Cardiovascular disease risk management during COVID: In-person versus virtual visits

Abstract

Objectives:

Limited research has assessed how virtual care (VC) impacts cardiovascular disease (CVD) risk management, especially in community clinic settings. This study assessed change in community clinic patients’ CVD risk management during the pandemic, and CVD risk factor control among patients who had primarily in-person or primarily VC visits.

Study Design/Methods:

Data came from an electronic health record shared by 52 community clinics, in index (03/01/2019–02/29/2020) and follow-up periods (07/01/2020–02/28/2022). Interrupted time-series analyses compared follow-up period changes in slope and level of population monthly means of 10-year reversible CVD risk score, blood pressure (BP), and hemoglobin A1c (HbA1c) among patients whose completed follow-up period visits were primarily in person versus primarily VC. Propensity score weighting minimized confounding.

Results:

There were 10,028 in-person and 6,593 VC patients in CVD risk analyses, 9,874 in-person and 5,390 VC patients in BP analyses, and 8,221 in-person and 4,937 VC patients in HbA1c analyses. The VC group was more commonly younger, female, white, and urban. Mean reversible CVD risk, mean systolic BP, and percentage of BP measures ≥140/90 increased significantly from index to follow-up periods in both groups. Rate of change between these periods was the same for all outcomes in both groups, regardless of care modality.

Conclusions:

Among community clinic patients with CVD risk, majority in-person versus majority VC was not significantly associated with longitudinal trends in reversible CVD risk score or key CVD risk factors.

Precis:

A comparison of CVD risk management in community clinics during the COVID pandemic, when primary care was delivered mostly in person versus mostly virtually.

Objectives/Introduction

Management of cardiovascular disease (CVD) risk (i.e., risk of having a cardiovascular event such as heart attack or stroke) is a common focus of primary care. In response to the COVID-19 pandemic’s onset, many primary care practices began offering telephone and video visits, a care delivery mode called synchronous virtual care (hereafter, VC).^1–5 This spike in primary care visits delivered virtually^6–8 provided an opportunity to examine whether CVD risk management thus delivered yields different outcomes from those associated with care delivered in person. This is important because while VC rates declined since their peak, VC continues to be a common care delivery mode.⁹

Prior studies indicate that VC may be associated with management of CVD risk factors (e.g., BP and HbA1c) similar to in-person care.^10–15 However, we know of no prior evidence on overall CVD risk management associated with VC in primary care settings serving socioeconomically vulnerable populations, such as community-based safety net clinics.¹⁶ Understanding the relationship between VC and CVD risk management is important in these settings, as community clinics’ low-income, medically / socially complex patients have high rates of unmanaged CVD risk,^17,18 and transportation-related barriers to accessing care.¹⁹ We examined change over time in CVD risk management when received in-person versus through VC in the two years post-pandemic onset (March 2020 – February 2022), and compared to the year pre-pandemic (March 2019 – February 2020).

Methods

OCHIN, Inc., the study setting, is a national network of primary care safety net clinics; its members (96 community clinic organizations, running 493 clinic sites in 14 states, as of September 2018) share a centrally managed instance of the Epic© electronic health record (EHR). Shortly after the pandemic’s onset, OCHIN members’ VC capabilities were expanded to include embedded / facilitated VC functionality within the EHR, including video and telephone options; a rapidly increased number of patients received care via these modalities.²⁰

In 2018, the CV Wizard clinical decision support system was activated in 70 OCHIN clinics for a trial of its effectiveness; the analyses presented here used data from these clinics.^21–23 CV Wizard’s algorithms process EHR data to generate a 10-year reversible CVD risk estimate (hereafter called CVD risk) and evidence-based care recommendations for patients meeting specific risk criteria (i.e., aged 40–75 years with (a) >10% risk of cardiovascular event in the next ten years attributable to uncontrolled CVD risk factors, or (b) diabetes and ≥1 uncontrolled CVD risk factor, or (c) CVD and ≥1 uncontrolled CVD risk factor.²³ Study data on CVD risk score came from CV Wizard. All other data were extracted from the parent trial clinics’ EHR, and made research-ready by the Accelerating Data Value Across a National Community Health Center (ADVANCE) clinical research network of PCORnet.²⁴

