PDF icon Download PDF
Appl Clin Inform 2024; 15(03): 528-532
DOI: 10.1055/s-0044-1787648
Invited Editorial

Social Media's Lessons for Clinical Decision Support: Strategies to Improve Engagement and Acceptance

Christopher Sova
1   Infection Prevention and Hospital Epidemiology, Duke University Hospital, Durham, North Carolina, United States
,
Eric Poon
2   Duke Health Technology Services, Durham, North Carolina, United States
,
Robert Clayton Musser
3   Department of Medicine, Duke University Health System, Durham, North Carolina, United States
,
Anand Chowdhury
4   Pulmonary, Allergy and Critical Care Medicine, Duke University School of Medicine, Durham, North Carolina, United States
› Author Affiliations
› Further Information
Also available ateRef
 
 PDF Download Permissions and Reprints

Background

Clinical decision support systems (CDSSs) are an important feature within modern electronic health records (EHRs). There are currently two main ways that CDSSs work: (1) knowledge-based CDSS, which applies if-then rules to generate an output, and (2) non-knowledge-based CDSS, which utilizes machine-learning technology rather than expert medical knowledge to analyze data such as user-generated content from patient interactions to enhance decision making.[1] [2] Used appropriately, CDSS can improve processes of care and reduce costs in health care.[3] Despite their utility, these tools create challenges because of workflow disruption, alert fatigue, and cognitive overload.[4]

Social media platforms are ubiquitous and have combined behavioral economics, neuroscience, design, and marketing with data mining to engage users and influence behavior.[5] [6] [7] [8] [9] Platforms such as Facebook and TikTok use complex recommender systems that analyze user data to provide personalized recommendations to reduce information overload of the users.[10] Other authors have explored components of social media's arsenal, especially population-based strategies and behavioral economics.[11]

In this article, we explore two social media principles, microsegmentation and A/B testing, that could be applied to current-day systems to improve engagement and acceptance of CDSS among clinicians.


 

Translating Social Media to Clinical Decision Support Systems

Microsegmentation

Marketing firms use microsegmentation to break down a customer base into granular groups. Modern strategies include data-driven approaches that harvest user data from sources such as social media and deliver targeted messages back through the same or similar platforms.[12] Compared with population-based approaches, microsegmentation improves messaging to complex markets by customizing these messages based on customer goals, values, and behaviors.[13] [14] One strategy, computer-tailoring health communication, has impacted changes in health behavior.[15] Microsegmentation of EHR users for CDSS content delivery could optimize the Five Rights of clinical decision support (CDS) by aligning closer to the individual.[16]

Collecting user EHR interaction data to power CDSS recommender systems is a novel opportunity for CDSS, even without the degree of profiling used by social media advertisers. For example, interruptive alerts, while effective at directing a user's attention, are disliked by users to the point that they are the targets of initiatives to reduce burnout.[17] User interaction with CDSS could be used to generate behavior profiles such as high versus low adopters, which could be used to tune the degree of interruption to subgroups. Users who demonstrate reliability to digest noninterruptive information and act accordingly would not need to be subjected to more interruptive alerts, sparing them the cognitive intrusion. Alternatively, CDSS can prompt users who fail to reliably act upon noninterruptive information with more interruptive alerts. For instance, consider an interruptive alert that prompts the user to place an order based on an abnormal laboratory result. User A typically ingests laboratory information through chart review and reliably orders the appropriate intervention before exiting the patient's chart, while User B frequently fails to order the intervention within a reasonable amount of time despite the same availability of information. User A may be included in a group that receives standard information delivery, while User B may be included in a group to receive more conspicuous support. There is a spectrum of options to optimize effectiveness and minimize intrusiveness for User B, such as selecting the channel of alert (e.g., pop-up notification) and the trigger for delivery (e.g., time elapsed since result; [Fig. 1]).

Zoom
Fig. 1 CDSS user segmentation. This figure illustrates a potential process for user profile sub-segmentation to deliver CDSS messaging in a manner more congruent with the user. CDSS, clinical decision support system.

Users may move along this spectrum over time, balancing the ability of the tool to generate the desired outcome against intrusiveness to the user. As the quality of user interaction data increases, future support could be specific to clinicians' behavior and practice and delivered in a manner that harmonizes with their decision-making processes. Eventually, the users' predicted reception of specific tools will allocate them into different user segments, as a result of either a change in performance or periodic review.


 

A/B Testing

A/B testing is a software development strategy to evaluate design choices that are favored by users. These tests involve presenting one of multiple designs to users in a live environment and measuring metrics such as user feedback or likelihood of desired outcome.[18] A/B testing is frequently utilized by social media companies such as Meta and shows promise to optimize CDSS ([Fig. 2]).[18] [19]

Zoom
Fig. 2 CDSS optimization with A/B testing. This figure illustrates a basic A/B testing strategy to evaluate delivery options for the delivery of CDSS messages. CDSS, clinical decision support system.

