Artificial Intelligence for Insurance Fraud Detection
In its latest report, the National Health Care Anti-Fraud Association, NHCAA (1), estimated that fraud received by insurers in the U.S. healthcare sector represents between 3% and 10% of the system’s total spending. If we consider the more than 2.26 trillion dollars that are being invested annually, it represents losses of between 70 and 300 billion dollars a year.
What is more serious is that apart from the monetary losses, it has been shown that health system fraud also represents a risk to patient care. In October 2019, John Hopkins University (2) published a study that found that patients treated by Provider Institutions, who were later excluded from the system for fraud, were 14% to 17% more likely to die than those treated by providers who did not commit fraud.
In response, insurers have set about the task of training experts to identify the different cases of fraud that arise manually. To this end, they define a set of basic rules to determine when a process can be labeled as fraudulent. However, this manual process takes a great deal of time due to the high volume of cases and can present consistent errors (3).
Can Artificial Intelligence help detect health insurer fraud?
Intelligent algorithms can learn from historical cases and adapt in a complex way to identify patterns in fraud detection. Banks use this type of technology to detect money laundering or scams. Such a system can identify and correct errors and avoid unnecessary or ineffective interventions, saving a great deal of time, money, and effort.
All claims can be analyzed automatically as an algorithm, making the time required to identify a possible fraud only as long as it takes the algorithm to analyze the information (less than 1 minute usually). By prioritizing claims, operators can better invest their time in claims handling, dramatically decreasing operational errors associated with time constraints. Finally, fraud detection models have such sensitivity and accuracy that they can generate a 3-10% reduction in undetected fraud expense for health insurers.
How to train AI algorithms capable of detecting insurance fraud?
With Arkangel Ai’s Hippocrates functionality, it is possible to train fraud detection models without writing a single line of code. Hippocrates processes and analyzes the data available from the insurer and runs hundreds of experiments automatically until it develops an algorithm capable of detecting abnormal patient claims and directing them to a suspected fraud investigation with an agent. This algorithm has two main objectives:
- Automatically detect anomalies in claims. Where predictions are delivered in binary form: anomaly or normal, along with a percentage of confidence in the prediction.
- Automatically prioritize claims according to the confidence percentage and the cost associated with the claim (or/and any other high-risk variable to which we want to give weight).
What does Hippocrates need to develop an Ai algorithm capable of detecting insurance fraud?
Like any AI-based platform, Hippocrates requires a set of data to serve as input in its learning process. In this case, this information corresponds to a historical database (>24 months) of claims containing records of some variables such as: the cost associated with the claim, basic data of the beneficiaries (age, sex, etc.), ICD code of diagnosis of the claim, specialty of the treating doctor, entity providing the service, among other variables of interest for the detection of fraud, and of course the identification of whether or not the record corresponds to a fraudulent record.
Once the data is structured, Hippocrates proceeds to develop the algorithm in 5 phases.
- Compilation and preprocessing of the claims data.
- Analysis of the data with statistical algorithms.
- Training of artificial intelligence algorithms.
- Development of benchmarking metrics.
- Field testing with real claims samples.
Insurers considering the use of an AI system in claims management should prioritize a culture of Data management and structure a solid foundation for success in Fraud detection:
- Digitized original claims.
- An established claims management process.
- Structured and digitized documentation of the results.
Arkangel Ai can also assist you in structuring this data.
Artificial intelligence will play an essential role in digitizing processes that involve analyzing hundreds of data and variables with extreme precision. In this sense, it is not surprising that all major companies have deployed their innovation teams to implement technological solutions that add value to their services and products, and the health sector is no exception. In fact, healthcare is perhaps one of the sectors in which AI will have more potential impact.
**If you want to know more about Arkangel Ai software and its functionality Hippocrates AutoML leave us your (professional information) and one of our agents will contact you to accompany you in a one-on-one on-boarding of our technology and advise you on the project you have in mind.
(1) FRAUD TRENDS PANEL
(2) Association Between Treatment by Fraud and Abuse Perpetrators and Health Outcomes Among Medicare Beneficiaries
(3) 5 million Euro savings thanks to not having to pay-out proven claim frauds