The COVID-19 pandemic led to a rapid uptake in telehealth use beginning in March 20201–3 to provide safer care to patients and reduce their exposure to the virus. While telehealth was lauded as a flexible and safe means of maintaining health care access for patients during the pandemic, care continuity’s effect on telehealth use and telehealth’s impact on quality of care is not well understood. Vulnerable populations, such as low-income patients, minoritized patients, and patients with complex care needs, patient populations often served by community health centers (CHCs), may not have experienced the same improvements as other more advantaged populations.4 Care continuity has previously been associated with improved quality of care and patient care experiences.5,6 While there is some evidence that telehealth supported care continuity during the pandemic,7 it is unclear how telehealth use impacted the relationship between care continuity and quality of care.4

Care continuity or the extent to which patient care is dispersed or concentrated among clinicians,8,9 is a key factor in providing evidence-based care to adults with diabetes and/or hypertension.10–16 Care continuity has previously been linked directly to trust in clinicians5,17–21—a key determinant of high quality patient care experiences.17,20,22,23 The link between care continuity and trust in clinicians may have important implications for telehealth adoption. Patients with limited English proficiency and trusted care team members to help with interpretation and/or deliver care can improve patient-clinician communication during a telemedicine encounter.24 In general, high levels of trust in clinicians can be especially useful in the face of external shocks when new innovations into care delivery need to be introduced.25 Limited research has been conducted on care continuity for patients with chronic conditions in CHCs. CHCs face rigorous continuous improvement expectations, and data collection and monitoring that may impact care continuity.21 CHCs also faced greater workforce loss during the pandemic compared with other health care organizations, which may have negatively impacted patient-clinician relationships for CHC patients.26

As a result of shelter-in-place ordinances, adults with diabetes and/or hypertension were vulnerable because their routine care involves close monitoring and medication management. These patients are not only likely to be at higher risk of COVID-19-related complications,27 but are at risk for exacerbations due to reduced access to, and utilization of, care.28 We analyze data from before and during COVID-19 pandemic to assess the relationship between care continuity, telehealth use, and quality of care for patients with type II diabetes and hypertension. To our knowledge, this is the first study to examine the association of care continuity and telehealth use among adult CHC patients with chronic conditions.

Recent evidence about the association of telehealth use and diabetes and hypertension care indicates telehealth helped alleviate disruptions and decreases in quality of care during the COVID-19 pandemic, although there were disparities in age, race, and income in likelihood to utilize telehealth.29–31 Based on these findings and evidence about the impact of care continuity on quality of care,11 we hypothesize that care during the COVID-19 pandemic will be associated with lower continuity of care (Hypothesis 1) due the disruption of clinician-patient relationships in CHCs and that telehealth use will be positively associated with patients with greater care continuity during the pandemic (Hypothesis 2). We also hypothesize that care continuity will be positively associated with processes and outcomes of diabetes and hypertension care during the pandemic (Hypothesis 3), and that the association between care continuity and process measures of quality will be mediated by telehealth use (Hypothesis 4).

METHODS Data Source

We analyzed 2019 and 2020 data from California CHC members of the Oregon Community Health Information Network (OCHIN) Accelerating Data Value Across a National Community Health Center Network (ADVANCE) Collaborative.32 The goal of ADVANCE is to create a data network of CHCs to inform and disseminate research targeted at improving access, engagement, equity, and quality of care for patients of CHCs.32

Sample

The study population are CHC clinicians and adult patients from a cohort of patients with diabetes and/or hypertension (n=20,792) with ≥2 encounters/year from 2019 (March-December 2019) to 2020 (March-December 2020) among 166 California CHC sites in the OCHIN ADVANCE Collaborative’s electronic health record (EHR) data.32 We restricted the sample to adults with at least 2 encounters during each year of the study because the assessment of continuity requires multiple encounters.

Outcomes

For Hypothesis 1, the outcome measure is care continuity by year (pre-COVID-19 vs. during COVID 19) by the Modified Modified Continuity Index (MMCI), a measure of care dispersion, calculated using Equation (1):

(1)MMCI=1−kN+0.11−1N+0.1,

where k=number of clinicians seen in a period and N=total number of encounters to all clinicians in a period. MMCI is an established measure of care continuity used commonly in published studies of care continuity33–39; scores range from 0 to 1, where 1 is perfect continuity with all encounters to a singular provider and 0 is all encounters to different clinicians.

