Traditional and HIV-Specific Risk Factors Associated With Atrial Fibrillation Among Underrepresented Minority Groups With HIV

Abstract

Background

Evaluation of nonvalvular atrial fibrillation/atrial flutter (NVAF) in under-represented racial and ethnic minority groups (UREGs) living with HIV has not been adequately studied.

Objectives

The purpose of this study was to describe the incidence of NVAF, identify associated factors, and describe oral anticoagulation (OAC) patterns among UREGs with HIV.

Methods

This is a secondary analysis of data collected in PATHWAYS (Pathways to Cardiovascular Disease Prevention and Impact of Specialty Referral in Underrepresented Racial and Ethnic Minorities with HIV; NCT04025125), a retrospective population-based study of HIV care among UREGs with HIV. We investigated the independent associations of cardiovascular and HIV-specific risk factors with incident NVAF using Cox regression analysis and examined appropriate OAC use.

Results

From 2015 to 2019, 11,066 UREGs with HIV met entry criteria; 10,945 were without NVAF at baseline. On average, patients were 44 years of age, 67.2% male, 94.4% Black, and 8.5% Hispanic. Average follow-up was 3.4 years, and 63.4% were on antiretroviral therapy. Incidence of NVAF was 4.54 incident cases per 1,000 person-years with a cumulative incidence at one and 5 years after HIV diagnosis of 0.48% and 2.16%, respectively. Age, diabetes, heart failure, severe renal disease, and antiretroviral therapy regimens including a protease inhibitor and/or integrase strand transfer inhibitor were independently associated with incident NVAF. Of those with NVAF meeting qualifying CHA₂DS₂-VASc scores, only 44.2% received any OAC prescription for stroke prophylaxis.

Conclusions

In this cohort of UREGs living with HIV, both traditional and HIV-specific risk factors were associated with incident NVAF. Rates of appropriate OAC prescribing were low.

Central Illustration

HIV affects over 35 million individuals worldwide.^1,2 With the increased effectiveness and availability of antiretroviral therapy (ART), HIV has become a chronic disease, with increasing prevalence of chronic comorbidities of aging among people living with HIV (PLWH). PLWH are at increased risk of myocardial infarction,^3,4 ischemic stroke,^5,6 heart failure,^7,8 atrial fibrillation,9, 10, 11, 12 and sudden cardiac death¹³ compared to people without HIV after controlling for relevant factors, in part due to persistent immune dysregulation and chronic inflammation.^14,15 This makes risk factor modification and identification of high-risk populations increasingly important.

Atrial arrhythmias are common worldwide,^16,17 associated with higher health care utilization and costs^17,18 and a 1.5- to 2-fold increased risk of death and 2.4-fold risk of stroke.^19,20 Few studies to-date have investigated risk factors associated with nonvalvular atrial fibrillation and atrial flutter (NVAF) among PLWH. While studies provide some evidence that traditional risk factors and CD4+ T-cell count <200 are associated with atrial fibrillation,9, 10, 11 to our knowledge, no studies have investigated the effects of ART on incident NVAF nor focused their efforts on underrepresented racial and ethnic minority groups (UREGs) living with HIV. This is particularly important given the increased risk of ischemic stroke among Black and Hispanic individuals.^21,22

We analyzed data from the PATHWAYS (Pathways to Cardiovascular Disease Prevention and Impact of Specialty Referral in Underrepresented Racial and Ethnic Minorities with HIV) study, a multi-institutional database of UREGs living with HIV gathered retrospectively using data extracted from the electronic health record (EHR) of participating sites in the Stakeholders, Technology, and Research (STAR) Clinical Research Network. In this population, we examined the prevalence and incidence of NVAF among UREGs living with HIV. We also examined health care utilization and risk factors for incident NVAF in this population, focusing on HIV-specific factors such as disease severity and ART. Finally, we investigated oral anticoagulation (OAC) prescribing patterns among UREGs living with HIV and NVAF.

Methods

Patient population

We utilized data from the PATHWAYS study (NCT04025125), a multi-institutional retrospective cohort study of UREGs living with HIV in the Southern United States. The PATHWAYS database includes patient- and provider-level data from the EHRs from the STAR Clinical Research Network institutions mapped to the PCORnet Common Data Model.²³ PCORnet Common Data Model uses a standardized representation of EHR data domains and a curation process to ensure data quality. The institutions included Duke University Health System (Durham, North Carolina), Vanderbilt University Medical Center (Nashville, Tennessee), the Medical University of South Carolina (Charleston, South Carolina), and Wake Forest Medical Center (Winston-Salem, North Carolina). Data were extracted in July 2021.

Inclusion criteria included HIV diagnosis recorded between January 1, 2015, and December 31, 2019, adults aged 18 to 99 years at first date of HIV diagnosis during this time frame, and self-reported UREGs (self-reported as Black/African American, American Indian or Alaska Native, Asian, multiple race, and/or Hispanic [was not mutually exclusive from reported racial group]). Exclusion criteria included insufficient race or ethnicity data to confirm eligibility (n = 171) and a diagnosis or procedure indicating valvular heart disease (n = 48). Eligibility was based on earliest qualifying encounter when all criteria were met, which defined the index/baseline date for the overall population. Data for ascertaining follow-up data were obtained through December 31, 2020.

Outcomes

The primary outcome was incident NVAF, defined by database coding for the first encounter with a diagnosis of NVAF, based on the International Classification of Diseases (ICD) (defined by the presence of one or more of the following diagnosis codes: ICD-9 code 427.3 and ICD-10 code of I48) (Supplemental Table 1). We excluded patients with valvular atrial fibrillation/atrial flutter (ICD-9 and ICD-10 codes listed in Supplemental Table 2). Prior literature has demonstrated acceptable sensitivity and specificity for the diagnosis of NVAF based on ICD codes after extensive duplicate review of medical records to verify both the diagnosis and confirm that the NVAF was new onset.^10,24 For assessing factors associated with incident NVAF, we excluded those who had a diagnosis of NVAF at or before their index HIV diagnosis date (baseline NVAF, n = 121).

