Contextual information consists of the functional, social and financial information about patients.1 Because contextual information is idiosyncratic and patient specific, it is key to understanding the patient's unique situation. However, social and functional information is often not communicated across settings and is neglected in clinical decision-making despite the fact that functional information is a powerful predictor of patient outcomes and social information can dramatically impact the treatment and placement decisions. Increasing the availability of contextual information has been shown to improve care, enhance decision-making and improve team-work.2 Despite its importance, contextual information is either unavailable or difficult to locate in current EHRs.3
Many psychological studies suggest contextual information may be not be organized cognitively the same way as clinical information.4 Episodic and social information is likely represented in "story-form' and is personal, specific, and linked to a particular time and geography.5 Thus, sharing contextual information may be best done through stories or narrative, rather than through lists of data. Clinical information, in contrast, is abstracted, semantic, and de- personalized knowledge and might be best represented numerically. Yet, much of what constitutes current decision-support eliminates important context-specific information in the effort to minimize information overload. The result is the ironic effect that less information is processed slower than more information.
Because contextual information is used by all clinical roles, enhancing the availability and usefulness of this kind of information requires creating a shared information space in the medical record. The format and structure of this space is unknown. Getting clinicians on the "same page" should be the goal of EHR design.
The purpose of this study is to determine the important characteristics of contextual information necessary to support the most effective natural language extraction methods and the most clinically useful display.
The objectives of this study are to: 1) Conduct a cognitive task analysis (CTA) of case managers working in a Medical Home care environment; 2) Develop a use case and vignettes as simulation materials; 3) Assess the impact of a "story-form" model versus an "information" model on shared situational awareness in an experimental design study.
Design. The design was qualitative and used CTA (cognitive task analysis) techniques and observation with protocol analysis or "think-aloud" protocols.
Setting. All providers interviewed were employed at the SLC VAMC, and recruited from Home Based Primary Care (HBPC), the Geriatric Home-base discharge program (G-Help), and the SLC VAMC outpatient primary care clinics.
Participants. Seventeen providers were recruited from the SLC VAMC, including with 7 Social Workers, 6 Nurse Practitioners, two residents, 1 MD, and 1 Case Manger. All had worked greater than 2 years.
Analysis. The transcripts were analyzed using conventional content analysis techniques. Coding protocols were developed iteratively by the research team and recorded in the qualitative software program Atlas ti@. The descriptive phrases and the original quotations were aggregated into categories/codes through team consensus.
Design. The PI and a co-investigator identified two patient use cases. The patients had equivalent diagnoses and comorbidities, were relatively complex and equivalent in ages (older) and both have some functional deficits. Both had an event that triggered the providers to need to make a decision about the patient's care, such as the need for placement.
Design. The design was a 2-way within-subject randomized trial where participants were assigned to an order of presentation of two cases, each with either a narrative format (+goals) for describing functional information or a numerical presentation.
Dependent Variables. Participants read the short vignette/patient summary and were asked to identify: 1) the top 3 issues for the patient and 2) the top 3 factors they used in considering their plan. Items were coded as being either "medical" or "contextual" by the research team through a consensus process. The proportion of items identified in each category were the dependent variables.
Participants. 30 physicians were assigned to one of two orders of receiving the information.
Procedures. Volunteer residents, nurse practitioners, and staff physicians were approached singly and asked to fill out the form. Study involvement times averaged 10".
The results from this initial analysis consisted of an initial taxonomy of the characteristics of contextual information. These characteristics were intended to be as orthogonal as possible in terms of their abstract qualities, but the team discovered that a particular quotation from the interviews could easily be coded with more than one attribute. The six components are: 1) Informativeness, 2) Goal elucidation, 3) Temporality, 4) Source attribution, and 5) Quality (completeness and accuracy).
Two patient cases were created and each were modified to have two formats for presenting contextual information (one in narrative forma and the other in numerical form). Pilot testing suggested they were equally understandable.
The proportion of contextual related topics were computed for both issues and factors and differences between groups was analyzed using a within subject design to control for within-subject correlation. Overall, there was a significant difference between the narrative and numerical form with those receiving the narrative form for contextual information being more likely to list it as an issue (F1,56 = 5.21; p=0.03; M(narrative group) = 69% and M(numerical group)= 31%). This difference was higher for one vignette than the other. However, the difference between the proportion of contextual factors listed was not significantly different for factors considered important to their planning, although the difference was in the right direction. Reference to goals in the description of factors was significantly higher for the group receiving goal information ((F1,56 = 6.20; p=0.01; M(narrative group) = 74% and M(numerical group)= 26%).
The results of this project will be utilized immediately in the design of the presentation layer of iEHR, especially in the area of care coordination. Information about the important clinical characteristics of contextual information will support the submission of a full IRR on NLP methods to capture such information. Improvement in care coordination and cognitive support for the EHR is expected.
- Weir CR, Staggers N, Gibson B, Doing-Harris K, Barrus R, Dunlea R. A qualitative evaluation of the crucial attributes of contextual information necessary in EHR design to support patient-centered medical home care. BMC medical informatics and decision making. 2015 Apr 16; 15(1):30.
- Weir C, Dunlea R, Staggers N, Dooing-Harris K, Mccormick T, Barrus R. Comparing narrative versus numerical display of functional information: impact on sense-making. Studies in health technology and informatics. 2014 Jan 1; 205:609-13.
- Doing-Harris K, Barrus R, Weir CR. Integrating Qualitative Analysis with Automated Topic Identification. Presented at: American Medical Informatics Association Annual Symposium; 2012 Nov 3; Chicago, IL.
- Doing-Harris K, Weir CR. Investigating Nursing Handoffs with Automated Topic Identification. Poster session presented at: American Medical Informatics Association Annual Symposium; 2012 Nov 3; Chicago, IL.
- Weir CR, Pepper G. Medication Management and Communication Across Roles. Paper presented at: Healthcare Systems Ergonomics and Patient Safety Annual Conference; 2011 Jun 1; Olveido, Spain.