Clinical outcomes related to CVD risk were examined: CVD risk, systolic blood pressure (SBP), diastolic blood pressure (DBP), and hemoglobin A1c (HbA1c). Analyses compared change in these outcomes in a pre-pandemic index period (March 2019 – February 2020) and a follow-up period (July 2020 – February 2022). March 2020 through June 2020 was considered a washout period during which the study’s clinics established VC capacity and CV Wizard was set up for use in VC.

The analysis population included patients with ≥1 index period primary care visit at one of the 70 study clinics at which they met CV Wizard’s risk criteria (above), and ≥1 primary care visit in the follow-up period. For BP and HbA1c analyses, ≥1 relevant measurement in the index and follow-up periods was also required. BP data could be self-reported during VC or collected during in-person visits. HbA1c data represents a laboratory test where results were received in the patient’s EHR from a clinic or external laboratory. Blood draws for HbA1c (unassociated with an ambulatory visit) were not counted as in-person visits. Clinics with <10% of visits coded as VC in the follow-up period were excluded. Another clinic at which key outcomes of interest were rarely recorded during the follow-up period was excluded. Patients who died during the study period were excluded. For each outcome, population-level monthly means were calculated for the index (12 time points) and follow-up (20 time points) periods.

The independent variable was created by categorizing patients based on how many follow-up period primary care visits were in-person versus VC: (1) the majority of visits (>50%) were in-person or evenly split between in-person and VC (in-person group), or (2) the majority were VC (VC group). Analyses compared these groups regarding how index period trends in the monthly mean outcome measures changed in the follow-up period.

Interrupted time-series analysis (ITSA) was used to account for index period trends when assessing VC impacts on follow-up period outcomes. To evaluate the presence of immediate or over-time impacts, models included both a level term and a slope change term. Models used multilevel segmented regression. A random effect for clinic was included to account for clinic-level clustering. Monthly mean data over time was plotted to assess trends visually and assess non-stationarity and seasonality. Although some seasonality was apparent in the CVD risk and BP outcomes, these effects were minimal and Durbin-Watson statistics^25,26 were non-significant at 6- and 12-month lags. These statistics, and autocorrelation and partial autocorrelation function plots, determined the nature of autocorrelation; as strong serial autocorrelation was indicated, a first-order autoregressive structure was added to the model. A log transformation was applied to eliminate skew in outcome variables. Both improved model fit.

Inverse propensity score weighting was used to minimize potential confounding caused by index period differences between comparison groups. Weighting was based on patient age, race / ethnicity, gender, language, number of index period clinic visits, federal poverty level, most common payor across visits, and level of rurality of patient residence (determined using ZIP codes linked to USDA ERS Rural-Urban Commuting Area codes and assigned using University of Washington Rural Research Health Center recommended categories (https://depts.washington.edu/uwruca/ruca-uses.php)). To create the propensity score, these variables were put into a logistic regression model with VC / in-person category as the dependent variable. Stabilized inverse propensity score weights were used to reduce type I errors. After applying the weights, there was <10% absolute standardized difference between the groups in all covariates, indicating good balance.²⁷

We also assessed COVID status during the follow-up period in the comparison groups. Supplement Table 3 lists the diagnostic codes used for this purpose.

