TELEmedicine as an intervention for sepsis in emergency departments: a multicenter, comparative effectiveness study (TELEvISED Study)

This study is a multicenter (n = 25) retrospective propensity-matched comparative effectiveness study of tele-ED care for sepsis patients with sepsis. Initially, 30 hospitals were identified for participation, but data use restrictions only permitted 25 to participate. This manuscript is reported in accordance with the Template for Intervention Description and Replication checklist for Population Health and Policy (TIDieR-PHP) and the study was designed using the Strengthening the Reporting of Observational Studies in the Epidemiology (STROBE) statement: Guidelines for Reporting Observational Studies [48,49].

This study will include all adult (age ≥18 years) sepsis patients who presented to a participating rural ED between 1 August 2016 and 30 June 2019 (Table 1). Data collection/abstraction started on 10 October 2019 and is anticipated to be complete by 30 December 2020. Because of poor sensitivity in diagnosis code-based definitions of sepsis [50], we will use a multistep sepsis definition requiring identification of infection in the ED, organ failure in the ED, at systemic inflammatory response syndrome criteria in the ED and hospital diagnosis of both infection and organ failure (Figure 1). To identify hospital diagnosis of infection and organ failure, we will use the Fleischmann-Struzek approximation of sepsis using International Classification of Diseases, 10th edition, Clinical Modification (ICD-10-CM) discharge diagnoses or an explicit discharge sepsis code (R65.20 or R65.21) [4]. Infection in the ED is defined as the explicit documentation of a source of infection in the ED medical record. Organ failure in the ED is defined based on the Sequential Organ Failure Assessment (SOFA) score of ≥2 points (or a change of two points in those with pre-existing chronic disease) [51]. Systemic inflammatory response syndrome is defined according to the original Bone criteria as at least 2 of the following: body temperature greater than 38°C or less than 36°C; heart rate greater than 90 beats per minute; respiratory rate greater than 22 breaths per minute or pCO₂ less than 32 mmHg; or a white blood cell count greater than 12,000 cells/μl, less than 4000 cells/μl or the presence of greater than 10% immature neutrophils [52]. These criteria were selected to parallel the Sepsis-3 definition, but additionally required convincing evidence of recognized infection and objective organ failure in the ED [53]. All records include data on the ED visit and inpatient visits, even when patients are transferred between rural hospitals and tertiary centers. Any patients transferred to a hospital outside the network will be excluded for missing inpatient data.

As part of standard treatment, Avera eCARE uses established protocols for sepsis management. Standardized nurse-directed sepsis screening is used in all participating rural EDs (Figure 2), and tele-ED consultation is left to the discretion of the local treating clinician. Tele-ED providers are board-certified emergency physicians providing care via remote connection from SD, USA. The telemedicine hub uses computerized decision-support software during tele-ED consultations to guide recommendations for patient care, including fields prompting for sepsis guideline adherence elements (e.g., lactate measurement) and fields for recording completion times for each element of the SSC bundle elements. This real-time checklist aids hub staff in guiding local clinicians through elements of guideline-adherent sepsis care [47]. In addition to real-time clinical support, the telemedicine network sponsors regular continuing education events provided for nurses, advanced practice providers and physicians by remote education. Many clinicians also cite ongoing professional education provided in the context of clinical care as an additional benefit of tele-ED use [54]. The telemedicine service is currently provided in a self-sustaining financial model, and none of the research funding supporting this analysis supports telemedicine clinical operations.

Patients and members of the public were not involved in the development of the research question or study design, and they will not be involved in the conduct of the study.

