The Effect of Sepsis Recognition on Telemedicine Use in Rural Emergency Department Sepsis Treatment

Abstract

Background:

Provider-to-provider emergency department telehealth (tele-ED) has been proposed to improve rural sepsis care. The objective of this study was to measure the association between sepsis documentation and tele-ED use, treatment guideline adherence, and mortality.

Methods:

This analysis was a multicenter (n = 23) cohort study of sepsis patients treated in rural emergency departments (EDs) that participated in a tele-ED network between August 2016 and June 2019. The primary outcome was whether sepsis was documented explicitly in the clinical note impression in the local ED, and the primary exposure was rural tele-ED use, with secondary outcomes of time to tele-ED use, 3-h guideline adherence, and in-hospital mortality.

Results:

Data from 1,146 rural sepsis patients were included, 315 (27%) had tele-ED used and 415 (36%) had sepsis recognized in the rural ED. Tele-ED use was not independently associated with sepsis recognition (adjusted odds ratio [aOR]: 1.23, 95% confidence interval [CI]: 0.90–1.67). Sepsis recognition was associated with earlier tele-ED activation (adjusted hazard ratio 1.66, 95% CI: 1.28–2.15) and greater 3-h guideline adherence (aOR 1.37, 95% CI 1.03–1.83) Sepsis recognition was not independently associated with mortality (aOR 1.32, 95% CI 0.97–1.80).

Conclusions:

Although tele-ED care is a promising strategy to improve sepsis outcomes, its use was limited by under-recognition of sepsis in rural EDs.

Introduction

Sepsis contributes to 850,000 emergency department (ED) visits annually in the United States (U.S.).¹ Of those with community-onset sepsis, mortality is 13%, making sepsis a leading cause of death in U.S. hospitals and a significant public health problem.^2,3 Sepsis also exhibits a significant volume-outcome relationship, with hospitals that have lower sepsis volumes having higher sepsis mortality than those with high volumes.^4,5

Vague symptoms at presentation make sepsis diagnosis challenging. Up to one-third of patients present without classical symptoms, and older and immunocompromised patients are particularly at risk for delayed recognition—possibly affecting early treatment and clinical outcomes.^6–8 Variability in sepsis presentations has prompted the use of standardized tools to improve recognition, but universal recognition remains elusive.⁹

The Surviving Sepsis Campaign (SSC) publishes guidelines that recommend sepsis treatments, and timely high-quality sepsis care has been shown to improve clinical outcomes.^10,11 Telehealth has been proposed as one way to mitigate sepsis disparities in low-volume hospitals, bringing tertiary care expertise into low-volume EDs.¹² Telehealth has been used in EDs for acute conditions such as ischemic stroke, myocardial infarction, and trauma, and it may impact care in sepsis.^12–18 We previously reported that in a multicenter observational study, telehealth use did not improve clinical outcomes, but guideline adherence in that cohort was low.¹⁹

The objective of this study was to determine whether sepsis recognition in the ED is associated with ED-based telehealth (tele-ED) use, increased SSC bundle adherence, and decreased mortality. We performed a secondary analysis of the TELEmedicine as a Virtual Intervention for Sepsis in Emergency Departments (TELEVISED) study to identify (1) the proportion of rural sepsis patients recognized in local ED documentation, (2) predictors of local sepsis recognition, (3) the association between sepsis recognition, tele-ED use, and tele-ED timing, (4) association between sepsis recognition and SSC bundle adherence, and (5) the association between sepsis recognition and in-hospital mortality. We hypothesized that sepsis recognition would be associated with increased tele-ED use, higher SSC bundle adherence, and lower in-hospital mortality. This observation would be important because if the reason that tele-ED does not improve clinical outcomes is because of missed diagnoses, then developing telehealth-enabled surveillance could be an impactful future intervention.

