Hemoglobin Decline After Oral Anticoagulation Initiation and Subsequent Cancer Detection in Patients With Atrial Fibrillation

Abstract

Background

Previous studies evaluating cancer risk during oral anticoagulation (OAC) have focused on overt bleeding events, whereas the significance of hemoglobin (Hb) decline in the absence of clinically overt bleeding events after OAC initiation has not been evaluated.

Objectives

The purpose of this study was to evaluate the association between Hb decline after OAC initiation in the absence of clinically overt bleeding events and subsequent cancer risk in patients with atrial fibrillation (AF).

Methods

Using the Korean National Health Insurance Service database linked to serial health examinations, 6,789 patients with AF were identified who underwent 2 examinations within 1 year before and after OAC initiation. Patients with prior cancer, end-stage renal disease, overt bleeding, or early OAC discontinuation were excluded. Patients were classified by Hb change (Hb decrease ≥2 g/dL vs no significant decrease). The primary outcome was incident cancer; secondary outcomes included site-specific cancers, all-cause mortality, thromboembolic events, and bleeding outcomes.

Results

During a median follow-up period of 10.6 months, patients with Hb decreases ≥2 g/dL had a higher incidence of overall cancer than those without (6.7 vs 4.3 per 100 person-years; adjusted HR: 1.42; 95% CI: 1.10-1.97). The association was particularly evident for gastrointestinal cancers (adjusted HR: 1.78; 95% CI: 1.12-2.83). Cancer risk associated with Hb declines was greater in older patients, those with chronic kidney disease, and those receiving antiplatelet therapy (P for interaction < 0.05). Hb decline was also associated with higher all-cause mortality and major bleeding, with no differences in stroke.

Conclusions

In patients with AF initiating OAC, Hb decline ≥2 g/dL in the absence of clinically overt bleeding events is associated with increased cancer risk, particularly gastrointestinal malignancy, and higher mortality.

Central Illustration

Atrial fibrillation (AF) is the most common sustained cardiac arrhythmia and is associated with substantial risks for stroke, systemic embolism, and mortality. Oral anticoagulation (OAC) is a cornerstone of stroke prevention in patients with AF and is widely prescribed in contemporary clinical practice.^1,2 Although OAC effectively reduces thromboembolic risk, bleeding complications remain a major clinical concern and often influence long-term treatment decisions.^3,4

Previous studies evaluating cancer risk in patients receiving OAC have focused largely on overt bleeding events, particularly gastrointestinal bleeding, as clinical triggers leading to diagnostic evaluation and subsequent cancer detection.5, 6, 7 In this context, anticoagulation-related bleeding has been considered a mechanism by which previously unrecognized malignancies are “unmasked.”⁸ However, such studies have primarily addressed clinically overt bleeding events and have not systematically examined whether changes in hemoglobin (Hb) levels after OAC initiation, in the absence of clinically overt bleeding events, carry prognostic significance for future cancer risk.

In routine practice, a decline in Hb following OAC initiation is frequently encountered, even without documented bleeding events. Such Hb declines may reflect occult blood loss or underlying pathology that does not reach the threshold of clinically overt bleeding.⁹ Despite its common occurrence, the clinical implications of OAC-related Hb decline, particularly its potential role as an early signal of occult malignancy, remain poorly defined in patients with AF. Importantly, distinguishing whether Hb decline represents a benign laboratory fluctuation or a marker of clinically meaningful disease has direct implications for patient evaluation and surveillance.

Therefore, using a nationwide population-based cohort with linked serial health examination data, we evaluated the association between OAC-related Hb decline and subsequent cancer risk in patients with AF, focusing on cancer incidence overall and by cancer subtype. We also examined the relationships between Hb decline and other clinically relevant outcomes, including mortality, thromboembolic events, and bleeding complications, to clarify the clinical implications of Hb changes after OAC initiation.

Methods

Data source

This study was conducted using the Korean National Health Insurance Service (NHIS) database, which covers the entire Korean population as a single-payer, nationwide health insurance system.¹⁰ The NHIS database contains comprehensive information on demographics, diagnoses, procedures, prescriptions, and mortality. These data were linked to the National Health Examination database, which includes standardized laboratory measurements and clinical variables obtained during biennial health examinations.¹¹ The linkage enabled longitudinal assessment of Hb levels before and after initiation of OAC.

The study protocol was approved by the Institutional Review Board of Yonsei University Health System (4-2025-0273). The requirement to obtain informed consent was waived given that anonymized administrative data were used.

Study population

We identified 43,410 patients with AF, defined using International Classification of Diseases-10th Revision (ICD-10) codes I48.x on the basis of previously validated definitions in the NHIS database,¹⁰ who newly initiated OAC during the study period (2011-2020). Among these patients, 10,261 underwent 2 consecutive National Health Examinations within 1 year before and 1 year after OAC initiation. The index date was defined as the date of the second health examination after OAC initiation. The overall study design and timeline, including the timing of health examinations relative to OAC initiation and the definition of the index date, are depicted in the Central Illustration.

Central Illustration.

Hb Decline After OAC Initiation and Cancer Risk in Patients With AF

A hemoglobin (Hb) decline ≥2 g/dL after oral anticoagulation (OAC) initiation in the absence of clinically overt bleeding events was associated with increased risk for subsequent cancer, particularly gastrointestinal (GI) malignancy. AF = atrial fibrillation; CKD = chronic kidney disease; NHIS = National Health Insurance Service.

