Before the COVID-19 crisis, the primary motivation for health care systems to invest in and implement virtual care as an alternative to in-person care was to reduce both patient and provider inconvenience and costs.1,2 From the patient perspective, virtual care is associated with lower direct costs (out-of-pocket costs from copayments and deductibles) and lower opportunity costs (travel time, clinic wait times, and missed work or school). Before 2020, estimates of the uptake of virtual or telehealth modes of care remained below 5% among ambulatory visits.3 In March 2020, the COVID-19 pandemic, along with the expansion of Centers for Medicare and Medicaid's and most private insurers' payment rules covering telehealth benefits4 accelerated a shift in ambulatory health care delivery from in-person to virtual care across the United States.5–8

Noted disparities in virtual care use in diverse insured populations who have access to in-person and virtual care, emerged during periods before the COVID-19 pandemic and following the emergency shutdown.9,10 However, policy makers and health care system stakeholders lack information on how the rapid, and likely sustained partial transition to virtual care has affected engagement with primary care providers by older patients and disadvantaged racial, ethnic, and other population groups.11 The literature on virtual care implementation before the COVID-19 pandemic suggests that when patients have a choice of virtual versus in-person care, younger and healthier patients, less racially and ethnically diverse patients, who live in areas of higher socioeconomic status, and have few technology barriers have a higher propensity to select virtual care; older and sicker patients have a higher propensity to persist in choosing in-person care.12,13 During the COVID-19 pandemic, vulnerable subgroups, including the elderly, those with high comorbidity burdens and those living in poor neighborhoods, were most likely to be adversely impacted by COVID-19. Results were mixed, however, with respect to racial, ethnic, and socioeconomic inequities in use or choice of virtual or telehealth services.14–19

Consistent with the framework described by Handel20 and Brot-Goldberg et al,21 when patients have a choice of health care alternatives such as virtual versus in-person care, adverse and favorable health risk selection among the alternatives is likely. Aggressive implementation of deductible and high deductible plans and other measures of patient cost-sharing—which has a known patient risk selection effect—may exacerbate selection effects.22,23 Patient risk selection between virtual and in-person visits suggests persistent users of in-person care could generate distortions in workflows, access to care, and inefficiency by increasing in-person visit appointment length and provider panel sizes.20

It is important for health care systems, provider groups, and payors to understand if high-risk subgroups of patients are more-or-less likely to engage in primary care, and how these same patient factors can thus impact use of virtual primary care services during the progression of the COVID-19 crisis. Understanding key patient demographic, clinical, and insurance factors associated with both engagement with, and use of, alternative primary care options of care is crucial to ensuring that vulnerable patient groups continue to receive high quality health care, especially as the transition in care continues to evolve.

In this study we evaluate patient characteristics associated with both engagement with primary care providers, and the use of in-person versus virtual care, over the first 16 months of the COVID-19 pandemic. Specifically, we conducted a 2-part assessment of patient characteristics associated with use of virtual only encounters versus in-person care only, versus a mix delivery modes, as a function of the probability of use of any form of adult primary care (APC), using data from 3 diverse, integrated health care systems from January 1, 2020, through June 30, 2021. Acknowledging that the COVID-19 pandemic altered both care seeking behaviors and APC availability, we hypothesize that insured individuals who are male, younger, healthier, and who face higher cost-sharing will be less likely to engage in APC during the pandemic, but if they do, they will be more likely to use or select (when choice was available) virtual venues of care. We also hypothesize that APC and virtual care use will vary by race and ethnicity. We posit that patient factors associated with engaging in any APC will also be associated with the use of in-person only visits, virtual only visits, or a mixture of virtual plus in-person.

