Improving Emergency Department Flow: Reducing Turnaround Time for Emergent CT Scans

Abstract

Emergency departments across the U.S. are more congested than ever, and there is a pressing need to create capacity by improving patient flow. The long turnaround time of imaging tests, such as computed tomography (CT) scans, are a major reason for delays in treatment and disposition. Over an eight-month pre-intervention period during which 10,063 CT scans were ordered in our emergency department, the average time from a CT order to the availability of the radiologist’s final report was 5.9 hours (median=4.2 hours). We created a multi-disciplinary team of physicians, nurses, technicians, transporters, informaticians, and engineers to identify barriers and implement technical as well as human-factors solutions. In the corresponding eight-month period after the implementation of the intervention bundle, there was a 1.2 hour reduction in CT turnaround time, despite a 13.8% increase in the number of CT scans ordered (p<0.0001).

Introduction

The number of patients visiting emergency departments (EDs) continues to increase every year¹. Because of this increasing number of patients waiting for the treatment in the EDs, almost half of all EDs report operating at or above capacity. This is a particularly worrisome trend as ED overcrowding has been shown to lead to poor quality of care^2,3. One highly successful method of overcrowding reduction is to streamline patient care and improve patient flow^4,5.

An often-cited barrier to efficient ED flow and a reason for delay in patient treatment is the turnaround time for imaging tests^6-8. In our institution, we found that there was persistent dissatisfaction with the time it took to receive the results of a computed tomography (CT) scan in the adult emergency room. As our adult ED is quite large, seeing almost 100,000 patients annually, the number of CT scans performed in the ED is also quite high, approximately 18,000 annually.

The process for completing a CT scan from order to final radiologist assessment also called “the final read” involves multiple departments and frequent handoffs (Figure 1). At our institution, the average turnaround time (TAT) for a CT scan was 5.9 hours with a median time of 4.2 hours. To identify strategies and employ solutions to reduce this TAT, a multidisciplinary workgroup with representation from each discipline was created.

Figure 1.

A swim lane demonstrating the different disciplines and handoffs involved in the ED CT scan process.

Methods

A team of physicians, nurses, technicians, transporters, informaticians, and engineers was assembled to reduce the TAT for ED CT scans. The team worked to understand the current process, assess the barriers, and create both technological and human-centered solutions. A mapping of an ideal process, list of issues, and potential solutions were compiled through a mixture of full-day events, shadowing, data analysis, and semi-structured interviews. This process was implemented following a LEAN methodology which has been shown to help mitigate common issues in many arenas of healthcare including emergency departments⁹.

The group creation and intervention took place between July 2016 and July 2017. For this study, the post-intervention time period is defined as July 1^st 2017 - March 1^st 2018 and to control for seasonality the pre-intervention time period is defined as July 1^st 2015 - March 1^st 2016. With the goal of creating a more efficient emergency CT workflow, the team created and implemented solutions within four categories: visual analytics, communication, automation, and education.

Visual Analysis

It is difficult to understand such complex processes without access to data. Through an iterative design, a regularly updating interactive system was created to monitor each part of the process. Data were combined from multiple systems to draw a complete picture of the CT TAT process. The disparate data streams that were combined included the EHR ordering system, the EHR medication administration record, the transportation system, and the multiple imaging systems. The aggregation of the information was extremely important as it allowed for the identification of breakdowns in this process. Prior to this project, all of the departments involved were using their own systems and managing their metrics independently. By uniting these data into one common display, each individual piece of the CT process was shown to all disciplines and it was clear who should investigate each delay.

The features of the display were iterated on multiple times based on discussions with the team. Although a real-time automatically refreshing display would be ideal, due to some of the systems employing a one day lag, the display was implemented as an automated daily refresh. To allow for further investigations of certain CT scans, the page was made interactive with detailed information about patient location, disposition, and the exact timing of each part of the CT process.

Engaging informaticians as a core part of the team, as opposed to a resource for building visual displays, resulted in two very distinct benefits: an updated definition of what it means to be an outlier and the impact of non-emergent scans on TAT.

Within the first few months of the project, the team had reached consensus on outliers being defined as TAT > 12 hours. However, by compiling the data around the distribution of TATs, it became clear that defining a rigid threshold for outliers would not be prudent; in fact, including the outliers in all reported statistics and focusing specifically on those outliers could help the team understand larger mechanisms and issues that contribute to CT TAT delays.

