Abstract
Introduction
Myocardial infarction (MI) and stroke are two of the most prevalent time-sensitive conditions treated in U.S. emergency departments (EDs), collectively affecting more than 1.5 million Americans each year. 1 Selected patients with both myocardial infarction (MI) and stroke are treated with fibrinolytic reperfusion therapy, and shortening time-to-therapy is associated with better outcomes. 2 –5 Elaborate systems of care have been developed to regionalize care for these patients to high-volume tertiary centers, through efforts by The Joint Commission, the American Heart Association, and local efforts. 6 –8 Rural hospitals, however, continue to struggle to achieve time metrics for these patients, possibly due to inadequate local procedures of care and staffing limitations. 9 –13
Telemedicine has been proposed as one strategy to improve access to high-quality emergency care. 14 We have previously shown that telemedicine is effective in decreasing time to transfer for patients with traumatic injuries and shortening the time to provider evaluation in remote rural centers. 15,16 Penetration of emergency department (ED)-based telemedicine has been rapid, 17,18 due to both improvements in broadband internet access and perceived value to rural clinicians, 19,20 and applications in rural hospitals have been myriad. 21 –24
Most prior studies of ED-based telemedicine applications have been either very small or have been the result of specific prospective pilot projects. 14 Prior reports have also focused carefully on the impact of telemedicine on medical decisions (e.g., decision to give tissue plasminogen activator (tPA) for stroke) without focusing on timeliness of supporting processes in rural hospitals (e.g., time to get a head computed tomography [CT] scan performed and interpreted). 25 –28 We have previously shown that implementation of telemedicine networks change processes and performance in local hospitals regardless of how frequently telemedicine is used, perhaps due to the effect of staff learning, mentorship, or ongoing professional education. 15 The objective of this study was to measure the effect of telemedicine implementation and use on timeliness of critical actions for MI and stroke care, with a focus on how these actions are affected by the availability of telemedicine. Our primary outcomes were time from ED arrival to electrocardiogram (EKG) for MI patients and time from ED arrival to head CT interpretation for stroke patients.
Materials and Methods
Study Design, Setting, and Population
This observational cohort study included all ED patients with MI or stroke treated in a network of critical access hospitals between November 2007 and August 2015 that subscribed to a single ED-based telemedicine provider (Avera eCare, Sioux Falls, SD). This telemedicine network provides a high-definition real-time video connection between a local ED and a board-certified emergency physician and experienced ED nurse in a hub-and-spoke model, available 24 h daily by activating a button at the rural hospital. This contract-based service currently serves 155 rural hospitals in 12 states, and the network has been expanding since 2009. Cases for this analysis were included from 19 federally designated critical access hospitals in the upper Midwest, selected because they used a common electronic medical record (EMR) and were able to provide cases originating before telemedicine adoption. Data collection for this study ended in 2015 so that all cases were classified using ICD-9 before the implementation of ICD-10. Details of this network have been reported previously. 29,30
Cases for this study were generated from an electronic query of each site's EMR, identifying cases based on the International Classification of Diseases, 9th edition, Clinical Modification (ICD-9-CM) ED diagnosis codes. Cases were compared with a record of all telemedicine encounters for each institution to determine whether telemedicine was consulted for each patient. Cases were also compared with service agreement records to determine whether telemedicine services were available on the date of the ED encounter. Cases during the month of implementation at each hospital were excluded to account for contamination from implementation activities (e.g., site visits, staff training, technical installation).
Data were collected both by querying data fields in the EMR and by manual chart review for variables not accurately or completely recorded in the medical record. Chart review was done by a trained research team member using a standard case report form. Cases were excluded if interventions were recorded at a time outside the ED stay. If ED length-of-stay was missing, we included all cases that had interventions performed within the 90th percentile of the distribution of ED length-of-stay for the entire data set for the qualifying diagnoses (222 min for stroke, 216 min for MI). This resulted in 73 (9.7%) MI cases and 14 (2.2%) stroke cases excluded for missing data.
Definitions
MI cases included patients with ICD-9-CM codes: 410.X1 (acute myocardial infarction, initial episode of care), 429.4 (functional disturbances following cardiac surgery), and 997.1 (cardiac complications not elsewhere classified). A subset of the MI cohort, ST-elevation myocardial infarction (STEMI), included ICD-9-CM code 410.X (excluding 410.7 and 410.9) only.
