Association of Albuminuria With 1‐Year Risk of Heart Failure and Other Adverse Outcomes in Atrial Fibrillation

Abstract

Background

Albuminuria is associated with increased stroke risk in atrial fibrillation (AF), but its relationship with heart failure (HF) and other adverse outcomes in AF is less well understood.

Methods

Using linked administrative databases, we conducted a retrospective cohort study of individuals aged ≥66 years who were newly diagnosed with AF between April 2009 and March 2019 in Ontario, Canada. Albuminuria was assessed using (1) urine albumin‐to‐creatinine ratio (UACR, mg/g) and (2) dipstick proteinuria (negative, trace, 1+, 2+, ≥3+). Cause‐specific hazards regression estimated adjusted hazard ratios (HRs) for HF hospitalizations or emergency department visits, stroke hospitalizations, bleeding hospitalizations, and death over 1 year.

Results

We included 64 717 individuals with UACR data and 110 430 with dipstick proteinuria data. Relative to UACR 5 mg/g, the HRs for UACR 30 mg/g (below the microalbuminuria threshold) were 1.39 (95% CI, 1.28–1.50) for HF, 1.22 (95% CI, 1.07–1.40) for bleeding, and 1.35 (95% CI, 1.27–1.42) for death. A UACR of 30 mg/g versus 5 mg/g was associated with an HR of 1.16 (95% CI, 0.99–1.36) for stroke but the HR was significantly elevated at UACR values ≥65 mg/g. Increasing dipstick proteinuria was also associated with increases in the HR for adverse outcomes. A UACR of 30 mg/g was associated with greater HF risk (versus 5 mg/g) than all CHA₂DS₂VASc components except age >75 years and prior HF.

Conclusions

Albuminuria is associated with increased hazards of HF, stroke, bleeding, and death in patients with AF even at low UACR levels. Albuminuria may enhance risk stratification in AF beyond traditional scores.

Nonstandard Abbreviations and Acronyms

OLIS

Ontario Laboratories Information System

UACR

urine albumin‐to‐creatinine ratio

Clinical Perspective.

What Is New?

• Albuminuria, even at levels below the threshold for microalbuminuria, is independently associated with an increased 1‐year risk of heart failure, stroke, bleeding, and death in older adults with newly diagnosed atrial fibrillation (AF).

• The hazard ratio for heart failure associated with a urine albumin‐to‐creatinine ratio of 30 mg/g (below the microalbuminuria threshold) versus 5 mg/g exceeds that of most individual CHA₂DS₂VASc risk factors.

• Dipstick proteinuria also demonstrates a graded association with adverse outcomes, supporting the utility of its use for risk stratification if quantitative microalbuminuria is not feasible.

What Are the Clinical Implications?

• Albuminuria can serve as a simple, inexpensive, and widely available marker to enhance risk stratification for adverse cardiovascular outcomes in patients with AF.

• Screening for albuminuria at the time of AF diagnosis may help identify high‐risk individuals who could benefit from closer follow‐up or initiation of cardioprotective therapies such as renin‐angiotensin system blockers or sodium glucose transporter 2 inhibitors, which are not currently part of the standard of care for AF.

• Incorporating urine albumin‐to‐creatinine ratio into AF management may improve prediction models for heart failure and bleeding beyond current risk scores and support broader application of kidney‐protective therapies.

Atrial fibrillation (AF) is the most common sustained arrhythmia and portends substantially increased risks of stroke, heart failure (HF), and death. ¹ Stroke prophylaxis is a clear treatment priority in AF. ² , ³ It is less appreciated that the highest absolute risk increase associated with AF is for heart failure (HF), ¹ making it a much larger contributor to mortality in people with AF. ⁴ Despite this, there is less emphasis on prognosticating HF risk and instituting preventative interventions for HF after AF diagnosis.

Albuminuria is established as a marker of endothelial dysfunction that is associated with increased cardiovascular and renal risk. ⁵ Among people with AF, albuminuria is associated with an increased risk of stroke. Consequently, it is incorporated in the ATRIA (Anticoagulation and Risk Factors in Atrial Fibrillation) score for predicting the risk of stroke, ⁶ which has been demonstrated to perform better than the CHADS2 and CHA₂DS₂VASC scores. ⁷ , ⁸ However, the ATRIA score handles albuminuria as a categorical variable reflecting its presence or absence without accounting for its magnitude. Albuminuria is also associated with a higher risk of HF, bleeding, and cardiovascular mortality in all comers, ⁵ , ⁹ , ¹⁰ , ¹¹ , ¹² even at levels that are currently considered to be in the normal range. ⁵

Thus, albuminuria may simultaneously prognosticate for multiple adverse outcomes that are important to people with AF. Accordingly, we conducted a population‐based cohort study of people with AF aged >65 years to study the association between albuminuria and the hazard of stroke, bleeding, HF, and death to inform its potential use in risk prediction models.

METHODS

Data Sources

The data set from this study is held securely in coded form at ICES (formerly the Institute for Clinical Evaluative Sciences). Although legal data sharing agreements between ICES and data providers prohibit ICES from making the data set publicly available, access may be granted to those who meet prespecified criteria for confidential access, available at https://www.ices.on.ca/DAS/ (email: das@ices.on.ca). The full data set creation plan and underlying analytical code are available from the authors on request, understanding that the computer programs might rely on coding templates or macros that are unique to ICES and are therefore either inaccessible or require modification.

