Assessing the Utility of ChatGPT Throughout the Entire Clinical Workflow: Development and Usability Study.

BACKGROUND: Large language model (LLM)-based artificial intelligence chatbots direct the power of large training data sets toward successive, related tasks as opposed to single-ask tasks, for which artificial intelligence already achieves impressive performance. The capacity of LLMs to assist in the full scope of iterative clinical reasoning via successive prompting, in effect acting as artificial physicians, has not yet been evaluated. OBJECTIVE: This study aimed to evaluate ChatGPT's capacity for ongoing clinical decision support via its performance on standardized clinical vignettes. METHODS: We inputted all 36 published clinical vignettes from the Merck Sharpe & Dohme (MSD) Clinical Manual into ChatGPT and compared its accuracy on differential diagnoses, diagnostic testing, final diagnosis, and management based on patient age, gender, and case acuity. Accuracy was measured by the proportion of correct responses to the questions posed within the clinical vignettes tested, as calculated by human scorers. We further conducted linear regression to assess the contributing factors toward ChatGPT's performance on clinical tasks. RESULTS: ChatGPT achieved an overall accuracy of 71.7% (95% CI 69.3%-74.1%) across all 36 clinical vignettes. The LLM demonstrated the highest performance in making a final diagnosis with an accuracy of 76.9% (95% CI 67.8%-86.1%) and the lowest performance in generating an initial differential diagnosis with an accuracy of 60.3% (95% CI 54.2%-66.6%). Compared to answering questions about general medical knowledge, ChatGPT demonstrated inferior performance on differential diagnosis (ß=-15.8%; P<.001) and clinical management (ß=-7.4%; P=.02) question types. CONCLUSIONS: ChatGPT achieves impressive accuracy in clinical decision-making, with increasing strength as it gains more clinical information at its disposal. In particular, ChatGPT demonstrates the greatest accuracy in tasks of final diagnosis as compared to initial diagnosis. Limitations include possible model hallucinations and the unclear composition of ChatGPT's training data set.

User experience with a voice-enabled smartphone app to collect patient-reported outcomes in rheumatoid arthritis.

OBJECTIVES: Rheumatoid arthritis (RA) is a chronic disease, requiring frequent patient-provider interaction and self-monitoring. We developed a novel mobile health smartphone app with a voice-enabled feature to help patients virtually track disease activity and ask general questions about RA. METHODS: With a user-centered design (UCD) approach, we developed a voice-enabled app (VEA) which was then tested in two focus groups of patients (n=8) and one with providers (n=4). Voice enablement and a question and answer (Q & A) library function were previously requested by patients. Based on focus group feedback, the VEA was refined and tested with 26 patients for 56 days. The VEA asked patients to fill in daily patient-reported outcomes (PROs) and complete the trial with a satisfaction survey. RESULTS: Of the 26 patients in the VEA trial, 77% were female and 50% were aged 55 and older. Adherence to daily PROs during the 56-day trial was 66%, with <1% of PROs completed using the voice-enabled feature. PROMIS short forms and RADAI-5 PROs remained stable. Of the 22 satisfaction survey respondents, 86% were satisfied with their overall experience with the app and 18.5% were satisfied with voice enablement. The voice assistant had an 86% success rate at understanding and answering interactions regarding surveys and a 44% success rate regarding Q & A interactions. CONCLUSIONS: We developed a novel VEA through a UCD framework and conducted pilot testing. Adherence was moderate and RADAI-5 and PROMIS measures were stable. Based on satisfaction results, PROs may not be the best use of voice enablement technology.

Alert-based computerized decision support for high-risk hospitalized patients with atrial fibrillation not prescribed anticoagulation: a randomized, controlled trial (AF-ALERT).

