using the critical incident technique to better understand patient experiences of ambulatory care...
TRANSCRIPT
Using the Critical Incident Using the Critical Incident Technique to Better Understand Technique to Better Understand Patient Experiences of Patient Experiences of Ambulatory CareAmbulatory Care
Presented by:Kristin L. Carman, Ph.D.American Institutes for Research
Presented at:Academy for Health Services ResearchAnnual Meeting, San Diego, CA June, 2004
Project Team Roger Levine, PhD
Managing Research Scientist, AIR
Karen K. Shore, PhDSenior Social Scientist
Margarita Hurtado, PhDPrincipal Research Scientist
Kristin L. Carman, PhDPrincipal Research Scientist
Judy Mitchell, MSSenior Research Scientist
Steven A. Garfinkel, PhDManaging Research Scientist
San Keller, PhD• Principal Research Scientist
Funding Agency for Healthcare Quality
and Research• Part of the CAHPS II grant
Partners HMSA, Hawaii Humana, Chicago
Purpose of our projectPurpose of our project
To develop an A-CAHPS Survey
To use Critical Incident data in novel ways to address issues related to: Instrumentation
Quality improvement
Reporting, and
Cultural comparability
Critical Incident (CI) Critical Incident (CI) TechniqueTechniqueCritical Incident
“Incident”=an observable, specific behavior
“Critical”=means incident was crucial to the outcome of interest
Organized structure for data collection; focuses on observable behavior
Used to collect and analyze reports of behaviors associated with specific outcomes
Qualitative method; in-depth interviews
Methods and dataMethods and data
200 interviews; 40 providers and 260 patients
Patient respondents divided equally among four different racial/ethnic groups
Open-ended responses are transcribedEach interview usually generates 10
incidents; we’ll have 2000+ incidents
CI data management and CI data management and processingprocessing
Gathering extensive data
Using qualitative software to create and manage a very complex data base
Developed a very careful data processing protocol Raw data is open-ended responses, transcribed
Data is transformed into “incident write ups”
These two types of data are the sources for analyses
Specific GoalsSpecific Goals
Instrumentation goalsInstrumentation goals Develop a complete taxonomy of the
components of quality ambulatory health care, based on both patient and clinician perspectives Confirm that the domains measured by the draft
CAHPS® instrument are salient to patients and providers and can be assessed by patients
Determine whether the domains are salient and measurable for both men and women, for individuals with different levels of education, and across a range of racial and ethnic groups
Identify additional domains that should be measured
Instrumentation goals (cont’d)Instrumentation goals (cont’d)
Identify CAHPS® item content that can result in a spread of scores at the high end of the score distribution to minimize ceiling effects Generate objective patient reports of
health care experience
Create items for Ambulatory CAHPS® for which a positive rating would be rare
Instrumentation analysisInstrumentation analysis
Developing the taxonomy Randomly select at least 200 incidents
Two teams classify incidents into major categories, then subcategories
Iteratively validate and refine the taxonomy with additional set of incidents
Instrument analysis (cont’d)Instrument analysis (cont’d)
Using incidents as the unit of analysis, then respondents, we investigate statistical associations between personal characteristics and the way people conceive of quality of care Tabulation of respondent characteristics by
taxonomy categories which are quality of care themes (e.g., coordination of care)
Regress taxonomy categories on respondent characteristics
Instrument analysis (cont’d)Instrument analysis (cont’d)
Logistic Regression Simple and multiple
Dependent Variable=taxonomy category
Independent Variables= Individual demographics
Respondent: provider or patient
Quality improvement goalsQuality improvement goalsIdentify physician behaviors associated
with excellent and poor quality of care based on experiences of patients and clinicians Identify combinations or co-occurrences of
behavior
Identify key facilitators and barriers to quality of care, specifically related to CAHPS domains
Create tools for QI interventions to improve CAHPS scores