The secondary outcome of interest was OAC prescribing among those with NVAF, defined based on a prescription for any OAC (warfarin, factor Xa inhibitor (FXaI), and/or direct thrombin inhibitor) from index NVAF diagnosis date through 12 months after for the incident NVAF group and from 6 months prior to cohort onset through 6 months after for those with baseline NVAF. OAC analysis was restricted to men with CHA₂DS₂-VASc ≥2 and women with CHA₂DS₂-VASc ≥3 per current national guidelines.^25,26

Covariates

Covariates of interest were defined from available EHR records from contributing sites. These included baseline demographics, traditional cardiovascular risk factors (tobacco or alcohol use, hypertension, hyperlipidemia, atherosclerotic disease, stroke/transient ischemic attack, diabetes, heart failure, and renal disease defined from ICD-9 and ICD-10 diagnosis codes), health care engagement (number of ambulatory clinic visits in year prior, cardiology visit in year prior, and HIV in-care with at least 2 HIV medical care encounters at least 90 days apart within the year prior), markers of HIV disease severity (including CD4+ T-cell count in cells/mm³ and viral load expressed in RNA copies/mL), medication use (based on prescribing records, including ART involving different classes and cardiovascular risk-modifying agents such as lipid-lowering treatment and antihypertensives), and history of major bleeding (diagnosis codes and procedure codes). Demographics were assessed at index date, cardiovascular risk factors were assessed at index date or within 2 years prior, laboratory measures were assessed at index date or at the closest date within 2 years prior or up to 10 days after, clinic visits were assessed within the year prior to index date, and medications other than OACs, such as ART, were assessed at index date or at the closest date within 18 months prior or up to 6 months after. Prior major bleeding was defined as any of the following within 2 years prior to baseline: 1) a hospital admission with diagnosis of intracranial hemorrhage; 2) a hospital admission with diagnosis of gastrointestinal bleeding with blood product transfusion within ±7 days of admission; 3) a hospital admission with diagnosis of bleeding in other locations (not intracranial) with blood product transfusion within ±7 days of admission); or 4) a control of bleeding procedure (inpatient or outpatient) with blood product transfusion within ±7 days of procedure (Supplemental Table 3).

Statistical analysis

Baseline demographics, comorbidities, HIV characteristics, medications, and measures of health care engagement were summarized using descriptive statistics. Differences between patients with and without NVAF at index HIV diagnosis were assessed using Wilcoxon rank sum tests for continuous variables and chi-squared tests for binary and categorical variables. We computed the cumulative incidence of NVAF at one and 5 years of follow-up handling death as a competing risk and estimated the incident event rate using a loglinear Poisson model.

Cox regression analysis was used to examine the univariable associations between baseline characteristics at HIV diagnosis and incident NVAF, adjusting for site and year of HIV diagnosis. Covariates were not time-varying. To determine which candidate covariates to include in the multivariable models, we considered HIV-related factors of interest and likely confounders, all selected a priori, as well as all variables with a P < 0.05 in our univariable analyses. We then used a stepwise variable selection process among these, sequentially adding variables if P < 0.05 and dropping variables if P ≥ 0.05 at each step. Site and year of cohort entry (HIV diagnosis date) were forced into the model to account for study design. Patients with missing data were excluded from univariable models. For the multivariable analysis, all patients had complete data for the candidate covariates. Model discrimination was summarized using Harrell’s concordance index.

Next, baseline OAC use was summarized using descriptive statistics among patients with NVAF and eligible CHA₂DS₂-VASc scores. Differences between patients with NVAF receiving OAC and not were assessed using Wilcoxon rank sum tests for continuous variables and chi-squared tests for binary and categorical variables. Logistic regression analysis examining factors associated with receipt of OAC was unable to be performed due to insufficient population size. All analyses were performed at the Duke Clinical Research Institute using SAS version 9.4 (SAS Institute).

Institutional ethics review and approval

The Duke University Health System obtained local Institutional Review Board (IRB) approval (Pro00101663 [site IRB], Pro 00101104 [coordinating center IRB]) and waiver of informed consent. All sites relied on Duke for IRB review and approval.

Results

Between January 1, 2015, and December 31, 2019, 11,114 individuals were identified as UREGs living with HIV. Of these, 48 (0.43%) were excluded due to valvular heart disease or prior heart valve procedure giving a final study population of 11,066 UREGs living with HIV (Figure 1). Of these, 121 (1.09%) already had a diagnosis of NVAF at the date of their first documented HIV diagnosis and 10,945 did not have NVAF.

Figure 1.

Patients Meeting Inclusion Criteria

Flow diagram of study inclusion and exclusion criteria, giving 10,945 patients without NVAF at index HIV diagnosis (baseline). NVAF = nonvalvular atrial fibrillation and atrial flutter.

At baseline, patients were on average (mean ± SD) 44 ± 13 years of age, 67% male, 94% Black, and/or 8.5% Hispanic (Table 1). Patients were engaged in medical care, with an average of 3.7 ± 4.4 clinic visits in the prior year and 6.4% of patients having visited cardiology clinic in the prior year. Average CD4+ T-cell count was 535 ± 350 cells/mm³, median (IQR) HIV viral load was 40 (4-15,849) copies/mL, and 63% were on any ART. Average CHA₂DS₂-VASc score was 0.94 ± 1.1.

Table 1.