Patients were excluded if they had histories of any malignancy (n = 2,253), clinically overt bleeding events between the first and second health examinations (n = 388), discontinuation of OAC before the index date (n = 824), end-stage renal disease (n = 5), or known coagulation and platelet disorders (n = 2). Clinically overt bleeding events were identified using hospitalization-based claims, with detailed ICD-10 codes provided in Supplemental Table 1. After applying these criteria, the final study population consisted of 6,789 patients (Figure 1).

Figure 1.

Study Design and Cohort Selection

Flow diagram of the study population showing inclusion and exclusion criteria and derivation of the final analytical cohort. AF = atrial fibrillation; Hb = hemoglobin; OAC = oral anticoagulation.

Hb decline after OAC initiation

Hb levels from the 2 health examinations were used to assess changes in Hb after OAC initiation. Hb change was calculated as the difference between Hb levels at the second and first examinations. A decline in Hb of ≥2 g/dL was selected to represent a clinically meaningful change.^12,13 This threshold was chosen because a decrease in Hb of this magnitude exceeds minor laboratory variability and is commonly used in clinical and research settings to indicate a significant Hb reduction in bleeding-related outcome definitions. Given that the aim of the present study was to evaluate a clinically relevant Hb decline after the initiation of OAC, the ≥2 g/dL threshold was considered pragmatic and readily interpretable for clinicians. On the basis of this threshold, patients were categorized into groups with Hb decrease ≥2 g/dL and with no significant Hb decrease (including those with declines of <2 g/dL, stable Hb, or increased Hb).

Covariates

Covariates were assessed at the time of the second health examination (index date) and prespecified according to clinical relevance. Because the second health examination served as the index date from which outcome follow-up began, covariates measured at this time were used to characterize the patients’ clinical status at the start of follow-up, when Hb decline after OAC initiation had become observable.

Clinical comorbidities were identified using ICD-10 codes from inpatient and outpatient claims records, on the basis of validated definitions in the Korean NHIS database.¹⁰ Comorbid conditions included heart failure, hypertension, diabetes mellitus, prior stroke or transient ischemic attack, prior myocardial infarction, chronic kidney disease (CKD), chronic obstructive pulmonary disease (COPD), and history of liver disease. Detailed definitions and diagnostic codes used to identify comorbidities are provided in Supplemental Table 1.

Laboratory, anthropometric, and lifestyle variables obtained from the health examination included body mass index, systolic and diastolic blood pressure, Hb, estimated glomerular filtration rate, total cholesterol, smoking status, and alcohol consumption.

Medication use was determined from prescription claims records and defined as exposure to the medication for at least 30 days between OAC initiation and the index date. Concomitant medications included antiplatelet agents, renin-angiotensin system blockers, and proton pump inhibitors.

Outcomes

The primary outcome was incident cancer diagnosed after the index date. Cancer diagnoses were identified using ICD-10 codes for malignant neoplasms recorded in inpatient and outpatient claims during follow-up. Patients with any history of malignancy before the index date were excluded to ensure ascertainment of newly diagnosed cancer.

Secondary outcomes included site-specific cancers, all-cause death, ischemic stroke, and bleeding events. Cancer subtypes were categorized as gastrointestinal; genitourinary; breast, lung, or thoracic; hematologic; thyroid; and other cancers on the basis of ICD-10 diagnosis codes. Major bleeding was defined as intracranial hemorrhage, gastrointestinal bleeding, or clinically overt bleeding requiring blood transfusion.¹² Detailed definitions and diagnostic codes for cancer outcomes and other clinical events are provided in Supplemental Table 2.

Patients were followed from the index date until the occurrence of the outcome of interest, death, or the end of the study period, whichever came first. Longitudinal OAC use after the index date was assessed using prescription claims and categorized according to the proportion of days covered, with a proportion ≥80% defined as adherent use on the basis of prior claims-based studies in AF patients.¹⁴ OAC discontinuation was also identified from prescription claims during follow-up.

Statistical analysis

Baseline characteristics were summarized according to Hb decline group using descriptive statistics. Continuous variables are expressed as mean ± SD or median (Q1-Q3), as appropriate, and categorical variables as counts with percentages. Between-group comparisons were performed using Student’s t-test for normally distributed continuous variables, the Wilcoxon rank sum test for skewed continuous variables, and the chi-square test for categorical variables.

For the primary analysis, time-to-event outcomes were evaluated from the index date. Incidence rates were calculated as events per 100 person-years. Cause-specific Cox proportional hazards regression models were used to estimate HRs and 95% CIs for the association between Hb decline after OAC initiation and subsequent clinical outcomes. Because Hb decline after OAC initiation was an observed clinical signal rather than an assigned treatment or intervention, a multivariable-adjusted Cox model was selected as the primary analysis to estimate its association with subsequent outcomes in the overall study population. Multivariable cause-specific Cox models were adjusted for prespecified covariates on the basis of clinical relevance. The covariates included in the adjusted models are listed in the footnotes of the corresponding tables. Kaplan-Meier survival curves were used for all-cause death and compared using the log-rank test, whereas cumulative incidence curves were used for cancer and other nonfatal outcomes and compared using Gray’s test.