The Ontario Health Insurance Plan provides coverage for medically necessary hospital care and physician services for all residents of the province. Prescription medications are covered for people aged ≥66 years under the Ontario Drug Benefit program. The plan is administrated using an Ontario Health Insurance Plan number unique to each individual, enabling linkage of administrative health data sets for research purposes. The Canadian Institute for Health Information Discharge Abstract Database records data on hospitalized patients and the National Ambulatory Care Reporting System collects data on emergency department visits. Physician billing records are stored in the Ontario Health Insurance Plan physician claims database, and drug dispensation records are stored in the Ontario Drug Benefit database. The Registered Persons Database maintains dates of birth and death. The Ontario Laboratories Information System is an information repository of laboratory test data, including creatinine, urine albumin‐to‐creatinine ratio (UACR), and dipstick urinalysis results. These data sets were linked using unique encoded identifiers and analyzed at ICES, which is an independent nonprofit organization with a mandate to collate and analyze publicly funded administrative health services in Ontario, Canada. The use of data in this project was authorized under section 45 of Ontario's Personal Health Information Protection Act, which does not require review by a research ethics board; individual patient consent is not required.

Cohort Creation

We used a validated algorithm that considered inpatient and outpatient codes for determination of AF in administrative data sets (specificity, 99.1% [95% CI, 98.9%–99.3%]) ¹³ to create a cohort of patients aged ≥66 years who were diagnosed with AF between April 1, 2009, and March 31, 2019. The index date was that of first AF documentation during the accrual period. Exclusion criteria included participants with limited life expectancy (age>105 years, long‐term care residence), non‐Ontario residence, and any AF diagnoses in the 5 years before the index date. The lookback period of 5 years was selected based on prior studies identifying this time frame as sufficient for excluding prevalent AF cases. ¹⁴ We excluded people with valvular AF given differences in their risk profile for adverse events compared with people with nonvalvular AF. Finally, we also excluded people who did not have available data for creatinine or albuminuria (UACR or dipstick proteinuria) in the 365 days before AF recognition.

The primary exposure was albuminuria, which was studied in 2 separate approaches: (1) UACR as a continuous measure (mg/g; divide by 8.85 to convert to mg/mmol), and (2) dipstick proteinuria, which was categorized as 1 of 5 possible values: negative, trace, 1+, 2+, or ≥ 3+ (given small number of people with 4+ proteinuria). Urinary protein:creatinine ratio was not used as it was very rarely measured (<0.5% of people with AF). For people with multiple measurements, we used the value closest to the index date. We constructed one cohort for assessment of UACR (consisting of people with available UACR data) and a separate cohort of people with available proteinuria measurements from dipstick urinalysis.

We studied 4 adverse outcomes using algorithms that have previously been validated for use in Ontario administrative data sets: (1) hospitalizations or emergency department visits for HF (henceforth referred to as “hospital presentations for HF”), ¹⁵ (2) hospitalization with a most responsible diagnosis of stroke, ¹⁶ (3) hospitalization with a most responsible diagnosis of bleeding, ¹⁷ and (4) all‐cause mortality. ¹⁸ , ¹⁹ Follow‐up was limited to 1 year following the index date because patients with AF frequently acquire additional stroke risk factors over time ²⁰ and because albuminuria levels are dynamic so are expected to have less robust associations with longer‐term outcomes.

Statistical Analysis

Parallel analyses were conducted separately in the UACR cohort and the dipstick urinalysis cohort. Baseline characteristics were summarized using the median with the 25th/75th percentiles (Q1–Q3) for continuous variables and counts (with percentages) for categorical variables. Statistical significance of baseline differences between UACR categories (0–4 mg/g, 5–9 mg/g, 10–29 mg/g, 30–299 mg/g, and ≥300 mg/g) and dipstick categories were assessed using median regression for continuous variables, the Cochran–Armitage test for dichotomous variables, and Jonckheere–Terpstra test for categorical variables with >2 categories. The rate of crude outcomes per 100‐person‐years was computed for each exposure category; the significance of differences determined using Poisson regression. Cells with <6 individuals are suppressed to reduce the risk of reidentification as per ICES contractual obligations with data providers.

We used cause‐specific hazards regression to estimate the association between the baseline measure of proteinuria (UACR or dipstick proteinuria) and the hazard of the 4 adverse outcomes of interest while adjusting for baseline characteristics and time‐varying anticoagulation status. This was chosen over a Fine–Gray subdistribution hazard model, as this study aimed to describe potential pathogenetic association between proteinuria with adverse outcomes rather than generating a model to estimate absolute incidence of adverse outcomes. Furthermore, this analysis required adjustment for time‐varying anticoagulation status, which cannot be done with Fine–Gray regression. ²¹ Death was treated as a competing risk when studying the outcomes of stroke, bleeding, and HF. ²² Patients were censored at 365 days if they were event free at that time.