Another issue that was uncovered through iterative data inspection was one of non-emergent scans. While examining individual CT scans, it became clear that a number of scans ordered on ED patients would not change the management or disposition of the patient because they were ordered for patients who already had admit orders placed and the scans were for cancer staging or pre-surgical workup purposes. The team decided that there was no reasonable way to deny patients’ these CT scans if it would help them during the admission, although 80% of physicians surveyed agreed that these scans should not be performed in the ED. However, a compromise was reached - any scans ordered after an admission order (and therefore can be counted as a scan which doesn’t change disposition or management of the patient) would be de-prioritized by the CT technicians. Given this de-prioritization, the decision was made not to include any scans ordered on patients after their admit decision in any metrics or displays.

Communication

It quickly became clear that each of the groups responsible for different portions of the CT process were not engaged in structured communication. The consistent lack of communication would lead to frustrations across the team. To solve this issue, three different strategies for multi-disciplinary communication were deployed.

Weekly Meetings

Weekly meetings with at least one representative from each discipline (informatics, engineering, ED physicians, ED nurses, radiology, CT technicians and nurses) were scheduled to maintain progress on the interventions and discuss successes of the week as well as any impediments that were faced.

Huddles

Four-time daily huddles were established to be attended by ED nursing, transportation, and CT staff. The huddles were scheduled for 10am, 5pm, 10pm, 3am and meant to last 5 minutes. The purpose of the huddles was to inform each other of current state and encourage real-time resolution of any problems. The huddles were structured to answer 3 categories of questions (Table 1).

Table 1.

The subsections of the huddle worksheet. These were designed as a one-page paper to be filled out during each huddle. The huddle sheet is then to be sent out to the entire workgroup, keeping the group abreast of the status of the CT flow.

Category	Details	Rationale for Inclusion
Huddle Basics	Date/time of huddle, huddle attendees	Record keeping purposes
Queue Details	ED census, # of CT scans in queue, # of patients in queue, # of patients who need IV contrast, # of stroke pages, # of trauma cases	If total patients > 15, patients who need IV > 6, any stroke pages, or any trauma cases then the team is prompted to consider a surge protocol.
Reasons for Delay	# of patients drinking contrast, # of patients who have received contrast, when will each patient be ready, # patients waiting over 3 hours and why Structured communication between the teams to discuss potential delays or assistance needs

The queue details category can help trigger “diversion” - a specific surge protocol developed to ease the CT queue. The surge protocol outlines a process for diverting resources from the main adult ED scanner to a less often used scanner that is usually reserved for pediatric patients. The surge protocol dictates that if the CT team has two members operating the scanner (which it does during most business hours), and there is no one in the pediatric scanner, the two-person team is separated and one CT technician is sent to the second scanner to perform the quickest types of scans (CT head without contrast). This thereby relieves the first scanner and reduces the overall queue.

Secure Messaging Across Disciplines

During the time of the intervention, a mobile application for secure messaging entitled “Mobile Heartbeat” was deployed across the organization to improve communication. We employed this tool for the CT TAT project as well. A particularly helpful feature of the application is the ability to define dynamic roles such as “ED Radiologist on Duty” or “CT Technician on Duty”. This allows for the multi-disciplinary group to seamlessly correspond without explicitly changing their messages to different team members during shift changes.

Automation

As the CT scan process is complicated and involves many decision points across multiple disciplines, there was a push to automate any part of the process that we could. To reduce the number of handoffs and necessary steps, two automation interventions were implemented: 1) the creation of a new order set and 2) the use of auto-protocoling for a pair of tests, a way to automate certain simple radiologist decisions in the workflow.

Order Set Creation

To maximize the impact of this step, we focused the order set specifically on Abdomen and Pelvis scans, which account for approximately 30% of all CTs. The order set improves the accessibility of each type of CT Abdomen and Pelvis scan, displays all relevant laboratory tests needed to decide on oral contrast ordering, allows for seamless ordering of additional laboratory tests, and automatically populates the medication administration record if oral contrast is ordered.