AIMS: Despite widely available risk stratification tools, safe and effective anticoagulant options, and guideline recommendations, anticoagulation for stroke prevention in atrial fibrillation (AF) is underprescribed. We created and evaluated an alert-based computerized decision support (CDS) strategy to increase anticoagulation prescription in hospitalized AF patients at high risk for stroke. METHODS AND RESULTS: We enrolled 458 patients (CHA2DS2-VASc score ≥1) with AF who were not prescribed anticoagulant therapy and were hospitalized at Brigham and Women's Hospital. Patients were randomly allocated, according to Attending Physician of record, to intervention (alert-based CDS) vs. control (no notification). The primary efficacy outcome was the frequency of anticoagulant prescription. The CDS tool assigned 248 patients to the alert group and 210 to the control group. Patients in the alert group were more likely to be prescribed anticoagulation during the hospitalization (25.8% vs. 9.5%, P < 0.0001), at discharge (23.8% vs. 12.9%, P = 0.003), and at 90 days (27.7% vs. 17.1%, P = 0.007). The alert reduced the odds of a composite outcome of death, myocardial infarction (MI), cerebrovascular event, and systemic embolic event at 90 days [11.3% vs. 21.9%, P = 0.002; odds ratio (OR) 0.45; 95% confidence interval (CI) 0.27-0.76]. The alert reduced the odds of MI at 90 days by 87% (1.2% vs. 8.6%, P = 0.0002; OR 0.13; 95% CI 0.04-0.45) and cerebrovascular events or systemic embolism at 90 days by 88% (0% vs. 2.4%, P = 0.02; OR 0.12; 95% CI 0.0-0.91). CONCLUSION: An alert-based CDS strategy increased anticoagulation in high-risk hospitalized AF patients and reduced major adverse cardiovascular events, including MI and stroke. CLINICALTRIALS.GOV IDENTIFIER: NCT02339493.

The opportunities and shortcomings of using big data and national databases for sarcoma research.

The rarity and heterogeneity of sarcomas make performing appropriately powered studies challenging and magnify the significance of large databases in sarcoma research. Established large tumor registries and population-based databases have become increasingly relevant for answering clinical questions regarding sarcoma incidence, treatment patterns, and outcomes. However, the validity of large databases has been questioned and scrutinized because of the inaccuracy and wide variability of coding practices and the absence of clinically relevant variables. In addition, the utilization of large databases for the study of rare cancers such as sarcoma may be particularly challenging because of the known limitations of administrative data and poor overall data quality. Currently, there are several large national cancer databases, including the Surveillance, Epidemiology, and End Results database, the National Cancer Data Base of the American College of Surgeons and the American Cancer Society, and the National Program of Cancer Registries of the Centers for Disease Control and Prevention. These databases are often used for sarcoma research, but they are limited by their dependence on administrative or billing data, the lack of agreement between chart abstractors on diagnosis codes, and the use of preexisting documented hospital diagnosis codes for tumor registries, which lead to a significant underestimation of sarcomas in large data sets. Current and future initiatives to improve databases and big data applications for sarcoma research include increasing the utilization of sarcoma-specific registries and encouraging national initiatives to expand on real-world, evidence-based data sets.

Comparing Characteristics of Patients Who Connect Their iPhones to an Electronic Health Records System Versus Patients Who Connect Without Personal Devices: Cohort Study.

BACKGROUND: While individual access to health records has traditionally been through paper and other physical media, there has been a recent push toward digitizing this process. Direct patient access to health data through application programming interfaces (APIs) is an important part of current United States policy initiatives, and Apple has created the product "Health Records on iPhone" to leverage APIs for this purpose. OBJECTIVE: The objective of this study was to examine the characteristics of patients at our institution who connected their personal iPhone devices to our electronic health records (EHRs) system through "Health Records on iPhone", as compared to patients at our institution who used our patient portal but did not connect a personal device to our system. METHODS: We examined adult patients at our institution who had authorized an iPhone device to download their health data from the Partners HealthCare EHR via APIs through "Health Records on iPhone" from February 18, 2018 (the date this feature was enabled at our health system) until February 17, 2019. We compared these patients to adult patients who used our portal at least once during this period but did not authorize an iPhone device to download their data via APIs. RESULTS: Variables associated with an increased likelihood of using "Health Records on iPhone" included male gender (adjusted OR 3.36; 95% CI 3.11-3.62; P<.001) and younger age, particularly below 50 years of age. With each decade of age over 50, people were less likely to be "Health Records on iPhone" product users. Asian patients were more likely to use the product than Caucasian patients (adjusted OR 1.32; 95% CI 1.16-1.51; P<.001), though there was no significant difference between African Americans and Caucasians (adjusted OR 1.15; 95% CI 0.94-1.41; P=.17). Patients who resided in higher ZIP code income quartiles were more likely to be users than those in the lowest quartile. CONCLUSIONS: Early results from the implementation of patient-facing APIs at a single institution suggest that there are opportunities for expanding these technologies to ensure all patients are aware of, and have access to, their health data on their personal devices. More work is needed on expanding these technologies to different patient populations.