Methods

STUDY DESIGN AND SETTING

This analysis is a cohort study conducted as a secondary analysis of the TELEVISED study, which was a retrospective, multicenter (n = 23) study designed to measure the association between tele-ED use and clinical outcomes conducted in rural EDs between August 1, 2016 and June 30, 2019. Participating hospitals were rural as defined by the Federal Office of Rural Health Policy and took part in a single provider-to-provider tele-ED network.²⁰ Detailed methods of the study and characteristics of included sites have been reported elsewhere, and the results of the main study have been previously published.^19,21 This study was approved by local institutional review boards under waiver of informed consent and it is reported in accordance with the Strengthening Observational Studies in Epidemiology statement.²²

SELECTION OF PATIENTS

An electronic health record (EHR) data query was used to identify all adults (age ≥18 years) meeting the definition of sepsis in participating rural EDs during the study period. All of the following four inclusion criteria were required to meet our operational definition of ED sepsis as we have used previously (to account for explicit sepsis diagnosis codes in the ED being insensitive): (1) inpatient discharge diagnosis of sepsis by International Classification of Diseases, 10th edition, Clinical Modification (ICD-10-CM) codes21, (2) infection suspected in the ED based on documentation in the ED clinical note (EHR), (3) organ failure in the ED (defined as a Sequential Organ Failure Assessment Score [SOFA] of two or greater or a change of at least two in patients with chronic disease), and (4) at least two systemic inflammatory response syndrome (SIRS) criteria in the ED.^23,24 We used such an expansive definition because there could be patients diagnosed with sepsis using a single definition that were excluded from our study (since no single definition was sufficient for enrollment)—but all of the patients included met all available definitions, including organ failure and clinical screening criteria in the ED. Notably, patients could meet this criteria without sepsis being the ED diagnosis—even if formal sepsis diagnostic criteria were met.^23,25 We excluded any patient with no local diagnoses clearly documented in the EHR, as these medical records were likely incomplete. Data were abstracted by trained data abstractors using a structured data collection instrument according to the methods outlined by Kaji et al.²⁶

INTERVENTIONS

All EDs participated in a single hub-and-spoke tele-ED network and used a standard nurse-directed screening protocol. The local screening protocol was positive if (1) the local nurse “suspected infection” and (2) met two of the following criteria: temperature <96.8 °F or >100.9 °F, pulse >90 beats per min, systolic blood pressure <90 mmHg, mean arterial blood pressure <65 mmHg, oxygen saturation <90%, respiratory rate >20 breaths/min, lactate >2 mmol/L, white blood cell count <4,000 cells/μL or >12,000 cells/μL. Local patients were screened by a triage nurse, and positive screens were reported in real time to the local treating clinician. Tele-ED use was determined at the discretion of the local provider, and it was encouraged for patients screening positive (regardless of whether patients were treated by physicians or advanced practice providers [APPs]), and we described the full screening protocol previously.²⁷

The tele-ED network was an on-demand monthly subscription-based provider-to-provider real-time consultation service based in Sioux Falls, SD, that could be activated by local providers or nurses 24 h per day by pressing a button on the wall of the patient treatment room. Tele-ED consultation was provided by a board-certified emergency physician and an experienced ED nurse through a high-definition video connection, and it could be used for any clinical scenario in participating EDs. These consultants had access to the local EHR and were able to review records, place orders, provide real-time on-camera guidance, and arrange interhospital transfer when necessary.

EXPOSURES AND OUTCOMES

Our primary exposure was whether sepsis was recognized in the rural ED (determined by EHR documentation), and our primary outcome was tele-ED use. Secondary outcomes were timing of tele-ED activation, SSC 3-h bundle adherence (i.e., measure lactate, draw blood cultures before antibiotics, administer broad-spectrum antibiotics, and administer 30 mL/kg crystalloid fluid bolus if lactate is >4 mmol/L or systolic blood pressure is <90 mmHg), and in-hospital mortality.¹⁰

DETERMINATION OF SEPSIS RECOGNITION

All patients in this analysis had sepsis, but sepsis recognition was determined locally in the index ED. To determine whether patients had local sepsis recognition, we conducted free-text analysis of ED diagnoses by two independent reviewers, who were trained using sample cases. All recorded diagnoses from the ED “impression/diagnosis” (not including differential diagnoses documented only in the medical decision-making) were reviewed and classified into one of four mutually exclusive groups: (1) sepsis, (2) infection but no diagnosis of sepsis, (3) signs or symptoms that could be consistent with sepsis but without diagnosis of sepsis or infection, and (4) other unrelated diagnoses only. In this classification, sepsis included sepsis, septic shock, severe sepsis, or any other form of systemic response to infection (including SIRS as a diagnosis), but not including documentation of localized infection alone (e.g., septic arthritis). For example, a patient with an ED impression of “pneumonia” would be classified as “infection,” a patient with an ED impression of “hypoxia, hypotension, and acute kidney injury” would be diagnosed as “signs or symptoms consistent with sepsis,” and a diagnosis of “atrial fibrillation with rapid ventricular response” would be diagnosed as “other unrelated diagnoses only.” Notably, patients were included in the overall study if local clinicians documented in their medical decision-making that they suspected infection (e.g., as a differential diagnosis), but cases were not classified as infection unless an infection was listed in the “impression” portion of the note. While sepsis may have been recognized by the clinician but not explicitly documented (i.e., we did not expect 100% recognition), we selected this explicit definition to limit bias and detect differences in recognition, as other authors have used.⁶ Disagreements between the two reviewers were resolved by a third reviewer. All cases without documented recognition of sepsis (but who met the diagnostic criteria for sepsis) were classified as having no local sepsis recognition.