Auto Protocoling

Part of the current process involves waiting for a protocol to be decided upon by the radiologist, before proceeding with the CT scanning (see 3rd row of Figure 1). Through iterative discussions it became clear that for certain types of scans, this step was unnecessary as the protocol was self-evident. For a CT of the head without contrast (accounting for 43% of the scans ordered), the team implemented an auto-protocoling feature. Auto-protocoling automatically assigns a protocol immediately after the scan is ordered, without populating in the radiologist queue for manual review. To implement the same workflow for another CT scan with similarly self-evident protocoling (a scan for the detection of a renal stone), a new order had to be created. Prior to the intervention, the request for detection of a renal stone was placed by ordering a CT Abdomen and Pelvis scan, with a free-text comment for the radiologist describing suspicion of a renal stone.

Education and Feedback

Throughout the full-day events, shadowing, data analysis, and semi-structured interviews a pattern of misunderstanding surrounding different parts of the CT process emerged. A list of misunderstandings compiled by the workgroup was converted into multiple choice questions and a survey was sent out through Google Forms to each of the five disciplines: ED physicians, ED nurses, radiologists, CT nurses, and CT technicians. The survey had two purposes:

• A)To test the knowledge gaps within each discipline. This part of the survey covered topics surrounding contrast procedures (IV gauges, sites, allergies, timing, etc.) and CT risk factors (pregnancy, creatinine levels, etc.),
• B)To provide a self-reported baseline for different pain points such as non-emergent scan ordering, the existence of adequate comments in the CT order for protocoling, and communication between providers.

Based on part A of the survey results, an individualized education packet was created for each personnel type and was presented at departmental meetings and huddles. To test the success of the education effort along with the success of resolving the known pain points, the survey was re-administered in 6 months.

Results

Overall, there was a statistically significant reduction in the CT TAT (Figure 2). The mean TAT was reduced by 70 minutes and the median TAT was reduced by 20 minutes (t-test, p<.0001). In addition, we were able to significantly reduce the number of scans where the turnaround time was over 12 hours. In the pre-intervention time period, there were 4.3% scans that took over 12 hours to result. In the post-intervention time period that number went down to 3.2% (chi-sq, p < 0.001).

Figure 2.

A violin and box plot demonstrating the distribution of total CT turnaround times during the pre- and post-intervention periods. The intervention was able to reduce the mean turnaround time by over 1. 1 hours and median by 0.4 hours. The y-axis is on a log scale and is truncated at 48 hours for visualization purposes but no outliers were removed for calculations.

Despite growing numbers of CT scans performed for ED patients (over 13% more scans in the post period than in the pre-period), we also saw significant reductions in each part of the process: preparing the patient, scanning, and scan interpretation (Table 2).

Table 2.

The differences to the overall TAT as well as the TAT for each part of the process was statistically significant, each with a p-value < 0.0001, using a t-test. The multi-pronged approach was able to reduce the mean and variation in each step of the process.

	Pre-Period 7/1/15 –3/1/16 N = 10,063 scans	Post-Period 7/1/17 – 3/1/18 N = 11,451 scans
	Mean (SD)	Mean (SD)
Total TAT: CT Order to Final Read (hr)	5.9 (12.9)	4.7 (9.4)
Preparing the Patient: CT Order to CT Scan Begin (min)	155 (135)	143 (119)
Scanning: CT Scan Begin to CT Scan Complete (min)	22 (14)	19 (13)
Scan Interpretation: CT Scan Complete to Final Read (min)	174 (763)	121 (550)

Visual Analytics

The final version of the metrics automatically refreshes once a day, displaying all of the available data from 2015 to the present day (Figure 3). The users are able to filter by any aspect of the CT process such as: time of order, time it took to conduct the final read, existence of a transporter, specific MRNs, scan type, which scanner the scan was completed on, etc. With each filter, the bar chart adjusts along with the medians displayed above. The display also allows for hovering over any scan to examine the patient, their disposition, location, and timing of each part of the CT process. The bar plot is sorted from longest overall TAT to shortest, thereby highlighting the scans with the most concerns first.

Figure 3.

Screenshot of the daily metrics implemented through Tableau software. The page consists of 3 different sections: the overall medians for each part of the process, the color-coded bar chart showing the time each part took for each scan, and the ability to filter on a variety of axes (not shown).

