Identification of Incident Uterine Fibroids Using Electronic Medical Record Data

Authors

Onchee Yu, Group Health Research Institute, Seattle, WAFollow
Renate Schulze-Rath, Bayer Pharma AG, Berlin, Germany
Susan Reed, University of Washington, Seattle, WA
Jane Grafton, Group Health Research Institute, Seattle, WA
Kelly Hansen, Group Health Research Institute, Seattle, WA
Delia Scholes, Group Health Research Institute, Seattle, WA

Publication Date

8-10-2017

Keywords

observational studies, clinical decision-making, biostatistics

Abstract

Background: Uterine fibroids are the most common benign tumors of the uterus that are associated with considerable morbidity in women. Diagnosis codes have been used to identify symptomatic fibroid cases, but their accuracy, especially for incident cases, is uncertain. This study assessed the accuracy of diagnosis codes in identifying incident fibroids and developed algorithms to improve incident fibroid case-finding using additional electronic data.

Methods: Women aged 18–65 years who received an ICD-9 diagnosis code for uterine fibroid during 2012–2014 were identified from electronic databases at Group Health Cooperative, an integrated health care system in Washington State. Women with a fibroid history or hysterectomy were excluded. Medical records were reviewed on a random sample of 617 women to confirm incident fibroid status. Additional data on demographics, symptoms, treatment, imaging, health care utilization, comorbidities and medication were collected. Classification and regression tree analysis incorporating these additional data were used to develop algorithms to identify incident fibroid. We focused on an algorithm with high sensitivity (ie, maximizing the inclusion of true incident cases) and another with high specificity (ie, avoiding incorrect inclusion of noncases as incident cases). Algorithm performance was assessed by calculating sensitivity, specificity and positive predictive value (PPV) using medical record as gold standard.

Results: Among the 617 women, mean age at diagnosis was 48 years. Medical record review confirmed 583 (95%) fibroid cases and 482 incident cases, a 78% PPV for incident cases based on diagnosis codes alone. Incorporating additional electronic data, the algorithm classified 395 incident cases among women with at least 2 pelvic ultrasounds on and prior to diagnosis date. Of these, 344 were correctly classified as incident cases, yielding an 87% PPV. Sensitivity was 71% and specificity 62%. A second algorithm further classified women based on a fibroid code of 218.9 in 2 years after diagnosis and lower body mass index yielded 93% PPV, 53% sensitivity and 85% specificity.

Conclusion: Identification of incident uterine fibroids through ICD-9 diagnosis codes alone was good with moderate PPV. Algorithms using additional electronic data improved incident fibroid case finding with higher PPV, and either higher sensitivity or higher specificity to meet different study aims.

Recommended Citation

Yu O, Schulze-Rath R, Reed S, Grafton J, Hansen K, Scholes D. Identification of incident uterine fibroids using electronic medical record data. J Patient Cent Res Rev. 2017;4:201-2.