DEFINITIONS

Fever was defined as temperature >38.0°C and hypothermia was defined as temperature <36.0°C. Hypotension was defined as systolic blood pressure <90 mmHg. Elevated lactate was defined as lactate >2 mmol/L. Tachycardia was defined as heart rate >100 beats/min.

“Time zero” for time-based analyses was defined as the time of the initial set of vital signs in the ED when patients first met SIRS criteria for bundle adherence metrics and the time of ED arrival for time-to-tele-ED activation.

STATISTICAL ANALYSIS

We calculated the percentage of locally recognized sepsis cases as the percentage of total included patients with an ED diagnosis of sepsis. Then we measured the association between recognized sepsis and variables of interest, using univariate logistic regression. Variables of interest were selected a priori and included source of infection, fever, hypothermia, hypotension, tachycardia, elevated lactate, surgery during hospitalization, SOFA score, and ED provider type (APP [either solo provider or provider with on-site physician] vs. physician only).

We used multivariable generalized estimating equations (GEE) with a logit link to identify predictors of locally recognized sepsis using our a priori variables. We then removed variables based on statistical criteria using the Quasi-Likelihood Information Criterion to achieve parsimony, we required tele-ED use to remain in the model and we included hospital identifier as a random effect to account for clustering.²⁸ All covariates were categorical, and we confirmed model goodness of fit.²⁹ We screened for collinearity and interaction terms. Then we constructed another GEE model using similar methods to measure the association between sepsis recognition (exposure) and 3-h SSC bundle adherence (outcome) and a separate model to measure the association between sepsis recognition (exposure) and in-hospital mortality (outcome).

Among patients who had tele-ED used, we compared time-to-tele-ED activation using Kaplan–Meier curves and the log-rank test to identify whether patients with recognized sepsis had faster tele-ED activation. Data for this analysis were right censored if patients left the ED or died. Then, we used a Cox proportional hazards model adjusted for the same covariates to measure the association (adjusted hazard ratio [aHR]) between sepsis recognition (exposure) and time-to-tele-ED (outcome). We assessed the proportional hazards assumption of the adjusted model using a global test based on the Schoenfeld residuals.

Analyses were conducted as complete case analysis without imputation, but missing values on the variables of interest were rare. All statistical procedures were conducted using two-tailed tests, p < 0.05 was considered statistically significant, and analyses were conducted using SAS v 9.4 (SAS Institute, Cary, NC).

Results

BASELINE CHARACTERISTICS

We included 1,146 patients (Fig. 1), and 315 (27%) had tele-ED used. Patients had a median age of 72 years (interquartile range [IQR] 62–82) and 517 (45%) were female. The most common source of infection was pneumonia (n = 619, 54%), and the median ED sequential organ failure assessment (SOFA) score was 4 (IQR 3–5). A total of 347 (30%) patients were transferred to another hospital, and 111 (10%) patients died in the hospital. Baseline characteristics are detailed in Table 1.

Fig. 1.

Selection of participants.

Table 1.

