Osaka’s Blood-Type Paradox: Why Type A Tracked Lower Cardiovascular Mortality on Dialysis

Three times a week in a typical dialysis life, blood leaves the body. A pump carries it past pressure monitors and through thousands of hollow fibers; dissolved wastes and excess water cross a membrane, and the cleaned blood returns. The machine can replace part of what failed kidneys once did continuously. It cannot erase the cardiovascular terrain created by kidney failure.

Into that landscape, an Osaka research team has introduced a surprisingly ordinary letter. In a prospective study of 1,671 people receiving maintenance hemodialysis at 17 facilities, patients with blood type A experienced lower adjusted all-cause and cardiovascular mortality than patients with type O. The result is arresting because decades of research outside dialysis have generally placed type O on the lower-risk side of cardiovascular comparisons.

The apparent reversal is the story. But so are the boundaries around it. Blood type was not assigned; it was inherited. Treatment was not randomized. The study cannot show that the A antigen protected anyone, nor that type O caused a death. Its confidence intervals only just excluded no difference. Adding ABO type to an existing mortality model barely improved its ability to distinguish individual outcomes. The investigators themselves say prevention and treatment should not change on the basis of blood group.

What the Osaka team actually studied

The work grew from the Osaka Dialysis Complication Study, or ODCS, a multicenter cohort established when Osaka Metropolitan University was still Osaka City University. In 2012, 1,696 prevalent—meaning already treated—maintenance-hemodialysis patients entered the study. Attending nephrologists submitted annual case-report forms to the university data center through 2017. Seven participants lacked ABO information and 18 had other key missing baseline variables, leaving 1,671 for this analysis.

The cohort’s median age was 68. Thirty-seven percent were women, 41% had diabetic kidney disease and 38% had a prior cardiovascular event. The median “dialysis vintage,” the time already spent on dialysis at baseline, was 64 months. Blood group came from medical records and had been determined serologically by agglutination—the same visible clumping principle that made ABO typing possible more than a century ago.

ABO group	Patients	Share of cohort	Primary comparison
A	650	38.9%	Lower adjusted all-cause and cardiovascular mortality than O
B	358	21.4%	No statistically significant difference from O
AB	176	10.5%	No statistically significant difference from O
O	487	29.1%	Reference group in the four-category models

The distribution—roughly 40:20:10:30—closely resembled Japan’s general population, reducing concern that an unusually composed dialysis cohort manufactured the pattern. Baseline groups were broadly similar, although age, alkaline phosphatase and use of anticoagulants and statins differed. The models therefore did not rely on a simple count of deaths. They adjusted for age, sex, time on dialysis, diabetic kidney disease, previous cardiovascular disease, smoking, hypertension, dyslipidemia, anemia-related factors, calcium–phosphate and parathyroid variables, body mass index, albumin and C-reactive protein.

The anatomy of 464 deaths

Over a median 1,826 days—almost exactly five years—464 participants died. Investigators classified 278, or 59.9%, as cardiovascular: 123 sudden deaths, 52 strokes, 44 deaths from coronary artery disease, 38 from congestive heart failure, 20 from peripheral artery disease and one from aortic dissection. The 186 noncardiovascular deaths included 123 infections, 36 malignancies and 27 deaths after trauma or fracture.

Those categories matter statistically. Someone who dies from infection can no longer die later from a heart attack. For cardiovascular mortality, the team therefore used a Fine–Gray competing-risk model, treating noncardiovascular death as a competing event; it reversed the arrangement for noncardiovascular mortality. Type A versus type O produced a cardiovascular subdistribution hazard ratio of 0.723, with a 95% confidence interval of 0.535 to 0.978 and a P value of 0.035. In plain language, at any point in the model’s follow-up framework, the type-A group had an estimated 27.7% lower cardiovascular hazard after adjustment.

For death from any cause, a Cox model gave an adjusted hazard ratio of 0.780 for A versus O, a 22.0% lower relative hazard. Its 95% confidence interval ran from 0.619 to 0.981, with P = 0.034. B and AB did not differ significantly from O on either principal comparison. ABO group did not show a significant relationship with noncardiovascular mortality.

