but from a model's perspective, it improves prediction. When the model assigns this patient's expected chance of dying based only on patient characteristics, it will now account for this diagnosis of atherosclerosis from the inpatient claim. Adding this diagnosis to the patient's other characteristics, the model would assign an expected probability of 30‐day mortality of 11.2%.

A risk‐adjustment model is only intended to account for a patient's risk factors on admission. If that is the case, how can the same patient be assigned such disparate probabilities of death solely because of choice of hospital? It is conceivable that, as described above, the presence of atherosclerosis diagnosis on an inpatient claim is not a risk factor, but rather a proxy for a hospital's quality of diagnosis and treatment. The finding that the rate of atherosclerosis diagnosis is highest among patients who did receive PCI or underwent CABG may be further evidence of this proxy effect.

What does this portend for quality measurement? When comparing hospitals, risk adjustment models are used to determine the expected number of deaths based on their patients’ characteristics on admission. Once the expected number of deaths is given by the model, the observed number of deaths is divided by it to form an “O/E” ratio (or in the case of Medicare's Hospital Compare, a stabilized observed number of deaths “P” to form a “P/E” mortality ratio12). A hospital with a lower O/E ratio is considered better than a hospital with a higher ratio. An accurate count of expected deaths is essential to this calculation, because an underestimation spuriously inflates the hospital's ratio.

PCI hospitals appear to code more patients as having atherosclerosis because they perform more PCI, and because the risk adjustment model says, implausibly, that the presence of atherosclerosis reduces risk, then the risk model underestimates the expected number of deaths at PCI hospitals and therefore inflates their O/E ratio compared with non‐PCI hospitals. It is implausible that atherosclerosis reduces risk, as the models suggest. More plausible is that an atherosclerosis diagnosis simply indicates subsequent performance of a PCI or a more thorough diagnostic examination, either or both of which may help reduce mortality. We would suggest removing atherosclerosis from these risk‐adjustment models.

There are important limitations to our study. This analysis was restricted to retrospective review of administrative claims for older fee‐for‐service Medicare beneficiaries, and therefore may not be generalizable to hospital benchmarking that includes younger patients. Moreover, the admissions spanned 2009 to 2011, and it is possible that with changes in coding practices and the use of diagnostic tests and treatments in intervening years, a more recent analysis could yield different findings. However, more recently reported Medicare models for 2013 to 2016 admissions still show that diagnosis of coronary atherosclerosis or angina is associated with a 0.65 odds of mortality.12

Although biologically implausible, risk‐adjustment for the notation of atherosclerosis finds it to be associated with a greatly reduced risk of death in the models used to provide a hospital's expected number of deaths against which it is benchmarked. It would appear that present adjusted comparisons of hospital AMI outcomes that include atherosclerosis in the risk‐adjustment model are underestimating the quality of PCI hospitals and overestimating the quality of non‐PCI hospitals. Risk‐adjustment models must not include a patient characteristic (such as atherosclerosis) whose notation may reflect the style of practice of a hospital (the availability and use of PCI), and not an implausible biological effect of a patient characteristic, such as the implied beneficial effect of atherosclerosis.

Sources of Funding

This work was supported by Agency for Healthcare Research and Quality grant R01HS023560.

Disclosures

None.

Table S1. Creation of Study Cohort

Table S2. Characteristics of the Study Cohort

Table S3. Hierarchical Models With and Without Atherosclerosis

Table S4. Logistic Models Predicting 30‐Day Mortality

Table S5. Directly Standardized Analysis of Logistic Models Comparing Outcomes at PCI Hospitals and Non‐PCI Hospitals

Table S6. Adjusted Outcome Rates Using Hierarchical and Logistic Models

Table S7. Effect of Removal of Atherosclerosis From Hierarchical Models Between 100 Largest PCI Hospitals and 100 Largest Non‐PCI Hospitals

Click here for additional data file.