Characteristics of Participants

	ALL CASES (n = 1,146) N (%)	NO LOCAL SEPSIS RECOGNITION (n = 731) N (%)	LOCAL SEPSIS RECOGNITION (n = 415) N (%)	DIFFERENCE (95% CI)
DEMOGRAPHICS
AGE, YEARS
<30	9 (0.8)	5 (0.7)	4 (1.0)	0.3 (−1.4, 0.8)
≥30 and <40	38 (3.3)	26 (3.6)	12 (2.9)	0.8 (−1.4, 2.8)
≥40 and <50	68 (5.9)	43 (5.9)	25 (6.0)	0.1 (−3.0, 2.8)
≥50 and <60	116 (10.1)	64 (8.8)	52 (12.5)	3.7 (−0.1, 7.5)
≥60 and <70	291 (25.4)	178 (24.4)	113 (27.2)	2.8 (−2.5, 8.1)
≥70 and <80	267 (23.3)	174 (23.8)	93 (22.4)	1.4 (−3.7, 6.5)
≥80 and <90	274 (23.9)	185 (25.3)	89 (21.5)	3.8 (−1.3, 8.9)
≥90	83 (7.2)	56 (7.7)	27 (6.5)	1.2 (−1.9, 4.3)
Age, median (IQR)	72.0 (62.4–82.1)	73.6 (62.9–82.9)	70.1 (61.4–80.8)	3.5 (1.0, 6.1)
Female	517 (45.1)	332 (45.4)	185 (44.6)	0.8 (−5.2, 6.8)
Body mass index, kg/m2
<18.5	39 (3.6)	28 (4.0)	11 (2.9)	1.1 (−1.1, 3.3)
≥18.5 and <25	283 (26.4)	177 (25.5)	106 (28.0)	2.5 (−3.1, 8.1)
≥25 and <30	290 (27.0)	191 (27.5)	99 (26.1)	1.4 (−6.9, 4.1)
≥30 and <35	198 (18.5)	128 (18.4)	70 (18.5)	0.1 (−4.8, 5.0)
≥35	263 (24.5)	170 (24.5)	93 (24.5)	0 (−5.4, 5.4)
Latinx	13 (1.2)	6 (0.8)	7 (1.8)	1.0 (−0.5, 2.5)
Race
White	1,016 (89.4)	652 (90.1)	364 (88.4)	1.7 (−2.1, 5.5)
African American	8 (0.7)	3 (0.4)	5 (1.2)	0.6 (−0.4, 2.0)
American Indian or	92 (8.1)	57 (7.9)	35 (8.5)	0.6 (−2.7, 3.9)
Alaska Native
Other	20 (1.8)	12 (1.7)	8 (1.9)	0.2 (−1.4, 1.8)
PATIENT CHARACTERISTICS
Heart rate, beats per min, median (IQR)	106 (94–120)	103 (93–1,169)	112 (97–127)	9 (6.0, 12.0)
Temperature (degrees Celsius), median (IQR)	37.5 (36.7–38.4)	37.4 (36.4–38.3)	37.8 (36.7–38.7)	0.3 (0, 0.6)
Lactate, mmol/L, median (IQR)	1.90 (1.30–2.90)	1.60 (1.20–2.40)	2.60 (1.90–3.90)	1.0 (0.8, 1.2)
Triage systolic blood pressure, mmHg, median (IQR)	130 (109–149)	131 (112–151)	126 (104–147)	5.0 (0.5, 9.5)
SOFA Score
2–3	560 (48.9)	372 (50.9)	188 (45.3)	5.6 (−0.4, 11.6)
4–7	521 (45.5)	332 (45.4)	189 (45.5)	0.1 (−5.9, 6.1)
≥8	65 (5.7)	27 (3.7)	38 (9.2)	5.5 (2.4, 8.6)
SOFA Score, median (IQR)	4 (3–5)	3 (3–5)	4 (3–5)	1 (0.7, 1.3)
Tachycardia	714 (62.3)	420 (57.5)	294 (70.8)	13.4 (7.7, 19.0)
Hypotension	94 (8.2)	35 (4.8)	59 (14.2)	9.4 (5.7, 13.1)
Source of Infection
Respiratory	619 (54.0)	441 (60.3)	178 (42.9)	17.2 (11.5, 23.3)
Genitourinary	209 (18.2)	125 (17.1)	84 (20.2)	3.1 (−1.6, 7.8)
Abdominal	69 (6.0)	43 (5.9)	26 (6.3)	2.9 (−2.5, 3.3)
Device-Related	2 (0.2)	0 (0)	2 (0.5)	0.5 (−0.2, 1.2)
Skin/Soft Tissue	109 (9.5)	74 (10.1)	35 (8.4)	1.7 (−0.5, 1.8)
Meningitis	13 (1.1)	7 (1.0)	6 (1.5)	0.5 (−0.9, 1.9)
Unknown	125 (10.9)	41 (5.6)	84 (20.2)	14.6 (10.4, 18.8)
Surgery during hospital stay	77 (6.7)	43 (5.9)	34 (8.2)	2.3 (−0.8, 5.4)
Interhospital Transfer	347 (30.3)	196 (26.8)	151 (36.4)	9.6 (4.0, 15.2)
Length of hospital stay (days), median (IQR)	4.9 (3.2–8.3)	4.9 (3.2–8.0)	5.0 (3.1–9.0)	0.1 (−0.5, 0.6)
Provider Type
Physician	975 (85.1)	635 (86.9)	340 (81.9)	5.0 (0.6, 9.4)
APP with physician	112 (9.8)	56 (7.7)	56 (13.5)	5.8 (2.0, 9.6)
APP alone	59 (5.2)	40 (5.5)	19 (4.6)	0.9 (−3.5, 17.1)
Arrive by Ambulance	561 (50.0)	341 (47.4)	220 (54.7)	7.3 (−1.2, 13.4)
COMORBIDITIES
COPD	361 (31.5)	259 (35.4)	102 (24.6)	10.8 (5.4, 16.2)
Cirrhosis	30 (2.6)	16 (2.2)	14 (3.4)	1.2 (−0.8, 3.2)
Solid organ transplant	12 (1.1)	7 (1.0)	5 (1.2)	0.2 (−1.1, 1.5)
Cancer	339 (29.6)	221 (30.2)	118 (28.4)	1.8 (−3.9, 7.1)
Diabetes	426 (37.2)	281 (38.4)	145 (34.9)	3.5 (−2.3, 9.3)
Chronic Dialysis	45 (3.9)	28 (3.8)	17 (4.1)	0.3 (−2.1, 2.7)
Asthma	138 (12.0)	93 (12.7)	45 (10.8)	1.9 (−1.9, 5.7)
Hypertension	824 (71.9)	531 (73.0)	293 (70.6)	2.4 (−3.0, 7.8)
OUTCOMES
Tele-ED Use	315 (27.5)	171 (23.4)	144 (34.7)	11.3 (5.8, 16.8)
SSC 3-h bundle adherence	150 (13.1)	88 (12.0)	62 (14.9)	2.9 (−1.3, 7.1)
In-Hospital Mortality	111 (9.9)	64 (9.0)	47 (11.5)	2.5 (−1.2, 6.3)