METHODS Setting

The setting for this study is the integrated health care delivery systems of Kaiser Permanente regions of Colorado (KPCO), Georgia (KPGA), and Mid-Atlantic States (KPMAS). All 3 study sites transitioned to a "Virtual First" model of ambulatory care, consistent with CDC recommendations,24 shortly after the March 13, 2020 declaration of the COVID-19 outbreak related National Emergency.25 At that time, KPCO provided comprehensive medical services to 502,355 adult enrollees (60.7% White, 4.4 % Black, 15.1% Hispanic) in the Denver/Boulder area; KPGA provided comprehensive medical services to 328,181 adult enrollees (28.5% White, 41.4 % Black, 4% Hispanic) in the Atlanta area; and KPMAS provided comprehensive medical services to 801,986 adult enrollees (25.3% White, 33.2% Black, 12.1% Hispanic) in the Washington, DC tri-state area. Before the COVID-19 pandemic, all 3 sites had implemented virtual care in the forms of scheduled telephone (all 3 sites), scheduled video (KPGA and KPMAS), and synchronous chat (KPCO only), but 85% of all APC encounters were delivered in-person, at a Kaiser Permanente (KP) medical office. Given the variation in the population characteristics, the implementation and availability of Telehealth or virtual venues of care both before and after the beginning of the COVID-19 crisis [Fig. 1 and Supplemental Digital Content (SDC), https://links.lww.com/%CE%BBR/C549], along with variation in local availability of in-person elective and emergent health care services determined by local infection rates and "stay-at-home" orders, we stratified our analyses by site.

FIGURE 1:

Distribution of adult primary care visits: 2019 versus 2020–June 30, 2021—by site. In-person, phone, and video visits require scheduling either via electronic health record patient portal, or by contact through a call-center. Synchronous Chat is available on-demand via electronic health record patient portal. KPCO indicates Kaiser Permanente regions of Colorado; KPGA, Kaiser Permanente regions of Georgia; KPMAS, Kaiser Permanente regions of Mid-Atlantic States.

Study Design and Data Sources

This retrospective cohort study was conducted from January 1, 2020 through June 30, 2021. We developed identical standard common data models at each of the 3 sites based on harmonized data derived from administrative and electronic health records (EHRs). Variables include patient demographics, procedures, diagnoses, census tract–derived measures of socioeconomic status based on the Area Deprivation Index (ADI),26,27 pharmacy dispenses, health plan benefit data, completed APC visits, and visit type. We excluded procedure-only, non–virtual-analog encounter such as a blood pressure assessments, as well as visits that were specific to COVID-19 for testing, screening, and vaccination. The unit of analysis was person-month for each of the 18 months of the observation period. We restricted the data to individuals who were enrolled in a KP health plan from January 1, 2020, through June 30, 2021, who were age 19 to 89 years at time of the first of the month of interest and until the end of study, death, or disenrollment. The study was reviewed and approved by the Institutional Review Board (IRB) at KPGA, the IRB of record for the project.

Variables Dependent Variables

We first evaluated a dichotomous variable indicating an APC visit of any type, for each month. For individuals with 1 or more APC visits in any month, 3 additional dichotomous variables were created indicating 1 or more in-person only visits, 1 or more virtual only visits, or a mixture of virtual plus in-person visits. We did not stratify the analyses by type of virtual visit (eg, scheduled telephone, video, synchronous chat) given the low density of use of some of these types of visits during many of the months of interest. We did not limit visits to those with the individuals’ linked or assigned primary care provider,28 given that KP patients always have the option to select “first available” visit. We also did not capture how the visit was scheduled (either call center or patient portal),13 nor were we able to observe what visit venue was available during the specific month.

Independent Variables

The covariates age, ADI, distance to primary care clinic, deductible health plan, and outpatient copay were calculated as of the first of each eligible person-month. Sex, and self-reported race and ethnicity, ADI, and distance to clinic, were assumed not to vary over time. The census tract based-ADI score was calculated by quartile (1–4), with quartile 1 being least deprived.27 The Charlson Comorbidity Index (CCI)29–31 was calculated for each month looking back 1 year from the first of each month. Two measures of access to care, Distance to primary care clinic and prior mail-order–based pharmacy dispense (eg, a proxy for prior EHR use), as described in Roblin et al,32 were included as moderators in the analyses. Dichotomous flags for each month and each quarter of the analysis period were calculated from the visit date.

Statistical Analyses

All analyses were conducted separately for each site-specific dataset. Descriptive statistics using frequencies and proportions were used to describe the distribution of categorical variables stratified by no visit, in-person visits, virtual visits, and a mix of virtual and in-person, by month for the 18-month observation period. Differences in the distributions by visit type were compared using χ2 tests. For multivariate analyses, we employed a 2-stage model to evaluate factors associated with APC use by visit type (eg, in-person, virtual, mixed), conditional on the probably of using any APC in any month.