Sensitivity analyses were conducted to assess the robustness of the primary findings. First, competing risk regression using the Fine-Gray method was performed for incident cancer outcomes, treating all-cause death as a competing event with results presented as the subdistribution HR (sHR) with its 95% CI. Second, propensity scores for Hb decline were estimated using logistic regression with the same clinically relevant covariates used in the primary multivariable-adjusted model. These covariates were prespecified on the basis of clinical relevance and were not selected solely according to between-group differences. Propensity score overlap weighting was then applied to improve covariate balance between groups.¹⁵ After overlap weighting, covariate balance was assessed using standardized mean differences, with an absolute standardized mean difference <0.1 considered indicative of adequate balance (Supplemental Figure 1). As an additional sensitivity analysis, Hb change was modeled as a continuous variable using a multivariable Cox proportional hazards model with a piecewise linear spline function and a knot at Hb change of 0 g/dL to evaluate differential associations with incident cancer risk for Hb decrease vs increase. Finally, to address potential ascertainment bias from cancers diagnosed shortly after the index date, we conducted an additional sensitivity analysis excluding individuals diagnosed with cancer within 3 months after the index date. The same Cox proportional hazards models used in the primary analysis were applied.

To evaluate whether the association between Hb decline and incident cancer was dependent on the selected cutoff, we performed additional threshold-based sensitivity analyses using alternative Hb decline thresholds of ≥1.0, ≥1.5, and ≥3.0 g/dL, in addition to the primary threshold of ≥2.0 g/dL. In each analysis, patients were classified according to whether they met the corresponding Hb decline threshold, and Cox proportional hazards models were fitted using the same covariates as in the primary analysis.

Prespecified subgroup analyses were performed across clinically relevant subgroups, such as age, sex, major comorbidities, and concomitant medication use, with effect modification assessed using interaction terms.

A 2-sided P value <0.05 was considered to indicate statistical significance. All analyses were performed using R version 4.5.2 (R Core Team).

Results

Study population and baseline characteristics

The final study population consisted of 6,789 patients with AF who newly initiated OAC and had available serial health examination data. Of these, 529 patients (7.8%) were classified into the Hb decrease ≥2 g/dL group and 6,260 (92.2%) into the no significant Hb decrease group (Figure 1).

Baseline characteristics by Hb decline group are presented in Table 1. Baseline Hb levels were slightly higher in the Hb decline ≥2 g/dL group (15.2 ± 1.6 g/dL vs 14.2 ± 1.6 g/dL), whereas follow-up Hb levels were lower (12.4 ± 1.8 g/dL vs 14.3 ± 1.6 g/dL). Patients in the Hb decline ≥2 g/dL group were more often male and had higher prevalences of hypertension, diabetes mellitus, CKD, and COPD compared with the no significant Hb decline group. Use of antiplatelet agents, renin-angiotensin system blockers, and proton pump inhibitors was also more frequent in the Hb decline ≥2 g/dL group. The median follow-up duration from the index date was 10.6 months (Q1-Q3: 4.7-22.8 months) and was similar between the 2 groups.

Table 1.