The CHA₂DS₂VASC score components at baseline were used as model covariates: age (as a continuous variable), sex, heart failure, hypertension, diabetes, prior stroke or transient ischemic attack, and vascular disease (composite of ischemic heart or peripheral vascular disease). Other model covariates included estimated glomerular filtration rate (eGFR, calculated using the 2009 Chronic Kidney Disease Epidemiology Collaboration equation without race adjustment), prior bleeding (as documented in prior hospitalizations or emergency department visits), and baseline use of blockers of the renin‐angiotensin system (includes angiotensin‐converting enzyme inhibitors or angiotensin receptor blockers, which can decrease albuminuria) and statins (given their association with better outcomes in AF). ²³ Furthermore, we adjusted for anticoagulation with warfarin or direct oral anticoagulants as a time‐varying covariate based on drug dispensation records in the Ontario Drug Benefit. People were considered to be receiving anticoagulation on the days following dispensation of an anticoagulant and considered off anticoagulation for the period where the number of days between dispensed prescriptions exceeded the number of days supplied by the earlier prescription. All continuous variables (UACR, age, eGFR) were modeled using restricted cubic splines with 5 knots placed at the 5%, 27.5%, 50%, 72.5%, and 95% percentiles to avoid forcing a linear association between these predictor variables and outcomes. ²⁴

We report the results for the association of UACR with each outcome graphically by illustrating the hazard ratio (HR) for each level of UACR compared with the reference level of 5 mg/g, ²⁵ in keeping with the analytical approach of the Chronic Kidney Disease Prognosis Consortium. ²⁶ Subsequently, the analysis was stratified by baseline diabetes status and we determined the statistical significance of an interaction term between diabetes and UACR. We also tested for statistical significance of an interaction term between UACR and the CHA₂DS₂VASC score for each outcome of interest, with a plan for stratified analyses based on the CHA₂DS₂VASC score if applicable (CHA₂DS₂VASC score 1–3 versus 4–6). We also assessed if the association between UACR and outcomes varied by calendar year by testing the significance between of an interaction term between UACR and a categorical variable indicating year of AF diagnosis with 3 possible values: 2013–2015, 2016–2018, or 2019–2022.

All analyses were performed using SAS Enterprise Guide 7.1 (SAS Institute Inc., Cary, NC). Statistical significance was defined as a 2‐tailed P value <0.05.

RESULTS

UACR Analyses

As illustrated in Figure S1, there were 261 569 potentially eligible people with AF, among whom 64 717 (24.7%) individuals had a direct measurement of UACR occurring at a median of 115 (48–214) days before AF recognition. Table S1 provides baseline characteristics of participants with/without missing UACR data. People with missing UACR were older, more likely female, and more likely to have missing data on eGFR (with higher eGFR in patients with missing UACR who had available eGFR data). They were less likely to have diabetes, hypertension, ischemic heart disease, or congestive heart failure.

The baseline characteristics of included participants are listed in Table 1. Among patients with available UACR data, 15 282 (24%) had a UACR<5 mg/g. Compared with people with a UACR<5 mg/g, those with elevated values were older and more likely to have established cardiovascular disease and cardiovascular risk factors. People with elevated UACR were more likely to have had both prior stroke/transient ischemic attack and prior bleeding. Median eGFR levels were lower in people with elevated UACR. The prevalence of CHA₂DS₂‐VASc score components increased with higher UACR. Patients with abnormal UACR were more likely to be treated with blockers of the renin‐angiotensin system or warfarin but were less likely to be treated with direct oral anticoagulants.

Table 1.

Baseline Characteristics of the Cohort Included for Analysis of the Association of UACR With Adverse Outcomes, After Stratification Based on the Level of UACR