95% CI, 95% confidence interval; APP, advanced practice provider; COPD, chronic obstructive pulmonary disease; IQR, interquartile range; SOFA, Sequential Organ Failure Assessment; SSC, Surviving Sepsis Campaign.

LOCALLY RECOGNIZED SEPSIS

In our sample, 415 (36%) had sepsis recognition documented in the local ED. Other nonsepsis local diagnoses included infection (n = 624, 54%), signs, symptoms, or findings consistent with sepsis (n = 85, 7%), and an unrelated diagnosis only (n = 22, 2%). Sepsis was more likely to be diagnosed in patients who presented with fever (41% vs. 33%, difference 8%, 95% confidence interval [CI] 2–14%), hypotension (63% vs. 34%, difference 29%, 95% CI 19–39%), or higher SOFA scores (53% with SOFA 7 or greater vs. 34% with SOFA 2–6, difference 19%, 95% CI 9–29%). Sepsis recognition was higher in those treated by an APP (with or without onsite physician supervision) compared with those treated by a physician alone (44% vs. 35%, difference 9%, 95% CI 1–17%). Tele-ED was used more frequently in those with sepsis recognition compared with those without (45% vs. 23%, difference 13%, 95% CI 7–19%). In multivariable modeling, fever, hypotension, tachycardia, elevated lactate, and APP-directed care (both alone and with an onsite physician) remained predictive of increased odds of locally recognized sepsis and pneumonia (compared with all other infection sources) was associated with decreased local recognition of sepsis. Tele-ED use was not associated with sepsis recognition in multivariable modeling (adjusted odds ratio [aOR]: 1.23, 95% CI 0.90–1.67, Fig. 2).

Fig. 2.

Factors predicting local recognition of sepsis. This forest plot depicts the association between signs and symptoms and local recognition of sepsis. Gray circles represent unadjusted values and black squares represent adjusted values. Variables used in statistical adjustment include all those variables shown in this forest plot.