Baseline Characteristics According to Hb Change After Anticoagulant Initiation

	Overall (N = 6,789)	Hb Decrease ≥2 g/dL (n = 529)	No Significant Hb Decrease (n = 6,260)	P Value
Age, y	66.5 ± 10.2	67.1 ± 10.7	66.4 ± 10.1	0.11
<65 y	2,579 (38.0)	182 (34.4)	2,397 (38.3)	0.050
65-74 y	2,695 (39.7)	208 (39.3)	2,487 (39.7)
≥75 y	1,515 (22.3)	139 (26.3)	1,376 (22.0)
Male	4,393 (64.7)	389 (73.5)	4,004 (64.0)	<0.001
Follow-up duration, mo	10.6 (4.7-22.8)	11.9 (5.3-24.1)	10.6 (4.7-22.6)	0.079
Body weight, kg	66.7 ± 11.8	66.7 ± 12.5	66.7 ± 11.8	0.98
BMI, kg/m2	25.0 ± 3.3	24.8 ± 3.5	25.1 ± 3.3	0.12
<25 kg/m2	3,482 (51.3)	284 (53.7)	3,198 (51.1)	0.54
25-30 kg/m2	2,798 (41.3)	206 (39.1)	2,592 (41.4)
≥30 kg/m2	509 (7.5)	39 (7.4)	470 (7.5)
Smoking				0.44
Current smoker	731 (10.8)	49 (9.3)	682 (10.9)
Former smoker	1,992 (29.3)	163 (30.8)	1,829 (29.2)
Never smoker	4,066 (59.9)	317 (59.9)	3,749 (59.9)
Alcohol				0.55
≥3 drinks/wk	808 (11.9)	58 (11.0)	750 (12.0)
1 or 2 drinks/wk	1,272 (18.7)	93 (17.6)	1,179 (18.8)
Nondrinker	4,709 (69.4)	378 (71.5)	4,331 (69.2)
CHA2DS2-VA score	4 (2-5)	4 (3-5)	3 (2-5)	<0.001
1	746 (11.0)	44 (8.3)	702 (11.2)	<0.001
2 or 3	2,626 (38.7)	174 (32.9)	2,452 (39.2)
≥4	3,417 (50.3)	311 (58.8)	3,106 (49.6)
Heart failure	3,836 (56.5)	304 (57.5)	3,532 (56.4)	0.68
Hypertension	6,056 (89.2)	497 (94.0)	5,559 (88.8)	<0.001
Diabetes mellitus	1,816 (26.7)	190 (35.9)	1,626 (26.0)	<0.001
Prior stroke/TIA	2,606 (38.4)	221 (41.8)	2,385 (38.1)	0.10
Prior myocardial infarction	586 (8.6)	53 (10.0)	533 (8.5)	0.27
Vascular disease	1,627 (24.0)	136 (25.7)	1,491 (23.8)	0.36
CKD	1,637 (24.1)	159 (30.1)	1,478 (23.6)	0.001
COPD	1,989 (29.3)	180 (34.0)	1,809 (28.9)	0.015
Dyslipidemia	6,099 (89.8)	482 (91.1)	5,617 (89.7)	0.35
Liver disease	3,135 (46.2)	264 (49.9)	2,871 (45.9)	0.081
Osteoporosis	2,147 (31.6)	159 (30.1)	1,988 (31.8)	0.45
Laboratory tests
Initial Hb, g/dL	14.3 ± 1.6	15.2 ± 1.6	14.2 ± 1.6	<0.001
Follow-up Hb, g/dL	14.2 ± 1.7	12.4 ± 1.8	14.3 ± 1.6	<0.001
Hb change, g/dL	−0.2 ± 1.3	−2.8 ± 0.9	0.1 ± 1.1	<0.001
Fasting glucose, mg/dL	108.2 ± 30.0	109.0 ± 37.3	108.1 ± 29.3	0.55
Total cholesterol, mg/dL	171.3 ± 38.2	158.1 ± 36.6	172.4 ± 38.2	<0.001
HDL-C, mg/dL	51.4 ± 13.3	49.3 ± 14.2	51.6 ± 13.3	<0.001
LDL-C, mg/dL	94.6 ± 35.8	84.1 ± 31.7	95.5 ± 36.0	<0.001
eGFR, mL/min/1.73 m2	73.9 ± 18.6	72.1 ± 21.1	74.1 ± 18.4	0.018
Medications
Warfarin	2,694 (39.7)	223 (42.2)	2,471 (39.5)	0.24
DOAC	4,095 (60.3)	306 (57.8)	3,789 (60.5)
Antiplatelet agent	1,711 (25.2)	157 (29.7)	1,554 (24.8)	0.016
Statin	3,430 (50.5)	274 (51.8)	3,156 (50.4)	0.57
Beta-blocker	3,272 (48.2)	263 (49.7)	3,009 (48.1)	0.49
RAS blocker	2,893 (42.6)	255 (48.2)	2,638 (42.1)	0.008
DHP CCB	1,048 (15.4)	84 (15.9)	964 (15.4)	0.82
Digoxin	915 (13.5)	82 (15.5)	833 (13.3)	0.18
Non-DHP CCB	764 (11.3)	55 (10.4)	709 (11.3)	0.56
Class Ic AAD	1,109 (16.3)	73 (13.8)	1,036 (16.5)	0.11
Class III AAD	665 (9.8)	55 (10.4)	610 (9.7)	0.68
Proton pump inhibitor	1,575 (23.2)	149 (28.2)	1,426 (22.8)	0.006

Values are mean ± SD, n (%), or median (Q1-Q3).

AAD = antiarrhythmic drug; BMI = body mass index; CCB = calcium-channel blocker; CKD = chronic kidney disease; COPD = chronic obstructive pulmonary disease; DHP = dihydropyridine; DOAC = direct oral anticoagulant medication; eGFR = estimated glomerular filtration rate; Hb = hemoglobin; HDL-C = high-density lipoprotein cholesterol; LDL-C = low-density lipoprotein cholesterol; MI = myocardial infarction; RAS = renin-angiotensin system; TIA = transient ischemic attack.

Incidence of cancer according to Hb decline

Cumulative incidence curves showed a higher incidence of cancer in patients with Hb decreases ≥2 g/dL compared with those without significant Hb decreases (P = 0.010, Gray’s test) (Figure 2, Central Illustration). The 1-year cumulative incidence of cancer was 7.2% (95% CI: 4.7%-9.8%) in the Hb decline ≥2 g/dL group and 4.0% (95% CI: 3.4%-4.6%) in the no significant Hb decline group. The separation between the 2 curves was evident early during follow-up and persisted throughout the observation period.

Figure 2.

Cumulative Incidence of Cancer According to Hb Decline

Cumulative incidence curves showing the incidence of cancer in patients with hemoglobin (Hb) decreases ≥2 g/dL and those without significant Hb decreases.

In multivariable Cox proportional hazards models, a decline in Hb ≥2 g/dL was associated with a significantly higher risk for cancer (adjusted HR: 1.42; 95% CI: 1.10-1.97; P = 0.038). The corresponding incidence rates were 6.7 and 4.3 per 100 person-years in the Hb decline ≥2 g/dL and no significant Hb decline groups, respectively, with an absolute rate difference of 2.4 (95% CI: 0.3-4.5) per 100 person-years (Table 2).

Table 2.

Risk for Incident Cancers According to Oral Anticoagulation–Related Hb Decline

Cancer Type	Hb Decrease ≥2 g/dL (n = 529)			No Significant Hb Decrease (n = 6,260)			Absolute Rate Difference per 100 PYs (95% CI)	Adjusted HRa (95% CI)	P Value
	Number of Events	PYs	Event Rate	Number of Events	PYs	Event Rate
All cancers	41	614	6.7	299	6,943	4.3	2.4 (0.3-4.5)	1.42 (1.10-1.97)	0.038
Gastrointestinal cancers	21	614	3.4	128	6,943	1.8	1.6 (0.1-3.1)	1.78 (1.12-2.83)	0.015
Genitourinary cancers	8	614	1.3	106	6,943	1.5	−0.2 (−1.2 to 0.7)	0.69 (0.33-1.42)	0.31
Breast/lung/thoracic cancers	4	614	0.7	36	6,943	0.5	0.1 (−0.5 to 0.8)	1.22 (0.43-3.47)	0.71
Hematologic cancers	2	614	0.3	4	6,943	0.1	0.3 (−0.2 to 0.7)	6.70 (1.02-43.9)	0.047
Thyroid cancers	2	614	0.3	9	6,943	0.1	0.2 (−0.3 to 0.7)	2.25 (0.44-11.4)	0.33
Other cancers	4	614	0.7	16	6,943	0.2	0.4 (−0.2 to 1.1)	2.37 (0.77-7.22)	0.13

Event rates are expressed per 100 person-years.