	0–4 mg/g	5–9 mg/g	10–29 mg/g	30–299 mg/g	≥300 mg/g	P value for trend
Sample size	N=15 282	N=7907	N=15 932	N=18 970	N=6626
Albumin/creatinine ratio value, mg/g, median (Q1–Q3)	1.8 (1.8–1.8)	8.0 (6.2–8.8)	17.7 (14.1–23.9)	72.5 (45.1–131.7)	748.7 (449.1–1550.5)	<0.0001
Age, y, median (Q1–Q3)	77 (71–82)	78 (73–84)	78.0 (72–83)	79.0 (73–84)	78.0 (72–83)	<0.0001
Female sex, n (%)	7306 (47.8%)	3574 (45.2%)	7345 (46.1%)	7592 (40.0%)	2335 (35.2%)	<0.0001
Income quintile
1, n (%)	3083 (20.2%)	1651 (20.9%)	3345 (21.0%)	4334 (22.8%)	1527 (23.0%)	<0.0001
2, n (%)	3223 (21.1%)	1720 (21.8%)	3474 (21.8%)	4189 (22.1%)	1549 (23.4%)
3, n (%)	3101 (20.3%)	1597 (20.2%)	3255 (20.4%)	3836 (20.2%)	1308 (19.7%)
4, n (%)	2948 (19.3%)	1437 (18.2%)	3006 (18.9%)	3389 (17.9%)	1174 (17.7%)
5, n (%)	2904 (19.0%)	1483 (18.8%)	2821 (17.7%)	3179 (16.8%)	1049 (15.8%)
Diabetes, n (%)	8067 (52.8%)	4432 (56.1%)	9803 (61.5%)	13 480 (71.1%)	5192 (78.4%)	<0.0001
Hypertension, n (%)	13 182 (86.3%)	6917 (87.5%)	14 343 (90.0%)	17 679 (93.2%)	6373 (96.2%)	<0.0001
Congestive heart failure, n (%)	3366 (22.0%)	1756 (22.2%)	4427 (27.8%)	7336 (38.7%)	3302 (49.8%)	<0.0001
Ischemic heart disease, n (%)	5184 (33.9%)	2666 (33.7%)	5536 (34.7%)	7158 (37.7%)	2760 (41.7%)	<0.0001
Peripheral vascular disease, n (%)	246 (1.6%)	136 (1.7%)	378 (2.4%)	616 (3.2%)	440 (6.6%)	<0.0001
Stroke or transient ischemic attack, n (%)	377 (2.5%)	238 (3.0%)	465 (2.9%)	710 (3.7%)	292 (4.4%)	<0.0001
Prior bleeding diagnoses in‐hospital, n (%)	1772 (11.6%)	1013 (12.8%)	2212 (13.9%)	3196 (16.8%)	1340 (20.2%)	<0.0001
Chronic kidney disease stage, n (%)
Stages 1 and 2, n (%)	10 063 (65.8%)	5231 (66.2%)	9279 (58.2%)	8437 (44.5%)	1551 (23.4%)	<0.0001
Stage 3, n (%)	4647 (30.4%)	2396 (30.3%)	5714 (35.9%)	8004 (42.2%)	2578 (38.9%)
Stage 4, n (%)	510 (3.3%)	247 (3.1%)	838 (5.3%)	2116 (11.2%)	1557 (23.5%)
Stage 5 or dialysis, n (%)	62 (0.4%)	33 (0.4%)	101 (0.6%)	413 (2.2%)	940 (14.2%)
Estimated glomerular filtration rate value, median (Q1–Q3)	68.7 (54.1–82.5)	69.1 (54.3–82.6)	64.8 (48.9–80.1)	56.1 (39.3–74.8)	38.1 (22.4–58.2)	<0.0001
Days between UACR test date and index date, median (Q1–Q3)	129.0 (55.0–231.0)	129.0 (54.0–228.0)	122.0 (52.0–221.0)	104.0 (42.0–200.0)	88.0 (37.0–174.0)	<0.0001
Medication use in the 6 mo before index date
Angiotensin‐converting enzyme inhibitor, n (%)	6128 (40.1%)	3333 (42.2%)	6936 (43.5%)	8717 (46.0%)	2854 (43.1%)	<0.0001
Angiotensin receptor blocker, n (%)	4722 (30.9%)	2372 (30.0%)	5089 (31.9%)	6473 (34.1%)	2689 (40.6%)	<0.0001
Warfarin, n (%)	2329 (15.2%)	1213 (15.3%)	2922 (18.3%)	3988 (21.0%)	1391 (21.0%)	<0.0001
Direct oral anticoagulant, n (%)	2601 (17.0%)	1349 (17.1%)	2722 (17.1%)	3275 (17.3%)	802 (12.1%)	<0.0001
Loop diuretics, n (%)	17 876 (27.6%)	3106 (20.3%)	1476 (18.7%)	3787 (23.8%)	6333 (33.4%)	<0.001
Thiazide diuretics, n (%)	22 240 (34.4%)	5142 (33.6%)	2752 (34.8%)	5624 (35.3%)	6477 (34.1%)	0.624
Mineralocorticoid receptor antagonist, n (%)	3369 (5.2%)	694 (4.5%)	340 (4.3%)	760 (4.8%)	1162 (6.1%)	<0.001
Angiotensin receptor‐neprilysin inhibitor, n (%)	15 (0.1%)	8 (0.1%)	19 (0.1%)	33 (0.2%)	13 (0.2%)	0.017

UACR indicates urine albumin‐to‐creatinine ratio.

Table S2 provides crude outcome data for people with/without missing UACR data. People with missing UACR data had higher crude incidence of death but lower crude incidence of bleeding and HF than people with available UACR. There was no significant difference in crude incidence of stroke between people with/without missing UACR data. In the 365 days after the index date, there were 4500 (7.0%) hospitalizations or emergency department visits for HF, 1004 (1.6%) stroke hospitalizations, 1389 (2.1%) bleeding hospitalizations, and 8764 (13.5%) deaths among people included in the study. The crude incidence of these adverse outcomes was higher in patients with abnormal UACR values, as summarized in Table S3.

After adjusting for baseline patient characteristics and time‐varying anticoagulation status, increases in UACR were associated with a higher hazard of all adverse outcomes (Figure 1). For all outcomes, there was a rapid rate of increase in the adjusted HR up to a UACR of ~60 mg/g versus a UACR of 5 mg/g followed by a slower rise to a UACR of 300 mg/g. For instance, a UACR of 30 mg/g (6× increase) was associated with an HR of 1.39 (95% CI, 1.28–1.50) for HF, 1.22 (95% CI, 1.07–1.40) for bleeding and 1.35 (95% CI, 1.27–1.42) for death, relative to 5 mg/g. Conversely, a UACR of 300 mg/g (60×) was associated with an HR of 1.84 (95% CI, 1.69–2.00) for HF and 1.46 (95% CI, 1.26–1.71) for bleeding. There was a minimal increase in the HR for death beyond UACR levels >60 mg/g. The HR for death at 60 mg/g versus 5 mg/g was 1.48 (95% CI, 1.38–1.58) whereas the HR for death at 300 mg/g versus 5 mg/g was 1.57 (95% CI, 1.48–1.67). A UACR of 30 mg/g versus 5 mg/g was associated with an HR of 1.16 (95% CI, 0.99–1.36) for stroke but the HR was significantly elevated at UACR values ≥65 mg/g. When compared against other risk factors, the HR for HF and death associated with a UACR of 30mg/g versus 5 mg/g was higher than the HR for HF associated with all CHA₂DS₂VASc risk factors other than age and prior HF. Moreover, the HR for all 4 outcomes associated with UACR 300 mg/g versus 5 mg/g was higher than most CHA₂DS₂VASc risk factors (see Table 2).

Figure 1. Association between urinary albumin‐to‐creatinine ratio and stroke, major bleeding, hospitalization for heart failure, and death.

