Key results
In this study, we found that the 76% of hospital-confirmed Covid-19 patients were identified by dispatch nurses, while 82% were identified by ambulance nurses. We identified a relatively high 30-day mortality rate of 24% among Covid-19 patients, but a relatively low rate of emergent response (30%) and transport (14%). Covid-19 patients presented with a diverse set of complaints, and a majority of Covid-19 patients (54%) presented with primary complaints not typical of Covid-19. With the exception of body temperature, initial vital signs had little predictive value for the identification of Covid-19 patients.
Limitations
This retrospective analysis has several important limitations. Documentation completeness issues likely entail an under-estimation of the true sensitivity of prehospital suspicion of Covid-19, given that suspicion is often likely to be communicated upon handover between care providers verbally, or in free-text notes. We chose to limit this analysis to structured documentation (i.e., the selection pre-defined options) however, as only this form of data has the potential to be used in automated alerting and monitoring systems.
We found that overall testing rates among patients arriving at the hospital was low (23%), constituting a potential source of misclassification bias. Furthermore, testing was unequally distributed, with for instance 52% of patients admitted to the hospital tested for ongoing SARS-CoV-2 infection, but only 8% of patients discharged from the ED being tested. Some portion of these untested patients were in all likelihood asymptomatic or mildly symptomatic carriers of SARS-CoV-2 [18]. While false negative patients in this group are unlikely to have clinically severe Covid-19-related symptoms, they nonetheless represent a disease vector which must be identified and isolated to prevent contagion. Further misclassification may result from errors in the tests themselves, though the combined oro-nasopharyngeal sampling method employed in the region is considered to be perhaps the most sensitive available sampling approach for the diagnosis of ongoing SARS-CoV-2 infection [19].
To address the risk of misclassification-bias, we performed a sensitivity analysis investigating predictive value within only the cohort of patients whose SARS-CoV-2 status was confirmed via PCR testing. Specificity was significantly lower in the sensitivity analysis, likely owing to the exclusion of a large number of negative cases. Estimates of sensitivity also tended to be lower in this analysis, but the difference could not be confirmed statistically. This loss of fidelity among patients identified via testing could be due to the inclusion of a larger proportion of patients with asymptomatic or mildly symptomatic SARS-CoV-2 infections, and/or the exclusion of clinically relevant positive cases identified by means other than PCR-testing.
The inclusion criteria for this study select for patients cared for through the typical prehospital care pathway of ambulance transport to the ED, and many patients referred to alternate forms of care by dispatchers and ambulance crews, as is common in the studied region [20], were not included owing to the lack of relevant outcome data for these patients. It may also be that some cases of Covid-19 were hospital-acquired, and that prehospital care providers could not have been expected to identify them. The data from this single-center study should be considered together with data collected in other settings to form a more general picture of the clinical impact of Covid-19 in the domain of prehospital care.
Interpretation
While the volume of Covid-19-related calls peaked in mid-April, the number of hospital episodes and admissions remained stable, owing to an increase in the number of patients directed by dispatchers and ambulance crews to non-ED destinations. In terms of pre-hospital acuity, Covid-19 patients had the lowest proportion of high priority (i.e., lights and sirens) dispatch responses of any comparison group (30%), and similar levels of high priority transport compared to the baseline population (14%). Expressed another way, the prevalence of Covid-19 did not differ between patients transported with a high- or low priority. Covid-19 patients were most likely to have signs of abnormal breathing (47%), and to receive supplemental oxygen (36%), but least likely to receive prehospital medications (28%) compared with other patient cohorts. Outcomes for Covid-19 patients were substantially worse than any other cohort, with 16% of patients cared for in the ICU, and a 30-day mortality rate of 23%.
It should be noted that prehospital priority was associated with patient outcomes in all comparison groups, as found in previous studies [21]. Within the cohort of Covid-19-positive patients, the 30-day mortality rate of Covid-19 patients transported to the hospital with lights and sirens was for instance 40%, compared with 21% for patients transported with a low priority. These findings thus suggest a discrepancy between prehospital prioritization and outcomes, and not a complete lack of association. This discrepancy could point to a difficulty in recognizing the early symptoms of severe Covid-19 infections, or to a more rapid deterioration among these patients after handover to definitive care.
Dispatch nurses were more selective in identifying suspected Covid-19 cases than their counterparts on the ambulance, with overall suspicion rates of 16% vs. 24%. This resulted in a higher specificity (87% vs 79%), but a lower sensitivity (76% vs 82%). The negative predictive value of prehospital suspicion shifted from 95% at the beginning of the pandemic to 99% in the latter phases, likely due in large part to changes in the overall prevalence of Covid-19 in the population. A priori, we suspected that we might find an increasing trend in predictive value over time due to improved guidelines and care providers learning more about the presentation of Covid-19 patients. While we did observe a small initial uptick in sensitivity from April to May, we cannot confirm the existence of such an effect with a sufficient level of confidence based on these observational data, and this should be examined in further studies in other contexts.
While the level of predictive value found in this study is useful for the purposes of for instance system-level monitoring, the authors consider the level of sensitivity found here to fall short of the level required to rule out the need for isolation precautions on an individual basis. These results suggest that during periods of significant community transmission of Covid-19, provider suspicion alone is not sufficient to rule out the potential for contagion, and some level of isolation precautions should be adopted when interacting with all patients.
Our results demonstrate that roughly half of Covid-19 positive patients present with primary complaints associated with Covid-19 (infection, fever, upper airway complaints, or difficulty breathing), while significant numbers of Covid-19 patients present to prehospital care providers with non-typical complaints and may be difficult for care providers to identify. Simultaneously, only a minority of patients presenting with “typical” complaints were indeed later diagnosed with Covid-19. Previous studies have employed the documentation of a Covid-19-associated prehospital primary complaint as a proxy measure for the impact of Covid-19 at the system-level [8, 9], and these results cast some doubt on the validity of this approach of gauging the true impact of Covid-19.
While patient vital signs demonstrated a limited capacity to detect the presence of SARS-CoV-2, they are likely to play a more substantial role in identifying severe cases with poor predicted outcomes [21,22,23]. The low overall testing rates and the preference towards testing higher-acuity patients found in this study entail a substantial risk for non-random misclassification-bias which would be imparted to models seeking to identify Covid-19 positive patients within a general population where testing is not widespread or randomly distributed [24]. Approaches for mitigating such systematic misclassification bias in testing should be investigated in further research.
Generalizability
Our results were in line with a previous investigation by Fernandez et al. of the sensitivity of Covid-19 suspicion documented in free-text notes written by ambulance crews, but given the substantial difference in overall prevalence of Covid-19 between these studies (5% here vs. 1% in Fernandez et al.) comparisons should be made only with caution, particularly with regards to specificity [11]. In examining the characteristics of Covid-19 patients presenting to the prehospital care system in this region, we found a 24% rate of 30-day mortality, which is congruent with preliminary investigations of hospitalized patients diagnosed with Covid-19 in Denmark, Norway, and Italy [23, 25, 26]. The dispatch and ambulance personnel investigated here consisted of nurses with 3–4 years of formal education, which is a relatively high level of education by international standards. The idiosyncratic approach taken by Swedish public health authorities [27], the 23% overall testing rate, and 11% test positivity rate should also be considered when generalizing these results to other settings.