Dissertations and Theses

Date of Degree


Document Type


Degree Name

Doctor of Public Health (DPH)


Health Policy and Management


Marianne (Mimi) C. Fahs

Committee Members

Levi Waldron

Alexis Pozen

Kenneth Boockvar

Zachary Grinspan

Subject Categories

Health and Medical Administration | Health Services Administration | Health Services Research | Medicinal and Pharmaceutical Chemistry | Public Health


adverse drug event, oral anticoagulants, incidence, risk factors, predictive modeling



Incidence, Risk Factors, and Prediction of Gastrointestinal and

Intracranial Bleeding in a Cohort of Older Veterans Prescribed Oral Anticoagulants


Angela L. Laurio

Advisor: Marianne (Mimi) C. Fahs, PhD, MPH


The objectives of this dissertation were to: (1) describe and compare the incidence and odds of gastrointestinal and intracranial bleeding in veterans age 50 to 89 who were prescribed warfarin, direct oral anticoagulants, or no oral anticoagulants; (2) identify risk factors for gastrointestinal and intracranial bleeding among older veterans prescribed oral anticoagulants, and to calculate the relative risk of bleeding over time through time-to-event analysis; and (3) develop and compare models to predict risk of gastrointestinal and intracranial bleeding among veterans age 50 to 89 who were prescribed warfarin or direct oral anticoagulants using traditional and machine learning methods. These objectives were designed to assist healthcare organizations meet goals for the reduction of anticoagulant-related adverse drug events.


The three studies in this dissertation were carried out using a retrospective cohort design and data from the Veterans Health Administration, American Community Survey, and National Center for Health Statistics. In Study 1, incidence and odds of gastrointestinal and intracranial bleeding were calculated using the full cohort of subjects and compared across three groups: those with no prescription for oral anticoagulants, those with a prescription for a direct oral anticoagulant, and those with a prescription for warfarin. In Study 2, time-to-event analysis for gastrointestinal or intracranial bleeding was conducted and independent risk factors identified for subjects with a prescription for an oral anticoagulant (warfarin or direct oral anticoagulant). In Study 3, predictive models were developed using traditional algorithms and machine learning tools to predict risk of gastrointestinal or intracranial bleeding for subjects with a prescription for oral anticoagulants.

The primary independent variable for all three studies was the subject’s prescription category: no oral anticoagulant, warfarin, or direct oral anticoagulant. The primary dependent variable was a dichotomous variable indicating the presence of an ICD-9 or ICD-10 code for gastrointestinal or intracranial bleeding in the electronic health record. Diagnosis codes for bleeding were associated with any type of clinical encounter, including inpatient admissions, outpatient visits, or emergency room visits. Each subject’s index date was the date of the first outpatient clinical encounter from October 1, 2010 to September 30, 2011 for the first study, and the first OAC prescription date on or after October 1, 2010 for the second and third studies. The cohort was drawn from patients with at least two primary care visits between October 1, 2010 and September 30, 2011 at a VISN-2 facility and assigned a primary care provider. Subjects included in the study did not have a prescription for an oral anticoagulant in the six months prior to index date.


Study 1 found that veterans who were not prescribed oral anticoagulants experienced an average of 9 to 10 times the number of gastrointestinal bleeding events, and an average of 7 to 8 times the intracranial bleeding events as would be expected in the general population. Using either a no blackout or 5-day blackout period approach, this study found lower incidence rates per 100 person-years for gastrointestinal and intracranial bleeding among veterans prescribed oral anticoagulants than previous studies on both veterans and non-veterans. This study also found lower odds of bleeding among veterans prescribed an oral anticoagulant than previous studies; this was the case regardless of approach. Finally, an important finding of Study 1 was the significant difference in incidence between the no blackout period and the 5-day blackout period approaches.

Study 2 found that as the time from oral anticoagulant prescription increased, the proportion of veterans age 50 and older who did not experience a gastrointestinal or intracranial bleeding event after being prescribed warfarin was similar to those prescribed direct oral anticoagulants. Findings were similar veterans with a diagnosis of atrial fibrillation. This finding was reversed in subjects age 75 or older, but all of these differences were small and not statistically significant. This study also found that prescriptions for antidepressants or statins were the strongest risk factors for gastrointestinal or intracranial bleeding for all subjects, while history of bleeding was the strongest risk factor for subjects with a diagnosis of atrial fibrillation and for subjects age 75 or older.

Study 3 found that a logistic regression model using the traditional ORBIT algorithm performed the best out of 12 models developed to predict gastrointestinal or intracranial bleeding risk. A logistic regression model with a subset of five variables performed second best. Two machine learning algorithms also performed fairly well in predicting bleeding risk: a lasso model and a CART model. None of the models performed well in all five of the measures evaluated.


The greatest gains in preventing adverse drug events associated with oral anticoagulants will likely be realized with increased sharing and use of electronic health data, and the ability to discover predictive models that perform well on a variety of evaluation measures. Findings of the three studies presented here were mixed in terms of being consistent with previous research, and additional research is necessary to understand these differences. Reducing the occurrence of gastrointestinal and intracranial bleeding following prescription of oral anticoagulants is an on-going challenge, and while the studies presented in this dissertation shed some light on older veterans and risk of gastrointestinal and intracranial bleeding, there are no easy answers to solving this complex clinical problem.



To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.