Patient-centered clinical decision support challenges and opportunities identified from workflow execution models

Abstract

Objective

To use workflow execution models to highlight new considerations for patient-centered clinical decision support policies (PC CDS), processes, procedures, technology, and expertise required to support new workflows.

Methods

To generate and refine models, we used (1) targeted literature reviews; (2) key informant interviews with 6 external PC CDS experts; (3) model refinement based on authors’ experience; and (4) validation of the models by a 26-member steering committee.

Results and Discussion

We identified 7 major issues that provide significant challenges and opportunities for healthcare systems, researchers, administrators, and health IT and app developers. Overcoming these challenges presents opportunities for new or modified policies, processes, procedures, technology, and expertise to: (1) Ensure patient-generated health data (PGHD), including patient-reported outcomes (PROs), are documented, reviewed, and managed by appropriately trained clinicians, between visits and after regular working hours. (2) Educate patients to use connected medical devices and handle technical issues. (3) Facilitate collection and incorporation of PGHD, PROs, patient preferences, and social determinants of health into existing electronic health records. (4) Troubleshoot erroneous data received from devices. (5) Develop dashboards to display longitudinal patient-reported data. (6) Provide reimbursement to support new models of care. (7) Support patient engagement with remote devices.

Conclusion

Several new policies, processes, technologies, and expertise are required to ensure safe and effective implementation and use of PC CDS. As we gain more experience implementing and working with PC CDS, we should be able to begin realizing the long-term positive impact on patient health that the patient-centered movement in healthcare promises.

Introduction

Traditional clinical decision support (CDS) focuses on providing clinicians with patient-specific preventive, diagnostic, treatment, or management guidance to help them provide the highest quality, safest care possible. Recently, there has been a push toward patient-centered clinical decision support (PC CDS) that exists on a continuum reflecting the degree to which CDS interventions (a) are based on patient-centered outcomes research (PCOR) findings, (b) incorporate patient-generated health data (PGHD)¹ including patient-reported outcomes (PROs),² patient preferences, or social determinants of health (SDOH), (c) are delivered directly to patients/caregivers via apps or portals, or (d) support shared decision-making.³ As with traditional CDS, some PC CDS interventions have shown positive outcomes, but gaps in our understanding of the sociotechnical factors that influence PC CDS design, development, implementation, use, and evaluation currently limit them from reaching their full potential.

The goal of this aticle is to use workflow execution models⁴ to highlight new considerations for PC CDS policies and procedures that healthcare systems, clinicians, electronic health record (EHR) developers, app developers, and others need to develop to support new and evolving workflows. Similar workflow models have been used successfully to develop guidelines for health information technology design in chronic care, for example.⁵ In this manuscript, we describe the use of newly created workflow execution models to explore 3 illustrative use cases for PC CDS: (1) collection and use of patient-reported outcomes (PROs) data, which are reports from patients about their health, quality of life, or functional status associated with the health care or treatment they have received; (2) collection and use of patient-generated health data (PGHD) other than PROs such as physiologic data from devices and wearables to improve the patient context; and (3) encouraging or facilitating a shared decision-making discussion where clinicians and patients make decisions together using the best available evidence. For each of these use cases, we explored the processes and tasks that must be accomplished by humans, computer applications, or a combination of the 2, to deliver high-quality, meaningful health outcomes.

Background

PC CDS developers and evaluators can use workflow execution models to track, measure, and monitor the necessary data, information, and knowledge as the PC CDS moves through the clinical decision support phase of the PC CDS lifecycle.^6,7 Such models are designed to highlight the possibilities, as well as the limitations, of our current understanding of the complex PC CDS space. For a PC CDS intervention to have the desired impact on either the healthcare delivery system or patient health, each model component must be designed and built to address patient and clinician needs, function as designed, and be used as expected. As the healthcare delivery system moves toward a more computer-enabled workflow system, it becomes more important to include the tasks and processes that the computer, which must be designed and developed, can do. These computer-enabled tasks support or replace tasks previously performed by clinicians or in some cases represent new tasks necessary to create new healthcare delivery processes, such as PC CDS. These computer-enabled tasks represent new work and new workflows for clinicians. The overarching goal of workflow execution models is to take into consideration the potential methods and sequencing of generating, sending, receiving, and acting on PC CDS interventions.