The graphs plot the hazard ratio for stroke, major bleeding, hospitalization for heart failure, and death associated with albumin: creatinine ratio, relative to a value of 5 mg/g, from a cause‐specific regression model accounting for age, sex, heart failure, hypertension, diabetes, prior nonischemic stroke or transient ischemic attack, vascular disease, estimated glomerular filtration rate, prior bleeding, baseline statin use, baseline use of blockers of the renin‐angiotensin system, and time‐varying anticoagulation status. ED indicates emergency department; and HF, heart failure.

Table 2.

Hazard Ratios Associated With Categorical Components of the CHA₂DS₂VASc Risk Score for Comparison With the Hazard Ratios Associated With UACR 30 mg/g and 300 mg/g (Relative to 5 mg/g)

Baseline risk factors	Outcomes
	Heart failure			Death			Stroke			Bleeding
	HR	LL CI	UL CI	HR	LL CI	UL CI	HR	LL CI	UL CI	HR	LL CI	UL CI
Heart failure	4.01	3.76	4.29	1.98	1.89	2.07	1.22	1.06	1.40	1.42	1.27	1.59
Hypertension	1.23	1.07	1.42	0.93	0.86	1.01	1.47	1.12	1.93	1.30	1.03	1.63
Diabetes	1.22	1.13	1.30	1.26	1.20	1.32	1.20	1.04	1.37	1.07	0.95	1.21
Stroke	0.97	0.83	1.14	1.26	1.14	1.40	2.86	2.29	3.57	1.38	1.08	1.76
Vascular disease	1.23	1.15	1.30	0.96	0.92	1.01	0.99	0.86	1.13	1.23	1.10	1.37
Female sex	1.13	1.06	1.20	0.82	0.78	0.85	1.15	1.01	1.30	0.77	0.69	0.86
UACR 30 mg/g vs 5 mg/g	1.39	1.28	1.50	1.35	1.27	1.42	1.16	0.99	1.36	1.22	1.07	1.40
UACR 300 mg/g vs 5 mg/g	1.84	1.69	2.00	1.57	1.48	1.67	1.60	1.34	1.92	1.46	1.26	1.71

The regression models were further adjusted for age and estimated glomerular filtration rate (as restricted cubic splines), prior bleeding, baseline use of blockers of the renin‐angiotensin system and statins, and time‐varying anticoagulation exposure. HR indicates hazard ratio; LL, lower limit; UACR, urine albumin‐to‐creatinine ratio; and UL, upper limit.

The findings from the stratified analysis by baseline diabetes status are illustrated in Figure S2. There was a statistically significant interaction by diabetes status for death (P<0.001) and to a lesser extent for bleeding (P=0.012). The HR for mortality and for bleeding associated with increasing UACR levels beyond 5 mg/g was significantly higher in people without diabetes relative to people with diabetes (ie, higher UACR conferred a higher relative hazard of death in people without diabetes than it did in people with diabetes). The findings from the stratified analysis by baseline CHA₂DS₂VASC Score are provided in Figure S3. There was a statistically significant interaction between UACR and CHA₂DS₂VASC Score for the outcomes of death and HF (P<0.001 for both interaction terms). For both outcomes, the HR associated with increasing UACR levels was higher in people with CHA₂DS₂VASC score 1 to 3 than it was for people with scores 4 to 8. The HR for other outcomes associated with UACR was not meaningfully different in the 2 categories of CHA₂DS₂VASC score. There was no significant interaction between calendar year of AF diagnosis and the association between UACR and outcomes (P=0.59 for stroke, 0.47 for bleeding, 0.34 for HF, and 0.13 for death).

Dipstick Proteinuria Analyses

Table S4 provides baseline characteristics of participants with/without missing dipstick data. There were fewer differences in baseline characteristics between participants with/without available dipstick data than observed with UACR data. People with missing dipstick data were more likely to have missing eGFR data, with higher median eGFR was higher among people with missing dipstick data who have available eGFR data. People with missing dipstick data were less likely to have diabetes and hypertension.

A total of 110 430 (42.2%) people had a semiquantitative estimate of proteinuria via urinary dipstick occurring at a median of 64 (interquartile range, 12–176) days before the index date. Of these, 34 719 (32.5%) people had trace proteinuria or greater. Their baseline characteristics are listed in Table 3. The median age was lower in people with 3+ proteinuria relative to no proteinuria. Higher grade proteinuria was positively associated with male sex, residence in lower income neighborhoods, prior bleeding, and presence of CHA₂DS₂VASC risk factors.

Table 3.