Baseline Characteristics at HIV Index Diagnosis by NVAF Status

	Overall Population (N = 11,066)	No Atrial Fibrillation at Baseline (n = 10,945)	Atrial Fibrillation at Baseline (n = 121)	P Value
Demographics
Year of index HIV diagnosis, n (%)				0.162
2015	6,126 (55.4)	6,067 (55.4)	59 (48.8)
2016	1,513 (13.7)	1,498 (13.7)	15 (12.4)
2017	1,229 (11.1)	1,209 (11.0)	20 (16.5)
2018	1,082 (9.8)	1,072 (9.8)	10 (8.3)
2019	1,116 (10.1)	1,099 (10.0)	17 (14.0)
Age (y) at index HIV diagnosis, mean (SD)	43.6 (13.2)	43.5 (13.1)	56.0 (11.8)	<0.001
Female, n (%)	3,631 (32.8)	3,599 (32.9)	32 (26.4)	0.134
Racea, n (%)
American Indian or Alaska Native	62 (0.6)	59 (0.6)	3 (2.6)
Asian	139 (1.3)	138 (1.3)	1 (0.9)
Black or African American	9,799 (94.4)	9,688 (94.4)	111 (96.5)
Native Hawaiian or other Pacific Islander	10 (0.1)	10 (0.1)	0 (0.0)
White	254 (2.4)	254 (2.5)	0 (0.0)
Multiple race	116 (1.1)	116 (1.1)	0 (0.0)
Hispanicb, n (%)	928 (8.5)	924 (8.5)	4 (3.3)	0.041
Insurance typec, n (%)
HRSA RWHAP	1918 (17.3)	1908 (17.4)	10 (8.3)	0.008
Medicare	2,536 (22.9)	2,476 (22.6)	60 (49.6)	<0.0001
Medicaid	2,136 (19.3)	2,113 (19.3)	23 (19.0)	0.934
Private insurance	3,635 (32.8)	3,604 (32.9)	31 (25.6)	0.089
Other source	242 (2.2)	240 (2.2)	2 (1.7)	1.000
Rural (zip code area)d, n (%)	1,428 (13.0)	1,406 (12.9)	22 (18.3)	0.079
Social Deprivation Indexe, mean (SD)	70.3 (23.4)	70.3 (23.4)	70.7 (24.9)	0.522
Comorbidities
Current smoker, n (%)	3,883 (35.1)	3,848 (35.2)	35 (28.9)	0.153
Alcohol abuse, n (%)	548 (5.0)	533 (4.9)	15 (12.4)	<0.001
Hypertension, n (%)	3,510 (31.7)	3,418 (31.2)	92 (76.0)	<0.001
Hyperlipidemia, n (%)	1,646 (14.9)	1,603 (14.6)	43 (35.5)	<0.001
Coronary artery disease, n (%)	394 (3.6)	351 (3.2)	43 (35.5)	<0.001
Peripheral artery disease, n (%)	215 (1.9)	198 (1.8)	17 (14.0)	<0.001
Stroke/TIA, n (%)	276 (2.5)	259 (2.4)	17 (14.0)	<0.001
Type II diabetes, n (%)	1,245 (11.3)	1,205 (11.0)	40 (33.1)	<0.001
Congestive heart failure, n (%)	427 (3.9)	373 (3.4)	54 (44.6)	<0.001
Severe renal disease, n (%)	506 (4.6)	485 (4.4)	21 (17.4)	<0.001
AIDS (CCI), n (%)	2,591 (23.4)	2,555 (23.3)	36 (29.8)	0.098
CHA2DS2-VASc score, mean (SD)	0.94 (1.13)	0.92 (1.10)	2.69 (1.76)	<0.001
Clinical measures
Number of clinic visits/prior year, mean (SD)	3.7 (4.4)	3.7 (4.3)	6.9 (7.6)	0.001
Cardiology clinic visit in prior year, n (%)	711 (6.4)	659 (6.0)	52 (43.0)	<0.001
HIV in-care flag (prior to index date), n (%)	5,738 (51.9)	5,670 (51.8)	68 (56.2)	0.336
Body mass index (kg/m2)f, mean (SD)	28.0 (7.1)	28.0 (7.1)	27.7 (7.1)	0.731
Kidney function (eGFR)g, mean (SD)	84.3 (29.2)	84.6 (29.0)	59.9 (30.8)	<0.001
HIV viral load (copies/mL)h, median (IQR)	40 (4, 15,849)	40 (4, 15,849)	40 (4, 1,585)	0.172
CD4+ T-cell count (cells/uL)i, mean (SD)	534.9 (350)	535.4 (350)	490.2 (368)	0.138
Antiretroviral therapy (ART)
Any ART, n (%)	7,017 (63.4)	6,944 (63.4)	73 (60.3)	0.479
NRTI	6,579 (59.5)	6,511 (59.5)	68 (56.2)	0.464
NNRTI	1861 (16.8)	1842 (16.8)	19 (15.7)	0.742
PI	3,778 (34.1)	3,742 (34.2)	36 (29.8)	0.306
INSTI	4,004 (36.2)	3,954 (36.1)	50 (41.3)	0.237
Other ART	48 (0.4)	46 (0.4)	2 (1.7)	0.097
Other medications, n (%)
Lipid-lowering treatment	929 (8.4)	899 (8.2)	30 (24.8)	<0.001
Antihypertensive treatment	2,288 (20.7)	2,208 (20.2)	80 (66.1)	<0.001
Anticoagulation	248 (2.2)	208 (1.9)	40 (33.1)	<0.001
Warfarin	149 (1.3)	126 (1.2)	23 (19.0)	<0.001
Dabigatran	6 (0.1)	4 (0.0)	2 (1.7)	0.002
Any DOAC	114 (1.0)	95 (0.9)	19 (15.7)	<0.001
Apixaban	69 (0.6)	54 (0.5)	15 (12.4)	<0.001
Edoxaban	1 (0.0)	1 (0.0)	0 (0.0)	1.000
Rivaroxaban	51 (0.5)	45 (0.4)	6 (5.0)	<0.001
Antiplatelet agent	920 (8.3)	872 (8.0)	48 (39.7)	<0.001
Aspirin	870 (7.9)	826 (7.5)	44 (36.4)	<0.001
Clopidogrel	161 (1.5)	149 (1.4)	12 (9.9)	<0.001
Prasugrel	9 (0.1)	9 (0.1)	0 (0.0)	1.000
Ticagrelor	9 (0.1)	9 (0.1)	0 (0.0)	1.000

