provides descriptive statistics showing the dramatic increase in treatment intensity during my study period of 1989–1998. With rates of reimbursement for a given treatment relatively constant, the rapid growth in inpatient expenditures is the result of increased treatment intensity during the study period (McClellan 1997
). The table reveals sharp increases in the rates of all three major invasive procedures used to treat AMI patients—catheterization, angioplasty, and bypass surgery. Along with the increase in the volume of catheterizations, there also was a move toward earlier intervention, with the fastest growth occurring in 0- and 1-day CATH rates. In addition to the diffusion of these invasive procedures for treating AMI, the study period was marked by the diffusion of noninvasive technologies as well, such as thrombolytic drugs, ACE inhibitors, β
-blockers, and aspirin. All of these technologies have contributed to the steady drop in 1-year mortality rates (McClellan and Noguchi 1998
). The increased survival rate has led to only slightly higher 1-year heart failure and AMI readmission rates. Over my study period, patients have grown older, and less likely to live in large MSAs, with racial and gender composition roughly constant.
Trends in Characteristics of Elderly Medicare AMI Patients Admitted through the Emergency Room
presents separate descriptive statistics for weekend and weekday patients admitted during the entire 10-year study period. This table provides evidence suggestive of a weekend effect on the rapidity and probability of procedure use. Weekend patients are significantly less likely to receive immediate catherization, angioplasty or bypass surgery on the day of their admission with AMI (p < 0.001). The gap in procedure rates is largest 1 day subsequent to the initial hospitalization, as Saturday patients continue to experience further weekend induced delays in treatment. Data not presented in the table show that weekend–weekday differences in immediate (both 0 day and 1 day) procedure rates became more pronounced over the duration of the study period, as these invasive procedures came to be used both more frequently and earlier in the treatment of AMI. The weekend–weekday differences in procedure rates decrease with time from the initial hospitalization, with most of this narrowing occurring during the first week. Only small reductions in rates of angioplasty and catheterization persist at 1 year. In turn, weekend patients experience higher 1-year mortality (p < 0.10), though the effects on cardiac readmission rates is somewhat mixed.
Descriptive Statistics for Elderly Medicare AMI Patients Admitted on Weekends and Weekdays (1989–1998)
also shows no evidence of systematic differences in the demographic characteristics or observable measures of ex ante health of weekend versus weekday patients. Weekend patients are no more likely to have been hospitalized in the prior year, and have mean prior year inpatient expenditures and Charlson indices that are statistically indistinguishable from those of weekday patients. For the purpose of this table only, I matched hospital characteristic data from the American Hospital Association Survey to patient-level data based on each patient's initial hospital of admission. These hospital characteristics show that there are no systematic differences in the quality of hospitals to which weekend versus weekday patients are admitted. While weekday patients are more likely to be admitted to large, teaching hospitals, weekend patients are more likely to be admitted to hospitals with catheterization laboratories.
In recent work, Dobkin (2003)
argues that nonuniform incidence, resulting from patients' and/or physicians' preferences for weekday admission, indicates the presence of selection that will generate upward biased estimates of the impact of weekend admission on health outcomes. reveals that the proportion of patients admitted on the weekend (263,068/922,074 = 28.5 percent) is exactly what would be expected if the true incidence of AMI were uniform. Although there is a slight spike in the number of AMI patients admitted on Mondays, the absence of a weekend drop-off suggests that this is unlikely to be the result of less severely ill patients experiencing weekend symptom onset deferring admission until Monday. Dobkin acknowledges that AMI may be a condition where the true incidence is in fact nonuniform, as researchers have offered explanations for the increased Monday incidence ranging from heavy weekend drinking (Evans et al. 2000
) to the stress associated with the start of the work week (Willich et al. 1994
and summarize the primary results of interest—the effect of weekend admission on procedure rates, expenditures and outcomes—under several specifications. presents estimates and standard errors of the effects of weekend admission on 0, 1, 2–7, and 365 day cardiac procedure rates, while presents the effects of weekend admission on inpatient and outpatient expenditures and health outcomes. These models all correspond to Equations (1)
and control for patient health status, patient demographics, hospital fixed effects, and differential time trends across MSAs of differing sizes. The standard errors are based upon an estimator of the variance–covariance matrix that is consistent with the presence of heteroscedasticity.
