📄 medicine

Admission Creatinine, Not Anticoagulant Class, Predicts In-Hospital Mortality in Anticoagulated Patients With Upper Gastrointestinal Bleeding

In a retrospective multicenter study of anticoagulated patients hospitalized with upper gastrointestinal bleeding, the specific class of anticoagulant did not predict in-hospital mortality or other clinical outcomes, whereas admission serum creatinine was identified as an independent predictor of death.

Original authors: Ali Sohail, Uday Sankar Akash Vankayala, Bivin George, Muhammad Abdurrahman Butt, Jessica Meng, Liliane Deeb

Published 2026-07-07

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Ali Sohail, Uday Sankar Akash Vankayala, Bivin George, Muhammad Abdurrahman Butt, Jessica Meng, Liliane Deeb

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine your body is a busy city, and your blood vessels are the roads. Sometimes, due to age or illness, these roads need "traffic controllers" (anticoagulants) to keep the blood from clotting and blocking the way. But sometimes, these controllers can get a little too enthusiastic, causing a "flood" in the form of an upper gastrointestinal bleed (a serious bleed in the stomach or the first part of the small intestine).

This study asked a very specific question: Once the flood has started and the patient is in the hospital, does it matter which brand of traffic controller they were using?

Here is the breakdown of the findings in simple terms:

The Big Question: Does the "Brand" of Blood Thinner Matter?

The researchers looked at nearly 2,000 patients who came to the hospital with stomach bleeding. About 15% of them were taking blood thinners. They split these patients into three groups based on the "brand" of medicine they were taking:

The New Guys (DOACs): Modern drugs like apixaban or rivaroxaban.
The Old School (Warfarin): The classic blood thinner.
The Bridge (Heparin products): Often used for specific short-term needs or cancer.

The Finding: It turns out, the "brand" didn't matter at all for how the patient did in the hospital. Whether they were on the new drugs, the old drugs, or the bridge drugs, the outcomes were almost identical.

Death rates: The same.
Time in the ICU: The same.
How long they stayed in the hospital: The same.
Need for surgery or scopes: The same.

The Analogy: Imagine two cars crash into a wall. One car has a red bumper, the other has a blue bumper. The study found that the color of the bumper doesn't determine how badly the driver is hurt or how long they stay in the repair shop. What matters is the crash itself and the driver's condition, not the paint job.

The Real Clue: The Kidney Gauge

If the type of medicine didn't predict who would get sicker or pass away, what did?

The study found that the single most important predictor was how well the patient's kidneys were working when they walked into the hospital.

The Metaphor: Think of the kidneys as the city's water filtration plant. If the filtration plant is clogged or broken (high creatinine levels in the blood), the city's water system gets backed up.
The Result: For every small increase in the "clog" (creatinine level), the odds of the patient dying in the hospital went up by 50%.

The researchers concluded that if you want to know how a patient with a bleeding stomach will do, don't ask, "What brand of blood thinner were they taking?" Instead, ask, "How are their kidneys?"

Why This Matters (According to the Paper)

The authors suggest that doctors shouldn't panic about the specific type of blood thinner a patient was on when they arrive at the ER. Instead, they should focus immediately on the patient's kidney function.

Why? High creatinine (a sign of kidney trouble) tells a story:
1. The kidneys might not be clearing the drugs out of the body, making the bleeding worse.
2. The patient might be so dehydrated from blood loss that their kidneys are shutting down.
3. The patient might have underlying liver or kidney disease that makes the whole situation more dangerous.

The Caveats (What the Study Didn't Say)

The paper is careful to point out a few limits:

It's a look back: They looked at old records, not a new experiment.
Small groups: There were very few people on Warfarin or Heparin compared to the new drugs, so the numbers for those groups are a bit shaky (like trying to guess the weather based on only three days of data).
Missing details: They didn't have data on exactly how much blood was lost or if doctors used special "antidotes" to stop the bleeding.

The Bottom Line

If you have a stomach bleed and you are on blood thinners, the specific type of pill you were taking isn't the main reason you might get better or worse. The most critical sign of your immediate future is the health of your kidneys. The "traffic controller" brand is less important than the condition of the "road system" (your body) itself.