Methods

We developed the workflow execution models using 4 methods: (1) a targeted literature review; (2) key informant interviews with 6 external CDS experts; (3) model refinement based on authors’ experience; and (4) validation of the models by a 26-member steering committee.

Literature review

In several cases, we found the beginnings of prototype workflow execution models for each of the 3 types of PC CDS (CDS based on PRO data, CDS based on PGHD, and CDS that led to shared decision-making) in the literature.^8–10 We iteratively augmented these early working model prototypes by adding in key workflow steps based on descriptions of the activities of people, the content of policies, the sequence of processes, and the steps in procedures reported in other publications that focused on specific PC CDS intervention implementation or evaluations.^11–13 We stopped searching for additional workflow steps to add to our models when we were no longer identifying new steps (ie, often called saturation in qualitative research).

Key informant discussions

We conducted 6 key informant interviews with 3 clinicians experienced in using PC CDS, 2 informaticians responsible for the design, development, implementation, and evaluation of PC CDS interventions, and 1 device and remote monitoring platform developer. The key informant discussions used early iterations of the workflow execution models to help focus the discussion on descriptions of current practices, policies, and approaches being used by healthcare systems to receive, curate, and manage PGHD data including PROs. Key informant discussions focused on challenges and gaps with current policies, or process activities.

Model refinement

The authors of this article (A.B., P.D., E.A.L., D.S., J.S., A.W.) have extensive experience in designing, developing, implementing and using various PC CDS interventions and applications. The authors conducted several virtual meetings to brainstorm, iteratively refine, and reorganize the workflow models and identify challenges to and opportunities for future success.

Steering committee input

After we reached consensus on the content of the models, the sequence of actions, and their visual display, we shared our models and findings with the AHRQ-funded Clinical Decision Support Innovation Collaborative (CDSiC) project steering committee which is a 26-member multi-stakeholder group that includes representatives from federal agencies, academic medical centers, informaticians, health information technology vendors, patient advocates, researchers/research organizations, and health systems. These external experts reviewed each of the models and compared the descriptions of the steps along with the sequence of steps outlined, to their own experiences in designing, developing, implementing, and using similar patient-centered clinical decision support interventions. When necessary, additional refinements to the description or sequence of steps in the models were made.

Results

We identified the following 3 workflow execution models: collection and use of PROs, collection and use of PGHD other than PROs, and identification of opportunities for shared clinical decision making.

Model 1: collection and use of PROs

Table 1 lists the various workflow activities and actors involved in a hypothetical PC CDS depression screening example that is based on PROs along with example measures that could be used to assess each activity. In this scenario, the eligible patients receive the Patient Health Questionnaire (PHQ-9), which is a 9-question screening tool for depression. All asymptomatic adults 19 years or older who do not have a diagnosed mental health disorder or recognizable signs or symptoms of depression or suicide risk are eligible to be screened.¹⁴ The use of a self-administered questionnaire helps ensure that the eligibility screening process is free from bias (See Figure 1).

Model 2: collection and use of PGHD

The second model involves PGHD that is used to drive specific PC CDS interventions. Table 2 provides an overview of the workflow components necessary to collect and use PGHD in the CDS process along with example measures that could be used to assess each activity. This example focuses on PGHD from a patient-controlled medical device, ie, a home blood pressure cuff. This information has been solicited and subsequently stored automatically in the healthcare organization’s EHR or some external database. We are deliberately not focusing on PGHD that are unsolicited by the clinician or their organizations. Unsolicited PGHD may arrive in the clinician’s in-basket or email in the event that a patient simply sends some physiologic measurements as free text in an email message or via an email attachment. While this unsolicited PGHD can be helpful to the clinician, it cannot be used by existing CDS interventions since it is not structured or easily stored in a structured manner within the EHR (See Figure 2).