An Internet of Things Buttons to Measure and Respond to Restroom Cleanliness in a Hospital Setting: Descriptive Study.

BACKGROUND: Restroom cleanliness is an important factor in hospital quality. Due to its dynamic process, it can be difficult to detect the presence of dirty restrooms that need to be cleaned. Using an Internet of Things (IoT) button can permit users to designate restrooms that need cleaning and in turn, allow prompt response from housekeeping to maintain real-time restroom cleanliness. OBJECTIVE: This study aimed to describe the deployment of an IoT button-based notification system to measure hospital restroom cleanliness reporting system usage and qualitative feedback from housekeeping staff on IoT button use. METHODS: We deployed IoT buttons in 16 hospital restrooms. Over an 8-month period, housekeeping staff received real-time notifications and responded to button presses for restroom cleaning. All button presses were recorded. We reported average button usage by hospital area, time of day, and day of week. We also conducted interviews with housekeeping supervisors and staff to understand their acceptance of and experience with the system. RESULTS: Over 8 months, 1920 requests to clean restrooms in the main hospital lobby and satellite buildings were received. The hospital lobby IoT buttons received over half (N=1055, 55%) of requests for cleaning. Most requests occurred in afternoon hours from 3 PM to midnight. Requests for cleaning remained stable throughout the work week with fewer requests occurring over weekends. IoT button use was sustained throughout the study period. Interviews with housekeeping supervisors and staff demonstrated acceptance of the IoT buttons; actual use was centered around asynchronous communication between supervisors and staff in response to requests to clean restrooms. CONCLUSIONS: An IoT button system is a feasible method to generate on-demand request for restroom cleaning that is easy to deploy and that users will consistently engage with. Data from this system have the potential to enable responsive scheduling for restroom service and anticipate periods of high restroom utilization in a hospital.

Effect of Restriction of the Number of Concurrently Open Records in an Electronic Health Record on Wrong-Patient Order Errors: A Randomized Clinical Trial.

Importance: Recommendations in the United States suggest limiting the number of patient records displayed in an electronic health record (EHR) to 1 at a time, although little evidence supports this recommendation. Objective: To assess the risk of wrong-patient orders in an EHR configuration limiting clinicians to 1 record vs allowing up to 4 records opened concurrently. Design, Setting, and Participants: This randomized clinical trial included 3356 clinicians at a large health system in New York and was conducted from October 2015 to April 2017 in emergency department, inpatient, and outpatient settings. Interventions: Clinicians were randomly assigned in a 1:1 ratio to an EHR configuration limiting to 1 patient record open at a time (restricted; n = 1669) or allowing up to 4 records open concurrently (unrestricted; n = 1687). Main Outcomes and Measures: The unit of analysis was the order session, a series of orders placed by a clinician for a single patient. The primary outcome was order sessions that included 1 or more wrong-patient orders identified by the Wrong-Patient Retract-and-Reorder measure (an electronic query that identifies orders placed for a patient, retracted, and then reordered shortly thereafter by the same clinician for a different patient). Results: Among the 3356 clinicians who were randomized (mean [SD] age, 43.1 [12.5] years; mean [SD] experience at study site, 6.5 [6.0] years; 1894 females [56.4%]), all provided order data and were included in the analysis. The study included 12 140 298 orders, in 4 486 631 order sessions, placed for 543 490 patients. There was no significant difference in wrong-patient order sessions per 100 000 in the restricted vs unrestricted group, respectively, overall (90.7 vs 88.0; odds ratio [OR], 1.03 [95% CI, 0.90-1.20]; P = .60) or in any setting (ED: 157.8 vs 161.3, OR, 1.00 [95% CI, 0.83-1.20], P = .96; inpatient: 185.6 vs 185.1, OR, 0.99 [95% CI, 0.89-1.11]; P = .86; or outpatient: 7.9 vs 8.2, OR, 0.94 [95% CI, 0.70-1.28], P = .71). The effect did not differ among settings (P for interaction = .99). In the unrestricted group overall, 66.2% of the order sessions were completed with 1 record open, including 34.5% of ED, 53.7% of inpatient, and 83.4% of outpatient order sessions. Conclusions and Relevance: A strategy that limited clinicians to 1 EHR patient record open compared with a strategy that allowed up to 4 records open concurrently did not reduce the proportion of wrong-patient order errors. However, clinicians in the unrestricted group placed most orders with a single record open, limiting the power of the study to determine whether reducing the number of records open when placing orders reduces the risk of wrong-patient order errors. Trial Registration: clinicaltrials.gov Identifier: NCT02876588.