Effects of Weekend Admission on Treatment Rates for Elderly Medicare AMI Patients Admitted through the Emergency Room, 1989–1998
Effects of Weekend Admission on Medicare Expenditures and Patient Health Outcomes for Elderly Medicare AMI Patients Admitted through the Emergency Room, 1989–1998
The first row of (specification 1a), shows evidence of substantial effects of weekend admission on the rapidity of intensive procedure use. Most of the weekend-effect induced delay in receiving CATH and PTCA occurs in the 7 days after the initial admission. Weekend admission leads to a decline in the probability of receiving CATH (PTCA) on the day of admission by 2.12 (1.18) percentage points with the weekend induced reduction in procedure rates rising further to 3.47 (1.52) at 1 day. Of that decline, 2.40 (0.80) percentage points are recovered in the 2–7 days after admission, with an additional 1.01 (0.33) percentage points recovered in the next 358 days (i.e., days 8–365).3
Expressed as a share of the number of patients in 1998 who received a CATH (PTCA) on the day of admission, weekend hospitalization delays the treatment of 19.1 = 2.12/11.1 (16.6 = 1.18/7.1) percent of patients. Most of the weekend-effect delay in receiving CABG occurs after the first week; βP7
for CABG is small and statistically insignificant, while (βP365
) remains large. There are no significant differences in the cumulative 365-day CATH and CABG rates for weekend versus weekday patients, although weekend patients are significantly less likely to receive PTCA within 1 year of admission with AMI.
The first row of shows that these differences in treatment have significant consequences for both expenditures and health outcomes. Patients admitted with AMI on the weekend have approximately 0.55 percent lower inpatient expenditures in the year after their AMI, reflecting, in part, the lower volume of intensive procedures. Patients hospitalized on weekends also have 4.40 percent lower outpatient expenditures in the subsequent year. Weekend heart attack patients are more likely to die than their weekday peers in all time periods subsequent to the initial hospitalization; at 1 year, patients admitted on the weekend experience 0.38 percentage points higher mortality from AMI, which is statistically significant (p
< 0.01). Expressed in relation to average AMI mortality in 1998 (36.3 percent), patients admitted on the weekend face a 1.0 percent increase in 1-year mortality. The effects of weekend admission on cardiac complications are mixed. While weekend admission has a positive and significant effect on readmission with CHF (p
< 0.01), weekend patients are less likely (p
< 0.01) to be readmitted with AMI in the year following hospitalization with the initial AMI.4
To address the issue of selection bias arising from delays in admission, the second rows of and (specification 1b) present estimates that group weekend and Monday admissions together. According to , the effect of a Saturday/Sunday/Monday admission on the rapidity and probability of intensive treatment is smaller, but still statistically and economically significant. The effect of a Saturday/Sunday/Monday admission on 1-year mortality is reduced, but remains statistically significant. As in the weekend versus weekday models, Saturday/Sunday/Monday patients experience higher 1-year rates of readmission with CHF, but slightly lower 1-year AMI readmission rates. The effect on inpatient expenditures of Saturday/Sunday/Monday versus Tuesday–Friday admission is small and statistically insignificant.
I take an additional step to ensure that my results are not driven by selection bias attributable to patients experiencing less severe symptom onset on the weekend deferring admission until the following Monday. If emergency departments were less likely to diagnose patients with AMI who present with chest pain on the weekend, and these patients remain symptomatic and are hospitalized with AMI on Monday this could lead to bias. The direction of this bias is uncertain as, if deferred admission were harmless, estimates of the weekend effect based on admission date would overstate the true effect. However, if deferred admission were harmful, estimates of the weekend effect based upon the date of the AMI admission would understate the true effect. To explore this issue, I use a 20 percent sample of outpatient claims to identify patients who were treated in the ER with chest pain (but not hospitalized) in the 3 days before their AMI admission. Fewer than one in 700 AMI patients had such a visit before their hospitalization, and the incidence of such visits is roughly constant across days of the week. I re-estimate all of the models in specification 1a, indexing each patient's admission date using the earlier of their actual AMI admission date or the date of any ER visit in the 3 days prior. This reassignment of index admission dates using prior ER visits with chest pain had no impact on any of the results.