The hazard function associated with tele-ED activation was 54% earlier in those with recognized sepsis versus those without (unadjusted HR: 1.54, 95% CI 1.20–1.98) and this difference persisted when adjusted for ED disposition and provider type (aHR: 1.66, 95% CI 1.28–2.15, Fig. 3), suggesting that recognition was associated with earlier tele-ED activation.

Fig. 3.

Time to tele-ED use, stratified by sepsis recognition. This figure depicts the cumulative probability of tele-ED consultation, stratified by sepsis recognition. Data are right censored if patients left the ED or died. ED, emergency department, APP, advanced practice provider.

SSC BUNDLE ADHERENCE AND IN-HOSPITAL MORTALITY

SSC 3-h bundle adherence was similar between those with and without locally recognized sepsis in bivariate analyses (15% vs. 12%, difference 2.9%, 95% CI −1.3–7.1%). After covariate adjustment, sepsis recognition was associated with greater 3-h bundle adherence (aOR 1.37, 95% CI 1.03–1.83). In-hospital adjusted mortality in our multivariable model was not significantly lower in those with local sepsis recognition (aOR 1.32, 95% CI 0.97–1.80, Fig. 4).

Fig. 4.

SSC 3-h bundle adherence and in-hospital mortality, stratified on sepsis recognition. This figure depicts the association between local sepsis recognition, SSC bundle adherence (3 h), and in-hospital mortality. On the left panel, the raw adherence numbers are reported (gray bars represent recognized sepsis, and black bars represent unrecognized sepsis). On the right panel, the unadjusted and adjusted odds ratios of local sepsis recognition and each outcome are presented (log scale, white circles represent unadjusted analyses, and black squares represent adjusted analyses). Error bars represent 95% CIs. Note that the point estimate and CI for the unadjusted estimate for the 3-h bundle is hidden by the symbol, but it is OR 0.97 (95% CI 0.95, 1.00). † = adjusted for fever and SOFA score. CI, confidence interval; OR, odds ratio; SOFA, Sequential Organ Failure Assessment; SSC, Surviving Sepsis Campaign.

Discussion

In our analysis of a multicenter cohort study in rural EDs, we found that tele-ED use was more common in those recognized as having sepsis. However, adjusting for factors associated with sepsis recognition, tele-ED use was not independently associated with sepsis recognition. Furthermore, we observed that a minority of sepsis patients in participating rural hospitals were recognized as having sepsis—regardless of tele-ED use. Even in those with recognized sepsis, tele-ED was used in only a minority of cases. In our prior report of the TELEVISED study, we showed that tele-ED use was not associated with improved clinical outcomes or SSC bundle adherence.¹⁹ The finding from this analysis is relevant because under-recognition could be a key explanation for why our prior work did not show the differences in processes and outcomes we hypothesized we might observe. If tele-ED were contributing to increased recognition and earlier treatment, we hypothesize that sepsis recognition would have been independently associated with tele-ED use.

Sepsis diagnosis—especially for patients with low illness severity—can be difficult, and under-recognition is an obstacle to appropriate and timely sepsis care^7,30 In a cohort study of 654 ED sepsis patients, 37% presented with vague symptoms, and the absence of explicit, traditional symptoms of sepsis was associated with delays in antibiotic therapy and higher mortality⁶ Others have associated misdiagnosis with delayed treatment and worse clinical outcomes.³¹ This finding parallels our results in that sepsis recognition was associated with improved guideline adherence, but we additionally found that the failure to recognize sepsis was associated with a lower frequency of tele-ED use.¹⁸

Many patients in whom sepsis was not recognized were appropriately diagnosed as having pneumonia, respiratory failure, or kidney injury—all diagnoses that these patients clearly had. One of the reasons sepsis diagnoses may be missed is that sepsis may be a secondary syndrome diagnosed in someone who may appropriately have a primary diagnosis of infection.³² Biomarkers have been proposed as one strategy to improve recognition of the constellation of findings that suggest a host immune response and may define sepsis, and EHR-based algorithms and ICU-based telehealth screening have also been proposed.^33–35