Stroke cases included both patients with acute ischemic stroke and with intracranial hemorrhage, because early evaluation is similar before the results of intracranial imaging are known. The stroke cohort consisted of patients with ICD-9-CM codes: 430 (subarachnoid hemorrhage), 431 (intracerebral hemorrhage), 433.X1 and 434.X1 (occlusion of cerebral arteries with ischemic infarction, cerebral embolism, and cerebral thrombosis). The study population included only stroke patients who presented to the ED within 3 h of symptom onset (no cases presented between 3 and 4.5 h), based on eligibility for intravenous tPA. 2 A subset of the entire cohort was defined for only those patients with acute ischemic stroke, excluding subarachnoid and intracranial hemorrhage.
Telemedicine availability was defined as an ED visit during a period with an active agreement for telemedicine services at the hospital. Telemedicine use was defined as an ED visit during which a video telemedicine encounter was initiated with the hub.
Time to first available provider was defined as the time from ED presentation to either being evaluated by a local provider (recorded as the time a local provider “signed up” for a patient in the EMR) or a telemedicine physician starting an evaluation, whichever came first.
Key Outcomes
Cardiac analysis
The primary outcome for the cardiac analysis was time-to-EKG, measured as the duration of time between ED arrival and completion of the EKG, as documented in the EMR. The secondary outcome was time-to-fibrinolysis for those patients receiving fibrinolysis (among those with STEMI). Because we had no data for patients after inter-hospital transfer and none of the participating hospitals had cardiac catheterization labs, percutaneous coronary intervention was not studied as an outcome. A post hoc analysis was performed for MI patients to measure the time of first provider evaluation, because this time was likely related to EKG interpretation.
Stroke analysis
The primary outcome for the stroke analysis was time-to-head CT interpretation. The time of head CT interpretation was defined as the time the first radiology report for the head CT (even if that report was a preliminary report) was available in the EMR. The secondary outcomes were time-to-tPA among those who received tPA for acute ischemic stroke and the proportion of those with acute ischemic stroke presenting within 3 h who received tPA. No contraindications for tPA were available from the medical record.
Analysis
Bivariate analyses were performed in both cardiac and stroke cohorts to compare demographic, discharge, and care-associated characteristics with telemedicine availability and use. Categorical variables were compared with the chi-squared test, parametric continuous variables were compared with the independent samples t-test, and nonparametric continuous variables (time-to-event analyses) were compared with the Wilcoxon rank sum test.
To measure the relationship between telemedicine and timeliness parameters, generalized estimating equations (GEE) were used, clustered on the presenting hospital to account for hospital-level differences in how quickly interventions were performed. Both telemedicine availability and telemedicine use were included as dichotomous variables of interest in the final model, because preliminary models including them separately yielded the same effect estimates. The GEE model used an exchangeable covariance structure and an identity link, and no additional covariates were included in the model. Primary and secondary outcomes were evaluated in different models to understand the effect of telemedicine on different parameters of timeliness of care.
For time-to-event outcomes, we decided a priori to analyze time with a log-transformation to satisfy the assumptions of regression analyses.
Statistical significance was set at α < 0.05 using two-tailed tests, and all analyses were conducted using SAS 9.4 (SAS Institute, Cary, NC). This study was approved by the local institutional review board under waiver of informed consent and is reported using the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) Statement. 31
Results
Of all patient visits to participating EDs, 756 records (0.6%) were included in the cardiac cohort and 638 records (0.5%) were included in the stroke cohort. Forty-one percent (n = 218) of cardiac patients had telemedicine used, and 26% (n = 200) had STEMI of which 68 (34%) had telemedicine used. Of STEMI patients, 18.6% (n = 34) received fibrinolytic therapy.
Twenty-three percent (n = 140) of stroke patients presented to the ED within 3 h (analysis population), and 43 of those (39%) had telemedicine used. Most stroke patients were transferred to another inpatient facility (61.5%), and this proportion was much higher in telemedicine cases (81%, Table 1). Of all ischemic stroke patients, 9.3% (n = 20) had tPA administered.
Discharge, and Care-Associated Characteristics of Telemedicine Utilization in the Emergency Department Among Myocardial Infarction and Stroke Cohorts
Cells with a value less than 5 have been suppressed.
CT, computed tomography; ED, emergency department; EKG, electrocardiogram; IQR, interquartile range; STEMI, ST-elevation myocardial infarction; tPA, tissue plasminogen activator.