Taken together, these results suggest that AMI patients admitted through the emergency room on the weekend receive lower quality health care, with the weekend effect on mortality not merely an artifact of selection. The 0.55 percent reduction in inpatient expenditures on the weekend amounts to approximately $105 per weekend patient (based on the sample average 1-year inpatient expenditure of $19,196). Using the point estimate of the effect of weekend admission on 1-year mortality, this suggests that the more intensive treatment provided to patients admitted through the ER during the week is cost effective assuming a value per year of life saved of approximately $27,631 or greater.5
A number of researchers (see, e.g., Duan 1983
; Manning 1998
) have noted that log-transformed models may lead to significantly biased inference on the raw scale. In light of these concerns, I examine the sensitivity of my cost-effectiveness results using the two primary approaches for dealing with the retransformation problem—smearing estimates and generalized linear models (GLM). Using the homoscedastic smearing coefficient (Duan 1983
), I find that weekend admission is associated with a $102 reduction in 1-year inpatient expenditures. When I employ separate smearing coefficients for weekend and weekday patients to account for heteroscedasticity on the log scale, the effect of a weekend hospitalization on inpatient expenditures is significantly attenuated. Lastly, estimates from the expenditure model using GLM with a γ-distribution and a log link (Blough, Madden, and Hornbrook 1999
) indicate that weekend patients have $123 lower 1-year inpatient expenditures than their weekday peers. Under all
of these alternative specifications, the treatment provided to weekend patients remains cost-ineffective.
To further explore the relationship between delayed provision of treatment and health outcomes, I estimate models where I examine separately the effects of Saturday and Sunday admissions. The coefficient estimates from these regressions (specification 1c) are reported in and . Saturday and Sunday patients experience similar declines in 0-day rates of all three cardiac procedures, suggesting the provision of immediate treatment is roughly similar on both weekend days. Patients admitted on Saturdays experience significantly larger declines in 1-day procedure rates. However, Saturday admissions make up most of the decline in 1-day procedure rates on days 2–7, whereas Sunday admissions do not. Even at 365 days, Sunday patients experience slightly larger decreases in procedure rates. Patients admitted on Saturdays and Sundays experience similar declines in 1-year inpatient expenditures relative to weekday patients, while the decrease in 1-year outpatient expenditures is much larger for patients initially hospitalized on Sunday. The mortality effects in all periods subsequent to the AMI admission are significantly larger for Sunday admits. The effect of a Sunday admission on mortality is significant (p < 0.001) in all periods, rising from −0.37 percentage points at 30 days to −0.53 percentage points at 1 year. Only at 1 year do I observe statistically higher mortality for patients admitted on Saturday versus patients admitted during the week.
The final four rows of and present estimates from models that allow the effects of weekend admission to vary with patients' ex ante health status. In , the effects of weekend admission on 0- day and 1-day procedure rates are negative for (base group) patients without a prior year's hospital admission, while the interaction effects between weekend admission and prior year admission are positive and smaller in absolute value than the base group effects. Conversely, in models of 2–7-day rates, the base group effects of weekend admission are positive, while the interaction effects are negative and roughly similar in magnitude to the interaction effects for 1-day rates. Interaction effects on 365-day cardiac procedure rates are insignificant. Taken together, these results indicate that sicker patients are less likely to experience delays in treatment, but when they do, those delays are similar in length to those experienced by healthy patients.
Specification 2 in examines how the effects of weekend admission on expenditures and outcomes vary with observable patient health status at the time of initial admission with AMI. Weekend admission leads to a statistically significantly larger decrease in inpatient expenditures for sicker patients versus healthier patients. In contrast, the decrease in outpatient expenditures resulting from weekend admission is larger for observably healthier patients. Sicker patients experience similar increases in 1-year mortality and CHF as a result of weekend admission, although lower rates of 1-year readmission with AMI.