A/B testing could be used to assess the impact of microsegmentation and provide data to power the recommender engines that subsegment users. Surrogates of cognitive load available from the EHR that could be considered for optimization via A/B testing include adherence to suggestions, time-to-task completion, and reduced task-switching.[20] [21] Optimizing such targets would shorten the cognitive work between information gathering and action while reducing distraction. In addition to potential increased efficacy of CDSS, this approach could have the benefit of matching information presentation with cognitive preference and avoid unnecessary depletion of cognitive reserves for users.[22]


 

Evaluation Plan

Appropriate evaluation of health information technology is critical to minimize the disruption of care delivery and prevent adverse outcomes. There are several frameworks and toolkits to consider, such as the AHRQ Health IT Evaluation Toolkit, and it is important to have an evaluation plan that includes quantitative and qualitative metrics.[23] [24] We highlight some metrics to consider.

Assessing these interventions involves evaluating their effects on clinician practice patterns, clinician acceptance, impact on cognitive load, and user well-being. Improvements in practice of care could be measured by assessing adherence to best-practice guidelines. The use of a like/dislike button and the opportunity to provide free-text response could provide insight into clinician acceptance of certain tools. Cognitive load and usability could be evaluated using established tools such as the NASA Task Load Index, the System Usability Scale, or an assessment of within-EHR errors.[25] [26] Lastly, validated survey instruments such as the Maslach Burnout Inventory, Single Item Burnout Measure, or Copenhagen Burnout Inventory may be administered to evaluate impact on health care worker burnout.[27]


 

 

Challenges

The implementation and uptake of new technology by a population requires significant effort, as illustrated by the Gartner hype cycle[28] and the Diffusion of Innovations Theory.[29] We identify challenges to these strategies using the eight-dimensional sociotechnical model proposed by Sittig and Singh.[30] Factors from each of these dimensions could contribute to disuse and compromised care, and these risks could be mitigated by a governance structure to ensure safety and efficacy before widespread adoption of these methods.[31]

Hardware and Software Computing Infrastructure

Custom infrastructure is required for these tools to interface with an organization's resources designed to deliver health care.


 

Clinical Content

Specific user-interaction data need to be selected to power the tools that drive microsegmentation. The relevance of these data and interventions may differ between users.


 

Human–Computer Interface

The iterative refinement of information delivery may complicate the user's ability to use tools necessary to deliver care.


 

People

Human resources are required for the development, implementation, and support of hardware and software. Building a detailed profile of clinicians' electronic behaviors may reveal undesirable variations in care delivery that are not attributable to patient-related variables. In addition to traditional adoption challenges, these strategies may be unpalatable to users due to bad press surrounding social media companies' nefarious use of machine-learning algorithms, and there could be backlash against profiling, similar to other industries that monitor employees with machine learning.[32] Partnerships among patients and caregivers to co-develop CDSS lead to a better understanding of actual needs and values of care delivery and may lead to more impactful and meaningful CDS.[33] Organizations should consider opportunities to engage patients and families to help design, implement, and evaluate these tools.


 

Workflow and Communication

Dynamically changing how information is delivered to health care workers comes with workflow disruption, and changes may force the person into a process that does not match their clinical workflow.


 

Internal Organizational Policies, Procedures, and Culture

Beyond HIPAA's mandates to protect patient privacy, institutions need to safeguard employee privacy as well. Institutions would need to establish protocols to identify, investigate, and potentially address significant variations in care discovered incidentally.


 

External Rules, Regulations, and Pressures

The secure monitoring, use, and storage of patient-identifiable information would need to be ensured. Ownership of logged user data would need to be clarified, particularly regarding regulations on consumer privacy such as the California Consumer Privacy Act[34] and the European Union's General Data Protection Regulation.[35]


 

System Measurement and Monitoring

In addition to the evaluation plan section of this article, machine-learning techniques require each use case to undergo model development, validation, and monitoring.[36]


 

 

Conclusion

CDSS is a vital component of the modern EHR with proven success. However, CDSS tools can frustrate clinicians due to irrelevance, redundancy, or poor delivery. Social media combines advanced data-driven techniques with digital marketing that captures users' attention and influences behavior. Data-driven microsegmentation and A/B testing to optimize the user experience may lead to better engagement and acceptance of CDSS. There are several challenges to successful implementation which could be mitigated with proven frameworks, toolkits, and governance structures.


 

 

Conflict of Interest

None declared.

Acknowledgments

The authors would like to acknowledge the special contributions of Dr. David Theil, MD, MMCi, to this work.


Address for correspondence

Christopher Sova, MMCi, BSN, RN, CIC
Infection Prevention and Hospital Epidemiology, Duke University Hospital
DUMC 3674, Durham, NC 27710
United States   

Publication History

Article published online:
03 July 2024

© 2024. Thieme. All rights reserved.

Georg Thieme Verlag KG
Rüdigerstraße 14, 70469 Stuttgart, Germany


  Permissions and Reprints
Zoom
Fig. 1 CDSS user segmentation. This figure illustrates a potential process for user profile sub-segmentation to deliver CDSS messaging in a manner more congruent with the user. CDSS, clinical decision support system.
Zoom
Fig. 2 CDSS optimization with A/B testing. This figure illustrates a basic A/B testing strategy to evaluate delivery options for the delivery of CDSS messages. CDSS, clinical decision support system.