Fraud in the health systems of Chile: a detection model
Loading...
Date
2009
Journal Title
Journal ISSN
Volume Title
Publisher
PAN AMER HEALTH ORGANIZATION
Abstract
Objectives. To develop a model for detecting cases of organized fraud in Chile based on data from the legal forms for medically authorized leave (formulario legal de licencia medica curativa-MAL) and to establish the relevance of this data to fraud detection.
Methods. A binomial logistic regression model was employed using four variables from the MAL form, a national requirement for illness-related work absences: the number of legal absences taken by a single person, the number of days authorized by the prescribing doctor, the total cost per illness, and a dichotic variable reflecting whether or not the diagnosis is one that can be proven. The analysis involved 4 079 MAL forms that had been submitted in 2003 to a private health provider and of which 356 were already identified as fraudulent by a panel of medical fraud experts.
Results. The model successfully identified 99.71% of the fraudulent medical authorizations and 99.86% of the non-fraudulent, according to the criteria of the panel of fraud experts. Three of the variables employed had statistically-significant independent predictive power. The positive predictive value of the proposed model was 98.59%, while its negative predictive value was 99.97%.
Conclusions. The binomial logistic model that was developed uses four variables that are common to all MAL forms in use by Chile's public as well as private insurers, permitting separation of fraudulent from non-fraudulent requests to be more accurate, more timely, and at a cost lower that of an expert panel.
Methods. A binomial logistic regression model was employed using four variables from the MAL form, a national requirement for illness-related work absences: the number of legal absences taken by a single person, the number of days authorized by the prescribing doctor, the total cost per illness, and a dichotic variable reflecting whether or not the diagnosis is one that can be proven. The analysis involved 4 079 MAL forms that had been submitted in 2003 to a private health provider and of which 356 were already identified as fraudulent by a panel of medical fraud experts.
Results. The model successfully identified 99.71% of the fraudulent medical authorizations and 99.86% of the non-fraudulent, according to the criteria of the panel of fraud experts. Three of the variables employed had statistically-significant independent predictive power. The positive predictive value of the proposed model was 98.59%, while its negative predictive value was 99.97%.
Conclusions. The binomial logistic model that was developed uses four variables that are common to all MAL forms in use by Chile's public as well as private insurers, permitting separation of fraudulent from non-fraudulent requests to be more accurate, more timely, and at a cost lower that of an expert panel.
Description
Keywords
Fraud, insurance, health, decision support techniques, Chile