EMERGENCY RESPONSE SERVICES: Mortality – The Case of Cardiac Arrest 8

In all of the specifications, we find that older patients are less likely to die (they may also be more likely to use ambulance services in less severe situations), while patients for whom cardiac arrest and defibrillation are reported are more likely to die. Likewise, we see a very strong effect of severity as measured by the Glasgow score: sicker patients are significantly more likely to die than patients with less severe symptoms. Privately insured patients are more likely to die than Medicare patients.

The second and third specifications consider the effects of response time and patient characteristics on mortality. We have already shown that response time varies with the severity of the patient as well as the kind of hospital to which the patient will eventually be admitted. Thus, it will be somewhat difficult to interpret the effects of the response time variables in the reduced-form mortality regression. We then propose a preliminary strategy for instrumental variables: we use county-level characteristics, and in particular the level of 911, as instruments for response time. We have already established that such characteristics affect the response time; it remains to argue that the level of 911 is uncorrelated with the unexplained variation in patient mortality (when patient-specific variables are included as controls in the regression). Our approach excludes all county-level demographic information from the regression; in future work, it may be possible to include zip-code level demographic data to capture any heterogeneity which might have been correlated with excluded county-level demographics.

Our instrumental variables results, while preliminary in nature, are suggestive. They show that shorter response times reduce the probability of death. The main coefficient which changes in sign as a result of the instrumental variables approach is the coefficient on TIME_TO_HOSP. It is not surprising that the coefficient changes in sign, since it is most sensitive to the severity of individual patients (in particular, patients with non-urgent symptons are transported to the hospital without lights and sirens). It is interesting to note that the instrumental variables strategy is successful despite the fact that higher levels of 911 are (unconditionally) correlated with both lower response times and higher average mortality rates.

We do not attempt an instrumental variables strategy for the technology of the hospital, though this is a potential area for future work. In our reduced-form specification, it is difficult to separate out the potentially beneficial effect of going to a better hospital from the effect due to the differential allocation of more severely ill patients and non-emergency patients to better hospitals.