A result that survived checks—but not the need for replication

The authors tested whether the signal vanished under alternative assumptions. Comparing A with all non-A types, an exploratory analysis returned a cardiovascular hazard ratio of 0.733. Results were similar across prespecified clinical subgroups, after propensity-score matching, and after additional adjustment for antiplatelet and anticoagulant use, beta-blockers, renin–angiotensin drugs, statins, alkaline phosphatase, previous heart-failure hospitalization and dialysis adequacy. Separating sudden from nonsudden cardiovascular death did not overturn the direction.

That consistency makes a random modeling accident less likely. It does not convert observation into mechanism. The A-versus-non-A comparison was not prespecified, and no formal correction was made for the additional comparisons. The cohort contained people who had already survived long enough on hemodialysis to enroll, creating possible survivor selection. Covariates were captured in 2012, not repeatedly updated. Cause of death was adjudicated from clinical records across multiple centers. Genetic subtypes such as A1 and A2 were not measured.

Most revealing is the prediction test. Adding the four ABO categories improved model fit statistically, yet increased the C-statistic by only 0.0025. That is a minute gain in discrimination. A marker can illuminate biology at the population level while adding almost nothing useful to a decision about the person in the chair.

Why type O usually appears protective

ABO is more than a label printed on a donor card. The gene at chromosome 9q34 encodes a glycosyltransferase—an enzyme that adds a final sugar to a precursor structure. A and B alleles make enzymes with different sugar preferences. The common O allele produces an inactive enzyme, leaving the H precursor largely unmodified. Those carbohydrate signatures appear not only on red cells but on other cells, tissues and secretions.

One of ABO’s clearest connections to vascular biology runs through von Willebrand factor, or VWF. When a blood vessel is injured, this large protein helps platelets adhere and also carries coagulation factor VIII. People with type O have, on average, about 25% lower plasma VWF and factor VIII levels than people with non-O types, partly because their VWF is cleared more quickly. Across nondialysis populations, non-O blood has accordingly been associated with more venous thrombosis and some arterial cardiovascular outcomes, while O can be associated with more bleeding in particular settings.

The Osaka result runs against that familiar current. Even in hemodialysis, previous work has found lower VWF and factor VIII in type O, and higher VWF has predicted worse outcomes. Yet it was type A, not O, that tracked lower cardiovascular mortality. The investigators did not measure VWF or factor VIII, so those pathways cannot be excluded. But adjustment for anticoagulant and antiplatelet therapy did not erase the result, and the expected VWF direction does not explain it neatly.

Alkaline phosphatase offered another hypothesis. In the 2012 Japanese assay, levels were markedly lower in the type-A group; high levels have been associated with mortality in dialysis cohorts. Additional adjustment for alkaline phosphatase again left the ABO finding standing. The honest mechanistic answer is therefore not a hidden “A-factor.” It is that the cause is unknown.

Dialysis remakes the cardiovascular map

Healthy kidneys regulate fluid, electrolytes, acid–base balance, blood pressure, red-cell production and mineral metabolism around the clock. Hemodialysis performs essential filtration intermittently. Between sessions, fluid and potassium accumulate; during treatment, volume and solute concentrations can change rapidly. The heart encounters repeated loading and unloading, while the vascular system carries the longer history of kidney failure.

Traditional risks—age, diabetes, smoking, hypertension and dyslipidemia—remain. Around them gather dialysis-specific or kidney-specific forces: uremic toxins; chronic inflammation and oxidative stress; anemia; malnutrition and frailty; calcium–phosphate disturbance and vascular calcification; left-ventricular hypertrophy; endothelial dysfunction; access-related circulatory changes; arrhythmia; and the hemodynamic stress of ultrafiltration. Blood also meets an artificial membrane again and again. Some factors favor clotting, others bleeding; some coexist in the same patient.