The visual is accessed an average of 5-15 times each day and the page views remain steady over time. The metrics are retrieved in a variety of ways: 1) on demand, 2) by subscription, and 3) during weekly meetings. Each member of the team can access the data on demand through a webpage. In addition, some team members have set up subscriptions that automatically send emails based on particular criteria. For example, a manager for the transport team has a subscription which automatically emails him once a week with a screenshot of the visual limited to any transport trips that took over 30 minutes. Finally, the visual is used to motivate weekly meetings; once the team assembles, the last week of data is displayed on a large monitor and it helps drive the conversation around the difficulties of the week.

In general, we found that ED physicians were the most active clinical users of the metrics, followed by radiologists and the transportation team. The CT RNs and Technicians were not often seen accessing the dashboard on demand.

One example of behavior change based on constant metric monitoring was evident in the transportation team. There was a large concern surrounding the flow of the transportation team who is assigned the task of moving patients from their ED locations to the CT scanner. It was shown that transporters often canceled assignments because patients were already taken by staff, there were multiple requests placed for the same patient, or a patient was not found. Through consistent monitoring the number of transport cancelations was reduced from 15% to 13.8% and the number of scans that occurred without documented transport decreased from 52% to 48%. Although the decreases demonstrate statistically significant moves in the right direction, there is still much work left to be done.

Communication

The three-level communication strategy has been one of the major successes of the project. The weekly meetings have been consistently attended by each discipline for over one year and have helped push through many of the other interventions. The huddles continue to occur multiple times a day and have been shown to be effective at realtime resolution of issues before they escalate. Finally, the mobile application for asynchronous communication have also helped streamline conversations. Specifically, we found that radiologists reporting that it was difficult to reach ED physicians went down from 43% to 13% and the perceived average wait time to reach an ED physician decreased from 1-10 to 1-5 minutes.

Automation

The CT Abdomen Pelvis order set implementation created a method to streamline the protocoling, laboratory test viewing and ordering, and contrast administration. As such, we measured the time savings for patient preparation before and after the intervention (ie. Time from ED Order Placement to Begin of CT Abdomen Pelvis scan). We found that the average time went down from 3.6 hours to 3.2 hours and the median decreased by 35 minutes. We also found consistent usage of the new CT Renal Stone order (which enables auto-protocoling) at a little over 50 times per month.

Education and Feedback

To gauge the effectiveness of the tailored education modules, along with the care team’s perception of other intervention effectiveness, we disseminated surveys to members of each discipline asking how the changes affected their work. We received fairly high response rates on the pre- and post- surveys across the departments, ranging from 21% to 47%. The high response rates, demonstrate engagement across the board and willingness to reform the current CT process. However, we found the results of the education initiative were quite mixed. In some areas, such as CT risk factors, there was a significant amount of learning across the disciplines. However, with other topics such as correct gauge sizes for IV, the number of respondents with the right answer stayed the same or decreased.

Nevertheless, the education surrounding two processes did demonstrate significant improvements: non-emergent testing and comments while placing orders.

Non-emergent Testing

The education and discussing surrounding non-emergent CTs performed in the ED clearly increased understanding of the scale of the issue: prior to the education, 49% of ED providers predicting the correct amount of non-emergent scans performed per month (200-300), and after the education 60% predicted correctly. Not only did understanding of the problem increase, the number of non-emergent scans decreased as well (Table 3).

Table 3.

The data demonstrates a statistically significant decrease in the number of scans ordered for patients already admitted to the hospital (chi-squared p-value < 0.001). ED providers and radiologists both seem to perceive a decrease in inpatient scans; however, the survey results are not statistically significant. The survey explicitly defines non-emergent scans as those which “would not change the management or disposition of the patient”.

CT Order Comments

The time it took for radiology to protocol different exams was identified as a pain point throughout this process. The role of protocoling is to determine the correct type of test and contrast necessary to answer the clinical question posed by the ED physicians. To correctly assess the differential diagnoses being tested by the ED provider, the radiologists rely on the comments written in the order, but if those are not sufficient they either read the patient’s chart or initiate communication with the ED. Following the education provided to the ED team surrounding the importance of the comments section, we found quantitative and qualitative improvements in the comments submitted with each order (Table 4).

