Patterns of Telemedicine Use and Glycemic Outcomes of Endocrinology Care for Patients With Type 2 Diabetes

This cohort study evaluates patterns of telemedicine use and their association with glycemic control among adults receiving endocrinology care for type 2 diabetes.


Introduction
1][12][13] Examination of clinical outcomes is critical, as patients with T2D may be more diverse, clinically complex, and face additional barriers to accessing care than clinical trial participants. 14,15[18][19] An understanding of which patients with T2D have continued to use telemedicine and how their glycemic outcomes vary across different clinical scenarios is necessary to identify which patients can successfully manage their diabetes with telemedicine alone and which may need additional support or in-person care to reach treatment goals.Endocrinology practitioners have expressed concern that patients with increased medical complexity may be less well served by telemedicine care. 19,202][23] In this study, we aimed to address this evidence gap by evaluating (1) the characteristics of adults with T2D who persisted in using telemedicine-only vs those who switched to in-person or mixed endocrinology care after initial telemedicine use early in the COVID-19 pandemic, (2) the association of these care modalities with glycemic outcomes, and (3) how factors that contribute to clinical complexity, including insulin regimen and comorbidities, are associated with glycemic outcomes across these different care modalities.

Study Design and Clinical Setting
This retrospective cohort study included adults with T2D who were seen via telemedicine for either an initial or follow-up visit between May 1 and October 31, 2020, in the endocrinology division of a large health system, which includes more than 30 practitioners across 8 clinics in rural and urban counties.Similar to other care settings, most patients with T2D in this health system are managed by their primary care practitioners.However, patients may receive care from an endocrinologist through referral by a health care practitioner or self-referral, with associated costs varying according to their specific heath insurance plan.In this clinical setting, use of telemedicine vs in-person care was based on individual patient and practitioner preference and availability; there were no blanket policies determining visit modality.To reduce bias in assessment of ongoing telemedicine use and glycemic outcomes, we focused on patients with capability to use telemedicine at baseline.This study was approved by the University of Pittsburgh institutional review board and determined to be exempt from informed consent as it involves secondary research on data collected as part of routine care.
The results are reported in accordance with Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines. 24

Study Cohort
Patients were included if they had at least 2 hemoglobin A 1c (to convert from percentage to proportion of total hemoglobin, multiply by 0.01) values and 1 subsequent encounter in the division of endocrinology during the follow-up period, which extended through May Related Health Problems, Tenth Revision [ICD-10] code E11.X).Patients had to be over age 18 years, have a primary care practitioner within the health system to ensure adequate capture of comorbidities, and at least 1 prescription for an antihyperglycemic medication at baseline, to exclude patients seen in endocrinology for another condition who also have T2D controlled without medication.Patients with type 1 diabetes, gestational diabetes, end-stage kidney disease, and dementia were excluded.Details on cohort creation and inclusion and exclusion criteria can be found in the cohort flowchart in Figure 1.

Measures
All data were extracted from electronic medical records.Telemedicine encounters were defined as completed synchronous visits with an endocrinology practitioner including audio-visual or audioonly communication.Encounters designated as phone calls, which are used to provide unscheduled support between visits, were excluded.Baseline factors including patient demographics and clinical characteristics were extracted as close to the initial qualifying visit date as possible starting May 1, 2020.The first HbA 1c value recorded in the study period was defined as the baseline value; follow-up HbA 1c values were collected through May 1, 2022, and had to be at least 10 weeks apart.Body mass index (BMI) was categorized into standard levels as shown in Table 1.Comorbidities of interest included cardiovascular disease and psychological conditions documented at least twice in outpatient encounters during the study period (see eTable 1 in Supplement 1 for ICD-10 codes).These were chosen to include 1 category of concordant comorbidities, which have management strategies similar to T2D (eg, coronary artery disease), and 1 category of discordant comorbidities with management unrelated to T2D (eg, bipolar disorder) to assess whether these had distinct associations with glycemic outcomes. 25,26Insulin prescription was based on active medication orders at baseline and categorized as no insulin, basal insulin only, or multiple daily injections (MDI; ie, both basal and prandial insulin).Social Deprivation Index (SDI), a composite measure of local area deprivation linked to health outcomes, was based on 5-digit zip code. 27,28[31][32] Race was extracted from the electronic medical record, based on patient self-report on clinical intake forms, and was included in this study to evaluate racial and ethnic variation in patterns of telemedicine use because race and ethnicity have previously been associated with factors, such as health insurance and physical environment, that affect both patterns of health care use and glycemic outcomes.Patients were separated into 3 categories: telemedicine only, in which all visits in the study period were conducted via telemedicine; in-person only, in which all visits after the initial telemedicine visit were in-person; or mixed follow-up, in which patients had both telemedicine and in-person visits following the initial visit in the study period.