ART = antiretroviral therapy; BMI = body mass index; CCI = Charlson Comorbidity Index; DOAC = direct oral anticoagulants; eGFR = estimated glomerular filtration rate; HRSA RWHAP = Health Resources and Services Administration Ryan White HIV/AIDS Program; INSTI = integrase strand transfer inhibitors; NNRTI = non-nucleoside reverse transcriptase inhibitors; NRTI = nucleoside reverse transcriptase inhibitors; NVAF = nonvalvular atrial fibrillation and atrial flutter; PI = protease inhibitors; TIA = transient ischemic attack.

^a686 with missing race data (680 without/6 with NVAF).

^b93 with missing ethnicity data (93 without/0 with NVAF).

^cMay have multiple insurance types or meds and may be counted in multiple rows.

^d55 with missing zip code data (54 without/1 with NVAF).

^e262 with missing Social Deprivation Index data (255 without/7 with NVAF).

^f2,741 with missing BMI data (2,762 without/21 with NVAF).

^g1,804 with missing eGFR data (1,797 without/7 with NVAF).

^h5,166 with missing HIV viral load data (5,099 without/67 with NVAF).

ⁱ3285 with missing CD4+ T-cell count data (3,244 without/41 with NVAF).

Of the 10,945 without NVAF at index HIV diagnosis, we identified 171 (1.6%) patients who developed NVAF over a follow-up of up to 5 years (mean 3.4 ± 1.1 years). Cumulative incidence of NVAF at one and 5 years after HIV diagnosis were 0.48% (95% CI: 0.36-0.63) and 2.16% (95% CI: 1.85-2.51) (Figure 2). The average incidence rate of NVAF was 4.54 (95% CI: 3.91-5.27) per 1,000 person-years.

Figure 2.

Cumulative Incidence of NVAF From HIV Index Diagnosis

Cumulative incidence of NVAF through 5 years after index HIV diagnosis. Abbreviation as in Figure 1.

In our univariable analysis (Table 2), HIV-related factors associated with incident NVAF included baseline CD4+ T-cell count <200 (HR vs ≥500 1.83; 95% CI: 1.20-2.80) and initial ART including a protease inhibitor (PI) (HR vs no PI 1.57; 95% CI: 1.15-2.15) and/or an integrase strand transfer inhibitor (INSTI) (HR vs no INSTI 1.48; 95% CI: 1.09-2.01). A cardiology visit in the year prior to HIV diagnosis and prior to any NVAF diagnosis was also associated with incident NVAF (HR: 4.15; 95% CI: 2.78-6.18). Additional associated factors included older age, Medicare, and comorbid diseases including hypertension, hyperlipidemia, coronary and peripheral artery disease, prior stroke/transient ischemic attack, congestive heart failure, and severe renal disease.

Table 2.

Association of Patient Characteristics With Incident NVAF in Univariable Models

	HRa (95% CI)	P Value
Demographics
Age (per 5-y increment)	1.43 (1.34-1.52)	<0.001
Female	1.09 (0.80-1.49)	0.580
Medicare	3.72 (2.74-5.04)	<0.001
Private insurance	0.68 (0.47-0.98)	0.039
HRSA RWHAP	0.58 (0.36-0.92)	0.021
Medicaid	1.27 (0.89-1.82)	0.182
Other source insurance	0.41 (0.06-2.96)	0.378
Rural (zip code area)b	1.41 (0.94-2.12)	0.096
Comorbidities
Congestive heart failure	11.53 (8.13-16.36)	<0.001
Coronary artery disease	7.54 (5.08-11.19)	<0.001
Severe renal disease	4.93 (3.25-7.47)	<0.001
Peripheral artery disease	4.40 (2.38-8.13)	<0.001
Hypertension	3.35 (2.46-4.56)	<0.001
Type II diabetes	3.30 (2.37-4.60)	<0.001
Hyperlipidemia	2.36 (1.69-3.29)	<0.001
Stroke/TIA	2.35 (1.16-4.80)	0.018
AIDS	1.54 (1.11-2.15)	0.010
Alcohol use disorder	1.04 (0.51-2.12)	0.922
Current smoker	1.00 (0.73-1.37)	0.992
Clinic visits
Cardiology clinic visit in prior year (y/n)	4.15 (2.78-6.18)	<0.001
Ambulatory visits in prior year (versus 0-1)c		0.014
2-6 clinic visits	0.92 (0.62-1.37)	0.682
>6 clinic visits	1.60 (1.04-2.47)	0.034
HIV in-care	1.14 (0.80-1.61)	0.471
Clinical measures
GFR (versus ≥60)d		<0.001
eGFR <30	6.34 (4.17-9.63)	<0.001
eGFR 30-60	3.16 (2.16-4.62)	<0.001
BMI (vs 18.5-25)e		0.226
BMI <18.5	1.90 (0.89-4.05)	0.099
BMI 25-29.9	0.97 (0.62-1.52)	0.906
BMI ≥30	1.28 (0.85-1.92)	0.241
HIV-specific factors
CD4+ T-cell count (versus ≥500)f		0.011
CD4+ T-cell count <200	1.83 (1.20-2.80)	0.005
CD4+ T-cell count 200-499	1.01 (0.68-1.51)	0.956
Viral load (vs <500 copies/mL)g		0.485
500-100,0000 copies/mL	0.72 (0.41-1.26)	0.252
>100,000 copies/mL	1.01 (0.46-2.22)	0.982
Entry ART classh (versus not on ART class)
PI	1.57 (1.15-2.15)	0.005
INSTI	1.48 (1.09-2.01)	0.012
NNRTI	1.24 (0.86-1.79)	0.244
NRTI	1.15 (0.83-1.61)	0.403
Other ART	2.53 (0.63-10.20)	0.193

Abbreviations as in Table 1.