Hb = hemoglobin; PYs = person-years.

^aHRs were adjusted for age, sex, body mass index, smoking status, drinking frequency, heart failure, hypertension, diabetes, prior stroke or transient ischemic attack, prior myocardial infarction, chronic kidney disease, chronic obstructive pulmonary disease, liver disease, estimated glomerular filtration rate, total cholesterol, type of oral anticoagulant (direct oral anticoagulant agent vs warfarin), and medications (antiplatelet agents, renin-angiotensin system blockers, and proton pump inhibitors).

The association between a decline in Hb ≥2 g/dL and incident cancer risk was generally consistent across sensitivity analyses. In the Fine-Gray competing risk analysis, a decline in Hb ≥2 g/dL was associated with higher risks for overall incident cancer (adjusted sHR: 1.41; 95% CI: 1.01-1.96; P = 0.043) and gastrointestinal cancer (adjusted sHR: 1.76; 95% CI: 1.10-2.82; P = 0.019) (Supplemental Table 3). In the propensity score overlap-weighted analysis, a decrease in Hb ≥2 g/dL remained associated with higher risks for overall incident cancer (weighted HR: 1.43; 95% CI: 1.03-1.99; P = 0.035) and gastrointestinal cancer (weighted HR: 1.72; 95% CI: 1.08-2.75; P = 0.023), consistent with the primary multivariable-adjusted Cox analysis (Supplemental Table 4).

In an additional sensitivity analysis excluding individuals diagnosed with cancer within 3 months after the index date, the association with overall incident cancer was attenuated but remained directionally consistent with the primary analysis (adjusted HR: 1.42; 95% CI: 0.99-2.03; P = 0.054), and the association with gastrointestinal cancer remained statistically significant (adjusted HR: 1.73; 95% CI: 1.02-2.95; P = 0.043) (Supplemental Table 5).

To further examine whether the findings were dependent on the selected Hb decline cutoff, threshold-based sensitivity analyses were conducted using Hb decline thresholds of ≥1.0, ≥1.5, ≥2.0, and ≥3.0 g/dL (Supplemental Table 6). The association between Hb decline and incident overall cancer was directionally consistent across thresholds, with adjusted HRs of 1.36 (95% CI: 1.08-1.71; P = 0.009), 1.46 (95% CI: 1.12-1.91; P = 0.005), 1.42 (95% CI: 1.10-1.97; P = 0.038), and 1.41 (95% CI: 0.82-2.42; P = 0.21), respectively. Event rates were consistently higher among patients meeting each Hb decline threshold than among those not meeting the threshold, and the absolute rate difference tended to increase with larger Hb decline thresholds.

Site-specific cancer outcomes

Cumulative incidence curves showed higher incidences of gastrointestinal and hematologic cancers in patients with Hb decreases ≥2 g/dL compared with those without significant Hb decreases (P = 0.003 and P = 0.011, respectively, Gray’s test) (Figure 3). No significant differences in cumulative incidence were observed for genitourinary cancers; breast, lung, or thoracic cancers; and thyroid cancers.

Figure 3.

Site-Specific Cancer Outcomes According to Hb Decline

Cumulative incidence curves for site-specific cancers comparing patients with hemoglobin (Hb) decreases ≥2 g/dL and those without significant Hb decreases. Panels show the cumulative incidence of (A) gastrointestinal; (B) genitourinary; (C) breast, lung, and thoracic; (D) hematologic; (E) thyroid; and (F) other cancers.

In adjusted Cox proportional hazards models, a decline in Hb ≥2 g/dL was associated with a significantly higher risk for gastrointestinal cancers (adjusted HR: 1.78; 95% CI: 1.12-2.83; P = 0.015) (Table 2). For hematologic cancers, although a higher risk was observed, the estimates were imprecise because of the small number of events. No significant associations were observed for the other cancer subtypes.

Results for site-specific cancer outcomes were directionally consistent in competing risk and propensity score–weighted analyses (Supplemental Tables 3 and 4). The proportional distribution of cancer subtypes according to Hb change is shown in Supplemental Figure 2.

Continuous analysis of Hb change

When Hb change was analyzed as a continuous variable using a multivariable Cox proportional hazards model with a piecewise linear spline function and a knot at Hb change of 0 g/dL, a decrease in Hb was significantly associated with a higher risk for cancer (adjusted HR per 1 g/dL decrease: 1.16; 95% CI: 1.04-1.30; P = 0.008), whereas no association was observed for Hb increase (adjusted HR per 1 g/dL increase: 1.00; 95% CI: 0.85-1.18; P = 0.99) (Figure 4, Central Illustration). The association between Hb change and cancer risk was confined to the decremental component of Hb change.

Figure 4.