Baseline Characteristics of the Dipstick Proteinuria Cohort

	Negative	Trace	1+	2+	≥3+	P value for trend
Sample size	N=75 711	N=9334	N=13 042	N=9416	N=2927
Age, y, median (Q1–Q3)	78.0 (72.0–84.0)	80.0 (74.0–86.0)	80.0 (74.0–86.0)	80.0 (73.0–85.0)	78.0 (72.0–84.0)	<0.0001
Female sex, n (%)	39 076 (51.6%)	4327 (46.4%)	5650 (43.3%)	3780 (40.1%)	1104 (37.7%)	<0.0001
Income quintile
1, n (%)	15 305 (20.2%)	2120 (22.7%)	2975 (22.8%)	2225 (23.6%)	678 (23.2%)	<0.0001
2, n (%)	15 946 (21.1%)	2045 (21.9%)	2859 (21.9%)	2117 (22.5%)	652 (22.3%)
3, n (%)	15 146 (20.0%)	1867 (20.0%)	2533 (19.4%)	1862 (19.8%)	593 (20.3%)
4, n (%)	14 410 (19.0%)	1673 (17.9%)	2364 (18.1%)	1624 (17.2%)	511 (17.5%)
5, n (%)	14 743 (19.5%)	1614 (17.3%)	2284 (17.5%)	1564 (16.6%)	485 (16.6%)
Diabetes, n (%)	26 633 (35.2%)	3981 (42.7%)	6006 (46.1%)	4985 (52.9%)	1790 (61.2%)	<0.0001
Hypertension, n (%)	64 282 (84.9%)	8189 (87.7%)	11 627 (89.2%)	8540 (90.7%)	2725 (93.1%)	<0.0001
Congestive heart failure, n (%)	19 289 (25.5%)	3170 (34.0%)	4676 (35.9%)	3895 (41.4%)	1395 (47.7%)	<0.0001
Ischemic heart disease, n (%)	24 654 (32.6%)	3111 (33.3%)	4429 (34.0%)	3356 (35.6%)	1029 (35.2%)	<0.0001
Peripheral vascular disease, n (%)	1488 (2.0%)	285 (3.1%)	435 (3.3%)	438 (4.7%)	229 (7.8%)	<0.0001
Stroke or transient ischemic attack, n (%)	2367 (3.1%)	409 (4.4%)	517 (4.0%)	465 (4.9%)	113 (3.9%)	<0.0001
Prior bleeding diagnoses in‐hospital, n (%)	8442 (11.2%)	1224 (13.1%)	1902 (14.6%)	1491 (15.8%)	577 (19.7%)	<0.0001
Chronic kidney disease stage, n (%)
Stages 1 and 2, n (%)	48 102 (63.5%)	5040 (54.0%)	6372 (48.9%)	3590 (38.1%)	791 (27.0%)	<0.0001
Stage 3, n (%)	24 318 (32.1%)	3499 (37.5%)	4959 (38.0%)	3680 (39.1%)	1026 (35.1%)
Stage 4, n (%)	3004 (4.0%)	654 (7.0%)	1337 (10.3%)	1374 (14.6%)	545 (18.6%)
Stage 5 or dialysis, n (%)	287 (0.4%)	141 (1.5%)	374 (2.9%)	772 (8.2%)	565 (19.3%)
Estimated glomerular filtration rate value (1 y before index), median (Q1–Q3)	67.8 (52.5–81.6)	62.5 (45.8–79.4)	59.2 (41.2–77.5)	51.0 (31.9–72.0)	39.3 (20.0–62.6)	<0.0001
Medication use 6 mo before index
Angiotensin‐converting enzyme inhibitor, n (%)	27 459 (36.3%)	3519 (37.7%)	4993 (38.3%)	3596 (38.2%)	1088 (37.2%)	<0.0001
Angiotensin receptor blocker, n (%)	20 203 (26.7%)	2504 (26.8%)	3533 (27.1%)	2727 (29.0%)	1007 (34.4%)	<0.0001
Warfarin, n (%)	12 251 (16.2%)	1435 (15.4%)	2210 (16.9%)	1497 (15.9%)	469 (16.0%)	0.827
Direct oral anticoagulant, n (%)	12 890 (17.0%)	1611 (17.3%)	2269 (17.4%)	1519 (16.1%)	387 (13.2%)	0.0005
Loop diuretics, n (%)	16 748 (22.1%)	2509 (26.9%)	3738 (28.7%)	3146 (33.4%)	1221 (41.7%)	<0.0001
Thiazide diuretics, n (%)	22 602 (29.9%)	2693 (28.9%)	3618 (27.7%)	2607 (27.7%)	778 (26.6%)	<0.0001
Mineralocorticoid receptor antagonist, n (%)	3257 (4.3%)	425 (4.6%)	583 (4.5%)	492 (5.2%)	152 (5.2%)	<0.0001
Angiotensin receptor‐neprilysin inhibitor, n (%)	73 (0.1%)	12 (0.1%)	15 (0.1%)	9 (0.1%)	8 (0.3%)	0.0842
D between dipstick protein test and index date, median (Q1–Q3)	86.0 (19.0–197.0)	23.0 (5.0–118.0)	32.0 (6.0–127.0)	23.0 (5.0–107.0)	25.0 (5.0–104.0)	<0.0001

Table S2 provides crude outcome data for people with/without missing dipstick data. People with missing dipstick data had lower crude incidence of death, bleeding, and HF but comparable crude stroke incidence to people with available dipstick data. Among people with available dipstick data, higher semiquantitative proteinuria estimates by dipstick were associated with a graded increase in the crude incidence of adverse outcomes, as illustrated in Table S3. After adjusting for baseline patient characteristics and time‐varying anticoagulation status, increasing dipstick proteinuria was associated with significantly higher hazard for all outcomes except for stroke, where the hazard was significantly increased only at >1+ proteinuria (Figure 2).

Figure 2. Association between dipstick proteinuria and stroke, bleeding, hospitalization for heart failure, and death.

The hazard ratio for each outcome associated with each level of urine dipstick proteinuria from a cause‐specific regression model accounting for age, sex, heart failure, hypertension, diabetes, prior nonischemic stroke or transient ischemic attack, vascular disease, estimated glomerular filtration rate, prior bleeding, baseline statin use, baseline use of blockers of the renin‐angiotensin system, and time‐varying anticoagulation status. HR indicates hazard ratio.

The findings from the stratified analysis by baseline diabetes status are summarized in Figure S4. Higher levels of dipstick proteinuria were positively associated with the risk of adverse outcomes regardless of diabetes status, but there was a statistically significant interaction by diabetes status for death (P<0.001) and to a lesser extent for bleeding (P=0.001). Accordingly, the HR for mortality and for bleeding associated with increasing dipstick levels was significantly higher in people without diabetes relative to people without diabetes. There was no meaningful difference in the HR for other outcomes between people with/without diabetes.