Technical Summary: Admission Creatinine, Not Anticoagulant Class, Predicts In-Hospital Mortality in Anticoagulated Patients With Upper Gastrointestinal Bleeding

Problem Statement
Direct oral anticoagulants (DOACs) have largely replaced vitamin K antagonists (e.g., warfarin) for stroke prophylaxis and venous thromboembolism treatment. While prior evidence indicates DOACs are associated with modestly higher rates of gastrointestinal bleeding compared to warfarin, it remains poorly characterized whether the specific class of anticoagulant influences the in-hospital trajectory of patients who have already experienced an established upper gastrointestinal bleeding (UGIB) event. Specifically, it is unclear if the pharmacologic class (DOAC, heparin product, or warfarin) independently predicts mortality, intensive care unit (ICU) utilization, length of stay (LOS), need for endoscopic intervention, or early readmission after adjusting for patient-level prognostic factors.

Methodology
The authors conducted a retrospective multicenter cohort study within the Northwell Health system involving adults aged 18–89 years admitted with UGIB between January 2019 and January 2024.

Cohort Selection: UGIB was identified via ICD-10 codes and confirmed via chart review to ensure the bleeding source was proximal to the ligament of Treitz. The analytic cohort included only patients with documented outpatient anticoagulant exposure at admission.
Exposure: Patients were stratified by anticoagulant class: DOACs (apixaban, rivaroxaban, dabigatran, edoxaban), heparin products (therapeutic enoxaparin, dalteparin, fondaparinux), or warfarin.
Outcomes: The primary outcome was in-hospital all-cause mortality. Secondary outcomes included ICU admission, LOS, therapeutic endoscopic intervention, and 30-day all-cause readmission.
Statistical Analysis: Continuous variables were compared using the Kruskal–Wallis test; categorical variables used chi-square or Fisher exact tests. Univariable predictors of mortality were screened at $p < 0.10$ . Due to a limited number of mortality events (22), the final multivariable logistic regression was constrained to two predictors per the events-per-variable rule. Anticoagulant class was forced into the model, and the second predictor was selected based on significance and model fit. Model performance was assessed via c-statistic (discrimination) and Hosmer–Lemeshow test (calibration).

Key Results

Cohort Characteristics: Of 2,009 UGIB admissions, 303 (15.1%) had documented anticoagulant exposure: 240 (79.2%) on DOACs, 40 (13.2%) on heparin products, and 23 (7.6%) on warfarin. The groups were comparable in age, sex, and comorbidity burden.
Mortality and Secondary Outcomes: In-hospital mortality occurred in 22 patients (7.3%) and did not differ significantly by anticoagulant class (DOAC 7.5%, heparin 5.0%, warfarin 8.7%; $p=0.83$ ). Similarly, no significant differences were found for ICU admission ( $p=0.27$ ), LOS ( $p=0.14$ ), endoscopic intervention ( $p=0.66$ ), or 30-day readmission ( $p=1.00$ ).
Predictors of Mortality: In the constrained multivariable model, anticoagulant class was not associated with mortality ( $p=0.92$ ). Conversely, admission serum creatinine was an independent predictor. Each 1 mg/dL increment in admission creatinine was associated with a 50% relative increase in the odds of in-hospital death (adjusted odds ratio 1.50, 95% CI 1.02–2.20; $p=0.041$ ). The model demonstrated acceptable discrimination (c-statistic 0.62) and calibration ( $p=0.82$ ).

Significance and Claims
The paper concludes that among anticoagulated patients hospitalized with UGIB, the specific class of anticoagulant does not independently predict in-hospital mortality or other clinical outcomes. The authors argue that once a clinically significant bleed has occurred, the acute hospital course is likely dominated by hemodynamic status, bleeding severity, and comorbid burden rather than the pharmacologic class of the agent.

In contrast, admission renal function (measured by serum creatinine) serves as a robust, independent prognostic marker. The authors suggest that creatinine integrates pathophysiologically relevant signals, including chronic kidney disease (which may limit drug clearance), acute kidney injury from hypovolemia due to blood loss, and hepatorenal physiology in cirrhotic patients. Consequently, the paper posits that admission creatinine may be a more clinically actionable prognostic marker than the specific anticoagulant class and warrants incorporation into early risk-stratification frameworks for UGIB.

Limitations Noted by Authors
The authors acknowledge several limitations: the retrospective nature of the study, lack of granular data on bleeding severity, transfusion volume, time to endoscopy, and reversal-agent use; the small sample sizes for heparin and warfarin subgroups leading to wide confidence intervals; and the constraint of the multivariable model to only two predictors due to the low number of mortality events. They note that prospective validation with propensity-score matching and more detailed data is required.