ABSTRACT
BACKGROUND: Cleveland, Ohio, is home to three major hospital systems serving approximately 80% of the Northeast Ohio population. The Cleveland Clinic, University Hospitals Health System, and MetroHealth are direct competitors for primary and specialty care, and patient overlap between these systems is high. Fragmentation of health data that exist in silos at these health systems produces an overestimation of disease burden due to double and sometimes triple counting of patients. As a result, longitudinal population-based studies across the Cleveland patient population are impeded unless accurate and actionable clinically derived health data sets can be created. METHODS: The Cleveland Institute for Computational Biology has developed the De-Duplicate and De-Identify Research Engine (DeDeRE) that, without any exchange of personal health identifiers (PHI) between health systems, will effectively de-duplicate the patients between one or more health entities. RESULTS: The immediate utility of this software for cancer epidemiology is the increased accuracy in measuring cancer burden and the potential to perform longitudinal studies with de-duplicated, de-identified data sets. CONCLUSIONS: The DeDeRE software developed and tested here accomplishes its goals without exposing PHIs using a state-of-the-art, trusted privacy preservation network enabled by a hash-based matching algorithm. IMPACT: This paper will guide the reader through the functions currently developed in DeDeRE and how a healthcare organization (HCO) employing the release version of this technology can begin sharing data with one or more additional HCOs in a collaborative and noncompetitive manner to create a regional population health resource for cancer researchers.See all articles in this CEBP Focus section, "Modernizing Population Science."
Subject(s)
Datasets as Topic , Health Information Exchange , Health Records, Personal , Neoplasms/epidemiology , Algorithms , Cities/epidemiology , Confidentiality , Humans , Ohio , SoftwareABSTRACT
BACKGROUND: Barrett's esophagus is often asymptomatic and only a small portion of Barrett's esophagus patients are currently diagnosed and under surveillance. Therefore, it is important to develop risk prediction models to identify high-risk individuals with Barrett's esophagus. Familial aggregation of Barrett's esophagus and esophageal adenocarcinoma, and the increased risk of esophageal adenocarcinoma for individuals with a family history, raise the necessity of including genetic factors in the prediction model. Methods to determine risk prediction models using both risk covariates and ascertained family data are not well developed. METHODS: We developed a Barrett's Esophagus Translational Research Network (BETRNet) risk prediction model from 787 singly ascertained Barrett's esophagus pedigrees and 92 multiplex Barrett's esophagus pedigrees, fitting a multivariate logistic model that incorporates family history and clinical risk factors. The eight risk factors, age, sex, education level, parental status, smoking, heartburn frequency, regurgitation frequency, and use of acid suppressant, were included in the model. The prediction accuracy was evaluated on the training dataset and an independent validation dataset of 643 multiplex Barrett's esophagus pedigrees. RESULTS: Our results indicate family information helps to predict Barrett's esophagus risk, and predicting in families improves both prediction calibration and discrimination accuracy. CONCLUSIONS: Our model can predict Barrett's esophagus risk for anyone with family members known to have, or not have, had Barrett's esophagus. It can predict risk for unrelated individuals without knowing any relatives' information. IMPACT: Our prediction model will shed light on effectively identifying high-risk individuals for Barrett's esophagus screening and surveillance, consequently allowing intervention at an early stage, and reducing mortality from esophageal adenocarcinoma. Cancer Epidemiol Biomarkers Prev; 25(5); 727-35. ©2016 AACR.