Statistical Analysis
We summarized baseline patient characteristics using frequency (percentage) for categorical variables and mean (SD) or median (IQR) for continuous variables.Baseline characteristics were compared among those who used telemedicine only, in-person only follow-up, and mixed follow-up using χ 2 tests for categorical variables and Kruskal-Wallis tests for continuous variables.To examine the outcomes of excluding patients without follow-up HbA 1c , we also compared characteristics of excluded patients with the modeled cohort.Each demographic variable that differed between the modeled cohort and excluded patients was included as a covariate in final model.
Our primary outcome was 12-month HbA 1c change, with a secondary outcome of HbA 1c change at 24 months to explore longer-term glycemic outcomes.We used a linear mixed model fitted via maximum likelihood estimation to assess HbA 1c change by follow-up care modality and clinical factors.Random effects for patient and practitioner were included, with the endocrinology practitioner of the initial encounter serving as the unit of the random practitioner effect.Variables of

Characteristics of Patients by Care Modality
There were 3778 patients in the final cohort, with a mean (SD) age of 60.3 (12.7) years, 58% female (2201 participants), 2% Asian (81 participants), 8% Black (300 participants), and 88% White (3332 participants) (Table 1).The plurality of patients (1547 participants [41%]) used mixed modalities after the initial telemedicine visit, with similar proportions of patients (1182 participants [31%]) using telemedicine only and in-person only (1049 participants [28%]) over the study period.Patients who used telemedicine only were younger and more likely to be women and Black than patients in the in-person and mixed follow-up groups.In addition, patients who used telemedicine only had less local area deprivation (SDI) and were more likely to be urban dwelling.The in-person follow-up cohort had lower baseline HbA 1c compared with the other cohorts.Patients in the telemedicine-only cohort were more likely to have a psychological comorbidity and not be prescribed insulin at baseline than patients in the other 2 cohorts.
Comparison of patients included in final models with those excluded due to lack of follow-up HbA 1c demonstrated significant differences in demographics (eTable 2 in Supplement 1).Excluded patients were more likely to be younger, women, Black, urban dwelling, not prescribed insulin, have lower baseline HbA 1c , and have fewer visits per year than included patients.

Patterns of Care Utilization
The proportion of endocrinology visits for T2D that were conducted via telemedicine was highest from May to October 2020 at 84% (10 987 of 13 031 visits), dropped to 63% from November 2020 rate of appointments for in-person and mixed follow-up groups with telemedicine) (Table 1).Patients who used telemedicine only also had fewer follow-up HbA 1c measurements per 12 months than those who used in-person and mixed follow-up.