Internet of Things Buttons for Real-Time Notifications in Hospital Operations: Proposal for Hospital Implementation.

BACKGROUND: Hospital staff frequently performs the same process hundreds to thousands of times a day. Customizable Internet of Things buttons are small, wirelessly-enabled devices that trigger specific actions with the press of an integrated button and have the potential to automate some of these repetitive tasks. In addition, IoT buttons generate logs of triggered events that can be used for future process improvements. Although Internet of Things buttons have seen some success as consumer products, little has been reported on their application in hospital systems. OBJECTIVE: We discuss potential hospital applications categorized by the intended user group (patient or hospital staff). In addition, we examine key technological considerations, including network connectivity, security, and button management systems. METHODS: In order to meaningfully deploy Internet of Things buttons in a hospital system, we propose an implementation framework grounded in the Plan-Do-Study-Act method. RESULTS: We plan to deploy Internet of Things buttons within our hospital system to deliver real-time notifications in public-facing tasks such as restroom cleanliness and critical supply restocking. We expect results from this pilot in the next year. CONCLUSIONS: Overall, Internet of Things buttons have significant promise; future rigorous evaluations are needed to determine the impact of Internet of Things buttons in real-world health care settings.

The Electronic Inbox-Benefits, Questions, and Solutions for the Road Ahead.

This Viewpoint looks at digital communication between patients and physicians, including approaches to provide adequate support for these efforts that balance patient needs with appropriate time investments from clinicians.

Assessing Documentation of Critical Imaging Result Follow-up Recommendations in Emergency Department Discharge Instructions.

To facilitate follow-up of critical test results across transitions in patient care settings, we implemented an electronic discharge module that enabled care providers to include follow-up recommendations in the discharge instructions. We assessed the impact of this module on documentation of follow-up recommendations for critical imaging findings in Emergency Department (ED) discharge instructions. We studied 240 patients with critical imaging findings discharged from the ED before (n = 80) and after (n = 160) implementation of the module. We manually reviewed hand-written forms and electronic discharge instructions to determine if follow-up recommendations were documented. Follow-up recommendations in ED discharge instructions increased from 60.0% (48/80) to 73.8% (118/160) post-module implementation (p = 0.03), a relative increase of 23%. There was no significant change in the rate of documented critical imaging findings in the discharge instructions (77.5% [62/80] before the intervention and 76.9% [123/160] after the intervention; p = 0.91). Implementation of a discharge module was associated with increased documentation of critical imaging finding follow-up recommendations in ED discharge instructions. However, one in four patients still did not receive adequate follow-up recommendations, suggesting further opportunities for performance improvement exist.

Impact of a Health Information Technology Intervention on the Follow-up Management of Pulmonary Nodules.

Lung cancer is the leading cause of cancer deaths in the USA. The most common abnormalities suspicious for lung cancer on CT scan include pulmonary nodules. Recommendations to improve care for patients with pulmonary nodules require follow-up management. However, transitions in care, especially for patients undergoing transitions to ambulatory care sites from the emergency department (ED) and inpatient settings, can exacerbate failures in follow-up testing and compromise patient safety. We evaluate the impact of a discharge module that includes follow-up recommendations for further management of pulmonary nodules on the study outcome and follow-up management of patients with pulmonary nodules within 1 year after discharge. After IRB approval, we collected data on all patients undergoing chest or abdominal CT exams over a 12-month baseline and 12-month intervention period at an academic medical center. The inpatient discharge module was implemented in November 2011; the ED module was implemented in May 2012. Multivariable logistic regression was performed to account for care setting, imaging modality, recommendations, and patient demographics. Implementation of a discharge module resulted in improved follow-up of patients with pulmonary nodules within 1 year after discharge (OR = 1.64, p = 0.01); the ED implementation resulted in better follow-up compared to the inpatient module (OR = 2.24, p < 0.01). Twenty-seven percent of patients with pulmonary nodules received follow-up management, which, although significantly improved from the 18% baseline, remains low. An electronic discharge module is associated with improved follow-up management of patients with pulmonary nodules, and may be combined with interventions to further improve management of these patients.