For Hypothesis 2, the outcome measure is telehealth use, defined as at least 1 telehealth encounter by a patient in each year as established through data collected from the EHR.

For Hypothesis 3, processes of care are measured by annual blood pressure and/or A1c testing. The relationship between MMCI and the final annual systolic and diastolic blood pressure level and A1c value of patients is examined to analyze if processes of care translate to improved intermediate outcomes of care. For Hypothesis 4, we conducted mediation analysis to examine the proportion of the relationship that is mediated by telehealth use between MMCI and the outcomes that have a significant relationship with both MMCI and telehealth use.

Main Independent Variables

For Hypothesis 1, the main independent variable is the year (2019 vs. 2020) of encounter. For Hypothesis 2, 3, and 4, the main independent variable is care continuity, measured by MMCI (range: 0–1).

Control Variables

Regression models controlled for patients’ sociodemographic characteristics, health status, encounters, and clinician types seen by patients during each year, categorized based on past research,40 which include single physician only, physician and nurse practitioner/physician’s assistant or registered nurse/medical assistant, combination of physician and a nurse practitioner/physician’s assistant and registered nurse/medical assistant, 2 different physicians, and 3 or more unique physicians. Comorbidities were determined from the EHR problem list and included body mass index (BMI), congestive heart failure, cardiovascular disease, coronary heart disease, depression, anxiety/posttraumatic stress disorder, general presence of a mental health condition, diabetic retinopathy, substance abuse, alcohol abuse, tobacco use, mobility impairments.41 The Charlson Comorbidity Index (range: 1–12), a validated, weighted index of comorbidities that considers the number and severity of each condition resulting in an integer starting from zero that represents risk of mortality, was constructed and also included as a control variable.

Statistical Analysis

First, a paired t test compared average levels of care continuity (MMCI) across periods defined as before (2019) and during (2020) the COVID-19 pandemic. Next, logistic regression models estimated the association of care continuity (MMCI) with: (1) telehealth use and (2) processes of care (blood pressure, hemoglobin A1c testing), net of control variables.33 Generalized linear regression models estimated the association of MMCI and intermediate outcomes (blood pressure, A1c control). Robust SEs accounted for patients clustering within CHC sites. Models were estimated separately for 2019 and 2020. The regression model of telehealth adoption by care continuity is presented in Equation (2):

(2)Y=β0+β1period+β2MMCI+β3careteamcomposition+β4numberofvisits+β5patientcharacteristics+ε,

Where β0 is an intercept term, β1 is a term indicating the period of the analysis as described above and β1=0 for analysis in the preperiod (2019) and β1=1 for analysis in the during-period (2020), β2 is the coefficient of care continuity, β3 is the coefficient for the clinician types seen in each period, β4 is the coefficient for the control variable for the number of encounters a patient had in a given period, β5 is the coefficient for control variables related to patient characteristics, which include Charlson score, income measured by percentage of federal poverty line, BMI, sex, and age and ε is an error term.

Equation (3) exhibits the regression model for pre-COVID-19 and during COVID-19 analysis of the association of care continuity and diabetes/hypertension management and the mediating impact of telehealth use is:

(3)Y=β0+β1period+β2MMCI+β3cliniciantype+β4THuse+β5patientcharacteristics+β62019baselinevalue+ε,

Where β0 is an intercept term, β1 is a term indicating the period of the analysis and β1=0 for analysis in the preperiod and β1=1 for analysis in the during-period, β2 is the coefficient of care continuity, β3 is the coefficient for the control variable that controls for the different configurations of clinician types that a patient saw in each period, β4 is the coffiecient indicating telehealth use in a period, β4=0 for no telehealth use, and β4=1 for patients with at least 1 telehealth encounter in the period, β5 is the coefficient for control variables related to patient characteristics, which include the Charlson score, income measured by percentage of federal poverty line, BMI, sex, and age, β6 is the coefficient for 2019 baseline values of intermediate outcomes in 2020 regressions, and ε is an error term.