Results

Patients from 52 clinics were included; 957 patients (0.5% of initial population) died during the study period and were excluded. CVD risk analyses included 10,028 (60.3%) in-person and 6,593 (39.7%) VC patients; BP analyses included 9,874 (64.7%) in-person and 5,390 (35.3%) VC patients; HbA1c analyses included 8,221 (62.5%) in-person and 4,937 (37.5%) VC patients. Characteristics of the CVD risk analysis population are in Table 1, including distribution of visits that were VC versus in-person; those of the BP and HbA1c subsets in Appendices 1–2. A higher percentage of the VC group patients were female, and a higher percentage were white, compared to the in-person group. A higher percentage of the VC group visits were to urban residents; a higher percentage were paid by Medicaid. As noted, methods were used to minimize confounding that this might cause.

Table 1 –

Patient characteristics of the CVD risk change analysis population

	Majority or equally in-person visits (in-person group)		Majority VC visits (VC group)		Difference p-value (chi-sq / t-test)
Patient characteristics – index period
Total (N, %)	10,028	60.3	6,593	39.7
Age at study start (mean, sd)	57.62	8.86	56.41	8.73	<.0001
Sex, (N, %)					<.0001
Female	5,066	50.5	3,640	55.2
Male	4,962	49.5	2,953	44.8
Race, (N, %)					<.0001
Asian	379	3.8	235	3.6
Black	1,981	19.8	653	9.9
White	6,615	66.0	4,901	74.3
Other	263	2.6	217	3.3
Unknown	790	7.9	587	8.9
Ethnicity, (N, %)					<.0001
Hispanic	3,269	32.6	2,307	35.0
Non-Hispanic	6,759	67.4	4,286	65.0
Primary language, (N, %)					0.0004
English	6,501	64.8	4,378	66.4
Spanish	2,798	27.9	1,837	27.9
Other	729	7.3	378	5.7
Federal Poverty Level, (N, %)
Majority of visits <138% FPL	6,456	64.4	4,252	64.5	0.7478
Majority of visits ≥138% FPL	2,339	23.3	1,555	23.6
All visits missing FPL	1,233	12.3	786	11.9
Had COVID diagnosis in follow-up period	1,020	10.2	818	12.4	<.0001
Visit characteristics
Visits in index period (mean, sd)
Number of total visits	5.40	4.84	6.03	5.21	<.0001
In-person - primary care	4.38	3.42	4.78	3.62	<.0001
Virtual care - primary care	1.33	0.78	1.60	1.17	0.0002
Visits in follow-up period (mean, sd)
Total visits	7.71	6.91	10.47	8.82	<.0001
in-person - primary care	4.57	3.98	2.97	2.28	<.0001
Virtual care - primary care	2.51	2.05	6.20	5.00	<.0001
Distribution of encounter modality (N, %)
0% (no VC encounters)	2,596	25.9	0	0.0	n/a
.01 to <25% VC encounters	1,419	14.2	0	0.0
25% to <35% VC encounters	2,114	21.1	0	0.0
35% to <50% VC encounters	1,601	16.0	0	0.0
50% to <65% VC encounters	2,298	22.9	1,738	26.4
65% to <75% VC encounters	0	0.0	1,514	23.0
75% to <85% VC encounters	0	0.0	1,403	21.3
85% to <95% VC encounters	0	0.0	485	7.4
95%–100% VC encounters	0	0.0	1,453	22.0
Both time periods
Total visits	146,818	54.3	123,626	45.7
Payor, (N, %)					<.0001
Medicaid	43,867	29.9	49,684	40.2
Medicare	57,835	39.4	43,595	35.3
Private	19,426	13.2	10,080	8.2
Uninsured	21,482	14.6	15,733	12.7
Other	4,208	2.9	4,534	3.7
Rurality (patient’s residence), (N, %)					<.0001
Urban	85,411	58.2	97,418	78.8
Large rural	39,082	26.6	23,046	18.6
Small rural	22,325	15.2	3,162	2.6

Table 2 displays unadjusted outcome measures for both comparison groups. In the index period: mean CVD risk and BP were higher in the in-person than VC group; mean HbA1c was higher in the VC group; the proportion of measurements at which CVD risk was elevated was not different between the groups; rate of uncontrolled BP was higher in the in-person group; and rate of measures with uncontrolled HbA1c was higher in the VC group. In most cases, though significant, these differences were small. Table 2 also shows the unadjusted changes in outcomes between the index and follow-up periods. After propensity score weighting, the parameter in the regression model representing the difference in index period intercept between groups was not significant.