A Custom-Developed Emergency Department Provider Electronic Documentation System Reduces Operational Efficiency.

STUDY OBJECTIVE: Electronic health record implementation can improve care, but may also adversely affect emergency department (ED) efficiency. We examine how a custom, ED provider, electronic documentation system (eDoc), which replaced paper documentation, affects operational performance. METHODS: We analyzed retrospective operational data for 1-year periods before and after eDoc implementation in a single ED. We computed daily operational statistics, reflecting 60,870 pre- and 59,337 postimplementation patient encounters. The prespecified primary outcome was daily mean length of stay; secondary outcomes were daily mean length of stay for admitted and discharged patients and daily mean arrival time to disposition for admitted patients. We used a prespecified multiple regression model to identify differences in outcomes while controlling for prespecified confounding variables. RESULTS: The unadjusted change in length of stay was 8.4 minutes; unadjusted changes in secondary outcomes were length of stay for admitted patients 11.4 minutes, length of stay for discharged patients 1.8 minutes, and time to disposition 1.8 minutes. With a prespecified regression analysis to control for variations in operational characteristics, there were significant increases in length of stay (6.3 minutes [95% confidence interval 3.5 to 9.1 minutes]) and length of stay for discharged patients (5.1 minutes [95% confidence interval 1.9 to 8.3 minutes]). There was no statistically significant change in length of stay for admitted patients or time to disposition. CONCLUSION: In our single-center study, the isolated implementation of eDoc was associated with increases in overall and discharge length of stay. Our findings suggest that a custom-designed electronic provider documentation may negatively affect ED throughput. Strategies to mitigate these effects, such as reducing documentation requirements or adding clinical staff, scribes, or voice recognition, would be a valuable area of future research.

Developing a Framework for Evaluating the Patient Engagement, Quality, and Safety of Mobile Health Applications.

Rising ownership of smartphones and tablets across social and demographic groups has made mobile applications, or apps, a potentially promising tool for engaging patients in their health care, particularly those with high health care needs. Through a systematic search of iOS (Apple) and Android app stores and an analysis of apps targeting individuals with chronic illnesses, we assessed the degree to which apps are likely to be useful in patient engagement efforts. Usefulness was determined based on the following criteria: description of engagement, relevance to the targeted patient population, consumer ratings and reviews, and most recent app update. Among the 1,046 health care-related, patient-facing applications identified by our search, 43 percent of iOS apps and 27 percent of Android apps appeared likely to be useful. We also developed criteria for evaluating the patient engagement, quality, and safety of mobile apps.

The Boston Marathon Bombings Mass Casualty Incident: One Emergency Department's Information Systems Challenges and Opportunities.

Emergency department (ED) information systems are designed to support efficient and safe emergency care. These same systems often play a critical role in disasters to facilitate real-time situation awareness, information management, and communication. In this article, we describe one ED's experiences with ED information systems during the April 2013 Boston Marathon bombings. During postevent debriefings, staff shared that our ED information systems and workflow did not optimally support this incident; we found challenges with our unidentified patient naming convention, real-time situational awareness of patient location, and documentation of assessments, orders, and procedures. As a result, before our next mass gathering event, we changed our unidentified patient naming convention to more clearly distinguish multiple, simultaneous, unidentified patients. We also made changes to the disaster registration workflow and enhanced roles and responsibilities for updating electronic systems. Health systems should conduct disaster drills using their ED information systems to identify inefficiencies before an actual incident. ED information systems may require enhancements to better support disasters. Newer technologies, such as radiofrequency identification, could further improve disaster information management and communication but require careful evaluation and implementation into daily ED workflow.

Clinical and financial impact of removing creatine kinase-MB from the routine testing menu in the emergency setting.