Cardiac: Electrocardiogram Time (Primary Analysis)
The median time-to-EKG (from patient arrival) in MI patients in this cohort was 20 min (interquartile range [IQR] 11–33). Time-to-EKG did not differ in patients diagnosed with MI after telemedicine was implemented in each hospital (log-transformed time 4% shorter, 95% confidence interval [CI] −3% to 10%) or when telemedicine was used (compared with contemporary cases when it was not used) (1% shorter, −4% to 7%, Table 2). In MI patients, the time to first available provider (either telemedicine physician, local physician, or local advanced practice provider) was shorter in telemedicine cases (10.2 min vs. 16.3 min, p < 0.001).
Association of Telemedicine with Time to tPA/CT Interpretation and Those Given tPA (Stroke Within 3 h of Symptom Onset) and Fibrinolysis and EKG Time and Fibrinolysis Rate (Myocardial Infarction and STEMI Cohorts)
Clustered on hospital.
A constant of 61 was added to times to allow for log transformation of zeros.
CI, confidence interval; MI, myocardial infarction; NA, not applicable; OR, odds ratio.
Cardiac: Fibrinolytic Administration and Timeliness (Secondary Analyses)
The median door-to-needle time (time from ED arrival to administration of fibrinolytic therapy) among patients with STEMI was 48 min (IQR 31–67.5). Fibrinolytic therapy was much more likely in STEMI patients who had telemedicine activated (aOR 6.2, 95% CI 2.3–16.3). Among those who received fibrinolytic therapy, timeliness was not associated with either telemedicine availability or telemedicine use (p = 0.576 and p = 0.884, respectively; Table 2).
Stroke: Head CT Interpretation (Primary Outcome)
The median time from ED arrival to head CT interpretation was 57.5 min (IQR 40–87). Adjusting for differences between hospitals, head CT interpretation time was faster in stroke patients that had telemedicine used (15% shorter time, 95% CI 4–26%) (Table 2). The implementation of telemedicine (irrespective of use) was not associated with a significant decrease in the time to head CT interpretation (13% shorter time, 95% CI −1% to 27%).
Stroke: Thrombolytic Administration and Administration Time (Secondary Outcome)
The median time to thrombolysis in stroke patients receiving tPA was 100.5 min (IQR 65–115.5 min). Among ischemic stroke patients, tPA administration was higher when telemedicine was used (aOR 3.5, 95% CI 1.5–8.2). Among those who received tPA, neither telemedicine availability nor use had an effect on timeliness of tPA administration (p = 0.180 and p = 0.667, respectively).
Discussion
Telemedicine has been used extensively for time-sensitive ED applications, but few studies have reported the value of ED-based telemedicine in changing the process of rural clinical care. For MI patients, multiple reports of the impact of early transmission of EKG data on medical decision making have suggested that timely prehospital activation of a cardiac catheterization lab can decrease door-to-balloon time and may decrease in-hospital mortality. 32 –34 The value of telemedicine in many of these studies is to rapidly identify ST-elevation, so moving the transmission or reading of 12-lead EKGs earlier in the transitions of care has been an effective strategy to improve time efficiency. 35 –37 Similar findings have been observed in stroke care. Tele-stroke applications have successfully increased the diagnosis of stroke and the appropriate and timely administration of tPA. 38 More widespread application of these networks has even decreased racial and ethnic disparities in access to high quality stroke services. 39
In contrast to some of these prior reports, however, we focused on time-based quality metrics very early in the evaluation of these patients: time-to-EKG and time-to-head CT interpretation. The reason that we selected these metrics was that they are not specifically influenced by telemedicine-based decision making, but they could be impacted by refining systems of care in response to the relationship with a telemedicine provider.
Similar to prior reports, we found that patients who have telemedicine used for cardiac and stroke care have thrombolytics administered more often. This observation could be either causal or reverse causal. For instance, providers who appropriately diagnose STEMI or acute stroke may activate telemedicine for help with inter-hospital transfer, decision making, or activating a remote cardiac catheterization lab (reverse causality, because the need for transfer leads to telemedicine use rather than telemedicine use causing inter-hospital transfer). We have reported previously in trauma patients that telemedicine use is strongly associated with severity of illness and inter-hospital transfer, but in this case telemedicine didn't cause inter-hospital transfer. It was simply a reflection of the appropriate identification of patients who needed to be transferred. 29 These observations could be evidence of the perceived value of telemedicine to rural clinicians—in fact, few qualifying cases of either STEMI or stroke received thrombolytic therapy without telemedicine involvement.