DISCUSSION

We studied a population‐based cohort of individuals with newly recognized AF to quantify the association of albuminuria with HF, stroke, bleeding, and death. We observed a prominent increase in the hazard of all 4 adverse outcomes as the UACR increased to 60 mg/g, with minimal increase in the hazard of death at higher UACR levels and less rapid increase in the hazard of other outcomes. The magnitude of the association between a UACR of 30 mg/g (the threshold for microalbuminuria) versus 5 mg/g and subsequent HF was higher than all CHA₂DS₂‐VASc risk factors other than age and preexisting HF. We also demonstrated that semiquantitative assessments of albuminuria using dipstick proteinuria are significantly associated with all adverse outcomes studied, after adjustment for shared risk factors and prior stroke, HF, or bleeding diagnoses. The HRs for death and HF were higher in people with lower CHA₂DS₂VASC scores and the HRs for death and bleeding were higher in people without diabetes, indicating higher utility of albuminuria for risk stratification in individuals otherwise considered to be at lower risk. The study findings are summarized in Figure 3.

Figure 3. Summary of study findings.

Association between albuminuria, measured as either UACR or with urine dipstick, and the risk of HF, stroke, bleeding, and death in patients with AF. AF indicates atrial fibrillation; HR, hazard ratio; and UACR, urine albumin‐to‐creatinine ratio.

There are multiple pathophysiological factors that may contribute to the development of albuminuria, which may complicate its use as a risk marker in highly comorbid conditions such as AF. For instance, the presence of albuminuria in people with AF may be due to common risk factors such as hypertension, HF, diabetes, and cardiovascular disease. Beyond these traditional risk factors, systemic inflammation—results in both kidney injury through podocytopathy and mesangial cell proliferation and cardiac injury through atrial remodeling and fibrosis—may also be a pathophysiological link for the development of both albuminuria and AF, as well as for the subsequent risk of HF and cardiovascular disease. ²⁷ Therefore, although albuminuria can serve as a marker of systemic vascular damage or microvascular disease, it is important to have a more detailed assessment of how the association of albuminuria and clinical outcomes may vary by clinical context and comorbidities. Our analysis specifically assessed the association between albuminuria and clinical outcomes in people with AF, with consideration for relevant comorbidities and treatment strategies employed in AF.

We hypothesized that albuminuria would also be useful for prognostication in the setting of AF for several reasons. First, albuminuria and AF share common risk factors such as hypertension, HF, diabetes, and systemic inflammation. Indeed, albuminuria is well established as an independent predictor of incident AF, ²⁸ , ²⁹ particularly for people with preserved eGFR. ²⁸ This makes trace or higher levels of albuminuria a common finding among people with AF. In our cohort of older people with newly diagnosed AF, 31.4% of people with available urine dipstick data had ≥1+ proteinuria. Admittedly, this overestimates the prevalence of albuminuria since the testing would have been done for suspicion of renal or urologic disease and dipstick proteinuria also include albumin and nonalbumin urinary protein. Nonetheless, it highlights the potential yield of testing for albuminuria in older people with newly diagnosed AF.

Second, older people with AF are at higher risk of the adverse cardiovascular outcomes that are associated with albuminuria: HF, stroke, and cardiovascular mortality. ⁴ , ¹³ , ³⁰ , ³¹ , ³² Although stroke is well recognized as a focus of care for people with AF, it is less appreciated that contemporary cohorts of people with AF are at higher risks of HF than they are of stroke. ⁴ , ³³ , ³⁴ , ³⁵ , ³⁶ It is also important to identify people at higher risk of bleeding after initiating anticoagulation, because it is associated with poor outcomes for people with AF. ³⁷ We hypothesized that the clustering of AF and albuminuria would result in a clinically relevant increase in the risk of adverse outcomes associated with both disease states. We demonstrated that a UACR level of 300 mg/g versus 5 mg/g had a higher magnitude of association with death, HF, stroke, and bleeding than any CHA₂DS₂VASc risk factor in our regression model except for older age plus established HF (for HF hospital presentations and death), prior stroke (for stroke), and prior bleeding (for bleeding).

Importantly, we observed an interaction between UACR and the CHA₂DS₂VASC score for the outcomes of HF and death; with a stronger association between UACR and these outcomes in people without diabetes compared with diabetes and in people with a CHA₂DS₂VASC score of 1 to 3 versus 4 to 8. These findings likely reflect the varied causes of albuminuria, which can affect its use as a risk marker. The presence of albuminuria suggests early endothelial dysfunction or podocyte or glomerular injury or may alternatively be due to other medical comorbidities. As people with a low CHA₂DS₂‐VASc score are healthier, there are fewer comorbidities that may lead to albuminuria and the association observed in this population more reflective of an independent risk marker. In contrast, in people with a high CHA₂DS₂‐VASc score, albuminuria may also be due to end organ damage from diabetes, hypertension, ischemic nephropathy, or other forms of kidney injury that may not be independently associated with clinical outcomes. The more varied causes of UACR in people with a high CHA₂DS₂‐VASc score results in random misclassification of participants with albuminuria and high baseline risk, ultimately diluting the association between UACR and clinical outcomes. People with lower CHA₂DS₂VASC are healthier and thus have a greater potential for avoidable morbidity and mortality while being more likely to tolerate interventions geared at reducing HF risk. This suggests that UACR may be an important early marker of risk and may inform the use of interventions aimed at prevention or early diagnosis of HF.