We conducted a formal mediation analysis42 to examine telehealth as a mediator, or variables that explains the relationship between care continuity and quality of care. We only examined the association of MMCI with the A1c testing during the pandemic, as it was the only significant association between telehealth use and study outcomes found in adjusted analyses. We chose this approach to enable estimation of effects described by nonlinear relationships. “PARAMED” package in STATA was used,43,44 which uses parametric regression models to estimate causal mediation effects. Percent mediation is then calculated by Equation (4) using natural indirect and direct effects45,46:

(4)(IndirecteffectDirecteffect+indirecteffect)×100%.

Mediation analyses were conducted for 2020 study outcomes for which the relationships between MMCI and the study outcome were statistically significant in adjusted analyses. Mediation analysis was not conducted for 2019 due to low uptake of telehealth during the period. All statistical analyses were performed using Stata 17.0.44

RESULTS

The analytic sample is predominantly female (58.1%) and identified as Hispanic/Latinx (52.7%). A plurality (43.5%) of the population preferred Spanish as their spoken language, 52.7% English, and 3.3% another language (Table 1). Homelessness (0.45%) were a small minority, and most patients had an assigned primary care physician (98.8%). Most (58.66%) of the sample was diagnosed with type II diabetes and 85.22% had hypertension. The average Charlson comorbidity score was 3.19 (SD=1.63).

TABLE 1 - Patient Demographics and Clinical Characteristics of the Analytic Sample, by Telehealth Exposure Patient demographics n (%) Telehealth [n (%)] No telehealth [n (%)] P No. patients† 20,792 (100) 4251 (20.45) 16,541 (79.55) Sex 0.021*  Female 12,069 (58.05) 2534 (59.61) 9535 (57.64)  Male 8723 (41.95) 1717 (40.39) 7006 (42.36) Race/ethnicity 0.08  Hispanic/Latino 10,955 (52.69) 2258 (53.12) 8697 (52.58)  White 6021 (28.96) 1191 (28.02) 4830 (29.20)  Asian 1374 (6.61) 288 (6.77) 1086 (6.57)  Black or African American 1345 (6.47) 290 (6.82) 1055 (6.38)  Native Hawaiian or Other Pacific Islander 128 (0.66) 20 (0.47) 108 (0.65)  American Indian or Alaskan Native 90 (0.43) 15 (0.35) 75 (0.45)  Multiple races 73 (0.35) 15 (0.35) 58 (0.35)  Unknown 806 (3.88) 174 (4.09) 632 (3.82) Patient preferred spoken language 0.132  English 10,960 (52.71) 2191 (51.54) 8769 (53.01)  Spanish 9139 (43.95) 1904 (44.79) 7235 (43.74)  Other 693 (3.33) 156 (3.67) 537 (3.25) Marital status <0.001*  Single 5034 (24.21) 864 (20.32) 4170 (25.21)  Married/domestic partner 4934 (23.73) 887 (20.87) 4047 (24.47)  Significant other 352 (1.69) 64 (1.51) 288 (1.74)  Separated/divorced 1172 (5.64) 202 (4.75) 970 (5.86)  Widowed 855 (4.11) 163 (3.83) 692 (4.18)  Unknown 8445 (40.62) 2071 (48.72) 6374 (38.53) Homelessness status 0.072  Yes 94 (0.45) 23 (0.54) 71 (0.43) Insurance <0.001*  Private 1816 (8.73) 376 (8.84) 1440 (8.71)  Medicaid 7374 (35.47) 1525 (35.87) 5849 (35.36)  Medicare 6562 (31.56) 1375 (32.35) 5187 (31.36)  Other public 1734 (8.34) 248 (5.83) 1486 (8.98)  Uninsured 3306 (15.90) 727 (17.10) 2579 (15.59) Assigned primary care physician 0.043*  Yes 20,541 (98.79) 4201 (98.82) 16,340 (98.78)  No 251 (1.21) 50 (1.18) 201 (1.22) Comorbidities  Type II diabetes 12,197 (58.66) 2887 (67.91) 9310 (56.28) <0.001*  Hypertension 17,718 (85.22) 3567 (83.91) 14,151 (85.55) 0.007*  Congestive heart failure 1000 (4.81) 231 (5.43) 769 (4.65) 0.033*  Cardiovascular disease 1297 (6.24) 308 (7.25) 989 (5.98) 0.002*  Congenital heart disease 1885 (9.07) 452 (10.63) 1433 (8.66) <0.001*  Diabetic retinopathy 1295 (6.23) 336 (7.90) 959 (5.80) <0.001*  Secondary diabetes 1011 (4.86) 273 (6.42) 738 (4.46) <0.001*  Mobility impairment 239 (1.15) 49 (1.15) 190 (1.15) 0.983  Substance abuse 2031 (9.77) 404 (9.50) 1627 (9.84) 0.515  Alcohol use 1068 (5.14) 202 (4.75) 866 (5.24) 0.203  Tobacco use 1836 (8.83) 361 (8.49) 1475 (8.92) 0.384  Depression 5251 (25.25) 1169 (27.50) 4082 (24.68) <0.001*  Anxiety/posttraumatic stress disorder 4050 (19.48) 854 (20.09) 3196 (19.32) 0.260  Other mental health condition 3199 (15.39) 687 (16.16) 2512 (15.19) 0.116 Age [mean (SD)] 57.8 (11.9) 57.8 (11.8) 57.8 (11.9) 0.703 Charlson score [mean (SD)] 3.19 (1.63) 3.46 (1.65) 3.12 (1.62) <0.001* Body mass index [mean (SD)] 31.76 (7.45) 32.16 (7.44) 31.66 (7.44) <0.001*

