Study design
Our team, which consisted of 3 primary care physician health services researchers, 1 fulltime primary care clinician, and 2 preclinical medical students, invited Internal Medicine physicians (n = 50), Family Medicine physicians (n = 39), nurse practitioners (n = 2), and physician assistants (n = 5) from 14 UMHS primary care practices to participate in a study about T2DM screening practices. We excluded trainees and clinicians with less than 0.5 full time equivalents devoted to outpatient clinical practice. Study invitation letters were sent by email, and informed providers that the primary aim of this study was to explore the factors that influence providers’ decisions to screen for T2DM.
Among clinicians who consented to study participation we conducted chart-stimulated recall (CSR) interviews, a methodology used to assess clinical decision-making processes [
12,
13]. During CSR interviews, a clinician uses his or her own documentation to answer questions and explain the rationale for specific clinical decisions. A significant strength of this approach – which has been used to examine physician decisions about screening for colorectal cancer [
14] and screening for prostate cancer [
15] – is that it examines a physician’s own recent clinical decisions, which could provide more valid data on their clinical decision making than assessments of how they might respond to hypothetical clinical scenarios.
Sample
Twenty-five physicians responded to our recruitment e-mail and agreed to participate in our study. We planned to conduct a minimum of 20 interviews with additional interviews to be conducted only if data saturation was not achieved at this point. Prior to each CSR interview, one investigator (DH) reviewed in the electronic health record (EHR) each physician’s clinical encounters from the previous 2 weeks to identify visits for non-diabetic patients 45 years of age and older who were candidates for screening for T2DM at their visit and thus eligible for inclusion in the CSR interviews. We focused on patients 45 years of age and older for 2 reasons. First, the American Diabetes Association (ADA) recommends screening all individuals over the age of 45 for T2DM regardless of other risk factors; therefore an age-based focus allowed us to easily identify patients for whom the ADA recommends periodic T2DM screening. Second, this age-based criterion was more feasible to implement in the context of the EHR than other more detailed criteria for which the ADA recommends screening for T2DM, such as being less than 45 years of age and overweight with at least one additional risk factor for T2DM [
16].
While both the ADA and the USPSTF recommend routine screening for T2DM among at-risk individuals, we used the ADA’s screening criteria because at the time of our interviews they were broader than the 2008 USPSTF criteria [
17]. The USPSTF has since issued revised screening guidelines (disseminated in October 2015), but these were released after completion of all CSR interviews. We did not sample patients who had already been screened for T2DM within the previous year, as most non-diabetic patients do not need to be screened more than annually [
16] and this study was not designed to explore reasons for over-screening.
To facilitate discussion of the range of factors that influence physicians’ decisions to screen for T2DM, we selected in approximately equal numbers encounters in which patients appeared (using the available information in the EHR) to have been screened for T2DM or not screened for T2DM. We classified the following as screening tests for T2DM, based on the ADA guidelines: hemoglobin A1c (HbA1c) tests, oral glucose tolerance tests (OGTTs), and fasting blood glucose (FBG) tests [
16]. Because blood glucose tests at our institution are not routinely labeled as “fasting,” we classified blood glucose tests (drawn either in isolation or as part of a metabolic panel) as being fasting if they were obtained before 11 a.m. or drawn at the same time as a lipid panel that had been labeled in the EHR as fasting.
We also sampled in approximately equal numbers encounters in which patients were seen for either health maintenance examinations (HMEs) or return visits (RVs). We focused on both HMEs and RVs because there is debate about the utility of annual HMEs [
18‐
20] and physicians have been encouraged to consider providing opportunistic preventive screenings at non-HME visits [
21‐
23]. We excluded urgent care visits because in our institution these are typically focused, single-problem visits in which patients are commonly seen by clinicians other than their own primary care clinician.
Using this approach, for each clinician we aimed to select 12 encounters (i.e., roughly 3 each for HMEs with screening, HMEs without screening, RVs with screening, and RVs without screening) from the previous 2 weeks to potentially be discussed during the CSR interview. A goal of 12 encounters for each physician provided the opportunity to skip a particular encounter during the interview in the event that the clinician did not recall the encounter, while still permitting discussion of at least 6 separate patient encounters which has previously been shown sufficient to yield a valid and reliable assessment of practice patterns [
24].
Interviews were conducted by trained preclinical medical students (DN, EM) and lasted approximately 30 min in duration. During the CSR interviews, physicians were first asked whether they screened each patient for T2DM and what factors influenced this decision. Our questions about reasons for the physician’s decision were guided by whether the physician stated they had or had not screened the patient for T2DM, rather than how the patient had been classified in our purposive sampling. For example, if it appeared from the EHR that the physician had screened the patient for T2DM, but the physician stated that they had not screened the patient for T2DM, then the physician was asked why they had not screened the patient for T2DM. Conversely, if a patient had been sampled as having not been screened (e.g., because they had a random glucose), but the physician said they had screened the patient for T2DM, we asked the physician why they screened the patient (even though the test chosen was not an ADA-recommended test). This approach allowed us to focus on physicians’ decision making processes irrespective of the outcome of that process.
When a screening test result was available for review (i.e., there was a lab result for a screening test that had been ordered at that visit), physicians were asked how they interpreted the result, whether and how they communicated the result to the patient, and what they communicated to the patient about the result. Following the CSR interview, physicians were also asked general questions regarding barriers to T2DM screening. Physicians had access to their EHR documentation during the interview. The interview guide was derived from previous CSR studies [
14,
15] and is provided in the Additional file
1.
For each interview, physicians received a $50 gift card. The study was approved by the Institutional Review Board at the University of Michigan Medical School.
Analysis
All interviews were audio-recorded and transcribed verbatim. Three members of the research team (DH, DN, JK) independently reviewed a subset of transcripts. The transcripts were de-identified prior to data analysis to minimize the potential for biased interpretation of the data. Codes and definitions were generated during consensus conferences using directed content analysis [
25]. Specifically, initial codes were created to reflect the main topics in the interview guide (e.g. decision to screen or not screen patients for T2DM), and additional codes were subsequently generated to reflect the patterns and themes that emerged from the data [
26]. Once the coding scheme was established, two investigators (DH, DN) independently coded each transcript. These investigators then met to review their coding and resolve all differences. Few new themes emerged after coding 12 transcripts and no new themes emerged after coding 16 transcripts. Given that we reached data saturation [
27,
28], we did not conduct additional interviews following the 20 interviews. All transcripts were coded in Dedoose, a software program for qualitative analysis.
The following data were abstracted from the EHR for each patient after the interviews were completed: age; sex; race; ethnicity; BMI; risk factors for T2DM documented in the patient’s problem list such as hypertension, hyperlipidemia, cardiovascular disease, prediabetes, history of gestational diabetes or polycystic ovarian disease.