Association for Public Policy Analysis & Management

Up to a third of U.S. medical spending may be wasteful in that it produces no health benefit and may at times even harm patients. Yet identifying wasteful spending and the factors that lead to it is difficult.

This study focuses on the decision to hospitalize elderly Medicare patients who present at the emergency room (ER) with respiratory conditions. We examine individuals aged 66 to 70 using hospital discharge records for New York State. Respiratory conditions are one of the most common classes of illnesses affecting these patients. Half of these patients are admitted to the hospital, meaning that clinicians are frequently differentiating between cases that require expensive hospital services and those that do not. Failing to hospitalize someone who needs to be admitted could lead to dire consequences. But in addition to generating higher costs, unnecessary hospitalization puts patients at risk of hospital acquired conditions and logistical disruption.

The ideal setting for measuring the effect of the marginal hospitalization on outcomes would involve randomly assigning patients with symptoms around the severity threshold for admission either to hospital or to home care. We attempt to approximate this research design by using variation in the patient’s nearest hospital’s propensity to admit patients with similar observable characteristics as an instrument for the actual admission decision.

We find considerable variation even within hospitals in the probability of admission which seems unlikely to be explainable by purely medical risk factors. For example, conditional on observable diagnoses and comorbidities, women, African-Americans, and Hispanics are less likely to be admitted; and there is a sharp spike in admission probabilities at age 70.

Second, we find that there is considerable variation across hospitals in the probability that patients with different estimated severity levels will be admitted. For patients in the fifth decile of estimated severity, the probability of admission varies from 20-80%. The admission rate at the nearest large hospital for patients in the individual’s predicted decile is a strong instrument for the individual’s own hospital admission. Since this instrument is based on other patients, it is not affected by the patient’s own health status.

Third, OLS estimates of the consequences of hospital admission suggest that admitted patients are more likely to die in the upcoming year. These estimates demonstrate the difficulty involved with trying to identify unnecessary hospitalizations given the observable data in administrative hospital discharge records. Clinicians are clearly choosing to admit the sicker patients among those presenting at the ER, indicating that they have additional information about the patient’s condition that unavailable to us.

When we instrument for patient admission, we find that admission increases the number of future hospital days by six, and significantly lowers the risk of death in the upcoming year. We hypothesize that when clinicians are deciding whether to admit the marginal patient, information such as whether the patient has been hospitalized in the past can sway them towards an admission decision, and that such information about their colleagues’ past decisions is helpful in reducing their patient’s risk of death.