PS1-24: Modeling the Relationship Between Self-Reported Urinary Incontinence Data and Diagnoses in an Electronic Health Record

Abstract

Background/Aims Self-reported data from large longitudinal mailed surveys provide data on disease symptoms in a population of interest. However, such studies do not give researchers an indication of whether patients have clinically significant symptoms. Access to electronic health record (EHR) data of a survey population provides diagnoses on subjects who have sought medical care. The General Longitudinal Overactive Bladder Evaluation (GLOBE) is a population-based study designed to collect longitudinal data on symptoms associated with urinary incontinence (UI). We used self-reported data from the GLOBE study to predict clinical UI based on ICD-9 diagnoses codes for all types of UI in the Problem List in the EHR. Methods We mailed baseline surveys to a random sample of 8,077 (3,648 baseline responses) female primary care patients aged 40+ years from a population of more than 400,000 patients. The surveys were completed by patients between March and May of 2006 while the clinical data collected from the EHR dated between 9/2/2002 and 6/1/2007. Using data from these two sources we created a model to predict which patients had a UI diagnosis in the EHR (dependent variable). Independent variables included age, age2, and Charlson index (from the EHR), and UI severity (range of 0-9), urine loss behavioral adaptations, duration of urine loss symptoms, marital status, education, BMI, and parity (from the survey). Results The mean age of participants was 59.04 (SD +/-13.13). The mean BMI and median parity were 28.51 (SD +/- 16.53) and 2.22 (SD +/- 1.41) respectively. Using a logistic regression model, factors that were statistically significant predictors of UI diagnosis (p<0.01) included UI score (OR 1.28 (95% CI 1.19-1.81)), urine loss behavior (OR 1.12 (95% CI 1.06-1.18)), Charlson Index (OR 2.22 (95% CI 1.52-3.23)) and duration of UI symptoms (OR 3.09 (95% CI 1.69-5.64)). The remaining variables (age, age2, marital status, education, BMI, and parity) were not statistically significant or associated with a UI diagnosis. Conclusions Self-reported UI severity, duration, behavioral change, and comorbidities increase likelihood of UI diagnosis. Understanding symptom and health characteristics of women with UI who seek care allows us to develop better tools to predict, target and potentially prevent UI.