Telemedicine in the Wake of the COVID-19 Pandemic

Increasing Access to Surgical Care

Paige K. Dekker, BA; Priya Bhardwaj, MS; Tanvee Singh, MPH; Jenna C. Bekeny, BA; Kevin G. Kim, BS; John S. Steinberg, DPM; Karen K. Evans, MD; David H. Song, MD, MBA; Christopher E. Attinger, MD; Kenneth L. Fan, MD


Plast Reconstr Surg Glob Open. 2021;9(1):e3228 

In This Article


All consecutive patients with outpatient appointments among 5 providers in the Plastic and Reconstructive Surgery Department between March 2, 2020, and April 10, 2020, were retrospectively reviewed. These providers were selected as (1) the 5 highest volume providers and (2) those with the highest percentage of patient visits at our hospital-based practice and not at affiliate hospitals or off-site outpatient clinics. In the District of Columbia, a public emergency was announced on March 11, 2020, and stay-at-home orders were enforced beginning on April 1, 2020. Our system began a multi-phase expansion of its telemedicine platform beginning on March 23, 2020. This date served as an anchor for data collection to capture patients seen by our service before and after the expansion of telemedicine in our system. Patient groups were primarily wound care, general reconstruction, and breast reconstruction.

Data Collection

Data on the following domains were collected from the electronic health record. Appointment characteristics included visit modality (in-person, phone appointment, and video appointment), reason for visit, new or established patient, and history of recorded procedure. Patient characteristics included demographics such as age, sex, race, insurance provider, urban/rural designation of the location of residence, SVI, and median income by location of residence. The primary outcome of interest was whether or not patients missed their appointment (show versus no-show).

For city/suburb/town/rural designation and median income by area of residence, Zone Improvement Program (ZIP) Codes were converted to Zip Code Tabulation Areas (ZCTA), as used by the Census Bureau. ZIP codes are designed to represent linear mail delivery routes, whereas ZCTAs represent more generalized spatial codes that are assigned by census block. For locality designation, the National Center for Education Statistics 2019 data were utilized.[18] For median income by ZCTA, the S1901 table from the American Community Survey in 2019 was obtained.[19]

The overall SVI is derived from census tract-level data, which account for increased granularity of neighborhoods. This is especially important, given the heterogeneous nature of communities in the District of Columbia and its surrounding areas. Census tracts were obtained from the Federal Financial Institutions Examination Council geocoding system, which is used by financial institutions to report information on mortgages as well as business and farm loans.[19,20] For addresses that were developed after the census was performed, census tract information was extrapolated based on the longitude and latitude of the address derived from the Google Maps API.

The SVI refers to the socioeconomic and demographic characteristics of a community that impact its resilience when faced with external stressors to human health, including disease outbreaks. The SVI ranks census tracts based on 15 social factors, grouped into 4 themes (Socioeconomic Status, Household Composition and Disability, Minority Status and Language, and Housing Type and Transportation), for an overall SVI score.[21] For this study, the overall SVI score was used. A pre-published study suggests that the SVI is associated with higher COVID-19 case fatality.[22]

Definitions and Exclusion Criteria

Patients from outside the District of Columbia, Maryland, and Virginia regions were eliminated from analysis. A "no-show" was defined as any appointment that a patient did not attend and was both (1) not intentionally rescheduled before the appointment date and (2) the patient was not hospitalized other reasons. Insurances were categorized as commercial (HMO/PPO), Medicaid, or Medicare. Self-pay and other insurances were excluded due to small numbers. Ethnicity was categorized as White, Black, or Other due to the low representation of American Indians, Asians, and Pacific Islanders in our region.[23]

Statistical Analysis

The statistical analysis was broadly separated into pre-lockdown and post-lockdown to determine the differences between no-show characteristics. Continuous variables were described by means and SDs. The student t-test was used to examine statistically significant differences between continuous variables when normality assumption was satisfied; the Wilcoxon rank sum test was used when normality assumption was not satisfied. Categorical variables were described by frequencies and percentages. Chi-square and Fisher exact tests (n < 10) were used as appropriate to examine statistically significant differences between categorical variables. To test for spatial autocorrelation and clustering of no-show appointments, Moran's I test for autocorrelation was used.[24,25] Two multivariate models before and after the COVID-19 lockdown were constructed with variables selected based on the purposeful selection method, as described by Hosmer and Lemeshow.[26–28] Multicollinearity was tested to assess the effects of certain variables on others within each model.[29] Statistical analysis was performed using STATA, v.15 (StataCorp, College Station, Tex.), with significance defined as P < 0.05.