Table 4.

The quantitative measure of more comprehensive commenting was obtained through counting the number of words within each comment, using the NLTK python package (t-test p-value p<0.0001). To assess the perception of improvement the survey defines adequate information for protocoling as “the radiologist would not need to get more information before the patient is scanned, either by calling or looking up notes”. None of the changes detected in the survey responses were statistically significant.

Discussion

Despite more patients arriving to the ED each year and more CT scans being ordered, a multi-disciplinary team was able to significantly reduce the time from CT order to CT final read. The ability to present real-time data and measure each aspect of the process was necessary to identify and resolve gaps. However, without proper strategies for resolution and appropriate context for the data, it would likely have proved much more difficult to inspire change.

There were a set of other interventions that were implemented at the same time, but were not discussed in this paper due to lack of space. The other interventions included: 1) shifting responsibility for the administration of oral contrast from CT nurses to ED nurses; 2) manual logs of patients who were unprepared for the scan when they arrived for the scanner (no IV placed, patients still wearing jewelry or clothing) to facilitate communication with the ED RN teams and the CT teams; and 3) education and usage of the “schedule ahead” transportation function to prevent transportation request batching.

A particularly challenging aspect of the work was that it was conducted in an emergency department. Compared to other hospital units, emergency departments are notably chaotic and thus difficult to streamline and control. Accordingly, we were unable to manage the presence of strokes or traumas which are very time sensitive and immediately rise to the top of the CT queue. The existence of multiple stroke calls can delay other CT scans for many hours and lead to very long TATs. However, notwithstanding these challenges we were able to reduce overall TAT.

Limitations

This study had a number of limitations. One of the major drawbacks is a potential influence of Hawthorne effect. Due to the ongoing tracking of the metrics and interactions of the care team with the intervention facilitators, the front-line staff know they are being observed and this knowledge may affect their behavior. However, as the post-intervention time period has spanned over 8 months, we believe that the Hawthorne effect is unlikely to fully explain the strength of the outcomes.

Despite the fairly high response rate, the number of physicians responding to the pre- and post-surveys remained small. Due to this limited sample size, we were unable to find any statistically significant results in the perception of the inpatient testing rate and the adequacy of the comments associated with CT orders. However, the survey results hinted at perceived improvements in both categories.

We also note that other institutions may measure TAT in a different way, which could reduce the generalizability of this work. As we are an academic medical center, radiology residents provide the preliminary read and the attendings provide the final read. Many other institutions may either not have preliminary reads as they don’t have residents or may act based on the preliminary read. We choose to act on final reads to prevent the unlikely scenario of incidental finding identification after a patient has already left the ED.

In addition to these issues, it is important to reemphasize that we implemented a large bundle of interventions. The bundle approach makes it particularly difficult to determine the cost-benefit of our work and to unambiguously determine which interventions most contributed to the reduction in TAT. We have tried to assess the effectiveness of individual components of the bundle but we are unable to measure the impact of some interventions such as weekly meetings with all team members, or each individual piece of the education strategy.

Future Work

There are two planned areas of future work: expansion and prediction. The success of this work has prompted discussions for expanding the findings to other campuses within the hospital as well as to other imaging modalities. As many of the same disciplines are involved in the emergency ultrasound process it will be the first direction of expansion. The translation of this work to additional campuses and other hospitals will be a larger undertaking as it may require a more in-depth cost-benefit analysis. However, the average cost of such a multi-modal intervention can be difficult to quantify given the high uncertainty in healthcare costs and well-established geographic variation¹⁰.

There is also an interest in predicting CT order surges in an effort to assist the team in implementing preventative measures instead of reactive ones. Throughout the course of the project, there has been an accumulation of data (specifically the 4-time daily huddle reports), which can be used as training data for a surge prediction model. Of course, the prediction would not replace the huddles but would only serve as one more input to the huddle discussion.

Conclusion

With a team of stakeholders involved in each step of the process, concrete and measurable goals, ample data, and a desire to make a change that would affect tens of thousands of patients each year, we were able to significantly reduce the time it took for a CT scan result to be returned to the provider to assist in their decision making. By reducing the TAT for such a critical emergency room test, we have helped improve the flow of an often-crowded ED.