Continuous Association Between Hb Change and Cancer Risk

Association between continuous hemoglobin (Hb) change and the risk for incident cancer estimated using a multivariable Cox proportional hazards model with a piecewise linear spline function and a knot at Hb change of 0 g/dL.

Subgroup analyses

Prespecified subgroup analyses of the association between a decrease in Hb ≥2 g/dL and incident cancer are shown in Figure 5. The association was generally consistent across clinically relevant subgroups. Significant effect modification was observed by age and CKD status, with a stronger association among patients aged ≥75 years (P for interaction = 0.009) and those with CKD (P for interaction = 0.012). A stronger association was also observed among patients receiving concomitant antiplatelet therapy (P for interaction = 0.041). No significant effect modification was observed according to sex, smoking status, alcohol consumption, CHA₂DS₂-VA score, heart failure, hypertension, diabetes mellitus, COPD, liver disease, type of OAC, or proton pump inhibitor use.

Figure 5.

Subgroup Analyses of the Association Between Hb Decline and Cancer Risk

Forest plot showing adjusted HRs for incident cancer associated with Hb decrease ≥2 g/dL across prespecified subgroups. Interaction analyses were performed to assess effect modification by subgroup variables. CKD = chronic kidney disease; COPD = chronic obstructive pulmonary disease; DOAC = direct oral anticoagulant medication; PPI = proton pump inhibitor; PYs = person-years; other abbreviations as in Figure 1.

Other clinical outcomes

Associations between Hb decline and other clinical outcomes are summarized in Table 3. Patients with Hb decreases ≥2 g/dL had higher risks for all-cause death and major bleeding compared with those without significant Hb decreases. In contrast, no significant difference was observed in the risk for stroke or systemic embolism between the 2 groups. A Kaplan-Meier curve for all-cause death and cumulative incidence curves for stroke/systemic embolism and major bleeding according to Hb decline group are shown in Supplemental Figure 3.

Table 3.

Risk of Clinical Outcomes During Follow-Up According to Oral Anticoagulation–Related Hb Decline

Outcome	Hb Decrease ≥2 g/dL (n = 529)			No Significant Hb Decrease (n = 6,260)			Absolute Rate Difference per 100 PYs (95% CI)	Adjusted HRa (95% CI)	P Value
	Number of Events	PYs	Event Rate	Number of Events	PYs	Event Rate
All-cause death	20	646	3.1	94	7,209	1.3	1.8 (0.4-3.2)	1.79 (1.08-2.95)	0.023
Stroke/systemic embolism	17	627	2.7	146	7,096	2.1	0.7 (−0.7 to 2.0)	1.15 (0.69-1.91)	0.60
Major bleedingb	23	622	3.7	106	7,122	1.5	2.2 (0.7-3.7)	2.03 (1.28-3.21)	0.003
Gastrointestinal bleeding	22	625	3.5	67	7,153	0.9	2.6 (1.1-4.1)	3.12 (1.91-5.09)	<0.001
Intracranial hemorrhage	0	646	0.0	27	7,186	0.4	−0.4 (−0.5 to −0.2)	NE	—
Other clinically overt bleedingc	14	632	2.2	77	7,139	1.1	1.1 (0.0-2.3)	1.97 (1.10-3.52)	0.022

Event rates are expressed per 100 person-years.

NE = not estimable (no events in the Hb decrease ≥2 g/dL group); other abbreviations as in Table 2.

^aHRs were adjusted for age, sex, body mass index, smoking status, drinking frequency, heart failure, hypertension, diabetes, prior stroke or transient ischemic attack, prior myocardial infarction, chronic kidney disease, chronic obstructive pulmonary disease, liver disease, estimated glomerular filtration rate, total cholesterol, type of oral anticoagulant (direct oral anticoagulant agent vs warfarin), and medications (antiplatelet agents, renin-angiotensin system blockers, and proton pump inhibitors).

^bMajor bleeding was defined as intracranial hemorrhage, gastrointestinal bleeding, or other clinically overt bleeding events requiring blood transfusion.

^cOther clinically overt bleeding includes bleeding events from respiratory, urinary, or other anatomical sites not classified as gastrointestinal or intracranial bleeding.

To explore whether these outcome differences could be related to differences in anticoagulation exposure after the index date, longitudinal patterns of OAC use were also evaluated according to Hb decline status. Adequate OAC use decreased over time in both groups, from 100% at initiation to 53.4% and 57.3% by the end of follow-up in patients with and without Hb declines of ≥2 g/dL, respectively, showing broadly similar temporal patterns between the 2 groups (Supplemental Figure 4).

Discussion

In this nationwide cohort of patients with AF initiating OAC, a decline in Hb of ≥2 g/dL without clinically overt bleeding events was independently associated with an increased risk of subsequent cancer diagnosis. This association was robust across multiple analytical approaches and was most pronounced for gastrointestinal malignancies. In addition, a decline in Hb was associated with higher risks for all-cause mortality and major bleeding, whereas the risk for stroke or systemic embolism did not differ by Hb change. These findings suggest that an unexplained Hb decline after OAC initiation represents a clinically meaningful signal rather than a benign laboratory fluctuation.

Comparison with previous studies

Several large population-based studies have reported substantial increases in site-specific cancer risk following overt bleeding during OAC. For example, a nationwide Danish cohort demonstrated a several-fold increase in the risk for colorectal cancer after lower gastrointestinal bleeding in anticoagulated patients,⁶ while cohort studies from Spain and Canada similarly reported markedly elevated risks for gastrointestinal, genitourinary, and respiratory cancers following bleeding from the corresponding organ systems.^5,7

However, these studies predominantly relied on bleeding as a discrete clinical event and therefore inherently focused on patients who crossed the threshold of clinically apparent hemorrhage. As a result, the potential role of occult blood loss or gradually progressive anemia in patients without documented bleeding events has remained largely unexplored.