^aAll models were adjusted for site and year of cohort entry (HIV diagnosis date).

^bExcludes 54 patients with missing rural (area code) data.

^cExcludes 446 patients with missing ambulatory clinic data in year prior.

^dExcludes 1,797 patients with missing eGFR data.

^eExcludes 2,741 patients with missing BMI data.

^fExcludes 3,244 patients with missing CD4+ T-cell count data.

^gExcludes 5,099 patients with missing HIV viral load data.

^hPatient could be included in multiple ART classes.

Candidate variables for stepwise selection included age, gender, smoking status, comorbid diseases, any cardiology visit in prior year, a diagnosis of AIDS (due to missing CD4+ T-cell count data, this was defined as the composite of either CD4+ T-cell count <200 or HIV and opportunistic infection at baseline or within prior 2 years),²⁷ and ART class (Supplemental Table 4). In our final multivariable analysis (Table 3), variables associated with incident NVAF included age (HR 1.37 per 5-year increment; 95% CI: 1.28-1.46), cardiovascular risk factors such as diabetes (HR: 1.51; 95% CI: 1.06-2.14), heart failure (HR: 5.83; 95% CI: 3.93-8.63), and severe renal disease (HR: 1.89; 95% CI: 1.19-2.99), and antiretroviral therapies including a PI (HR vs no PI 1.43; 95% CI: 1.03-1.98) or an INSTI (HR vs no INSTI 1.41; 95% CI: 1.03-1.94).

Table 3.

Association of Patient Characteristics With Incident NVAF in Multivariable Models

	HRa (95% CI)	P Value
Congestive heart failure	5.83 (3.93-8.63)	<0.001
Severe renal disease	1.89 (1.19-2.99)	0.007
Diabetes mellitus	1.51 (1.06-2.14)	0.022
PI	1.43 (1.03-1.98)	0.031
INSTI	1.41 (1.03-1.94)	0.034
Age (per 5-y increments)	1.37 (1.28-1.46)	<0.001

Abbreviations as in Table 1.

^aIn addition to covariates presented, model is also adjusted for site and year of cohort entry (HIV diagnosis date); Model was fit in 10,945 patients with no history of atrial fibrillation at cohort entry.

Among the 121 baseline NVAF and 171 incident NVAF cases, only 51 and 78 had qualifying CHA₂DS₂-VASc scores for OAC prescription, respectively. Of those with NVAF at baseline who met CHA₂DS₂-VASc criteria, 47.1% received an OAC prescription within ± 6 months of index HIV date and of those with NVAF first diagnosed during follow-up who met CHA₂DS₂-VASc criteria, 42.3% received an OAC prescription within 12 months of NVAF diagnosis. Combining those with both baseline and incident NVAF, only 44.2% received any anticoagulation prescription (Figure 3). Warfarin (23.3%) and FXaIs (23.3%) were equally common selections. Of FXaIs, the majority prescription was apixaban, at 21.7% of all patients. Among PLWH and newly diagnosed NVAF during the study period, there was a slight majority of FXaIs compared to warfarin prescriptions (24.4% vs 21.8%). Given the time window to assess OAC use, patients may have contributed to more than one class of OAC if their OAC prescription changed within the assessment window. A majority of the entire cohort were on antiplatelet agents (66.7%) and a minority had a prior major bleeding event (6.4%).

Figure 3.

Anticoagulation Rates Among UREGs Living With HIV and Qualifying CHA₂DS₂VASc^a

Anticoagulation prescribing rates for different classes of OAC (warfarin, factor Xa inhibitor, direct thrombin inhibitor, total) among UREGs living with HIV and qualifying CHA₂DS₂VASc score ^aFor NVAF at baseline, oral anticoagulation rate is based on prescription within ±6 months of index HIV date and for incident NVAF, oral anticoagulation rate is based on prescription within +12 months of NVAF incident diagnosis date. OAC = oral anticoagulation; UREGs = underrepresented racial and ethnic minority groups; other abbreviation as in Figure 1.

In our univariable analysis, CHA₂DS₂-VASc score, history of major bleeding, or prescription of an antiplatelet agent did not differ between those receiving OAC and those not (Table 4). Furthermore, engagement with care and HIV-specific characteristics also did not differ between those receiving OAC and those not. Due to insufficient number of individuals meeting criteria for OAC selection, we were unable to perform multivariable analysis to investigate baseline factors associated with the receipt of any anticoagulation therapy.

Table 4.

Baseline Factors and the Receipt of Anticoagulation Among Qualifying CHA₂DS₂-VASc