†Percentages displayed for “number of patients” are row percentages. All other percentages presented above reflect column percentages.

*Significant at P<0.05 level.

Overall, encounters declined during the COVID-19 pandemic with 263,633 encounters in 2019 and 103,634 in 2020. The types of clinicians that patients had encounters with changed from 2019 to 2020, with a larger proportion of patients seeing a single physician (2019: 21.7%, 2020: 25.4%), a physician and a nurse practitioner/physician’s assistant or registered nurse/medical assistant (2019: 7.0%, 2020: 8.7%), a combination of a physician and a nurse practitioner/physician’s assistant and registered nurse/medical assistant (2019: 4.7%, 2020: 5.7%), and 2 different physicians (2019: 31.4%, 2020: 39.5%) in 2020 compared with 2019. There was a reduction in the proportion of patients seeing 3 or more unique physicians in 2020 (2019: 35.3%, 2020: 20.8%).

COVID-19 Pandemic Impact on Care Continuity

Supporting Hypothesis 1, patients experienced reduced continuity of care in 2020 (MMCI=0.63, SD=0.36) compared with 2019 (MMCI=0.71, SD=0.28, P<0.001). Almost all patients (2019: 99.99% vs. 2020: 99.75%) had their blood pressure screened annually but only 69.78% versus 63.32% of adults with diabetes had their A1c tested in 2019 versus 2020.

Telehealth Use and Care Continuity During the Pandemic

Telehealth accounted for 0.33% of encounters in 2019 and increased to 9.55% in 2020 (Fig. 1). In our sample, 14.1% of clinicians used telehealth to provide care in 2020 out of n=16,597 clinicians represented in our analytic sample (data not shown). In adjusted analyses, higher MMCI scores were associated with higher odds of telehealth use in 2020 [odds ratio (OR)=1.94, marginal effect=0.20, z=70.78, P<0.001], but not 2019, which partially supports Hypothesis 2 (Table 2). Contrary to expectations, an inverse relationship was found between 2019 MMCI scores and telehealth use in 2020 (OR=0.82, marginal effect=0.20, z=70.73, P=0.003; Table 2).

FIGURE 1:

Proportion of encounters conducted via telehealth by year.