Model 3: shared decision-making

The third model focuses on shared decision-making, which involves a collaboration between patients and clinicians to arrive at healthcare decisions grounded in evidence, the expertise of the care team, and the patient's values, objectives, preferences, and individual circumstances.¹⁷ While patient participation in the clinical decision-making process is being increasingly promoted, many patient-related factors (eg, lack of medical knowledge, lack of confidence, co-morbidities, and other sociodemographic information) and clinician-related factors (eg, desire to maintain control, lack of time, personal beliefs, and lack of training in establishing and maintaining patient–caregiver relationships) make such interactions challenging.¹⁸ In an effort to facilitate shared decision-making between patients and clinicians, healthcare organizations have begun to develop CDS interventions that support shared decision-making processes. Table 3 provides an overview of the workflow components and actors necessary to implement shared decision-making in the healthcare delivery process along with example measures that could be used to assess each activity. For example, a patient can be sent easy-to-understand information prior to their visit to help them understand the basic clinical details of their condition. Data on patient’s needs, preferences, and goals can also be elicited from the patient either before or during the visit. This information can then be visualized and compared to reference ranges, if applicable, to help illustrate the options and potential outcomes and tradeoffs associated with selecting among different treatment options.¹⁹ In this way, PC CDS can be used both asynchronously or synchronously to facilitate the shared decision-making process (See Figure 3).

Discussion

Developing the PC CDS workflow execution models highlighted many new policies, processes, technologies, and expertise required to design, develop, implement, and use PC CDS safely and effectively within any healthcare organization. We identified these new aspects of PC CDS based on the authors’ collective experience, knowledge, and understanding of the sociotechnical constraints of implementing new, state-of-the-art health information technology interventions within complex healthcare delivery organizations. In addition, implementation of PC CDS adds additional sociotechnical constraints to previous clinician-focused CDS, such as the need to (1) extend the reach of the healthcare organization beyond the physical confines of the clinic, (2) interact with patients outside normal clinic working hours, (3) engage patients in clinical decision-making activities, and (4) support collection, documentation, and display of PGHD, including PROs from patients in support of patient-centered care. These 4 changes combine to place several new expectations on clinicians, healthcare delivery systems, payers, and technology companies.

Across the 3 illustrative use cases, we identified 7 major PC CDS-related challenges that must be addressed. Addressing these challenges provides significant opportunities and important considerations for the collection and use of new, patient-centric data that have implications for policies, procedures and operations for healthcare systems, researchers, administrators, health IT departments, and app developers to address. These include considerations around frequency of data collection, data validity, provenance, precision, accuracy, and reliability. Additional considerations focus on whether these are data that are solicited and requested by a health system and the processes they have in place to monitor and manage these data.

Next, we describe the 7 major PC CDS-related challenges and opportunities.

1. Policies, procedures, and people are required to ensure PGHD including PROs are documented, reviewed, and managed by appropriately trained clinicians, between visits and especially after working hours. The vast majority of a patient’s life occurs outside their infrequent, short interactions with the healthcare delivery system. The data collection components of PC CDS workflows bring patient data directly into the EHR, via manually entered or automatically collected physiologic data or from questionnaires soliciting patient-reported outcomes—requiring someone in the healthcare system to review and respond to critical information in an appropriate and timely manner. This will require new workflows regarding the data values that should trigger clinician notification and when and how these notifications should be made, data display options, and time to complete the new work. In addition, clinicians and healthcare delivery organizations are not currently prepared to handle the activities that should occur post-data collection, such as receiving and acting upon potentially life-threatening patient data between patient visits. To respond to this need, some healthcare organizations are starting to work with third-party remote patient-monitoring companies that offer services for patients with connected medical devices.²⁴ These companies are developing digital platforms and hiring relevant clinical staff to monitor patient contributed data, a more encompassing term used for “data, information, or insights created, collected by, or originating from a person regarding his or her health or care.”²⁵ Several of these companies have also developed the capability to exchange curated summaries of patient data with a healthcare system’s EHR.