Adult Primary Care Use

In the first stage, generalized estimating equations with a logit distribution with robust SEs accounting for multiple observations per patient (from 1 to 18 depending on how many months a patient was enrolled) and independent covariates noted above were used to estimate adjusted odds ratios (ORs) and 95% CIs. Inverse probability of treatment weights (IPTW)33 were generated to address the probability that an individual had 1 or more APC visits in any person-month.

Use of Adult Primary Care by Visit Type

Using the IPTW generated from the first model, we estimated a multinomial model using a general estimating equation accounting for multiple observations per-person as well as a generalized logit-link function to account for the 3 categories of in-person only, virtual only, and mixed use for the cohort of patients with an APC visit in any month. In addition to the covariates noted above, and dichotomous variables associated with each quarter of the observation period (rather than month) were included to reflect different stages of the pandemic. Given the overlap in covariates used in both stages, sensitivity analyses using weighted and unweighted split samples were conducted to address the potential for overfitting. In addition, we also conducted a several sensitivity analyses to examine the impact of a variety of model and variable specifications. Analyses were performed using SAS, version 9.4M6 (SAS Institute Inc.).

RESULTS Study Populations

The total of number of person-months for KPCO, KPMAS, and KPGA was 7,055,549, 11,014,430, and 4,176,934, respectively (Table 1). Consistent with our 2019 evaluation that employed similar data and methods (SDC, https://links.lww.com/%CE%BBR/C549), >84% of the population did not engage in APC during any month, but this distribution of visit months for in-person only, virtual only, and virtual with in-person, were similar across all regions. Statistically significant differences (P<0.001) in the distribution of all categorical variables, by visit mode, were noted for all 3 sites, but the overall person-month characteristics differed across the 3 sites with respect to key variables. KPCO is majority White with the highest proportion of Hispanic person-months across the 3 sites. The majority of KPGA and KPMAS enrollees were historically minoritized34 with the largest proportion self-identifying as Black. While comorbidity profiles are similar across sites, out-of-pocket cost-sharing and mail-order use varied by site, with KPCO the highest in both.

