Blood Thinners Linked to Better Survival in the Very Old With Atrial Fibrillation

Adults aged 80 and older with atrial fibrillation who stayed on anticoagulant therapy had substantially lower death rates than those who never took the drugs, according to a large retrospective cohort study published in the Journal of General Internal Medicine. The finding comes from one of the largest real-world datasets assembled for this age group, a population that has long been underrepresented in the randomized trials that shape prescribing decisions.

The research team, led by Gizem Kaya and including Karen E. Joynt Maddox, analyzed synthetic electronic health record data drawn from the BJC HealthCare network, a system of 14 hospitals with associated outpatient, emergency, and home health services. Starting from more than 318,000 patients aged 80 or older, they identified 53,401 with a diagnosis of atrial fibrillation or atrial flutter. After excluding people with mitral stenosis, mechanical heart valves, cirrhosis, or end-stage renal disease — conditions that could confound both stroke and bleeding risk — 45,305 patients remained for analysis. The study period ran from January 1, 2015, to May 1, 2026, and required at least three inpatient or outpatient encounters to minimize bias from loss to follow-up.

Patients were sorted into four groups based on their anticoagulation status: those who started the drugs at the index visit, those already on therapy who continued throughout follow-up, those who stopped during follow-up, and those never prescribed an anticoagulant. The never-treated group was the largest, at 18,949 patients, or 41.8 percent of the cohort — a striking figure given that these patients carried substantial stroke risk. The continuation group numbered 8,599 (19.0 percent), the initiation group 12,060 (26.6 percent), and the discontinuation group 5,697 (12.6 percent). Apixaban was the most commonly prescribed agent, taken by 13,294 patients (29.3 percent), followed by warfarin at 9.8 percent.

The cohort had a mean age of 85.04 years, and 53.5 percent were women. Hypertension was the most prevalent comorbidity at 81.2 percent, followed by hyperlipidemia at 71.1 percent. Notably, the continuation group carried the highest average CHA₂DS₂-VASc stroke-risk score (4.51 ± 1.30) while the never-anticoagulated group carried the lowest (3.79 ± 1.20), meaning the patients most likely to benefit from stroke prevention were also the ones most likely to be receiving it — and yet more than four in ten eligible patients received no anticoagulation at all.

The mortality gap was stark at every measured time point. Patients who continued anticoagulation had a 1.5 percent death rate at 30 days, 5.4 percent at six months, and 8.7 percent at one year, compared with 8.3 percent, 14.5 percent, and 17.7 percent in the never-treated group — all differences statistically significant. In the primary time-varying Cox model, which handled patients whose anticoagulation status changed over follow-up as multiple time intervals (50,140 person-intervals across 45,305 patients), anticoagulation was associated with a 25 percent lower hazard of all-cause mortality (hazard ratio 0.75, 95 percent confidence interval 0.71–0.79). Compared with continued therapy, one-year mortality was higher among those who initiated (HR 1.51, 95 percent CI 1.38–1.66), discontinued (HR 1.64, 95 percent CI 1.49–1.81), or never received anticoagulation (HR 2.22, 95 percent CI 2.05–2.41).

Because observational studies of anticoagulation are notoriously vulnerable to bias, the researchers deployed three complementary analytical defenses. The time-varying Cox model directly addresses immortal time bias, the artifact that arises when patients must survive long enough to be classified as continuing therapy. A 90-day landmark analysis excluded patients who died within three months of the index encounter and re-anchored follow-up from day 90; it still found higher mortality in the never group (HR 1.16, 95 percent CI 1.03–1.30), as well as in the discontinued (HR 1.17) and initiated (HR 1.19) groups. Finally, 1:1 propensity score matching comparing the continued and never groups — designed to balance measured baseline differences between patients — preserved the association (HR 2.20, 95 percent CI 2.01–2.42). The authors note that this consistency across methods built to counter the principal biases in observational anticoagulation research strengthens confidence that the association is not merely an artifact of study design.

