For 3 of the 4 conditions we studied, unadjusted survival rates were higher among patients receiving BLS despite these patients being older and having more comorbid conditions on average than those receiving ALS. After adjustment, these outcome differences persisted; we found similar or better health outcomes associated with prehospital BLS than ALS in all of our analyses for major trauma, stroke, and respiratory failure. Because these high-acuity conditions necessitate early optimization of care, one would have expected any advantage of ALS over BLS to manifest itself in these diagnoses. Although ALS may be expected to improve outcomes because of early treatment, the opposite may occur in practice if ALS is associated with delays in hospital management or iatrogenic injury (3, 5–7, 31). We used 2 methodological approaches to adjust for potential confounders of comparisons between BLS and ALS outcomes. One analysis used propensity score methods to balance observed characteristics. This approach is susceptible to confounding by any unobserved patient characteristics associated with survival and ALS use; however, because ambulance dispatch protocols prioritize ALS for the conditions we studied, such individual-level confounding is plausibly minimal. Our second approach used the instrumental variable of county-level variation in overall ALS prevalence to predict the likelihood that a patient would receive ALS. This approach is less susceptible to confounding by unobserved patient characteristics but is subject to confounding by associations between rates of ALS use and other county characteristics that affect mortality. However, our falsification tests suggest that such confounding is unlikely
...
With the exception of patients who had AMI, BLS was associated with outcomes similar to or significantly better than ALS using both methodological approaches. Survival after AMI was substantially better with BLS than ALS in instrumental variable analysis (between 4 and 8 percentage points across all time points), but the propensity score analysis found higher survival with ALS than BLS (between 1 and 2 percentage points at 90 days, 1 year, and 2 years). We did several sensitivity analyses (Appendixes 9 to 16 of the Supplement), none of which changed the direction or significance of our main findings. Our findings are consistent with other evidence for cardiac arrest (Appendix 17 of the Supplement) and trauma (4, 5, 7, 10 – 18). Little prior evidence, however, exists for patients with stroke, AMI, and respiratory failure.
...
Our study has several limitations. Selection bias may confound our findings if patients receiving ALS and BLS differ in unobserved illness severity or in the quality of hospital care that they receive. To address this issue, we did 2 types of analyses that are subject to different types of confounding. The propensity score analysis would be biased toward finding worse outcomes associated with ALS if ALS providers were dispatched to patients with greater unobserved illness severity. Interviews we conducted with 45 state emergency medical service representatives, however, indicate that, if available, ALS would routinely be dispatched for many of the conditions that we investigated. As a result, the decision to dispatch ALS providers may plausibly be uncorrelated with unobserved illness severity for conditions other than trauma; for trauma, we controlled for severity. Moreover, BLS patients were older and had more comorbidities than ALS patients, which suggests that any unobserved differences in severity may actually have biased our results against BLS. The instrumental variable analysis could be confounded if counties with poorer quality hospital care had higher ALS penetration. Our falsification tests found no association of ALS penetration with nonemergent surgical mortality or nonemergent intensive care unit mortality at the county level. Although these tests increase confidence in our results, we had no way of directly assessing the quality of care for emergency patients that was not susceptible to potential confounding by characteristics of ambulance services. Estimates and significance tests under the propensity score analysis could be subject to bias if ALS providers evaluated a patient and then selectively provided care and billed at the BLS level for less acute cases. Given substantial reimbursement differences between ALS and BLS, it is unlikely that ALS providers billed at BLS rates because Medicare allows billing at the ALS level if assessment by ALS-trained providers was considered necessary at dispatch. Further, analysis of survival differences in 2005 claims, which distinguish ALS claims billed at the BLS level, showed little sensitivity to whether this small group was categorized as ALS or BLS (Appendix 15 of the Supplement). Finally, significance tests under the instrumental variable analysis would still be valid in this case as long as higher rates of ALS claims reflect higher utilization rates of ALS providers, although estimates of the effect magnitude might be biased. Because our sample included only patients with hospital claims, another potential limitation may be that more BLS patients died at the scene or en route to the hospital. In sensitivity analyses that considered these cases, however, the direction and significance of our findings were unchanged (Appendixes 10 and 11 of the Supplement).