This is why risk relationships derived from the general population can weaken or even reverse in dialysis, a phenomenon often called “reverse epidemiology.” The phrase should not be mistaken for a law of nature. It is a warning that severe chronic illness alters confounding, selection, nutrition and physiology. The Osaka authors propose that inflammation, calcification, uremia and repeated blood–membrane interaction could modify the expression of ABO-linked risk. Their study did not directly measure that chain, so it remains a research agenda rather than an explanation.

Two histories converge in the dialyzer

ABO science began with a problem visible to the naked eye. In Vienna around 1900, Karl Landsteiner mixed red cells from laboratory colleagues with other people’s serum. Some combinations remained smooth; others clumped. He classified the patterns that became A, B and O, with AB identified soon afterward. The discovery explained why earlier transfusions could save one patient and kill another. It earned Landsteiner the 1930 Nobel Prize and transformed surgery, trauma care, transplantation and obstetrics.

Dialysis followed a different path from chemistry to engineering. Thomas Graham used the word “dialysis” in the nineteenth century to describe substances separating across a semipermeable membrane. John Abel, Leonard Rowntree and B. B. Turner demonstrated extracorporeal dialysis in animals in 1913. Georg Haas treated a human with an artificial kidney in 1924. During the Second World War, Willem Kolff built rotating-drum machines from improvised materials in the Netherlands; in 1945, one of his patients with acute kidney injury survived and recovered renal function.

Chronic treatment required reliable access to the bloodstream. In 1960, the Scribner shunt made repeated access possible, turning an emergency experiment into maintenance therapy—and immediately forcing society to confront who could receive a scarce machine. Membranes, water purification, access surgery, anemia treatment and monitoring steadily improved. Japan’s national dialysis registry has followed the resulting population annually since 1968.

The scale of the question in Japan

At the end of 2024, the Japanese Society for Dialysis Therapy counted 337,414 dialysis patients—2,725 per million population—with a mean age of 70.27. Diabetic kidney disease was the leading underlying condition. The registry recorded 38,348 deaths that year and an 11.3% crude annual mortality rate.

Japan’s cause-of-death pattern also prevents easy slogans. Infection was the largest single category in 2024 at 24.2%, followed by heart failure at 19.0% and malignancy at 7.4%. When the registry combined heart failure, stroke and myocardial infarction, cardiovascular deaths represented 27.3%—down from 54.8% in 1988, but still a large burden. The Osaka study used a broader cardiovascular definition that included sudden death, coronary and peripheral artery disease and aortic dissection, so its 59.9% share should not be compared directly with the registry’s 27.3% without acknowledging different definitions.

This distinction is more than statistical housekeeping. “Cardiovascular death” is a constructed endpoint. Which diagnoses are grouped, how sudden death is handled and what evidence is available after death all shape the number. In ODCS, sudden death alone accounted for 123 of 278 cardiovascular cases. That substantial component is one reason future work needs detailed rhythm, electrolyte, cardiac-imaging and dialysis-session data.

An earlier clue—and a much larger confirmation attempt

The Osaka team was not the first to see type A move in this direction. A 2023 single-center Japanese study followed 365 hemodialysis patients and recorded 73 composite cardio-cerebrovascular events or deaths. Type A was associated with a lower adjusted hazard than non-A blood. The study was smaller, combined fatal and nonfatal endpoints and did not provide a cause-specific mortality analysis.

ODCS supplied nearly five times as many participants, multiple facilities and 464 deaths, then separated cardiovascular from noncardiovascular mortality and treated the latter as a competing risk. That is a stronger observational design, not independent proof in a new country or treatment system. Both Japanese studies could share population genetics, practice patterns or unmeasured exposures. The most persuasive next result would come from an incident cohort—people enrolled when dialysis begins—in a different region, with prespecified replication and direct measurement of candidate pathways.

What this means for a patient today

It does not mean a patient with type A is “protected,” or that someone with type O is destined for a cardiovascular event. It does not justify changing dialysis frequency, fluid targets, anticoagulation, antiplatelet treatment, diet, vascular-access care, transplant eligibility or screening intensity. Those choices carry real benefits and harms and must rest on clinical history, examination, laboratory trends, dialysis-session data and shared decisions with the treating team.