In routine clinical practice, a decline in Hb after initiation of OAC is frequently observed even in the absence of clinically overt bleeding events. Such Hb declines may reflect subclinical blood loss or underlying pathology that does not reach the threshold of clinically recognizable bleeding. Despite its common occurrence, the clinical significance of OAC-related Hb decline as an early signal of occult malignancy has not been well defined in patients with AF.

Accordingly, our findings demonstrate that a clinically meaningful decline in Hb after OAC initiation is independently associated with subsequent cancer risk, even in patients without clinically overt bleeding events, thereby extending prior bleeding-based observations to a subclinical stage.

Hb decline following the anticoagulation “bleeding stress test”

Hb decline after OAC initiation can be viewed as a quantitative manifestation of the anticoagulation “bleeding stress test.” Anticoagulation does not induce malignancy but lowers the hemostatic threshold, thereby facilitating blood loss from pre-existing pathologic lesions that would otherwise remain clinically silent.^8,16 The short median follow-up duration from the index date in our study further supports the interpretation that subsequent cancer diagnoses more likely reflected the unmasking of pre-existing occult malignancy, rather than de novo cancer development caused by OAC. In this setting, repeated low-grade bleeding or impaired erythropoiesis often does not lead to clinically apparent hemorrhage but can accumulate over time and manifest as a gradual decline in Hb level.

This mechanism is particularly relevant for gastrointestinal malignancies, in which chronic occult bleeding often precedes overt bleeding by months or years,¹⁷ as well as for hematologic malignancies, where anemia may arise from bone marrow involvement, chronic inflammation, or ineffective erythropoiesis rather than frank bleeding.¹⁸ The cancer subtype–specific patterns observed in our study are consistent with these biological mechanisms and support the plausibility of Hb decline as an early indicator of underlying malignancy in anticoagulated patients.

Cancer subtype–specific associations

The association between Hb decline after OAC initiation and subsequent cancer risk differed across cancer subtypes. The excess risk was confined largely to gastrointestinal malignancies, with a similar but less precise signal observed for hematologic cancers, whereas no meaningful associations were observed for other cancer types. This pattern suggests that Hb decline preferentially reflects malignancies in which early reductions in Hb occur in a clinically silent manner, such as through chronic occult blood loss or impaired erythropoiesis. In contrast, cancers that present without early effects on Hb may be less sensitively detected through changes in Hb alone. Accordingly, Hb decline after OAC initiation appears to represent selective vulnerability among specific cancer subtypes rather than a nonspecific marker of overall cancer risk.

Because the index date was defined by the second national health examination, this examination may have influenced subsequent cancer ascertainment, particularly for gastrointestinal cancers. In Korea, the national health examination program includes Hb measurement and also offers cancer screening examinations, such as upper gastrointestinal endoscopy or upper gastrointestinal series for gastric cancer and fecal occult blood testing for colorectal cancer. Thus, the index examination may have served not only as the time point for defining Hb decline, but also as an opportunity for cancer screening or for initiating additional diagnostic evaluation. Accordingly, the early separation of the cumulative incidence curves, particularly for gastrointestinal cancer, should be interpreted cautiously, as some cancers diagnosed shortly after the index date may have been identified through diagnostic processes initiated at or around the second examination.

Vulnerable clinical subgroups

The association between Hb decline after OAC initiation and subsequent cancer risk was more pronounced in specific clinical subgroups. In particular, stronger associations were observed among older patients, those with CKD, and patients receiving concomitant antiplatelet therapy. The clinical significance of Hb decline differed according to underlying vulnerability and concomitant treatment.

Several mechanisms may underlie this heterogeneity. Advanced age and CKD are associated with both increased baseline cancer risk¹⁹ and reduced physiological reserve, which may amplify the impact of subclinical blood loss or impaired erythropoiesis. Concomitant antiplatelet therapy may further lower the threshold at which minor bleeding or pathologic changes translate into measurable Hb decline.²⁰ Rather than indicating differential effects of OAC, these subgroup-specific patterns likely reflect variation in susceptibility to the downstream consequences of the anticoagulation “bleeding stress test.”.

Relationship with mortality and bleeding outcomes

Beyond cancer risk, Hb decline after OAC initiation was associated with a distinct pattern of clinical outcomes. Patients with a Hb decline experienced higher risks for all-cause mortality and major bleeding, particularly gastrointestinal bleeding, whereas the risk for stroke or systemic embolism did not differ according to a Hb change. This dissociation between bleeding and thromboembolic outcomes suggests that a Hb decline does not simply reflect inadequate anticoagulation or reduced treatment effectiveness.

Longitudinal analyses of OAC use patterns further informed the interpretation of these findings. As shown in Supplemental Figure 4, adequate OAC use decreased over time in both patient groups, with broadly similar temporal patterns. Consistent with this observation, covariate-adjusted longitudinal analyses indicated that the presence of clinically significant Hb decline did not influence the rate of decrease in adequate OAC use. These findings suggest that the higher risks for mortality and bleeding observed in patients with Hb declines are unlikely to be driven by differences in OAC use patterns.