Data collection will be performed using manual chart abstraction in accordance with the methods described by Kaji et al. [55]. Research team members will access the health system electronic medical record in which charts from all participating hospitals are maintained. We will use the original hospital medical record instead of the telemedicine record, because we want to limit ascertainment bias for non telemedicine patients and we have observed previously that telemedicine is rarely referenced in the primary medical record in participating hospitals [56]. Data will be abstracted by three research assistants blinded to the use of telemedicine and details of the study hypothesis, using a standard data collection form with a priori-defined data elements and a coding guide using a Research Electronic Data CAPture (REDCap) secure database (Supplementary Material 1). The case report form was developed by a research team member with significant training in both sepsis data abstraction and sepsis clinical practice, and the case report form was pilot-tested to ensure that data collection aligned with the order in which data could be abstracted efficiently. Real-time error checking and data validation algorithms were built into the database to reduce data entry errors.

All research assistants will have an introductory training to the medical record by a project trainer who has prior experience with sepsis data abstraction and use of the medical record. The trainer will conduct education sessions regarding the data collection tools with each of the three research assistants through one-on-one mentorship over a 4-week dedicated training period. Ten training records will be abstracted by all research assistants, and the trainer will review the data collected for each data element and provide feedback prior to certifying research assistants to abstract additional records. The trainer will continue reviewing portions of data collection until data collection is reliable for every research assistant. After training activities are complete, a 10% sample of all records will be abstracted by both a research assistant and the trainer, and discordance will be reviewed with research assistants on an ongoing basis for training and quality control.

Abstractor inter-rater reliability measurements (Cohen’s kappa) will be conducted on the decision for inclusion in the study cohort and on key data elements (hospital length-of-stay, mortality, timing of each SSC intervention and Acute Physiology and Chronic Health Evaluation Score, 2nd edition [APACHE-II] score). The study team determined a priori that agreement of greater than 90% on cohort inclusion, determination of adherence with the SSC 3-h bundle and outcome variables was acceptable.

The primary exposure of interest is tele-ED consultation. Tele-ED use is defined as using video tele-ED during the rural ED visit for sepsis, and it will be defined by identifying the presence of a matching visit in the tele-ED call log maintained by the telemedicine hub. No telemedicine consultations occurring exclusively outside the ED will be included in the tele-ED cohort. Any sepsis case identified in our inclusion criteria for which no tele-ED consultation is identified will comprise a case without tele-ED use.

Additional variables are defined for use in the propensity score and risk-adjustment models. Illness severity will be measured using the APACHE-II score using the worst values of each parameter from the first 24 h of health system contact [63]. Comorbidities will be identified as present from the ED and hospital admission medical record among the following: hypertension, chronic obstructive pulmonary disease, asthma, cirrhosis, history of organ transplant, congestive heart failure, malignancy, HIV infection, diabetes mellitus, end-stage renal disease requiring hemodialysis. Insurance status will be recorded from patient financial records. Shock index will be calculated as (heart rate)/(systolic blood pressure), measured at ED triage [64]. Year will also be recorded to capture temporal trends in sepsis management and outcomes.

If data are missing on clinical outcomes, the record will be excluded. Participants not recorded in the call log will be assumed not to have had tele-ED used, and sepsis-related interventions not recorded will be assumed that these interventions did not occur.

This study is testing the hypothesis that telemedicine decreases mortality and hospital length of stay by increasing the probability that high-quality sepsis care is delivered, and that the improved quality of care improves clinical outcomes by decreasing organ dysfunction that leads to prolonged hospital stay. The impact of telemedicine is proposed to occur through direct patient care recommendations, training local providers who can provide better care to subsequent patients and improved protocols and standard care elements that improve future care (Figure 3). Because high-quality sepsis care may be delivered without telemedicine and certain patient-level characteristics (e.g., severity of illness, need for transfer) may be associated with telemedicine use, our analytic strategy is designed to reduce the impact of selection bias and confounding.