2. Policies, processes, and people are required to educate patients on how to use connected medical devices, and handle technical issues associated with them. The number and variety of remote patient monitoring devices and PRO questionnaires available for clinicians to order and/or send to their patients is increasing rapidly. In addition, the sheer number of possible devices makes it difficult for healthcare organizations to assess, add to their formulary and orderable catalogs, and train clinicians on how to order them.¹³ Some younger, tech-savvy patients may not require special education on basic smartphone usage, downloading, configuring, and connecting new apps or physiologic measurement devices (eg, blood pressure cuff, pulmonary function testing equipment, or glucose monitoring). However, healthcare organizations will need staff with dedicated time to educate other patients on how to set up and use these devices to facilitate data collection activities. These may be as simple as how to log in to the device or how to enable sharing of data, or as complex as how to calibrate a pulmonary function testing device. Finally, as with any new technology, unintended consequences, technical issues, delayed adoption, and difficulty learning or adjusting to new ways of working will occur.²⁶

In some cases, healthcare organizations are contracting with third-party remote patient monitoring companies that work directly with consumers to set up and manage technical issues that arise. To make the care transitions between the healthcare organization and the third-party companies as safe and seamless as possible, new policies on what setting changes can be made or suggested to the patients, especially those using a smartphone application, for example, will need to be developed. Often, these third-party remote patient monitoring companies establish agreements with the healthcare organization that specify the approved list of medical devices that can be ordered, supplied, and supported. Based on the approved list, these companies work directly with patients to set up and manage technical issues that arise with connected medical devices.

3. Policies, processes, and standards are required to facilitate collection and incorporation of PGHD, including PROs, patient preferences, and SDOH into existing EHRs. To date, EHR and app developers and patient-centered outcomes researchers have faced significant challenges in developing methods for collecting high-quality, error-free, patient-generated data and integrating it within existing EHR database structures. These challenges have included lack of agreement on what data should be collected from various devices (eg, raw data or summaries), where it should be stored, which data constitute the legal medical record, what standards should be used to encode the data, whether it should be integrated with clinician-collected data, and how its provenance should be recorded.²⁷

In addition, processes and procedures within healthcare organizations can vary based on the type of device used. For example, some healthcare organizations require clinicians to review and approve patient-provided data before it can be integrated into the medical record. In other cases, if patients are enrolled into a care management program, such as for hypertension monitoring, data submitted by the patient does not require any additional review/adjudication before it is integrated into the EHR. The timing of data collection can also vary and present challenges. For example, some organizations encourage patients to complete PRO questionnaires prior to a scheduled visit, while others wait for the patient to arrive at the doctor’s office. Currently, efforts are underway to develop²⁸ and/or encourage adoption of new controlled terminology standards including additions to ICD-10-CM,²⁹ LOINC/SNOMED CT,³⁰ and the Gravity Project.³¹

4. Policies, procedures, and people are required to troubleshoot potentially erroneous data received from devices. By definition, some percentage of automatically collected physiologic data will be erroneous. These erroneous data points may trigger CDS logic that sends a message to a clinician warning them falsely of a potentially life-threatening event. Other erroneous data may be in the normal range when the patient is, in fact, suffering a potentially life-threatening event. Either situation presents a potential liability for healthcare organizations and clinicians. Without a means to periodically test and fix these remote data collection devices, healthcare organizations are putting themselves in a precarious legal position and may not be able to respond in real-time to patient safety issues. The process for curating and adjudicating connected device data, including managing aberrant and potentially erroneous readings, is often included in the services remote patient-monitoring companies offer. In addition, several third-party remote monitoring companies are working on adding artificial intelligence capabilities into their data management platforms to look for anomalous trends in the raw patient data. Changes in these trends may signal erroneous data or an important change in the patient’s condition, both of which warrant additional attention and/or intervention.