	Whole Cohort (N = 129)	Received Anticoagulation (n = 57)	Did Not Receive Anticoagulation (n = 72)	P Value
CHA2DS2-VASc score, mean (SD)	4.30 (1.23)	4.39 (1.35)	4.24 (1.13)	0.740
Prior major bleedinga (yes), n (%)	5 (6.4)	3 (9.1)	2 (4.4)	0.645
Antiplatelet agent, n (%)	86 (66.7)	34 (59.6)	52 (72.2)	0.133
Aspirin	81 (62.8)	33 (57.9)	48 (66.7)	0.306
Clopidogrel	28 (21.7)	11 (19.3)	17 (23.6)	0.555
Prasugrel	1 (0.8)	0 (0.0)	1 (1.4)	1.000
Ticagrelor	3 (2.3)	1 (1.8)	2 (2.8)	1.000
Demographics
Age (y) at index date, mean (SD)	60.6 (11.3)	59.4 (11.3)	61.6 (11.3)	0.228
Female, n (%)	46 (35.7)	25 (43.9)	21 (29.2)	0.084
Raceb				0.773
Black/African-American	121 (96.8)	54 (96.4)	67 (97.1)
Asian	1 (0.8)	0 (0.0)	1 (1.4)
Other	3 (2.4)	2 (3.6)	1 (1.4)
Hispanic	4 (3.1)	2 (3.5)	2 (2.8)	1.000
Insurance typec
HRSA RWHAP	7 (5.4)	3 (5.3)	4 (5.6)	1.000
Medicare	89 (69.0)	36 (63.2)	53 (73.6)	0.202
Medicaid	24 (18.6)	14 (24.6)	10 (13.9)	0.122
Private insurance	31 (24.0)	14 (24.6)	17 (23.6)	0.900
Other insurance source	2 (1.6)	1 (1.8)	1 (1.4)	1.000
Clinical factors
Number of clinic visits in prior year, mean (SD)	9.0 (7.1)	9.7 (7.5)	8.4 (6.8)	0.349
Cardiology clinic visit in prior year, n (%)	64 (49.6)	26 (45.6)	38 (52.8)	0.419
HIV in-care flag (prior to index date), n (%)	105 (81.4)	46 (80.7)	59 (81.9)	0.857
Current smoker, n (%)	38 (29.5)	18 (31.6)	20 (27.8)	0.638
Body mass indexd (kg/m2), mean (SD)	27.7 (8.1)	28.4 (9.3)	27.2 (7.1)	0.707
Kidney functione (eGFR), mean (SD)	48.6 (29.5)	48.6 (31.8)	48.7 (27.9)	0.953
HIV viral loadf (copies/mL), median (IQR)	40 (4, 566)	40 (3, 794)	40 (5, 251)	0.659
CD4+ T-cell countg (cells/uL), mean (SD)	479.8 (428.0)	451.6 (390.4)	499.6 (454.7)	0.632
Antiretroviral therapy (ART)
Any ART	89 (69.0)	41 (71.9)	48 (66.7)	0.521
NRTI	81 (62.8)	36 (63.2)	45 (62.5)	0.934
NNRTI	26 (20.2)	9 (15.8)	17 (23.6)	0.271
PI	47 (36.4)	22 (38.6)	25 (34.7)	0.650
INSTI	66 (51.2)	35 (61.4)	31 (43.1)	0.038
Other ART	2 (1.6)	1 (1.8)	1 (1.4)	1.000

Abbreviations as in Table 1.

^a51 with missing bleeding data (24 of whom received/27 of whom did not receive anticoagulation).

^b4 with missing race data (1 of whom received/3 of whom did not receive anticoagulation).

^cPatients may have multiple insurance types, or meds, and may be counted in multiple rows.

^d8 with missing BMI data (4 of whom received/4 of whom did not receive anticoagulation).

^e2 with missing eGFR data (2 of whom received/0 of whom did not receive anticoagulation).

^f49 with missing HIV viral load data (28 of whom received/21 of whom did not receive anticoagulation).

^g27 with missing CD4+ T-cell data (15 of whom received/12 of whom did not receive anticoagulation).

Discussion

This large, retrospective cohort of UREGs living with HIV examines a high-risk and underserved population that is not proportionally represented in either real-world studies or randomized clinical trials to-date. To our knowledge, this is the first study evaluating NVAF incidence among UREGs living with HIV specifically, with incidence comparable to similar-aged populations regardless of HIV status or race and ethnicity.^9,28 We also found that both traditional risk factors such as age, diabetes, heart failure, and severe renal disease and a history of use of PIs and integrase strand inhibitors are associated with incident NVAF (Central Illustration). Furthermore, less than half (44.2%) of our OAC-eligible population was on appropriate OAC, with similar prescribing rates of warfarin compared to FXaIs.

Central Illustration.

Atrial Fibrillation Among Underrepresented Minority Groups With HIV

Traditional and HIV-specific risk factors associated with incident NVAF and appropriateness of oral anticoagulation among UREGs. CVD = cardiovascular disease; other abbreviations as in Figures 1 and 3.

In this study, we observed a cumulative incidence of NVAF of 2.16% at 5 years, with an average follow-up of 3.4 years, corresponding to an incidence rate of 4.54 (95% CI: 3.91-5.27) per 1,000 person-years. This rate is comparable to prior studies examining incident atrial fibrillation among both all-comers and PLWH of similar age regardless of race or ethnicity,^9,28 but lower than older-aged cohorts, supporting a well-known relationship between age and NVAF.^29,30 In a 2018 cohort of >500,000 individuals in the Geisinger Health System of similar average age to our cohort (mean age 47 years), an overall NVAF incidence rate of 6.82 per 1,000 person-years (95% CI: 6.65-7.00) was observed, but this incidence rate increased among individuals aged 85 years or older.²⁸ Among an older study population aged 65 years and older, Hsu et al⁹ found higher rates of NVAF regardless of race (15.5 events per 1,000 person-years, 95% CI: 13-18.5). Similarly, among a cohort of Medicare beneficiaries (median age 80 years), incident NVAF in 2007 was 28.3 per 1,000 person-years.²⁹

Few studies have investigated the effect of race in HIV and incident NVAF or focused their efforts on UREGs specifically as our study did. The Hsu et al study was a Veteran cohort comprised of 54% Black individuals. When stratified by age and race, Hsu et al found lower rates of NVAF among Black patients compared to White patients aged 45 years and older.⁹ On the contrary, a study by Sardana et al found the relative incidence of NVAF in HIV was higher in Black and Hispanic patients compared to White patients.¹² Despite Black individuals having a higher burden of most NVAF risk factors compared to White individuals, there is a well-known paradoxical race effect demonstrating higher risk of NVAF among White individuals compared to Black individuals among populations without HIV,^31,32 however, it is not known if this relationship holds among HIV populations. The conflicting data to date underline the need for significantly more work investigating NVAF incidence rates among Black and Hispanic patients with HIV and possible mechanisms of effect.