Glycemic Outcomes by Care Modality
There was no significant change in HbA 1c from baseline to 12 months in the telemedicine-only group (−0.06%; 95% CI, −0.26 to 0.14), while the in-person and mixed follow-up groups demonstrated HbA 1c improvement (estimated change for in-person group, −0.37; 95% CI, −0.59 to −0.15; estimated change for mixed group, −0.22; 95% CI, −0.38 to −0.07) (Table 2).There was no significant change in the secondary outcome of HbA 1c change from baseline to 24 months across any care modality (Table 2).Model-derived trajectories of HbA 1c over time differed within care modalities based on insulin regimen (Figure 2).In the telemedicine group, there was minimal estimated HbA 1c change over time among patients not on insulin and those on MDI, while HbA 1c increased steadily from baseline to 24 months for patients on basal insulin.In contrast, among the in-person and mixed follow-up cohorts, adjusted HbA 1c declined from baseline to 12 months with subsequent increase from 12 to 24 months for patients in all 3 insulin groups.

Glycemic Outcomes by Clinical Complexity
In all 3 cohorts, patients prescribed basal insulin had worse adjusted HbA 1c changes at 12 months compared with those not prescribed insulin (Table 3).Similar trends were seen at 24 months, but the estimated difference in HbA 1c change between patients prescribed basal insulin and those not prescribed insulin was significant only for the telemedicine group.For patients prescribed MDI compared with those not prescribed insulin, estimated HbA 1c change was worse at both 12 and 24 months across all modalities (Table 3).In difference-in-difference analysis, HbA 1c change from baseline to 12 months for patients prescribed MDI vs no insulin was significantly worse in the telemedicine group than the in-person group (estimated difference in HbA 1c change, 0.25% higher; 95% CI, 0.02 to 0.47; P = .03)but was not significantly different between the mixed and telemedicine groups (estimated difference in HbA 1c change, 0.15% higher for telemedicine only; 95% CI, −0.36 to 0.05; P = .05).Similarly, outcomes of telemedicine vs in-person care among patient cohorts defined according to insulin use demonstrate significantly worse HbA 1c changes at 12 months for telemedicine vs in-person care among patients on MDI (estimated difference in HbA 1c change, −0.47%; 95% CI, −0.78 to −0.15) for in-person vs telemedicine only (eTable 3 in Supplement 1).There was no significant association of cardiovascular or psychological comorbidities with HbA 1c changes at 12 or 24 months in any care modality cohort (Table 3).

Subgroup Analysis: Baseline HbA 1c of 8% or Higher
Among patients whose baseline HbA 1c was 8% or higher, those who used telemedicine only had no significant change in adjusted HbA 1c at 12 or 24 months (Table 2).However, patients who used in-person or mixed follow-up had significant improvement in adjusted HbA 1c at both 12 and 24 months, mirroring patterns observed for the overall cohort.Among patients who used telemedicine or mixed follow-up, those prescribed MDI had a worse estimated change in HbA 1c at 12 months compared with those not prescribed insulin; there was no significant difference among patients who used in-person follow-up only (Table 3).The estimated difference in HbA 1c change at 24 months for patients prescribed either basal or MDI compared with no insulin was not significant across any care modality in this subgroup.As in the overall cohort, cardiovascular and psychological comorbidities were not significantly associated with glycemic outcomes in any care modality.