5. Dashboards within EHRs and patient portals are required to display longitudinal patient-reported data. All of these new data will require new methods of integrating PGHD including PROs with existing EHR data while maintaining its provenance, displaying these data in a way that helps both clinicians and patients make the best decision, and developing metrics to summarize the data and use it for predictions.³² In some instances, the most important aspect of the data will be the changing values over time (ie, either improving or worsening). In others, the most recent value may be the most important. Currently, many organizations are in the process of designing and developing such PRO and PGHD longitudinal data displays.^32–34 Several remote monitoring device companies are developing data visualization and analytic services alongside triaging services. As has been learned in many past experiments, the more patient and clinician input that goes into dashboard design, and the more closely integrated dashboard displays are with the EHR, the more likely the Patient App or Patient Portal will be useful and used.

6. New, or revised, reimbursement policies are required to support these new models of care. Collection, management, curation, and use of PGHD in support of patient-centered care will likely require new or revised reimbursement models from federal, state, and commercial payers. These new reimbursement policies will be especially important for ambulatory clinicians who are used to working a regular 5 day per week, 8-hour shift, while the new PC CDS-related care takes place 24-hours per day, 7 days per week. During the COVID-19 pandemic, for example, Medicare and Medicaid programs and commercial payers were forced to expand their telehealth reimbursement models to allow different communication applications (eg, Zoom, Webex, Teams) to be used, clinicians to work across state lines without additional medical licensure requirements, and clinicians to bill for their time as if it were an in-person visit.³⁵ These changes may be a helpful blueprint for ways payment policy needs to change to prevent healthcare organizations from being forced to absorb these costs. The findings from our key informant discussions indicate that healthcare organizations are currently relying on billing for care coordination services to support PC CDS activities, but there are a significant number of resources and time that goes into working with new vendors and setting up processes and procedures that still goes unpaid. Without a reliable, payment model for these specific PC CDS activities, healthcare organizations will need to internally fund these activities.

7. Strategies to support patient engagement with remote devices. PC CDS interventions that rely on PGHD such as physiologic measurements and PROs will require sustained patient engagement in submitting data via remote devices, questionnaires, and other mechanisms. The findings from our key informant interviews indicate that keeping patients engaged in submitting data via remote devices has been a challenge for app developers and health systems alike. Specifically, patient use often declines after a few months. However, few studies have examined the factors that improve and sustain patient engagement, or what types of engagement lead to improved clinical or other outcomes.³ Informants mentioned one potential reason as the narrow focus on a specific clinical condition (eg, hypertension), when the reality is that patients often have multiple comorbidities and symptoms that they care about. Informants also mentioned that patients are increasingly accustomed to using the patient portal to interact with the health care system, since it is a more comprehensive source of information. Given this, patients may be less inclined to submit and view data via a separate device.

Conclusion

The lessons learned from our delineation of workflow execution models for different types of PC CDS point out several new policies, processes, technologies, and people within healthcare delivery organizations, payers, and third-party remote patient-monitoring companies that must work together to design, develop, implement, and test these new interventions. As we gain more experience implementing and working with these new PC CDS-focused workflow execution models, identify the necessary internal and external resources, and gain critical buy-in from clinicians and patients, we should be able to begin realizing the long-term positive impact on patient health we all hope for.

Author contributions

Dean F. Sittig wrote the first draft of the manuscript. All authors participated in the review and refinement of the manuscript. All authors approved the final version of the manuscript.

Funding

This work was supported by the Agency for Healthcare Research and Quality (contract no. 75Q80120D00018).

Conflicts of interest

The authors have no competing interests to declare.

Data availability

Not applicable.

References