TABLE 1 - Member-month Characteristics for Nonusers and Users of Adult Primary Care (All Visit Modes) 2020–June 2021—By Site n (%) n (%) n (%) Patient characteristics No visit KPCO In-person KPCO Virtual visit KPCO Virtual or in-person KPCO P KPCO No visit KPGA In-person KPGA Virtual visit KPGA Virtual or in-person KPGA P KPGA No visit KPMAS In-person KPMAS Virtual visit KPMAS Virtual or in-person KPMAS P KPMAS No. observations 6,127,630 (86.9) 464,994 (6.6) 349,204 (4.9) 113,721 (1.6) 3,539,328 (84.7) 319,705 (7.7) 243,592 (5.8) 74,309 (1.8) 9,448,363 (85.8) 740,093 (6.7) 652,459 (5.9) 173,425 (1.6) Age 19–34 1,495,426 (24.4) 74,397 (16.0) 82,711 (23.7) 23,966 (21.0) <0.001 990,160 (28.0) 52,998 (16.6) 46,483 (19.1) 12,396 (16.7) <0.001 2,731,054 (28.9) 156,538 (21.2) 139,983 (21.5) 29,952 (17.3) <0.001 Age 35–49 1,596,428 (26.1) 92,479 (19.9) 86,841 (24.9) 26,961 (23.7) 984,886 (27.8) 77,938 (24.4) 65,559 (26.9) 18,948 (25.5) 2,493,651 (26.4) 169,583 (22.9) 158,489 (24.3) 38,694 (22.3) Age 50–64 1,604,932 (26.2) 121,796 (26.2) 82,815 (23.7) 27,925 (24.6) 1,048,761 (29.6) 109,545 (34.3) 82,367 (33.8) 25,573 (34.4) 2,533,079 (26.8) 217,992 (29.5) 189,344 (29.0) 53,589 (30.9) Age 65 and older 1,430,844 (23.4) 176,322 (37.9) 96,837 (27.7) 34,869 (30.7) 515,521 (14.6) 79,224 (24.8) 49,183 (20.2) 17,392 (23.4) 1,690,579 (17.9) 195,980 (26.5) 164,643 (25.2) 51,190 (29.5) Female (yes) 3,229,193 (52.7) 266,828 (57.4) 227,444 (65.1) 73,122 (64.3) <0.001 1,931,010 (54.6) 190,381 (59.6) 158,473 (65.1) 48,148 (64.8) <0.001 5,043,237 (53.4) 427,139 (57.7) 401,579 (61.6) 105,819 (61.0) <0.001 White 3,920,230 (64.0) 310,964 (66.9) 237,115 (67.9) 75,208 (66.1) 1,125,539 (31.8) 105,119 (32.9) 78,483 (32.2) 23,000 (30.9) 2,593,797 (27.5) 189,567 (25.6) 182,623 (27.9) 42,813 (24.7) Black/African American 270,750 (4.4) 25,832 (5.6) 18,455 (5.3) 6841 (6.0) 1,559,390 (44.1) 161,952 (50.7) 126,606 (52.0) 39,918 (53.7) 3,260,526 (34.5) 313,892 (42.4) 277,062 (42.5) 78,467 (45.3) Asian 237,816 (3.9) 15,742 (3.4) 10,236 (2.9) 3123 (2.8) 191,096 (5.4) 16,123 (5.0) 10,048 (4.1) 3143 (4.2) 1,113,632 (11.8) 86,643 (11.7) 69,906 (10.7) 18,414 (10.6) Hawaiian/Pacific Islander 17,303 (0.3) 1321 (0.3) 860 (0.3) 324 (0.3) 2040 (0.1) 198 (0.1) 165 (0.1) 55 (0.1) 6483 (0.1) 386 (0.1) 368 (0.1) 82 (0.0) Native American 23,094 (0.4) 1869 (0.4) 1443 (0.4) 530 (0.5) 5486 (0.2) 578 (0.2) 433 (0.2) 135 (0.2) 14,187 (0.2) 1070 (0.1) 1006 (0.2) 256 (0.2) Multiple 46,638 (0.8) 3363 (0.7) 3128 (0.9) 975 (0.9) 18,268 (0.5) 1572 (0.5) 1306 (0.5) 367 (0.5) 309,394 (3.3) 29,037 (3.9) 23,702 (3.6) 6845 (3.9) Other 182,773 (2.9) 11,313 (2.4) 8550 (2.5) 2709 (2.4) 0 (0.0) 0 (0.0) 0 (0.0) 0 (0.0) 71,212 (0.8) 4742 (0.6) 4663 (0.7) 1085 (0.6) Missing 515,616 (8.4) 16,030 (3.5) 13,012 (3.7) 3428 (3.0) 495,033 (13.9) 19,704 (6.2) 17,617 (7.2) 4433 (6.0) 958,387 (10.1) 3729 (0.5) 11,724 (1.8) 920 (0.5) Hispanic (yes) 913,410 (14.9) 78,560 (16.9) 56,405 (16.2) 20,583 (18.1) <0.001 142,476 (4.0) 14,459 (4.5) 8934 (3.7) 3258 (4.4) <0.001 1,120,745 (11.9) 111,027 (15.0) 81,405 (12.5) 24,543 (14.2) <0.001 Charlson Comorbidity Index=0 3,529,055 (57.6) 249,209 (53.6) 204,750 (58.6) 60,619 (53.3) <0.001 