Quality improvement Quality improvement analysisanalysis Analyze interview and CI files Conduct additional coding of data Focus on which behaviors or actions
by clinicians co-occur to create positive (or negative) experiences for patients
Compare findings by respondent characteristics
Cultural Comparability goalsCultural Comparability goals
Identify variations in taxonomic structure for different racial/ethnic groups
If differences exist, identify the implications for: Supplemental domains, concepts and items
CAHPS domain labels and explanatory vignettes
Culturally appropriate interventions to improve care
Reports goals and analysisReports goals and analysis
Identify narratives (phraseology) that clearly and effectively explains CAHPS measures in reports
Next stepsNext steps
Complete interviews
Complete analyses
Disseminate findings
For more information, contact:For more information, contact:
Kristin L. Carman, PhDPrincipal Research ScientistAmerican Institutes for
Research1000 Thomas JeffersonWashington, DC 20007(202) [email protected]
Karen K. ShoreSenior Research ScientistAmerican Institutes for
ResearchADDRESSPalo Alto, CA (ZIP)PHONE [email protected]
[CI 1-2]
[CI 1-3]
[CI 2-1]
[CI 2-2]
[CI 3-1]
[CI 3-2]
[CI 1-1] Episode 1
Episode 3
Episode 2
Critical Incident behaviors
CI 1-1
Critical Incident Forms
CI 1-2
CI 1-3
CI 2-1
Verbatim TranscriptMarked by interviewer for Episode # andfor CI behaviors (not numbered—only numbered here for illustration) in Atlas/ti
CI Forms filled out by cut and paste fromTranscript, numbered to reflect source episode.This takes place in Word
CI 1-1
CI 1-2
CI 1-3
CI 2-1
CI forms cut into separatetext files for sorting into Taxonomy
Links are logical only: episode # in CI # allows analyst to trace back to
transcript
Transcript CIs for Interview Separate CIs
Database Creation
[CI 1-2]
[CI 1-3]
[CI 2-1]
[CI 2-2]
[CI 3-1]
[CI 3-2]
[CI 1-1] Episode 1
Episode 3
Episode 2
CI 1-1
CI 1-2
CI 1-3
CI 2-1
At this stage, the finalized concatenated CI file, with taxonomy codes embedded, is imported into Atlas/ti. Taxonomy codes are “autocoded” by searching for Taxonomy codewords. At this point, hypertext links can be created between CI forms and episodes in the transcript. If desired, Taxonomy codes may be applied manually to the transcript.
P-01-M-JM-1-1-ATAX-Communicates
P-01-M-JM-1-2-BTAX-Clarifies
The ID code includes #’s for episode and CI.Since ID #’s include identifiers for referents, they are assigned to demographic code families for referents
(See next page fordetail)
Transcript Concatenated CIs
Final Atlas/ti Database
Both transcript and concatenated CI filesare assigned to docfamilies for respondentdemographics
(See next page fordetail)
Taxonomy codes are
assigned by autocoding
Respondent demographics are represented by placing“primary documents” in “PD Families” or sets:
Example:
Gender::Male = {PD1; PD4; PD5; PD8…}Gender::Female = {PD2; PD3; PD6; PD7…}Age::20s = {PD1; PD2…}Age::30s = {PD4; PD8…}Age::40s = {PD3…}Age::50s = {PD6…}Age::60s = {PD5; PD7…}
Imported in a table:
Gender AgePD1 Male 20sPD2 Female 20sPD3 Female 40sPD4 Male 30sPD5 Male 60sPD6 Female 50sPD7 Female 60sPD8 Male 30s…
Dataset can be parsed according to Boolean combinations of set-memberships, for example, to restrict a query to documents that belong in both the Male and 20s sets.
Referent demographics are represented by placing CI ID codes into “Code Families” or sets. The logic is the same as for PD families, but at present there is no table import feature and assignment is made with the “code family manager” tool in Atlas/ti.
Again, set memberships can be used to focus queries ondifferent classes of referents.
RESPONDENT DEMOGRAPHICS REFERENT DEMOGRAPHICS
Strategies for associating data at episode or interview levels
Layered IDs:
Since ID’s are layered, a hierarchy of ID codes can be created in Atlas/ti. For example, a hierarchy could be structured as follows:
Interview ID +--Episode ID +--CI ID +--Referent ID
Code Families:
A code family that included all the CI ID codes for a given episode would enable searching by episode in the concatenated CI files.
Hypertext:
Hypertext links can be created from the CI forms to the episodes in the transcript, or even to the specific descriptions of behaviors from which the CIs are derived.