Discussion
In this retrospective study of adults who received endocrinology care for T2D in a large health system from 2020 to 2022, patients accessing care through telemedicine alone had worse glycemic outcomes compared with patients who transitioned to in-person or mixed care.These findings build on and contrast with prior studies conducted in the primary care setting, which demonstrated similar glycemic outcomes of telemedicine and in-person care for T2D. 10,11,13,33Patients with T2D who receive endocrinology care and have more complex care needs, including those who use insulin or have HbA 1c above goal, may not be well served by telemedicine care alone as currently implemented.
Telemedicine emerged as a prominent modality of diabetes care delivery during the COVID-19 pandemic, but utilization patterns have changed over time.Our findings on patient subgroups who  rely on telemedicine to access specialty diabetes care are consistent with prior work and identify new characteristics associated with ongoing telemedicine use.We found that younger, [34][35][36][37][38] female, 10,[35][36][37][38] and urban-dwelling [39][40][41] patients were more likely to use telemedicine only, similar to previous data in primary care and endocrinology settings.[38]42,43 We found that patients with less complex diabetes were more likely to use telemedicine only.In addition, our findings add new evidence that telemedicine may be particularly important for ensuring access to endocrinology care for patients who have psychological comorbidities, which are known to impact diabetes self-management and outcomes, and may require additional support to achieve treatment goals. 44,4512][13]33 There are several potential explanations for the observed lack of HbA 1c improvement among patients using telemedicine alone to access endocrinology care in our study.First, patient-level factors that may lead to preferential use of telemedicine care can also impact diabetes self-management and health care access in general.The telemedicine group was more urban dwelling, younger, more likely to be Black and female, and may face unmeasured competing demands to diabetes self-management, such as caregiving responsibilities, transportation barriers, or work schedules. 46Telemedicine-only users also had lower care utilization, including less frequent appointments and HbA 1c testing, which may lead to more clinical inertia 47 and less intensification of treatment by endocrinology practitioners.Although it is not clear whether lower care utilization was driven by patients, practitioners, or systemic barriers, prior studies of diabetes telemedicine have found similar results. 13,48,49Additionally, differences in patientpractitioner communication via telemedicine, including difficulty building rapport, 50,51 may lead to differences in both patient self-management and practitioner treatment decisions.
Another potential explanation for inferior glycemic outcomes in the telemedicine group is that strategies to support glycemic improvement that are available during in-person appointments have not consistently been translated to telemedicine care.Care elements which may be particularly influential for patients with elevated HbA 1c or complex insulin regimens, such as self-management education and support, sharing of home blood glucose data through device downloads or written logs, and educational resources for initiation of diabetes technology or new medications, may not currently be routinely delivered through telemedicine or available at the point-of-care during telemedicine visits.In our prior work in this care setting, practitioners described how inferior availability of glucose data limited their ability to intensify treatment through telemedicine. 19plementation of approaches to overcome these differences, such as team-based virtual care and technological tools to automate blood glucose data sharing, are needed to ensure all patients receive high-quality diabetes care regardless of care modality. 52

Strengths and Limitations
There are a number of strengths and limitations to this study.This is the first study, to our knowledge, to examine outcomes of telemedicine care specifically in the endocrinology setting and according to clinical factors that impact treatment complexity.Although demographic variables that differed between groups were included as covariates, cohorts were not balanced on potentially confounding baseline characteristics.Factors including treatment complexity and glycemic control, as well as geographic and transportation barriers, may have impacted whether patients received care via telemedicine or in-person care; thus, findings reflect glycemic outcomes for clinical patients who received care via each modality, and do not indicate causal associations.This study was performed in a single health system, and patients had to use at least 1 medication for diabetes and were predominantly white and urban; thus, results may not generalize to other settings with different infrastructure for telemedicine care, rural areas, or patient populations with more racial and ethnic diversity or who have diet-controlled diabetes.In addition, HbA 1c values were not consistently captured for patients who had testing done at facilities that do not communicate with the electronic medical records.However, demographic variables which differed between included patients and those excluded due to lack of follow-up HbA 1c value (11.6% of eligible patients) were controlled for in models, limiting the impact of this issue on our results.Finally, with loss to follow-up over time, there were fewer patients who had HbA 1c results at 24 months compared with 12 months; thus, these results should be interpreted as exploratory only.

Conclusions
In this cohort study of adults who received endocrinology specialist care for T2D in a large health system from 2020 to 2022, patients who accessed care through telemedicine alone had worse glycemic outcomes compared with patients who transitioned to in-person or mixed care.Since some patients with barriers to in-person endocrinology care will continue to rely on telemedicine to access

JAMA Network Open | Diabetes and Endocrinology
Patterns of Telemedicine Use and Glycemic Outcomes of Endocrinology Care for Type 2 Diabetes care, structured approaches to ensure routine delivery of high-quality team-based diabetes care are needed.Translation of successful strategies from clinical trials into routine telemedicine care, especially targeted toward adults with more complex diabetes, is critical to improve clinical outcomes for patients who rely on this care modality.

