Clinical Reminders Designed and Implemented Using Cognitive and Organizational Science Principles Decrease Reminder Fatigue

Setting

This project was conducted in 5 off-campus family medicine clinics affiliated with a large, midwestern academic medical center. Two sites serve suburban, largely educated populations; 1 is located in an urban underserved area; and 2 are in small towns/rural settings. The urban site and 1 rural site serve large numbers of Medicaid and uninsured patients. Two (the urban site and 1 small town) are residency education sites, but the majority of visits in all 5 sites are with attending physicians. A total of 47 attending physicians and 8 nurse practitioners or physician assistants practiced in the 5 sites during the time the data were collected. All the sites are connected to the medical center via high-speed networking and during the study period used the medical center's homegrown electronic medical record system (which at the time did not have CRS capability).

Design and Implementation

We consulted with a human factors engineer experienced in the design of information presentation systems in aviation and discussed our plans and ideas with colleagues in the University of Michigan's Decision Consortium.¹³ After the initial launch and approximately a year of experience with the system, we conducted a cognitive task analysis¹⁴ of how family physicians structured office visits and acquired patient-specific information before and during visits. The methods and findings are detailed elsewhere¹⁵; briefly, the Task Diagram tool was used, with specific focus on information handling during visits. The findings were used in ongoing improvement of implementation.

This combined formal and informal design process led us to understand that clinical reminders are not at all the simple intervention they might at first seem. Though often compared with checklists in aviation, reminders for preventive or chronic disease management services in primary care differ in important ways. They are often task-discordant, that is, they direct attention to matters other than those for which the patient presents. Task flow in clinical practice is context-dependent and highly nonlinear; decisions must be made with limited or unreliable evidence under changing circumstances,¹⁶ and addressing a reminder may require interaction with a team member who is not present when the reminder is presented—in contrast to the deliberately highly routinized processes of airliner piloting. Aviation processes are deliberately designed to reduce cognitive workload; that certainly cannot be said of primary care. In the primary care setting clinical reminders are powerful stimuli being introduced into an information and cognitive processing environment that is already very complex and challenging.^17,18 Viewed from this perspective, clinical nonresponse to reminders can be considered an adaptive response: actively filtering a stimulus that could produce a break-in-task event¹⁹ and hence disrupt the decision-making process for a patient's primary need.

Taking into consideration these concerns, the results of our cognitive task analysis, and the advice of our human factors engineer, we developed a set of 8 core principles to guide our design and implementation. These principles are listed in Table 1. Though the work by Patterson et al⁹ was not yet published as we were designing our intervention, our 8 principles converged independently on the need to address 6 of the 10 barriers to CRS implementation they documented (workload, time to remove inapplicable reminders, false alarms, training, ease-of-use issues, and administration benefiting more than providers from clinical reminder use). The team-oriented implementation approach also designed out the physician–nurse coordination issues that Saleem et al¹⁰ found impeded CRS operation. Finally, our principles also allowed us to reach independently an important conclusion from the major primary care improvement initiatives in the United Kingdom: Improvement targets should be chosen for clinical value, not to control costs.²²

The system that resulted from this design process, called “ClinfoTracker,” has been described in detail elsewhere.²³ The system was subsequently licensed by the University of Michigan in 2006 to a startup firm (Cielo MedSolutions, LLC) for commercialization, but it remained unchanged in daily use, supporting patient care throughout the study period. Figure 1 illustrates the printed version of the reminder form printed by the system and given to the care team in advance of each visit. The on-screen version appeared very similar, with the exception of the response options being radio buttons. It was available for the team's use as an ordinary web form; it was not embedded in the electronic medical record. At each visit the clinician and other team members updated the problem list (independent of billing diagnoses) and indicated their responses to reminders either on the paper form or on-screen. Whichever form was used was simply made available for use—there was no requirement to interact with it—and could be referred to whenever and as many times as the team members chose. The on-screen version became available in 2009 but was used by only 2 clinicians, although it was available to all. The paper form was always available, even to on-screen users, and was overwhelmingly preferred by clinicians.

• Download figure
• Open in new tab

Figure 1.

Printed version of the reminder form.

The process of building support and buy-in was initiated through discussions of evidence-based quality goals at monthly physician meetings. Discussion among the physicians guided the selection and specific details of reminders. These 2 activities addressed the first and second implementation principles in Table 1. We addressed the fourth implementation principle using feedback received during the monthly meetings and negotiations with administrators. For example, the asthma action plan reminder was launched then deactivated based on feedback. We worked with clinic directors to allocate registered nurse time and to obtain training for registered nurses before relaunching the reminder. We also discovered during monthly meetings that sites were using the reminder forms in novel ways to facilitate communications among team members, and we actively encouraged them to continue (third implementation principle). After the launch, we began publicly recognizing physicians with high response rates at the monthly meetings (awarding first a coffee mug and subsequently gift cards for coffee). Finally, and only after most clinicians were regularly achieving high response rates, a small ($1 per relative value unit), individual quarterly payment incentive was implemented for achieving a 90% rate of documenting consideration of the problem lists and reminders issued at each visit, averaged over 3 months. All that was necessary to document consideration was return of the paper form or use of the electronic form with at least 1 item of data: a new or changed problem list entry, an existing problem checked as seen at this visit, or any response to any (not all) of the reminders. The incentive amounted to about $12 for a typical 4-hour clinic session.

The initial launch included only reminders for screening and prevention services selected by the physicians from among those recommended by the US Preventive Services Task Force. Once those were in use and the logistics of printing, distributing, collecting, and entering data from the paper forms in 5 offices were working smoothly, sets of reminders for chronic disease management were launched, 1 set at a time every 2 to 3 months. The sets requiring the most practice change and additional training and resources were launched last (heart failure and asthma). The full set was in place after a year. The full list of reminders is presented in Table 2.

Data Collection and Analysis

The data for these analyses were extracted from the CRS. All patient encounters between January 1, 2006 (when the CRS was first rolled out across all the sites), and July 31, 2012, were extracted. Encounters at which no reminders were issued were excluded. Our unit of analysis was an individual reminder issued during an office visit through printing on an encounter form. A reminder could be to perform a specific action (eg, administer influenza immunization) or to evaluate a clinical situation (eg, blood pressure is out of goal range, consider change in therapy). Response options included the following:

1. Done: taking the advised action during the visit (eg, administering influenza vaccine, evaluating antihypertensive regimen)

2. Ordered: completing an order for the action (eg, placing a mammography requisition)

3. Not Cand.: documenting noncandidate status (eg, the patient had a hysterectomy for benign disease and does not need Papanicolaou tests)

4. Declined: documenting that the patient declined

5. Discussed: documenting that the issue was discussed but no decision was made

6. Not Addr.: documenting that the matter was not addressed (eg, patient presenting in crisis, discussion of colorectal cancer screening was inappropriate)

Logistic regression adjusted for clustering by visit was used to model the odds of a reminder being addressed as a function of patient age (in integer years), sex, time (months) since CRS launch, number of reminders issued during the visit, and number of active problems on the patient's clinician-verified problem list (as a proxy for patient complexity). Modeling was conducted using Stata software version 10.1 (StataCorp, College Station, TX). Response options 1, 2, 3, and 4 above were designated as “actions” in response to reminders, and option 5 was designated “consideration.” Models were computed both for outcomes of action and action plus consideration.