The outcome-specific associations, when considered alongside the longitudinal analyses of OAC use patterns after the index date, suggest that Hb decline identifies patients with increased clinical vulnerability and that the observed outcome differences are unlikely to be explained by differential OAC use following Hb change. The higher mortality observed among patients with Hb decline may reflect the combined effects of underlying malignancy, bleeding susceptibility, and comorbidity burden rather than excess thromboembolic risk.

Clinical implications

The present findings have important clinical implications for the care of patients with AF receiving OAC. In routine practice, Hb declines after OAC initiation are frequently encountered and are often interpreted as treatment-related or transient laboratory findings. Our results suggest that an otherwise unexplained Hb decline warrants closer clinical attention, including confirmation of anemia or ongoing Hb decline, review of bleeding symptoms and concomitant medications, and consideration of age-appropriate or symptom-directed diagnostic evaluation. However, these findings should not be interpreted as supporting indiscriminate cancer screening for all patients initiating OAC. Because a decline in Hb may not originate from the eventual site of malignancy, it should be regarded as a nonspecific but clinically meaningful warning signal rather than a cancer-specific diagnostic marker. Given the stronger association observed for gastrointestinal malignancies, evaluation for occult gastrointestinal blood loss may be particularly relevant when clinically appropriate.

Study limitations

First, although Hb decline after OAC initiation was associated with subsequent cancer risk, the specific causes of Hb decline could not be directly determined. A Hb decline may arise from occult blood loss, impaired erythropoiesis, nutritional deficiency, or underlying chronic disease, and residual confounding related to underlying disease burden may still have influenced the observed associations.

Second, this study did not include a comparator group of individuals without AF or individuals not receiving OAC who experienced a similar degree of Hb decline. Therefore, we could not determine how much of the observed association reflects the anticoagulated AF setting vs the background risk associated with Hb decline itself.

Third, Hb decline was assessed using 2 health examinations before and after OAC initiation, limiting the ability to characterize the timing, duration, or persistence of Hb changes. More granular longitudinal Hb measurements would be needed to clarify how temporal patterns of Hb change relate to cancer incidence.

Fourth, detailed information on cancer stage at diagnosis and postindex diagnostic pathways was not available, precluding assessment of whether Hb decline was associated with earlier stage cancer detection or specific diagnostic processes.

Fifth, although the second national health examination may have influenced subsequent cancer ascertainment, our data set did not reliably capture whether each individual actually underwent upper gastrointestinal endoscopy, upper gastrointestinal series, fecal occult blood testing, or subsequent diagnostic procedures at or around the index examination. Therefore, we could not directly evaluate the effect of screening or diagnostic work-up related to the index examination on subsequent cancer detection. Although a sensitivity analysis excluding cancers diagnosed within 3 months after the index date showed directionally consistent findings, residual ascertainment bias related to the index health examination cannot be fully excluded.

Sixth, because covariates were measured at the second health examination, some variables may have reflected interval changes after OAC initiation rather than purely pre-exposure baseline characteristics. In addition, OAC adherence and persistence were assessed using prescription claims; therefore, actual medication intake could not be confirmed, and misclassification of adherence or persistence may have occurred.

Finally, this study was focused on patients who newly initiated OAC in order to more clearly capture Hb changes temporally related to OAC exposure. Although this design strengthens the interpretation of Hb decline occurring after OAC initiation, it limits generalizability to patients who experience a Hb decline later during long-term anticoagulation. Accordingly, Hb declines arising after prolonged or stable OAC use could not be evaluated in this study.

Conclusions

Among patients with AF, a significant decline in Hb (≥2 g/dL) after OAC initiation in the absence of clinically overt bleeding events was associated with an increased risk for subsequent cancer and adverse clinical outcomes, particularly gastrointestinal malignancy and mortality. Hb decline after OAC initiation represents a clinically relevant signal of underlying vulnerability, not merely a benign laboratory fluctuation. In this context, unexplained Hb declines during OAC therapy should prompt heightened clinical vigilance and consideration of appropriate diagnostic evaluation for underlying malignancy, particularly in the gastrointestinal tract. Further studies are warranted to better define the clinical significance of Hb decline during OAC therapy and its role in risk stratification.

Perspectives.

COMPETENCY IN MEDICAL KNOWLEDGE: In patients with AF initiating OAC, an otherwise unexplained decline in Hb, even in the absence of clinically overt bleeding events, may represent a clinically meaningful signal associated with subsequent cancer, particularly gastrointestinal malignancy.

TRANSLATIONAL OUTLOOK: Further studies are needed to determine whether clinically guided assessment of unexplained Hb declines after OAC initiation can facilitate earlier detection of occult malignancy and improve clinical outcomes without promoting indiscriminate cancer screening.

Data Availability

All data and materials of the National Health Insurance Service are accessible to the public at the National Health Insurance Data Sharing Service website (http://nhiss.nhis.or.kr). Applications to use the National Health Insurance Database are reviewed by the inquiry committee of research support, and once approved, raw data are provided, on payment of a fee, to the authorized researcher at several permitted sites.

Funding Support and Author Disclosures

This research was supported by grants from the Patient-Centered Clinical Research Coordinating Center, funded by the Ministry of Health and Welfare, Republic of Korea (grant RS-2026-25520638). Dr Joung has served as a speaker for Bayer, Bristol Myers Squibb/Pfizer, Medtronic, and Daiichi-Sankyo; and has received research funds from Samjin, Yuhan, Medtronic, Boston Scientific, and Abbott Korea. Dr Sung has received research funds from Yuhan (no fees have been received directly or personally). All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.