Our findings build on prior studies on albuminuria in people with AF. In an analysis of 10 908 patients with AF in the ATRIA study, albuminuria increased the risk of thromboembolism (relative risk, 1.54 [95% CI, 1.29–1.85]). ³⁸ However, albuminuria was quantified on the basis of urine dipstick ≥1 versus no proteinuria and there was no assessment of whether increments in proteinuria were associated with a graded increase in thromboembolism risk. This strategy does not capture the increase in risk with even trace proteinuria on dipstick or the nonlinear relationship between albuminuria and clinical outcomes in people with AF. Our analysis shows that there is a statistically significant and substantial graded increase in stroke risk at low levels of UACR (corresponding to trace dipstick proteinuria), suggesting that risk prediction can be enhanced with more nuanced modeling of albuminuria levels.

Fewer studies have investigated the risk of adverse outcomes aside from stroke in patients with AF with albuminuria. A study of 12 403 patients with AF aged >65 years starting warfarin reported that ≥2+ urine dipstick proteinuria was associated with a risk of major bleeding after 30 days of warfarin treatment. ¹² In a small cohort of 2164 people with AF, dipstick proteinuria was associated with an increased risk of a composite of cardiovascular death and HF. However, this study included participants with established and new AF. ³⁹ Dipstick proteinuria does have limitations. For instance, although we noted a that dipstick proteinuria was associated with the clinical outcomes in our study, the gradient of risk in HF from trace to 1+ dipstick proteinuria did not align well with the rapid increase in risk that we observed with UACR levels that were below the microalbuminuria threshold (<30 mg/g). This is likely because dipstick proteinuria may not directly correlate with UACR, partly because dipstick testing provides a general classification of urine protein levels, combining UACR with other urine proteins. Further, dipstick protein tests are not standardized to creatinine to correct for dilution. Therefore, where possible, UACR should be obtained for risk assessment in people with AF.

Overall, these data suggest that albuminuria can be helpful for clinical decision making in people with newly diagnosed AF. Patients who have even minimal albuminuria should be triaged for closer follow‐up to reduce their risk of HF and bleeding in addition to their risk of stroke. The coexistence of AF and albuminuria may identify a high‐risk subgroup that may also derive greater absolute benefit from interventions for cardiovascular risk reduction that are earmarked for people with kidney disease. For instance, people with albuminuria may stand to benefit from better blood pressure control, given its association with HF and stroke in AF, as well as other adverse cardiovascular disease and renal outcomes (particularly in those with albuminuria). ⁴⁰ , ⁴¹ , ⁴² , ⁴³ There is a strong association between baseline albuminuria and HF, and albuminuria reduction is also associated with reduced HF risk. ⁴⁴ Given the impact of blockers of the renin‐angiotensin system on reducing albuminuria, ⁴² , ⁴⁵ and the benefit of sodium glucose cotransporter 2 inhibitors and finerenone in preventing HF and preserving renal function, ⁴⁶ , ⁴⁷ , ⁴⁸ these interventions merit particular consideration in this higher‐risk patient group.

Limitations

As an observational study, we cannot exclude residual confounding of the association between albuminuria and the adverse outcomes studied. Our analysis was limited to people who had measurements of creatinine and albuminuria within a year before AF diagnosis and to adults aged ≥66 years. We observed notable differences between people with/without available UACR data. Thus, our inclusion criteria mean that our findings may not be generalizable to younger patients with less comorbidities and who have a lower likelihood of cardio‐renal disease. The measurements of albuminuria were not standardized, and we selected the most recent urinary dipstick or UACR result available. We used a single measurement of UACR and eGFR, which does not exclude transient fluctuations in UACR or eGFR. This study was not conducted in a manner that allows the use of UACR for risk prediction, which is beyond the scope of the planned analyses. Finally, we could not access important clinical variables such as AF burden, pattern of AF (paroxysmal or permanent), ejection fraction, blood pressure, smoking, or natriuretic peptide levels.

Conclusions

Among older people with newly recognized AF, the presence of albuminuria as measured by UACR was associated with a substantial increase in the rate of stroke and bleeding hospitalizations, HF hospital presentations, and death. There was a clinically relevant increase in risk even at low levels of albuminuria that was independent of other CHA₂DS₂VASC risk factors and prior bleeding. A UACR of 300 mg/g versus 5 mg/g had a stronger association with all 4 outcomes than most CHA₂DS₂VASc risk factors. Accordingly, albuminuria can be considered as a component of risk evaluation for most people with newly diagnosed AF for further prognostication of their risk of bleeding, HF, and death.

Sources of Funding

This study was funded by the Heart and Stroke Foundation of Canada Polo Award, the Canadian Cardiovascular Society Atrial Fibrillation Research Award, and a Tier 2 Canada Research Chair in Cardiovascular Disease Epidemiology and Outcomes (to Dr Abdel‐Qadir). The funding sources had no role in the conduct of the study, the decision to publish, or the preparation of the article.

Disclosures

No relevant conflicts of interest.

Author Contributions

Dr Abdel‐Qadir had full access to all data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Concept and design: Odutayo, Abdel‐Qadir. Acquisition, analysis, or interpretation of data: All authors. Drafting of the article: Odutayo, Abdel‐Qadir. Critical revision of the article for important intellectual content: All authors. Statistical analysis: Odutayo, Fang, Abdel‐Qadir. Obtained funding: Abdel‐Qadir. Administrative, technical, or material support: Abdel‐Qadir. Supervision: Abdel‐Qadir.

Supporting information

Tables S1–S4

Figures S1–S4