In adjusted analyses, there was a significant and independent association with incident NVAF between traditional cardiovascular risk factors such as age, diabetes, heart failure, and severe renal disease, consistent with well-established effects.^29,33, 34, 35, 36, 37, 38, 39 While male sex,⁴⁰ smoking,^35,41,42 hypertension,33, 34, 35^,43,44 and coronary artery disease^34,45 have additionally been associated with the development of atrial fibrillation among individuals without HIV, we hypothesize our lack of finding may be suggestive of either unaccounted mediator effects or type 2 error. Furthermore, a cardiology visit in the year prior to HIV diagnosis date was no longer significant in the multivariable analysis. Important to recall, these patients did not yet have a known NVAF diagnosis so the univariate association was likely due to confounding by indication, such as history of congestive heart failure, which had the highest ratio in the final multivariable model (HR: 5.83; 95% CI: 3.93-8.63).

For HIV-specific risk factors, our study found that a history of antiretroviral therapies with a PI or an INSTI may be associated with incident NVAF. In previous studies, PI and INSTI exposure have been associated with a host of cardiovascular and cardiometabolic diseases through a variety of mechanisms, including chronic inflammation, immune reconstitution, lipodystrophy, gut microbial translocation, pancreatic cell dysfunction, and drug-drug toxicity or drug metabolism to name a few.⁴⁶ Namely, PIs have been associated with increased atherosclerotic cardiovascular disease,^47,48 heart failure,⁴⁹ dyslipidemia,^50,51 type 2 diabetes,52, 53, 54, 55 and metabolic dysfunction–associated steatotic liver disease,56, 57, 58 and INSTIs have been associated with weight gain,59, 60, 61, 62, 63, 64, 65, 66 hypertension,⁶⁷ and dyslipidemia.⁶⁸ Prior studies have varied in how ART exposure was assessed, examining cumulative use, recent use defined by some subset of time, and/or past use depending on the study. While cumulative exposure to ART, particularly PIs, certain nucleoside reverse transcriptase inhibitors such as abacavir, and INSTIs have been associated with increased atherosclerotic cardiovascular disease,^47,48,69, 70, 71 there are some data to suggest this relationship is not linear over time. The RESPOND study (The International Cohort Consortium of Infectious Disease and Outcomes of Antiretroviral Treatment) investigated exposure to INSTIs and incidence of cardiovascular disease, finding an initial excess incidence of cardiovascular disease that was later attenuated to normal around 24 months of use.⁷² While the present study did not address cumulative exposure to classes of ART over time, our finding that ever exposure to PIs and/or INSTIs may be associated with incident NVAF adds to the literature highlighting the complex interplay between ART, cardiovascular diseases, and the development of NVAF. Given the uptake in INSTI-based regimens specifically, further investigation and corroboration regarding this exploratory finding and possible association with incident NVAF, potential mechanisms of effect, and cumulative effects on risk of incident NVAF is needed.

Furthermore, while CD4+ T-cell count <200 was associated with incident NVAF in our univariate analysis, the surrogate variable of AIDS used in our multivariable analyses due to missing CD4+ data was not. To our knowledge, the relationship between opportunistic infections among PLWH and NVAF has not yet previously been established. However, previous studies provide some data associating various infections and incident NVAF73, 74, 75 as well as low CD4+ T-cell count <200 with incident^9,11 and prevalent¹⁰ atrial fibrillation among cohorts without a focus on UREGs. Furthermore, low CD4+ T-cell count has been linked to a whole host of other cardiovascular diseases, including myocardial infarction,76, 77, 78 coronary artery disease,⁷⁹ stroke,80, 81, 82 and heart failure.^8,83 Potential mechanisms include chronic immune activation and persistent inflammation leading to increased vessel inflammation, endothelial dysfunction, and atherosclerotic plaque formation, instability, and risk of rupture.^14,15,84, 85, 86 Even so, specific pathways by which HIV severity may lead to incident atrial fibrillation and the effect of opportunistic infections on NVAF merits further investigation.

In our study, less than half (44.2%) were on guideline-recommended OAC based on qualifying CHA₂DS₂-VASc scores, with a nearly even split between warfarin and FXaIs. Important to keep in mind, this prevalence represents prescriptions; actual fill history and adherence may be even lower. One possible explanation for the comparable prescribing patterns among warfarin and direct oral anticoagulants (DOACs) is the issue of drug-drug interactions with ART and OAC, with PI-based regimens more likely to have interactions with DOACs than modern INSTI-based regimens that do not include a pharmacologic booster, though studies have also shown providers are not always aware of drug-drug interactions with ART.87, 88, 89 Regarding the overall low rates of appropriate OAC prescribing, while many previous studies over a similar time frame have demonstrated suboptimal OAC prescribing among all-comers with NVAF, our rates of OAC prescribing appear to be even lower. In a large retrospective cohort study from 2011 to 2020 investigating OAC use among adults with NVAF, OAC used increased from 56.3% to 64.7%, with a steady rise in DOACs (4.7%-47.9%) and steady decline in warfarin (52.4%-17.7%).⁹⁰ A systematic review from 2010 to 2018 investigating temporal trends in OAC users in OAC-eligible AF patients corroborated this finding, noting a rise in appropriate OAC use from 42% to 78% across the time frame, including a rise in DOACs from 0% to 45% and fall in warfarin from 42% to 32%.⁹¹ While our study’s lower OAC prescribing rates may be partially explained by being unable to account for prescriptions outside the studied health systems, it may also be due to health care access and pharmacoequity barriers that under-represented minorities often face. A study from the Korean National Health Insurance Service database found suboptimal OAC use (31.8%) among patients with atrial fibrillation and HIV.⁹² Similarly, a retrospective cohort analysis of 69,553 hospitalized patients with atrial fibrillation from the GWTG-AFIB (Get With The Guidelines-Atrial Fibrillation) registry noted Hispanic and Black patients were less likely to be discharged with appropriate OAC use compared to White and Asian patients.⁹³ Furthermore, despite the uptrend of DOAC use, current studies suggest that Black patients are less likely than White patients to be prescribed DOACs.^93,94 Similarly, Black and Hispanic patients with NVAF are also thought to have a higher risk of stroke compared with other races.^21,22 As such, one key clinical question that remains is if an alternative risk calculator other than CHADS₂VASc score should be applied to individuals living with HIV for more appropriate stroke risk modification. Finally, our findings in the context of the current literature highlight the need for better understanding pathways to pharmacoequity in guideline-recommended NVAF management to improve overall outcomes.⁹⁵