It also does not validate Japan’s popular blood-type personality stereotypes. ABO antigens and clotting proteins are biological; claims that the same letters determine temperament have not held up in large Japanese and U.S. survey analyses. A statistically associated health outcome in a defined cohort cannot be stretched into a theory of character.

The experiment the result now demands

A useful replication would begin with patients starting hemodialysis and follow time-varying exposures rather than freezing the clinical picture at baseline. It would genotype the ABO locus, distinguish A1 from A2, and enroll diverse populations. It would measure VWF, factor VIII, ADAMTS13, endothelial and inflammatory markers, residual kidney function, calcification, nutrition, dialysis membrane and modality, access type, ultrafiltration stress, potassium trajectories and rhythm events.

Researchers would prespecify the A-versus-O and A-versus-non-A comparisons, harmonize death adjudication and publish absolute risks alongside hazard ratios. External validation would test whether ABO improves prediction enough to change a decision. Mechanistic work could then ask whether the antigen itself matters, whether it tags another inherited feature, or whether the signal is produced by a dialysis-specific interaction.

Negative replication would be informative. It could reveal that the narrow confidence intervals of one cohort did not travel. Positive replication could point investigators toward neglected cardiovascular pathways in kidney failure. Either outcome would be more valuable than turning the current association into premature advice.

A letter, a machine and an unanswered question

Landsteiner’s letters became powerful because they predicted a direct, reproducible event: incompatible blood clumped. Dialysis became powerful because engineers, clinicians and patients repeatedly tested whether a membrane and a circulation could sustain life. The new Osaka finding belongs to a more uncertain stage of science. It observes a pattern and asks what mechanism could have drawn it.

The pattern is coherent enough to deserve attention: type A carried lower adjusted all-cause and cardiovascular mortality, not lower noncardiovascular mortality; the result persisted through several analyses; a smaller Japanese study pointed in the same direction. The pattern is also too limited for action: observational, prevalent-cohort, baseline-only, almost entirely Japanese, and minimally helpful for individual discrimination.

That tension is not a weakness to conceal. It is the substance of the discovery. In the general population, type O often marks the lower-thrombotic-risk shore. In Osaka’s dialysis current, type A appeared downstream of fewer cardiovascular deaths. The task now is not to award one blood group a talisman. It is to learn what changed the river.

Sources and method

This report distinguishes association from causation and relative hazard from absolute risk. Study counts, models, confidence intervals, covariates, sensitivity analyses and limitations were checked against the full open-access article. National figures use the latest completed JSDT annual report available for this edition. Historical context relies on medical and scientific institutions and peer-reviewed reviews.

• Kurajoh et al., Kidney International Reports: ABO Blood Types and Mortality in Patients Undergoing Hemodialysis (full text)
• Journal DOI record: 10.1016/j.ekir.2026.106558
• Osaka Metropolitan University research release, July 21, 2026
• Nakayama et al., 2023 Japanese dialysis study of cardio-cerebrovascular outcomes
• Japanese Society for Dialysis Therapy: 2024 Annual Dialysis Data Report
• JSDT: 2024 patient dynamics and cause-of-death tables
• NIDDK: how hemodialysis works and what it replaces
• KDIGO: chronic kidney disease and arrhythmias conference report
• Peer-reviewed review: cardiovascular pathophysiology and screening in dialysis
• NCBI Bookshelf: The ABO Blood Group
• O’Donnell and Laffan: ABO, factor VIII and von Willebrand factor
• Rios et al.: ABO, VWF, ADAMTS13 and factor VIII in hemodialysis
• Nobel Prize Outreach: Landsteiner and the discovery of blood groups
• Peer-reviewed history of Willem Kolff and the artificial kidney
• Large Japanese and U.S. surveys finding no relationship between ABO type and personality

Funding disclosure: The Osaka study reported partial grant support from the Japan Kidney Foundation, Astellas Pharma, Chugai Pharmaceutical and Daiichi Sankyo. The article was published under a CC BY license. This report is medical journalism, not personal medical advice.