CONTEXT: Cardiac troponins T and I have replaced creatine kinase-MB (CK-MB) as the criterion standard for diagnosing myocardial injury. However, many laboratories still routinely perform a high volume of CK-MB testing in conjunction with troponin. PURPOSE: The purpose of this study is to study the clinical and financial impact of removing CK-MB from the routine emergency department (ED) test menu at a large academic medical center. METHODS: Creatine kinase-MB was removed from ED ordering templates and laboratory requisitions (ie, intervention), although the test could still be manually ordered. Data for creatine kinase (CK), CK-MB, and troponin T (TnT) specimens ordered during a 12-month period (6 months preintervention and 6 months postintervention) (n = 14571) was downloaded from our laboratory information system. All specimens with (1) normal TnT (ie, <0.01 ng/mL), (2) elevated CK-MB (ie, >6.6 ng/mL), and (3) elevated CK-MB index (ie, >5) were considered discrepant and independently reviewed by 2 ED clinicians for the presence of an acute coronary syndrome and for documentation of final diagnosis. Creatine kinase, CK-MB, and TnT ED volumes preintervention and postintervention were analyzed to assess laboratory cost savings. RESULTS: Of the 6444 cases included in the analysis, only 17 were discrepant. Of all 17 cases, no patients were diagnosed with acute coronary syndrome. After removing CK-MB from the templates and requisitions, CK-MB and CK volumes decreased by 80% and 76%, respectively, translating to annual reagent cost savings of approximately $47000. CONCLUSIONS: Creatine kinase-MB can be removed from the routine ED test menu without adversely affecting patient care. In addition, substantial cost savings can be achieved by reducing unnecessary CK-MB testing and associated CK orders.

The effect of electronic health record implementation on community emergency department operational measures of performance.

STUDY OBJECTIVE: We study the effect of an emergency department (ED) electronic health record implementation on the operational metrics of a diverse group of community EDs. METHODS: We performed a retrospective before/after analysis of 23 EDs from a single management group that experienced ED electronic health record implementation (with the majority of electronic health records optimized specifically for ED use). We obtained electronic data for 4 length of stay measures (arrival to provider, admitted, discharged, and overall length of stay) and 4 measures of operational characteristics (left before treatment complete, significant returns, overall patient satisfaction, and provider efficiency). We compared the 6-month "baseline" period immediately before implementation with a "steady-state" period commencing 6 months after implementation for all 8 metrics. RESULTS: For the length of stay measures, there were no differences in the arrival-to-provider interval (difference of -0.02 hours; 95% confidence interval [CI] of difference -0.12 to 0.08), admitted length of stay (difference of 0.10 hours; 95% CI of difference -0.17 to 0.37), discharged length of stay (difference of 0.07 hours; 95% CI of difference -0.07 to 0.22), and overall length of stay (difference of 0.11 hours; 95% CI of difference -0.04 to 0.27). For operational characteristics, there were no differences in the percentage who left before treatment was complete (difference of 0.24%; 95% CI of difference -0.47% to 0.95%), significant returns (difference of -0.04%; 95% CI of difference -0.48% to 0.39%), overall percentile patient satisfaction (difference of -0.02%; 95% CI of difference -2.35% to 2.30%), and provider efficiency (difference of -0.05 patients/hour; 95% CI of difference -0.11 to 0.02). CONCLUSION: There is no meaningful difference in 8 measures of operational performance for community EDs experiencing optimized ED electronic health record implementation between a baseline and steady-state period.

Hospital collaboration with emergency medical services in the care of patients with acute myocardial infarction: perspectives from key hospital staff.

STUDY OBJECTIVE: Evidence suggests that active collaboration between hospitals and emergency medical services (EMS) is significantly associated with lower acute myocardial infarction mortality rates; however, the nature of such collaborations is not well understood. We seek to characterize views of key hospital staff about collaboration with EMS in the care of patients hospitalized with acute myocardial infarction. METHODS: We performed an exploratory analysis of qualitative data previously collected from site visits and detailed interviews with 11 US hospitals that ranked in the top or bottom 5% of performance on 30-day risk-standardized acute myocardial infarction mortality rates, using Centers for Medicare & Medicaid Services data from 2005 to 2007. We selected all codes from the previous analysis in which EMS was most likely to have been discussed. A multidisciplinary team analyzed the data with the constant comparative method to generate recurrent themes. RESULTS: Both higher- and lower-performing hospitals reported that EMS is critical to the provision of timely care for patients with acute myocardial infarction. However, close collaborative relationships with EMS were more apparent in the higher-performing hospitals, which demonstrated specific investment in and attention to EMS through respect for EMS as valued professionals and colleagues, strong communication and coordination with EMS and active engagement of EMS in hospital acute myocardial infarction quality improvement efforts. CONCLUSION: Hospital staff from higher-performing hospitals described broad, multifaceted strategies to support collaboration with EMS in providing acute myocardial infarction care. The association of these strategies with hospital performance should be tested quantitatively in a larger representative study.