Table 2 –

Analysis outcomes by study period in each analysis population

	Majority / equally in-person visits (in-person group)		Majority VC visits (VC group)		Difference p-value (chi-sq / t-test)
	CVD risk change
Total N	109,027		91,988
CVD 10-year risk score as % (mean, SD)
Index period	9.19	10.14	8.99	10.08	0.0130
Follow up period	10.57	12.36	10.12	11.09	<.0001
CVD risk score >10% (N, %)
Proportion >10% index period	13,318	34.9	9,682	34.5	0.3320
Proportion >10% follow-up period	24,800	37.8	23,142	38.4	0.0247
	BP change
Total N	97,302		50,480
Systolic BP (mean, SD)
Index period	131.20	18.70	129.60	18.65	<.0001
Follow up period	133.20	19.33	131.00	19.28	<.0001
Diastolic BP (mean, SD)
Index period	76.42	10.71	75.87	10.83	<.0001
Follow up period	76.46	10.88	75.76	10.90	<.0001
Uncontrolled BP (N, %)
Measured BP≥140/90 index	13,987	30.9	8,071	28.0	<.0001
Measured BP≥140/90 follow-up	15,909	34.0	5,757	30.3	<.0001
	HbA1c change
Total N	43,885		25,699
HbA1c (mean, SD)
Index period	8.00	2.01	8.21	2.07	<.0001
Follow up period	8.01	1.99	8.23	2.00	<.0001
Uncontrolled HbA1c (N, %)
Measurements with HbA1c≥8 index	6,607	37.9	4,648	43.9	<.0001
Measurements with HbA1c≥8 follow-up	9,158	38.9	6,094	45.5	<.0001

Means for HbA1c are calculated after excluding 584 additional measurements where the value was only given as above or below a certain number - these were retained for use in categorical analyses but excluded from continuous.

These results also show the unadjusted overall changes in CVD risk between the index and follow-up periods. Mean CVD risk score increased in both groups (in-person: 9.2 to 10.6; VC: 9.0 to 10.1), as did proportion of measures with >10% risk (from 34.9% to 37.8% and from 34.5% to 38.4%, respectively). Mean SBP increased in both groups (from 131.2 to 133.2 and from 129.6 to 131.0, respectively); mean DBP was static in both groups; and proportion of BP measures ≥140/90 increased in both groups (from 30.9% to 34.0% and from 28.0% to 30.3%, respectively; these differences were significant at p<.0001. In both groups, mean HbA1c was static.

The in-person group had a significant upward index period trend in CVD risk (p=0.004) with no significant change for other outcome measures (p=0.51 for SBP, 0.46 for DPB, and 0.91 for HbA1c). Between the end of the index period and the start of the follow-up period (i.e., the washout period) there was a 2.2% (p=0.006) increase in this group’s mean SBP, and a 3.4% (p=0.04) decrease in mean HbA1c. Neither mean CVD risk nor mean DBP changed.

Differences in change in the outcomes of interest between the comparison groups during the washout period is relevant to understanding follow-up period results; Figures 1–3. During the washout period, change in CVD risk, SBP, and DBP was not significantly different between comparison groups. HbA1c increased 5.2% more in the VC than the in-person group (p=0.04).

Figure 1.

CVD risk

Figure 3.

HbA1c

The difference in slope of change over time between the pre-pandemic onset index period and post-pandemic onset follow-up periods was the same between groups for all outcomes, with no significant difference in longer-term impact.