Provider-to-provider telehealth has been viewed as a major advantage for managing rare, life-threatening conditions in low-volume centers.³⁶ However, our findings highlight one significant shortcoming of this approach—when tele-ED use is low, appropriate on-demand tele-ED activation requires accurate local diagnosis.³⁷ Under-recognition of sepsis has been commonly reported.^31,38,39 Similar patterns have emerged in stroke, myocardial infarction, and aortic dissection, where treatment paradigms are clear and under-recognition is associated with morbidity and mortality.^40–42 Like other time-sensitive acute care conditions, sepsis treatment has been operationally protocolized, so tele-ED’s greatest potential to improve guideline adherence and outcomes may be from its ability to identify sepsis and initiate a predefined treatment bundle. That reality may also be reflected in differential tele-ED use between physicians and APPs, since less experienced providers may be more willing to ask for expert consultation in making complex treatment decisions even for recommended care elements. The fact that remote diagnostic surveillance screening is not a feature of most current tele-ED programs is a significant concern, and it may be a call to action to implement EHR-based and telemetry-based strategies to incorporate into tele-ED networks screening technology for conditions not recognized by local providers. The effectiveness of nurse-directed screening in low-volume vs. high-volume hospitals is unclear, but one might hypothesize that triage tools may affect EDs differently. Informatics and machine learning-based tools continue to improve, and providing health system-level surveillance—even remotely—may be one way to amplify the effect of tele-consultation by remote experts.⁴³

LIMITATIONS

Our study has several limitations. First, we relied on EHR documentation to indicate recognition of sepsis or infection, and such documentation may miss cases accurately recognized (but not adequately documented) by the treating clinician. Lack of documentation in the medical record has been hypothesized to decrease case capture for the sepsis core measure, but as most of our sites do not voluntarily report sepsis quality measures,⁴⁴ we think this factor is not likely relevant in our sample. Second, our observational study cannot be used to imply causality because we do not know whether tele-ED consultation is caused by sepsis recognition, diagnosis is caused by tele-ED consultation, or if clinical features that drive tele-ED consultation are coincidentally associated with accurate sepsis diagnosis. Finally, our multicenter study was conducted in a single tele-ED network that used common procedures at all sites, which may affect the generalizability of our results.

Conclusions

In conclusion, sepsis recognition in rural EDs participating in a tele-ED network was not independently associated with tele-ED use, but recognition was associated with increased guideline adherence. Future work will seek to identify ways to improve rural sepsis recognition using automated methods and tele-ED surveillance, and to identify barriers to effective tele-ED use.

Authors’ Contributions

A.M.K. participated in study concept and design, analysis and interpretation of data, and drafting the article. J.P.V. and U.O. participated in data analysis, drafting portions of the article, and statistical expertise. C.W., K.C., B.F., A.Z., and K.W. participated in acquisition of data, interpretation of data analysis, and critically revising the article for important intellectual content. K.K.H. and M.B.S. participated in study design, data interpretation, critically revising the article for important intellectual content, and statistical expertise. B.M.F., L.M., A.B., K.D., E.A.P., K.M., E.C., and C.R.C. participated in study design, interpretation of the data, and critically revising the article for important intellectual content. M.P.J. participated in study design, data analysis and interpretation, critically revising the article for important intellectual content, and statistical expertise. M.M.W. participated in the study concept and design, analysis and interpretation of data, critically revising the article for important intellectual content, and acquisition of funding. N.M.M. participated in the study concept and design, acquisition of data, analysis and interpretation of data, drafting the article, and acquisition of funding.

Institutional Review Board Approval

This study was approved by the University of Iowa Institutional Review Board IRB-001, approval 201901748.

Disclosure Statement

L.M., A.B., and K.D. are employed by an organization that provides direct telehealth services. All other authors report no conflicts of interest.

Disclaimer

These contents are solely the responsibility of the authors and do not necessarily reflect the views of the Agency for Healthcare Research and Quality, the Health Resources and Services Administration, or any other funding agency. Funding agencies had no role in the study design, data analysis or interpretation, writing the report, or the decision to publish.

Funding Information

This study was funded by the Agency for Healthcare Research and Quality (K08HS025753), the Institute for Clinical and Translational Science at the University of Iowa through a grant from the National Center for Advancing Translational Sciences at the National Institutes of Health (UL1TR002537), and the University of Iowa Department of Emergency Medicine. N.M.M. is additionally supported by funding from the Rural Telehealth Research Center with funding from the Health Resources and Services Administration (U3GRH40003).