Since the analysis will be conducted using propensity-matched pairs, we propose a power calculation using the paired samples t-test. While hospital-free days is unlikely to be normally distributed, the difference in the outcomes for propensity-matched pairs is expected to be parametric. We defined a clinically meaningful difference in our primary outcome of length-of-stay as 10% based on prior reports [75,76]. We conducted a pilot study of sepsis patients in a smaller cohort of rural hospitals from the same network in February 2016 that reported mean 28-day hospital-free days to be 18.5 days [56]. We calculate that 234 pairs will provide 80% power and 312 pairs will provide 90% power to detect a difference of ≥10% in 28-day hospital-free days between the paired cohorts, assuming α = 0.05; ρ = 0.25 and mean 28-day hospital-free days of 18.5 days (SD 8.2; Figure 4).

In our 2016 pilot study, 40% of cases identified by ICD-9-CM codes met all inclusion criteria (Figure 1), and 34% used telemedicine [56]. Preliminary query of the medical record of participating hospitals suggests a total available sample of 4,240 records based on discharge diagnoses alone. This sample would predict approximately 576 qualifying sepsis cases for which telemedicine was used.

Based on prior related analyses, we expect to observe that tele-ED use is associated with improved adherence to SSC guidelines, and that this improvement in adherence will be associated with increased 28-day hospital free days [44,56]. We believe that is reasonable because of the strong published data that SSC bundle adherence is associated with improved outcomes [77–79], but directly demonstrating the causal model involving tele-ED is important, and mediation analysis will help clarify the pathway.

In our mediation analysis, we expect that our effect will be both through the causal pathway of SSC guideline adherence and also through the direct pathway (e.g., through a pathway other than through guideline adherence).

All data will be entered into a secure password-protected REDCap database accessible only to the study team. For each data element, values will be summarized to identify missingness, outliers and discrepancies. Prior to analysis, data will be deidentified, and protected health information will only be accessed on password-protected computers using a secure encrypted server. The study PI will have access to the final dataset and will take ultimate responsibility for its integrity.

The study protocol was reviewed and approved by the institutional review board (IRB) at both the University of Iowa and at Avera eCARE under waiver of informed consent.

The results of this study will be disseminated via publication in peer-reviewed journals, presentations at scientific meetings and dissemination of research briefs to participating rural hospitals. Aggregate data will be shared with investigators who make a written request to the study team, and cross-institution collaborations are encouraged. Because of the sparsely populated region where this study is being conducted, hospital-identifiable data could permit identification of individuals, so patient-level data will not be released. The study team will, however, collaborate with investigators to conduct additional analyses while maintaining the security of the dataset.

The main strength of this study design is that we are measuring real-world pragmatic impact of tele-ED consultation in a mature telemedicine network. Many telemedicine studies report the results of small single-center pilot programs, or they report the findings from demonstration projects that are not scalable [40,43,45]. The intervention we are testing is operating in a sustainable network at scale.

The second strength is our use of a multicenter network. Rural hospitals vary considerably, so including 25 rural hospitals of different sizes, ownership structures and in different states increases our external validity.

The third strength is our use of propensity score matching and mediation analysis to both account for selection in who received telemedicine and also to test a causal pathway for how patients might be impacted by tele-ED consultation. Testing the hypothesis that tele-ED is associated with clinical outcomes while also measuring the pathway of that proposed effect is robust, because it both adds credibility to any effect on the primary outcome and it allows for subsequent theories on how the relationship between a hub and a rural center might function.

Finally, our use of a clinical outcome is a strength. Our goal in this analysis is to prioritize patient-oriented clinical outcomes over process outcomes, because those clinical outcomes are most relevant to patients, providers and policymakers.

The main limitation of our study is its observational design. Because our primary data source is an electronic medical record, we are limited by the quantity and quality of the data recorded in the medical record. Despite our use of propensity score methods, there may be residual confounding by indication not adjusted by the statistical methods we propose. We are limited by our method of data collection, in that a group of data abstractors will be abstracting charts manually. We have designed robust quality control checks into this process, but some data collection errors may be unrecognized. Another limitation is the fact that participating hospitals subscribe to a single tele-ED network. The impact of that network may be different in structure, process and outcomes from other tele-ED networks [80]. All participating hospitals are in the Midwest, which may mean they are different from hospitals in other US regions.