cared for by paramedics) and the n o n t r e a t m e n t group (those transported by methods other than paramedic units) were similar on entry, the lack of random allocation to either group introduces a large potential for bias. Specifically, it is not unreasonable to suspect that patients who were treated by the paramedics might have been ,sicker" because those patients (or the families of those patients) might be more likely to call for assistance. If this were the case, valid c o n c l u s i o n s a b o u t the e f f i c a c y of the paramedic-administered treatment would be difficult to make. Moreover, although the authors contend that the use of multiple logistic and linear regressions allows a comparison of outcomes between the two relatively small study groups, we disagree. Specifically, the use of crude criteria to classify patients, such as entry Killip classification and estimated time until study entry, cannot overcome the bias introduced in this design by allowing patients and family to "assign" themselves to a given group. Additionally, patients who arrested prior to contacting paramedics or hospital personnel were included only if medical care was sought prior to the arrest. Once again, this introduces a bias against the paramedic-treated group. If only a handful of patients with symptoms of acute myocardial infarction went on to suffer cardiac arrest, without paramedic assistance, and were included in the nontreatment group, the conclusion of a lack of paramedic efficacy might be reversed. Often, self-transported patients suffering cardiac arrest en route to the hospital are misidentiffed as trauma mortalities and may represent a population that could benefit from paramedic-delivered ALS. Another difficulty with the study design involves the use of many different individuals to observe and record data points. These ranged from bystanders to paramedics to hospital personnel with varying levels of training. Use of the Killip classification as a criterion for stratification of patients is made questionable because the accuracy of observations such as magnitude of rales and auscultatory blood pressures may be influenced by training and experience. Also, when data acquisition and in-hospital treatment are not controlled, conclusions based on changes in Killip classification over time cannot be made. The application of statistical analysis to this body of data is thus highly problematic. The etiology and significance of the difference in mean delay in arrival to the hospital between groups (29 minutes longer for the paramedic-treated group) was not assessed. This delay may represent a system-specific problem that could influence interpretation of outcome data. Finally, the arbitrary cutoff of data collection at 48 hours in attempting to assess the preservation of left ventricular function may have prevented observation of difference between the treatment groups. Although Killip classification estimates the amount of left ventricular function after myocardial infarction, it is possible that subtle differences may be uncovered with further examination of the subjects. In particular, either longer standardized follow-up or the application of more invasive procedures, such as echocardiography or radionuclide im~ aging, may be helpful. 18:8 August 1989
We agree that the authors' data did not support the use of paramedic personnel for the treatment of acute myocardial infarction in their prehospital system, except in cases requiring immediate defibrillation. However, the fault may lie with the study design rather than the data. In the future, better-designed trials may be able to assess the utility of prehospital care for a variety of acute conditions. Until then, we must interpret the results of the current literature carefully and ensure that any limitations of the design or execution are fully appreciated.
Donald M Yealy, MD Paul T Hogya, MD Paul M Paris, MD, FACEP Ai]an B Wolfson, MD, FACEP The Division of Emergency Medicine University of Pittsburgh The Center for Emergency Medicine of Western Pennsylvania Pittsburgh
In Reply: We appreciate the interest in our study demonstrated by the comments from Yealy et al. Their letter generally reiterates the points covered in the discussion section of our article. As we stated, the two groups studied were not selected randomly, and statistically significant baseline differences were present between the groups. All known predictive factors for outcome were considered in analyzing the data by multiple logistic regression. However, we cannot exclude the possibility that patient self-selection in choosing mobile paramedic unit care reflects an underlying worse prognosis. Only a randomized study can resolve this question, but such a study has never been done. Clinical studies in this area are difficult, and there are no better studies available that support the multiple, timeconsuming procedures that paramedics perform. It is an unproven hypothesis that paramedic care of acute myocardial infarction patients (other than defibrillation) is beneficial. We do not currently advocate any change in paramedic procedures. However, mobile paramedic unit protocols should be examined critically to determine which procedures are effective and which are ineffective and delay hospital arrival. We hope that our article encourages the performance of well-designed, randomized trials that define the role of paramedics in the prehospital care of patients with acute myocardial infarction.
Nathan C Dean, MD Peter J Haug, MD Paula J Hawker, RN LDS Hospital University of Utah Salt Lake City
Annals of Emergency Medicine
910/167