The bleeding picture was more nuanced. Gastrointestinal bleeding occurred most frequently among anticoagulated patients — 13.0 percent in the continuation group and 11.9 percent in the initiation group, versus 6.5 percent in the discontinuation group and 6.6 percent in the never group. Yet intracranial hemorrhage, the feared complication that often drives clinicians to withhold blood thinners in the very old, was low and statistically comparable across all groups, ranging from 1.3 to 1.6 percent (p = 0.129). Ischemic stroke after the atrial fibrillation diagnosis was highest in the initiation group at 7.4 percent, followed by the continuation group at 7.0 percent — a pattern the authors attribute partly to greater baseline risk in anticoagulated patients and possibly to index-event bias, since anticoagulation is often started when atrial fibrillation is first identified at the time of an acute stroke.

One finding with direct relevance to clinical conversations deserves emphasis: the survival advantage narrowed over time. At five years, mortality converged across all four groups, ranging from 27.5 to 30.2 percent. The authors suggest this may indicate that the benefit of anticoagulation is concentrated in the short-to-medium term and is progressively outweighed by competing age-related mortality from dementia, frailty, and accumulated comorbidity. The study also identified other independent predictors of one-year mortality, including dementia (HR 1.71), leukemia (HR 1.63), prior non-gastrointestinal bleeding (HR 1.63), COPD (HR 1.47), thrombocytopenia (HR 1.44), heart failure (HR 1.37), anemia (HR 1.37), and cancer (HR 1.32), alongside age (HR 1.06 per year) and male sex (HR 1.06).

The underuse of anticoagulation in this cohort mirrors a persistent real-world pattern documented in prior large retrospective studies, some covering populations exceeding one million patients. The evidence gap has structural roots: in the pivotal DOAC trials, patients aged 80 and older were scarce. Only 12.7 percent of RE-LY participants were aged 80 to 84, while ARISTOTLE and ENGAGE AF-TIMI 48 included just 13.4 percent and 17.0 percent of patients over 80, respectively, and patients aged 85 and older made up only about 4.0 to 4.3 percent of the enrolled populations in RE-LY and ENGAGE AF-TIMI 48. Trial participants at advanced ages are also highly selected, limiting generalizability to unselected elderly patients with frailty, polypharmacy, renal impairment, and prior bleeding — exactly the patients clinicians hesitate to anticoagulate.

The authors are careful about what the study can and cannot show. As a retrospective analysis of administrative data, it is subject to coding errors and misclassification; its observational design demonstrates associations, not causation; and residual confounding from unmeasured factors such as lifestyle and laboratory values cannot be excluded. Selection bias is likely, since treatment reflected clinician and patient decisions rather than randomization — patients may have been started on anticoagulation only if deemed sufficiently healthy, while discontinuation may have followed intercurrent illnesses. Falls were likely undercaptured, limiting frailty assessment, and because the database was synthetic, INR values were unavailable, so warfarin control could not be assessed and the labile INR component was excluded from HAS-BLED scores. Cause-specific mortality was also unavailable. The data platform, MDClone, generates synthetic patient-level datasets that reproduce the statistical distributions and relationships of the underlying records without identifiable patient information, and multiple validation studies have shown close concordance with analyses of the original source data. Within those limits, the authors’ conclusion is measured: anticoagulation was associated with lower all-cause mortality despite increased gastrointestinal bleeding, supporting continued therapy in appropriately selected elderly patients rather than withholding it on the basis of age alone — with treatment decisions individualized to incorporate frailty, fall risk, and comorbidity.

Subject of Research: Anticoagulation use and all-cause mortality in adults aged 80 years and older with atrial fibrillation

Article Title: Anticoagulation and All-Cause Mortality in Adults Aged 80 Years and Older with Atrial Fibrillation: A Retrospective Cohort Study

Article References: Kaya, G., Sanchez Garcia, J., Schneider, M., Yadav, D., Ndukwe, M., Chen, C.-H., & Joynt Maddox, K. E. (2026). Anticoagulation and All-Cause Mortality in Adults Aged 80 Years and Older with Atrial Fibrillation: A Retrospective Cohort Study. Journal of General Internal Medicine. https://doi.org/10.1007/s11606-026-10920-7

Image Credits: AI Generated

DOI: 10.1007/s11606-026-10920-7

Keywords: atrial fibrillation, anticoagulation, all-cause mortality, older adults, gastrointestinal bleeding, intracranial hemorrhage, retrospective cohort study, DOACs, warfarin, stroke risk, synthetic health record data, bleeding risk