Figure 2 .
Figure 2. Adjusted Hemoglobin A 1c (HbA 1c ) Levels by Follow-Up Care Modality and Baseline Insulin Regimen Adjusted 12-month HbA 1c changes are model-based estimates derived from linear mixed modeling of repeated measures of HbA 1c adjusted for patient age, sex, race, ethnicity, social deprivation index, rurality, baseline HbA 1c , and body mass index; patients were nested within practitioners.To convert to proportion of total hemoglobin, multiply by 0.01.

JAMA Network Open | Diabetes and Endocrinology Patterns
of Telemedicine Use and Glycemic Outcomes of Endocrinology Care for Type 2 Diabetes 1, 2022.A diagnosis of T2D was based on encounter diagnosis codes (International Statistical Classification of Diseases and

Table 1 .
Baseline Characteristics of Patient Cohort interest were treated as fixed effects and included follow-up care modality, insulin treatment regimen, and composite binary indicators for presence of cardiovascular disease or psychological conditions.To allow HbA 1c response to vary over the follow-up period, we included a quadratic function of time, defined as number of months since baseline HbA 1c .To capture between-patient variability in trajectories of HbA 1c response, a random slope for time was also incorporated.We adjusted for patient age, sex, race, ethnicity, SDI, RUCA category (urban, suburban, rural), baseline HbA 1c , and BMI (with patients with missing BMI included in a discrete level) to control for potential confounding.To quantify HbA 1c change over time within follow-up care modalities and test whether HbA 1c change differed significantly between modalities, we included 2-way interactions between follow-up care modality and fixed effects for time.Least-squares means (LS means) estimation quantified HbA 1c change over time within follow-up modalities, and contrasts of LS means estimated differences in HbA 1c change over time between modalities.We conducted 2 additional analyses.First, a subgroup analysis limited to patients with baseline HbA 1c 8% or higher was performed to further explore outcomes specifically for patients with elevated HbA 1c .Second, difference-in-difference analyses contrasted LS means estimates of HbA 1c change for patients with insulin use and comorbidities vs those without between care modalities and between patients who used telemedicine vs mixed or in-person care between levels of insulin use.Two-and 3-way interactions between time, care modality, and insulin regimen, as well as interactions between time, care modality, and comorbidities, were included in the mixed models.All analyses assumed a type 1 error rate of .05 and were performed using SAS software, version 9.4 (SAS Institute Inc).Data were analyzed from June 2022 to October 2023.
b Other includes Indigenous American, Alaska Native, Samoan, Other Pacific Islanders, and those who indicated their race was unknown, not specified, declined to answer, or for whom race data was missing.cBody mass index is calculated as weight in kilograms divided by height in meters squared.main Patterns of Telemedicine Use and Glycemic Outcomes of Endocrinology Care for Type 2 Diabetes telemedicine-only group had fewer mean (SD) appointments per year (2.1 [0.8] appointments per 12 months) than those in the in-person and mixed follow-up groups (2.5 [1.1] appointments per 12 months and 2.9 [0.9] appointments per 12 months, respectively; rate ratio of appointments per 12 months comparing telemedicine to in-person, 0.803; 95% CI, 0.771-0.836;rateratio comparing telemedicine to mixed follow-up groups, 0.699; 95% CI, 0.674-0.725;P< .001forcomparison ofJAMA Network Open | Diabetes and Endocrinology

Table 2 .
Adjusted HbA 1c Change From Baseline for Each Follow-Up Care Modality a Adjusted 12-month HbA 1c changes are model-based estimates derived from linear mixed modeling of repeated measures of HbA 1c adjusted for patient age, sex, race, ethnicity, social deprivation index, rurality, baseline HbA 1c , and body mass index; patients were nested within practitioners.

Table 3 .
Adjusted Difference in HbA 1c Change for Those With Insulin Use or Comorbidities vs Without for Each Follow-Up Care Modality