TABLE 2 - Multivariable Regression Analyses: The Association of Care Continuity and Telehealth With Quality of Hypertension and Diabetes Care Telehealth use (2019, n=19,385; 2020, n=19,385) Hemoglobin A1c process (2019, n=11,373; 2020, n=11,373) Hemoglobin A1c value (2019, n=3384; 2020, n=3625) Systolic blood pressure (2019, n=19,360; 2020, n=17,738) Diastolic blood pressure (2019, n=19,360; 2020, n=17,738) OR 95% CI OR 95% CI Coefficient 95% CI Coefficient 95% CI Coefficient 95% CI 2019  Care continuity†‡ 0.84 0.6–1.06 1.98*** 1.55–2.53 −0.57** −0.99 to −0.15 −1.58 −3.17 to −0.01 −0.67 −1.57 to 0.23  Telehealth† — — 2.72*** 1.99–3.72 −0.15 −0.54 to 0.24 −0.44 −2.31 to 1.42 −0.88 −1.88 to 0.11 2020  Care continuity†‡ 1.94*** 1.55–2.43 1.50*** 1.23–1.82 −0.45** −0.78 to −0.12 −2.90*** −4.16 to −1.63 −1.44*** −2.16 to −0.72  Telehealth† — — 10.00*** 7.30–13.70 0.03 −0.24 to 0.30 −0.08 −2.06 to 1.89 0.84 −0.27 to 1.94

†Regression results control for the types of clinicians seen by the patient, the number of visits by the patient in each year, age, sex, annual income as a percentage of federal poverty line, body mass index, and the Charlson Comorbidity Index. The 2020 regressions also control for 2019 baseline values of their respective dependent variables. All regression models were estimated using robust SEs.

‡Care continuity is measured using the Modified Modified Continuity Index (MMCI).

**P<0.01.

***P<0.001.

Care Continuity, Telehealth Use, Monitoring, and Health Outcomes

Care continuity (MMCI 2019: OR=1.98, marginal effect=0.69, z=165.50, P<0.001; 2020: OR=1.50, marginal effect=0.63, z=147.73, P<0.001) and telehealth use (2019: OR=1.50, marginal effect=0.85, z=122.87, P<0.001; 2020: OR=10.00, marginal effect=0.90, z=155.57, P<0.001) were significantly associated with more consistent A1c testing in both periods (Table 2), supporting the first part of Hypothesis 3. Contrary to the second part of Hypothesis 3, MMCI, but not telehealth use, was significantly associated with lower A1c values in 2019 (β=−0.57, P=0.007) and 2020 (β=−0.45, P=0.008). Higher care continuity (MMCI) was associated with lower systolic blood pressure (β=−2.90, P<0.001) and diastolic blood pressure values (β=−1.44, P<0.001) in 2020.

Mediating Role of Telehealth in the Care Continuity and Quality Relationship

The mediation analyses found the pathway between care continuity and telehealth, telehealth and A1c testing, and care continuity and A1c testing to all be statistically significant (Fig. 2). Care continuity partially mediated the care continuity and A1c testing relationship in 2020 based on the 4 steps used to assess mediation effects.47 In 2020, 38.7% of the relationship between MMCI and A1c testing was mediated by telehealth use (direct effect: β=1.76; 95% CI: 1.45–2.12 indirect effect: β=1.11; 95% CI: 1.05–1.12), but telehealth use did not mediate the association of care continuity and other study outcomes, offering only partial support for Hypothesis 4.

FIGURE 2:

Results of causal mediation analysis with bootstrap SEs of the role of telehealth in the relationship between continuity of care and hemoglobin A1c testing.

DISCUSSION

Our analyses of care continuity, telehealth use, and quality of care among adults with diabetes and/or hypertension in CHCs before and during the pandemic reveals that care continuity and telehealth use are associated with quality of care in complex ways. Consistent with Hypothesis 1, care continuity for adults with diabetes and/or hypertension receiving care in CHCs during the early COVID-19 pandemic (2020) declined compared with prepandemic period (2019). Higher care continuity in 2020 was associated with higher telehealth use and A1c testing, as well as lower A1c scores and lower blood pressure in accordance with Hypothesis 2. Although continuity of care is generally associated with better patient outcomes, the findings of this study are consistent with the mixed effects found in the literature analyzing the impact of care continuity on quality of