Model uses available patient data to identify high-risk patients for screening, intervention
WEDNESDAY, Sept. 30 (HealthDay News) -- Readily available patient medical data can be used in a Bayesian model to estimate the future risk of a diagnosis involving domestic abuse, according to a study published Sept. 29 in BMJ.
Ben Y. Reis, Ph.D., of Children's Hospital Boston, and colleagues assembled anonymous patient data on 561,216 men and women over the age of 18 years and identified 5,829 patient cases that met a narrow definition of domestic abuse (ICD-9 codes explicit to abuse) and 19,303 patient cases that met a broader definition (including assault). Two-thirds of the patient data were used to develop separate Bayesian models predicting abuse risk for men and women and the remaining third was used for model verification.
The Bayesian models developed by the researchers achieved sensitive, specific predictions of risk of future abuse diagnosis (area under the receiver-operating characteristics curve of 0.88 for the narrow definition and 0.82 for the broad definition). Among the clinical data associated with a future abuse diagnosis were alcohol and substance-related mental disorders, affective disorders, poisoning, and mental health conditions and psychoses.
"Commonly available longitudinal diagnostic data can be useful for predicting a patient's future risk of receiving a diagnosis of abuse. This modeling approach could serve as the basis for an early warning system to help doctors identify high-risk patients for further screening," the authors write.