Study Limitations

First, our cohort derived from the PATHWAYS study comprised UREGs living with HIV only, without a specific HIV-negative, White comparison group. However, this is the first study to our knowledge to examine NVAF incidence among UREGs living with HIV. Second, our population on average was relatively young and predominantly male with nearly normal CD4+ T-cell count and low CHA₂DS₂-VASc scores on average. As a result, our estimates may not be generalizable to specific populations poorly represented in this cohort, such as the elderly, women, or patients without access to medical care. Third, covariates were assessed at or near date of index HIV diagnosis only in retrospective fashion and were not adjusted temporally nor was propensity matching performed. While adjustments in ART regimens over time were not accounted for, the temporal shift from PI- and NNRTI-based regimens to INSTI-based regimens had already largely occurred by the start time of our study cohort,^96,97 though some residual confounding may still exist. Given the potential implications of any association between ART subclass exposures and incident NVAF, it will be important to further investigate our exploratory finding with larger study populations and propensity-matched models. Similarly, medication adherence and intolerance patterns could not be captured and likewise, due to the possibility of patients receiving health care outside the health systems contributing to our database, there is the possibility of under capture of prescriptions, including for ART and OAC. Under capture of prescriptions may also have occurred for patients receiving care less frequently than our prescription windows (ie, <12 months for OAC prescription). We also cannot exclude the possibility of unknown confounding variables not included in our multivariable analyses. For example, we did not distinguish between inpatient vs outpatient atrial fibrillation diagnoses and thus cannot exclude confounders related to being hospitalized. Furthermore, our risk estimates may be underestimations among patients with worsened HIV control or increased comorbidities. Due to the small cohort meeting qualifying CHA₂DS₂-VASc scores for OACs, we were unable to perform multivariable analyses investigating risk factors associated with appropriate OAC prescriptions. Finally, although prior studies have confirmed the sensitivity and specificity of using ICD codes for NVAF diagnoses, this study did not adjudicate the atrial fibrillation events and thus there still is the possibility of inaccurate code entry like in all EHR-based studies.

Conclusions

In a large sample of UREGs living with HIV, traditional risk factors along with PIs and INSTIs were associated with incident NVAF. Furthermore, appropriate OAC prescribing among this high-risk population was low. These findings highlight the need for additional studies investigating mechanisms by which ART may increase NVAF risk, disparities contributing to inequities in anticoagulation prescribing, and whether or not anticoagulation improves outcomes in this high-risk population.

Perspectives.

COMPETENCY IN MEDICAL KNOWLEDGE: UREGs with HIV represent a high-risk group for cardiovascular disease. In addition to traditional CVD risk factors, unique aspects of living with HIV may further enhance risk of incident NVAF.

COMPETENCY IN PATIENT CARE: UREGs with HIV and nonvalvular atrial fibrillation/flutter may have additional barriers to appropriate OAC prescribing compared to all-comers with nonvalvular atrial fibrillation/flutter.

TRANSLATIONAL OUTLOOK: Data from this secondary analysis indicate both traditional and HIV-specific risk factors are associated with incident nonvalvular atrial fibrillation/flutter among UREGs with HIV. Unfortunately, rates of appropriate OAC prescribing in this population were suboptimal. Interventions to mitigate nonvalvular atrial fibrillation/flutter risk and downstream effects in this population will require interdisciplinary, team-based approaches.

Funding support and author disclosures

Research reported in this publication was supported by the National Institute of Minority Health and Health Disparities (R01MD013493, PI: Bloomfield); and Duke University’s CTSA grant (UL1TR002553) from the National Institutes of Health (NIH)’s National Center for Advancing Translational Sciences (NCATS) and powered by PCORnet. PCORnet has been developed with funding from the Patient-Centered Outcomes Research Institute (PCORI). Its contents are solely the responsibility of the authors and do not necessarily represent the official views of the NIH or views of organizations participating in, collaborating with, or funding PCORnet or of PCORI. Dr Pettit reports funding from the Tennessee Center for AIDS Research (P30 AI110527) and the Vanderbilt Institute for Clinical and Translational Research (UL1TR002243). Dr Longenecker reports advisory board affiliation with Theratechnologies. Dr Marsolo reports contracts to his institution from Novartis, BMS, Boehringer Ingelheim, Eli Lilly, Merck, Bayer, Amgen, and Pfizer. Dr Okeke has received consulting fees from ViiV Biosciences. Dr Thomas reports advisory board affiliations with Boston Scientific and Bristol Myers Squibb. Dr Shah reports financial interests and/or personal relationships with Amgen Inc, Janssen, National Institutes of Health, Esperion Therapeutics Inc, Amgen Inc, and Novartis. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.