Patient-Centered Guidelines for Geriatric Diabetes Care: Potential Missed Opportunities to Avoid Harm

Ellen M. McCreedy, PhD; Robert L. Kane, MD; Sarah E. Gollust, PhD; Nathan D. Shippee, PhD; Kirby D. Clark, MD


J Am Board Fam Med. 2018;31(2):178-180. 

In This Article


This study used a vignette survey of primary care clinicians, including internal medicine (IM) physicians, family medicine (FM) physicians, and nurse practitioners (NPs). A vignette was a 1-paragraph description of a common patient scenario. We systematically varied the patient characteristics to better understand whether clinicians individualize care and what patient factors drive these treatment decisions. Respondents were specifically asked whether they would add a second-line diabetes medication (intensify medication therapy) for a patient who had already been treated for 6 months with first-line metformin therapy.

Data and the Sample

A convenience sample was obtained primarily through the use of state licensure lists from Minnesota and Florida. We also had participants from 3 Agency for Health Care Research and Quality practice-based research networks (PBRNs) and a small state professional association meeting. To be eligible, physicians and NPs had to be actively practicing primary care medicine at least 75% of the time.

For the PBRNs, participants were contacted by the network director using a standardized e-mail invitation with an embedded link to complete the online survey. E-mail addresses of PBRN members were not released to investigators. For the clinician e-mails obtained through licensure lists, the study primary investigator (EMM) e-mailed clinicians directly using the same standardized e-mail invitation used by the PBRNs. All responses were collected anonymously using Qualtrics software. A small number of study participants completed the study in person while attending a state professional association meeting. This study was exempted from full review by the University of Minnesota Institutional Review Board Human Subjects Committee (study no. 1305E33481).

Vignette Design

The patient factors randomly varied in the vignettes were HbA1c level, age, disease duration, presence of cognitive impairment with IADL dependencies, and history of coronary artery disease with previous coronary artery bypass graft. Each of the 4 patient characteristics were presented at 2 HbA1c levels (Table 1), yielding a total of 16 possible vignette combinations. All vignettes depicted a hypothetical patient of the same sex, weight, kidney function, and ability to pay for medications. In addition, every patient had hypertension, mild neuropathic symptoms, no comorbid depression, and no activity of daily living dependencies. Each respondent viewed 4 randomly selected vignettes. Therefore, every clinician did not receive the same survey. A sample vignette and the response set are provided in the Appendix.


Treatment intensification was defined as adding any 1 of 5 classes of approved second-line medication therapies. A dichotomous variable was created to indicate intensification of medication therapy. Any choice to add another medication was categorized as treatment intensification.

In addition to the vignette characteristics, we investigated the effects of several clinician characteristics on the decision to intensify treatment. Clinician training was defined using an item that asked respondents to self-identify as "physician, family medicine," "physician, internal medicine," "nurse practitioner," or "other." Clinicians were not specifically asked about specializations. Other clinician and practice characteristics were investigated: year in which the respondent finished professional education, average duration of a routine visit (minutes), and percentage of practice that is Medicare (<25%, 25–75%, >75%).


We used bivariate analysis and multivariate regression to study the effect of vignette characteristics on the decision to intensify treatment. Because clinicians viewed >1 vignette, we used random effects probit, a regression model that accounts for correlations between responses from the same clinician, or clinician cluster effects. Vignette and clinician characteristics were entered into the model as fixed effects, with a random intercept for each clinician. The random effects model assumed that variation across respondents is not correlated with the independent variables included in the model (vignette factors). The Hausman test confirmed that this assumption held for the current analysis and that random effects was an appropriate model (P = .30, fail to reject null hypothesis that variation across respondents is correlated with predictors). We also were able to estimate the intraclass correlation, or the amount of total variation attributable to idiosyncratic clinician effects. All analyses were conducted using Stata software version 14.