The impact of telemedicine on head CT performance in potential stroke patients is interesting. Although there is no effect for timeliness of head CT interpretation after telemedicine is implemented in a hospital, head CT interpretation is still faster in cases where telemedicine is used.
This finding suggests that the effect may actually be from a provider ordering a head CT more quickly, rather than simply streamlining care from a systems-based perspective. This provider-level effect is likely, because the telemedicine interaction between the emergency physician and the hub are unlikely to impact the workflow of a radiology technician or a remote radiologist (since the radiologist is unaffiliated from the ED-based telemedicine service).
The effect on time-to-EKG, however, is modest. Presumably, the telemedicine provider is not directing the nurse or technician either to obtain an EKG or how to perform the EKG. It is also unlikely that the telemedicine provider is suggesting that a local provider order an EKG for a patient with chest pain faster than the local provider otherwise would have (and this may be accomplished by nurse-driven protocol). A difference that may exist, though, could be the timeliness of EKG interpretation. Although this is not a parameter that we could measure in this cohort, we have previously shown that time-to-provider is faster with the use of telemedicine in this network 15 —likely attributable to hospitals that do not have on-site providers 24 h daily. In these hospitals, the ability for a telemedicine provider to interpret an EKG more quickly and initiate therapy may continue to be faster, and in fact the time-to-provider was ∼6 min faster in telemedicine cases than in nontelemedicine cases (which aligns with our previously reported report 16 ).
We have observed in our prior work that X-ray and CT utilization increases in trauma patients after the implementation of a telemedicine program in rural hospitals. 15 We have attributed these changes to learning effects—the concept that a mentored provider interaction for a specific clinical case can be a very powerful strategy for changing clinician behavior. Based on the findings from this study, despite telemedicine interactions changing provider behavior, those changes may not extend to the entire healthcare team. We had hypothesized that if local staff sensed that telemedicine staff expect the EKG to be performed more quickly, they might prioritize the EKG earlier in the care of patient evaluation and the EKG may actually be performed sooner. Our findings might suggest that individual rural EDs may have already optimized their care processes and that further improvement in timeliness of diagnostics may require additional resources (staff, equipment) that are not provided by the telemedicine connection.
A surprising finding was the decreased use of thrombolytic therapy after telemedicine adoption, primarily for encounters for which telemedicine was available but not used. This observation could be secondary to improved appropriateness of tPA ordering after telemedicine implementation, increased diagnostic coding for stroke, or good risk stratification such that tPA ineligible patients did not have telemedicine consulted. The increase in tPA use in the telemedicine-activated patient encounters makes this finding more likely to be reflective of patient selection.
Our study has several limitations. First, the data were collected from retrospective medical records. While this allowed us to measure the impact of telemedicine robustly, it also restricted our available data to those parameters recorded in the medical record. Second, because it is retrospective, we do not have accurate severity of illness information. The NIH Stroke Scale and quantitative information about the severity of MI were not routinely recorded, and the medical record narrative was insufficiently detailed for retrospective scoring.
In conclusion, telemedicine influences the timeliness of the early evaluation for rural ED patients with stroke. Interestingly, the lack of effect on EKG timeliness may reflect local efficiency not influenced by telemedicine-enabled decision making. The proportion of eligible patients who receive fibrinolysis was increased, but the timeliness of fibrinolysis was not affected. These data provide important insights into the manner in which telemedicine influences ED-based care in rural hospitals, and support previous research that telemedicine can improve early time-sensitive care. Future work should better characterize the rural-hub relationship and refine strategies for better supporting rural systems of care.
Footnotes
Acknowledgments
This project was supported by the Health Resources and Services Administration (HRSA) of the U.S. Department of Health and Human Services (HHS) under grant G01RH27868 entitled Evidence-Based Tele-Emergency Network Grant Program and grant number 6 UICRH29074-01-01 entitled Telehealth-Focused Rural Health Research Center. This content and conclusions are those of the authors and should not be construed as the official position or policy of, nor should any endorsements be inferred by, the HRSA, HHS or the U.S. Government. The authors acknowledge Werner Berg, Katie Moore, Shiann Shipp, and Luke Mack for their help with data preparation, study planning, and analysis. This analysis was conducted by the Rural Telehealth Research Center at the University of Iowa.
Disclosure Statement
BS, AW and AB are employed by Avera eCARE, which provides emergency department based telemedicine services. All other authors have nothing to disclose.