We analyzed electronic medical record (EMR) data for all patients who were identified by our institutional population registry as having type 2 diabetes mellitus (DM) as of March 19, 2020. We abstracted the following variables from our institution’s electronic medical record (EMR) at the individual level: decade of age, gender, race, ethnicity, category of insurance (Medicare, commercial, Medicaid, or institutional managed care plan), systolic blood pressure, hemoglobin A1c, aspirin and/or statin prescription, smoking status, ambulatory visits to primary care, endocrinology, and other department of medicine visits, as well as hierarchical condition category (HCC) score. We also included income information from the American Communities Survey 2018 5-year estimates at the five-digit zip code level.
Design and Study Period
We used a quasi-experimental (difference-in-differences) design. We defined our post-period to include the 9 months following the declaration of “safer-at-home” orders in Los Angeles County on March 19, 2020.24 We defined two sequential pre-periods of the same duration in the 18 months prior to the declaration to identify if pre-pandemic quality of care trends were parallel.
Study Population and Exclusion Criteria
Our study population (N = 21,460) consisted of all patients identified by institutional clinicians as having type 2 diabetes mellitus when added to the electronic medical record problem list. This is verified for accuracy by the institutional quality team. Two physicians, one registered nurse, and two non-clinicians independently verify a random sample of patients identified as having type 2 diabetes mellitus. The study population was restricted to align with quality reporting standards, restricting to include adult patients under age 75 who are not currently enrolled in hospice care (N = 16,588).
We defined our outcome for quality of care for patients with diabetes using the same domains as indicators included in the CMS Diabetes Composite quality measure from the CMS Measures Inventory.25 For our primary outcome, we evaluated the probability of meeting any given component in the five-component composite measure. The components are the last recorded value of (1)systolic/diastolic blood pressure less than 140/90 mmHg, (2) hemoglobin A1c less than 8.0%, (3) active prescription for statins and/or(4) aspirin or other anti-platelet agents if not contraindicated, and (5) tobacco non-users or those that quit during the study period.23 We made pre-specified changes to the composite for the purposes of our analysis. Due to limitations in our EMR, we did not exclude patients who may have a contraindication for anti-platelets or consider other anti-platelet agents (i.e., clopidogrel) as meeting this measure criterion, and did not analyze diastolic blood pressure as the proportion of patients with diastolic-only hypertension is negligible. We also did not exclude patients who became pregnant during the study period and were not able to ascertain deaths in our cohort.
Exposure and Control Group
We identified all included patients with at least one telemedicine visit in the post-period to either a primary care clinician (N = 6,405) or an endocrinologist (N = 2,251) to define our total exposure group (N = 7,581). Telemedicine visits are defined at our institution using the (EPIC/Clarity) visit types as recorded in our EMR, and distinguish between telephone encounters, new video visits, and return video visits. Visit type codes are unique to the EMR instance but available on request.
We included the following covariates in addition to our primary regressor: decade of age, gender, race, ethnicity, category of insurance (Medicare, commercial, Medicaid, or institutional managed care plan), hierarchical condition category (HCC) score, all at the individual level. We also included income information from the American Communities Survey 2018 5-year estimates at the five-digit zip code level.
Our study protocol was approved by the UCLA Institutional Review Board via expedited review.
Using a binomial regression model, we estimated the effect of telemedicine use (at least one encounter in primary care or endocrinology) by comparing the likelihood of meeting any individual component of the diabetes composite outcome in the 9 months before and after March 2020 between the exposure and control group. All models included as covariates the sociodemographic and clinical characteristics described above. We performed several sensitivity analyses including (1) meeting all five component indicators of the diabetes composite measure, and sequentially fewer (4/5, 3/5, 2/5, and 1/5) indicators, (2) meeting four component indicators omitting the systolic blood pressure variable, which we hypothesized would be difficult to assess via telemedicine, and (3) a subgroup analysis of patients receiving Medicare benefits (either through disability or over the age of 65 years) and meeting 4/5 indicators given hypothesized challenges of older and disabled patients in accommodating new technology (Appendix 1). As a fourth sensitivity analysis to determine if there was a disproportionate quality impact on complex patients, we performed a subgroup analysis of patients meeting 4/5 outcome indicators among those with an HCC score of two or above, double that of the typical Medicare beneficiary, approximately in the top decile of complexity for our population (Appendix 1).
A significance level of 0.05 was used throughout, and analyses were performed using Stata 16c (StataCorp, College Station, TX). Our binomial regression model was specified by using the STATA command “binreg”, and all sensitivity analyses were using logistic regression (“logistic”).