2,027,542 (57.3) 177,280 (55.5) 133,864 (54.9) 39,518 (53.2) <0.001 5,406,942 (57.2) 427,378 (57.8) 379,149 (58.1) 90,348 (52.1) <0.001 Charlson Comorbidity Index=1 628,857 (10.3) 66,304 (14.3) 52,182 (14.9) 18,718 (16.5) 370,441 (10.5) 48,940 (15.3) 43,138 (17.7) 12,979 (17.5) 837,860 (8.9) 100,463 (13.6) 96,831 (14.8) 26,903 (15.5) Charlson Comorbidity Index=2+ 650,047 (10.6) 100,510 (21.6) 59,212 (17.0) 24,935 (21.9) 328,925 (9.3) 55,991 (17.5) 44,338 (18.2) 15,378 (20.7) 806,335 (8.5) 120,439 (16.3) 116,467 (17.9) 40,619 (23.4) Distance<5 miles 4,406,138 (71.9) 337,686 (72.6) 248,993 (71.3) 81,880 (72.0) <0.001 1,528,212 (43.2) 138,013 (43.2) 101,176 (41.5) 31,679 (42.6) <0.001 5,847,229 (61.9) 478,111 (64.6) 405,995 (62.2) 109,690 (63.3) <0.001 Distance 5 to ≤10 miles 1,067,839 (17.4) 81,221 (17.5) 62,923 (18.0) 20,209 (17.8) 1,187,188 (33.5) 110,085 (34.4) 82,658 (33.9) 25,325 (34.1) 2,109,469 (22.3) 168,449 (22.8) 153,830 (23.6) 40,418 (23.3) 10+ miles 640,475 (10.5) 45,173 (9.7) 36,591 (10.5) 11,425 (10.1) 819,699 (23.2) 71,347 (22.3) 59,512 (24.4) 17,221 (23.2) 1,459,157 (15.4) 91,063 (12.3) 90,713 (13.9) 22,757 (13.1) ADI first quartile (0–25) 3,730,556 (60.9) 273,756 (58.9) 209,171 (59.9) 65,209 (57.3) <0.001 925,780 (26.2) 79,371 (24.8) 55,172 (22.7) 16,580 (22.3) <0.001 4,981,977 (52.7) 376,018 (50.8) 338,553 (51.9) 85,596 (49.4) <0.001 ADI second quartile (26–50) 2,061,843 (33.7) 164,066 (35.3) 120,076 (34.4) 41,397 (36.4) 1,209,286 (34.2) 110,494 (34.6) 83,894 (34.4) 25,698 (34.6) 2,322,376 (24.6) 196,961 (26.6) 171,385 (26.3) 46,893 (27.0) ADI third quartile (51–75) 220,236 (3.6) 18,253 (3.9) 13,405 (3.8) 4780 (4.2) 848,307 (23.9) 78,427 (24.5) 62,635 (25.7) 19,190 (25.8) 1,205,778 (12.8) 106,295 (14.4) 92,200 (14.1) 26,038 (15.0) ADI fourth quartile (76–100) 75,409 (1.2) 6083 (1.31) 4363 (1.3) 1655 (1.5) 538,821 (15.2) 50,654 (15.8) 41,182 (16.9) 12,599 (16.9) 607,954 (6.4) 55,259 (7.5) 45,046 (6.9) 13,636 (7.9) Previous mail order prescription (%) 3,300,885 (53.9) 309,623 (66.6) 240,468 (68.9) 77,779 (68.4) <0.001 1,146,925 (32.4) 131,952 (41.3) 109,098 (44.8) 32,185 (43.3) <0.001 3,756,617 (39.8) 362,658 (49.0) 370,542 (56.8) 97,523 (56.2) <0.001 Deductible health plan 3,527,795 (57.6) 219,019 (47.1) 180,566 (51.7) 55,318 (48.6) 1,802,118 (50.9) 136,427 (42.7) 108,275 (44.5) 32,010 (43.1) 1,995,151 (21.1) 128,553 (17.4) 111,455 (17.1) 26,159 (15.1) Outpatient copay $0 to ≤$10 1,542,277 (25.2) 164,290 (35.3) 103,509 (29.6) 36,487 (32.1) <0.001 480,935 (13.6) 69,705 (21.8) 45,556 (18.7) 15,673 (21.1) <0.001 4,675,016 (49.5) 414,017 (55.9) 364,534 (55.9) 102,246 (58.9) <0.001 Outpatient copay $10 to ≤$20 1,492,032 (24.4) 118,296 (25.4) 88,721 (25.4) 28,603 (25.2) 984,048 (27.8) 85,914 (26.9) 65,918 (27.1) 19,694 (26.5) 2,983,855 (31.6) 225,038 (30.4) 197,346 (30.3) 49,393 (28.5) Outpatient copay $20 to ≤$30 1,940,533 (31.7) 114,895 (24.7) 100,338 (28.7) 30,756 (27.1) 859,079 (24.3) 61,049 (19.1) 51,758 (21.3) 14,641 (19.7) 1,037,233 (10.9) 65,910 (8.9) 61,013 (9.4) 14,135 (8.2) Outpatient copay >$30 783,050 (12.8) 39,415 (8.5) 33,844 (9.7) 9519 (8.4) 929,800 (26.3) 73,646 (23.0) 59,482 (24.4) 17,651 (23.8) 469,131 (4.9)