Mean CVD risk increased for both groups in the index period but leveled off for the in-person group during follow-up, while the VC group means decreased slightly but not significantly (Figure 1). The counterfactual dashed lines show that both groups had lower mean CVD risk at the end of the study period than that projected based on index period trends. For BP (Figure 2), little change was seen in the index period; during follow-up, mean SBP of the in-person group increased initially but then flattened, while the VC group’s mean SBP dropped slightly and continued on a modest upward trajectory, ultimately ending below the predicted counterfactual. For DBP, both groups trended downward in the follow-up period. Mean HbA1c was stable for both groups in the index period, with no significant change in the follow-up period (Figure 3).

Figure 2a.

SBP

Discussion

Among community clinic patients, mean CVD risk and SBP increased slightly following the pandemic’s onset, which aligns with prior research.^28–32 Pandemic-related drivers of risk include patients having difficulty with maintaining CVD risk reduction behaviors.¹ That SBP naturally increases with aging³³ likely explains some of the higher SBP over time observed here, and some of the CVD risk increase.

The increased CVD risk and SBP in the first few months post-pandemic onset gradually corrected after some time; DBP and HbA1c followed similar patterns. As DBP increases into the 50s and declines thereafter, given the age distribution of this population we would expect DBP to be flat or slightly decreasing over time.³⁴ That these measures improved more than expected based on pre-pandemic trends might reflect a post-pandemic onset correction to care over time in the study clinics.³⁵

The rates of increase of mean CVD risk and SBP were not significantly different between comparison groups, suggesting that in these circumstances CVD risk management was of similar quality regardless of care delivery mode. It is possible that VC visits were used selectively to tailor care to patients’ needs, and / or used more often for quick assessment / follow-up and to increase care accessibility, and in-person visits used more often when in-depth physical assessment and more specialized treatment were needed.

These findings align with prior research showing that CVD risk management through VC yields outcomes similar to those associated with in-person visits,⁵⁶ though little prior research involved community clinic populations.^{10,12,14,36–39} One study found VC associated with worse BP control, but not among patients with ≥1 recorded BP, perhaps reflecting poor documentation rather than inferior BP control.¹⁵ Synchronous VC has also been associated with outcomes similar to those from in-person care for hospitalizations and emergency department visits among patients with heart failure, but this evidence is limited⁴⁰ and results vary.⁴¹ Research is needed to better understand the circumstances in which CVD risk management via VC is most effective. For example, VC visits may be more effective when patients have an established relationship with a provider, and less ideal for new patient visits.

These results add knowledge by focusing on community clinic patients, whose rates of CVD and uncontrolled risk factors are high than in the general population.⁴² The finding that CVD risk management was as effective when delivered mostly via VC versus mostly in-person holds particular promise for community clinic patients, as a VC option for care receipt might mitigate some access barriers for these patients (e.g., transportation insecurity) without impacting care quality.⁴³ Transportation insecurity negatively impacts healthcare access and outcomes, especially among lower-income persons and those with chronic conditions.⁴⁴ However, VC also has the potential to increase disparities in access and health outcomes as persons without the resources needed to engage in VC (e.g., broadband internet) may not benefit from VC equitably.^45–54 Policies related to VC should ensure that its potential benefits are distributed equitably; this may involve expanding access to high-speed internet and other resources needed to use such technologies.

Limitations

Analysis results can be generalized only to community clinic patients who accessed care during the periods prior to and after the pandemic’s onset. The differences between comparison groups at the start of the follow-up period warrant consideration, as the groups were defined by the visits they had during that period, which could have been influenced by their CVD risk management status or practice decisions about when to use in-person or VC visits. Propensity score weighting addressed these differences, but contextual reasons why patients received VC versus in-person care may still have influenced mode of care.

In sum, the in-person group had somewhat higher mean CVD risk and SBP levels at the start of follow-up than the VC group, and somewhat lower HbA1c levels. This might have influenced care received in the follow-up period if clinics emphasized seeing patients in person based on their most recent BP. Yet the mean number of total visits in the follow-up period was higher in the VC group, indicating more points of engagement with the clinic. (The ITSA with inverse propensity score weighting is intended to mitigate such potential bias, as the in-person group was well-balanced with the VC group on outcome variables, as seen through the model output and patient characteristics, through weighting.) Furthermore, it was not feasible to assess whether CVD risk management was a focus of these visits. This is a minor limitation, as all included patients had high CVD risk, and CVD risk management is often discussed in community clinics regardless of the main purpose of the visit.

Analyses were limited to clinics that had previously volunteered for a study on CVD risk management, with the potential that this received particular emphasis in these clinics. However, CVD risk management is of importance to all primary care community clinics because of its health impacts and because management of specific CVD risk factors is a reported quality metric. These analyses also excluded clinics with low VC provision rates, which may differ from those willing / able to implement this change. Analyses also excluded patients who died during follow-up. Results may not be generalizable to the sickest persons.

Comparison group categories reflect whether the majority of a given patient’s visits were in-person or VC; results do not reflect the impact of receiving all care virtually. However, these categories reflect real-world practice, where a combination of care modes is likely. Research is needed to assess whether specific proportions of in-person to VC visits optimize outcomes, and whether there is a threshold of proportion of VC visits below which care outcomes are less optimized. Furthermore, we were unable to distinguish between different virtual care modalities (e.g., video or telephone), as this is not well-documented or easily determined using EHR data.

Results relating to BP were likely impacted by practices beginning to document BP from patient-reported measures in the post-pandemic period. Home-based measures are known to trend lower than those taken in-person visits.⁵⁵ Post-pandemic BP trends in both groups should be interpreted noting that they differed categorically from those pre-pandemic onset. It is also possible that BP measures in VC visits are lower than those taken in the office setting due to preferential patient reporting of lower BP measures.

While just 10–14% of the analysis populations had a COVID diagnosis in the follow-up period, it was statistically more likely in the VC group. In sensitivity analyses excluding persons diagnosed with COVID during follow-up, results were not meaningfully different than those in the original models; details available upon request.

All follow-up period data came from the first two years post-pandemic onset, a period in which primary care provision was substantially disrupted. Therefore, results do not indicate the comparability of in-person vs. VC in a more typical period. Nevertheless, this period of rapid change in care delivery provides an opportunity for initial assessments of CVD risk management outcomes in VC. Our team’s companion paper is assessing quality of CVD care preventive at in-person vs. VC visits in the same population analyzed here, which will illuminate this potential; further research is needed to understand these patterns.

Conclusions

In the two years post-pandemic onset, among community clinic patients with high CVD risk, having a majority of visits via VC was associated with CVD risk management outcomes that did not differ from those with a majority of visits in-person. This supports reimbursing safety net clinics for providing VC as part of increasing care accessibility, and policies designed to expand equitable access to VC. Research is needed on the optimal ratio of in-person to VC visits in terms of CVD risk management and other health elements in primary care settings.

Figure 2b.

DBP

Takeaway Points:

Among primary care community clinic patients with high cardiovascular disease (CVD) risk, in the period following the COVID pandemic onset, receipt of mostly in-person versus mostly virtual care was not associated with rate of change in CVD risk or CVD risk factor management. No significant difference in these measures over time was associated with mode of care. Mean CVD risk was lower at the end of the study period than projected based on pre-pandemic trends, regardless of care mode. This indicates that virtual and in-person care can yield similar outcomes when managing CVD risk among low-income patients.

• Results add knowledge by focusing on virtual care in community clinic patients, whose rates of CVD and uncontrolled risk factors are high, and for whom a virtual care option might mitigate common access barriers (e.g., transportation insecurity).

• Results indicate the importance of reimbursing safety net clinics for providing virtual care, and of policies that expand equitable access to virtual care.

• More research is needed on the optimal ratio of in-person to virtual visits in CVD risk management and other health elements managed in primary care settings, and on the circumstances in which CVD risk management via virtual care is most and least effective.