Document forensics in private health insurance– insurance for insurers

Private health insurers (PHIs) in Germany have a problem: some of their customers are committing insurance fraud and thereby affecting the other policyholders. In 2020, losses associated with insurance fraud amounted to around 2.1 billion euros. According to the association of private German health insurers, insurers spent just under 31 billion euros on benefits, which means that fraudulent claims accounted for around 7 percent of all payouts made by private health insurers.

How is this possible?

The digitization of the healthcare system not only provides new opportunities for patients and insurers, but also for fraudsters. Until a few years ago, most claims and bills were submitted by regular mail. Today, a photo – submitted to the PHI via a customer platform or app – is usually sufficient. For policyholders, submitting claims and receiving refunds is much faster and more convenient. For insurers, however, digitization is associated with a loss of control as they no longer have access to the original, physical documents. Instead, they receive pictures (sometimes of poor quality) taken with cell phones.

Falsifying documents has never been easier

For many years now, word processing programs have made it possible to duplicate original prescriptions or invoices and submit these documents to PHIs for refunds. On the other hand, it has been much more difficult for medical laypersons to create a medical invoice that appeared legitimate from a professional point of view. But in 2022, a new tool appeared that allowed fraudsters to do this as well: ChatGPT. For example, if you ask ChatGPT which indicators are used on a medical bill for the treatment of glaucoma, you get the following:

The numbers on a doctor's bill for the treatment of glaucoma can vary depending on the billing system and country. For example, in the German medical fee schedule (GOÄ) billing system, the following numbers may be relevant
be relevant:

• Number 1: Consultation and examination
• Number 5: Injections or infusions
• Number 6: Surgical measures
• Number 7: Laser treatment
• Number 8: Drug therapy
• Number 9: Follow-up and control examinations

Fortunately, ChatGPT does not yet provide a comprehensive proposition for an invoice. It does, however, facilitate further research. With the indicator “drug therapy” (for example), you can ask which drugs are used for the treatment of glaucoma. ChatGPT provides a valid answer for this question, including a comprehensive explanation.

Various medications are used to treat glaucoma. The choice of medication depends on the type of glaucoma, the severity, and the individual needs of the patient. Here are some commonly prescribed classes of medications:

1. prostaglandin analogs: these eye drops increase the outflow of fluid from the eye, thereby lowering intraocular pressure. Examples of prostaglandin analogs include latanoprost, bimatoprost, and travoprost.
2. beta-blockers: these eye drops reduce the production of aqueous humor in the eye, thereby lowering intraocular pressure. Examples of beta-blockers include timolol, betaxolol, and levobunolol.
3. alpha-agonists: these eye drops lower intraocular pressure by improving the outflow of fluid and reducing its production. Examples of alpha agonists include brimonidine and apraclonidine.
4. carbohydrate anhydrase inhibitors: these drugs reduce the production of aqueous humor in the eye, thereby lowering intraocular pressure. Examples of carbohydrate anhydrase inhibitors include dorzolamide and brinzolamide.
5. miotics: These eye drops constrict the pupil and open the outflow channel for aqueous humor, thereby lowering intraocular pressure. Example of a miotic is pilocarpine.

What’s more, it’s safe to say that AI-powered chatbots will be able to create better and better suggestions with each iteration.

So what can insurers do to protect themselves?

It’s true that falsifying invoices has never been easier. However, unlike (surreptitious) laypersons, physicians do not use standard word processors to create receipts, but specialized software. Duplicating and modifying the content of invoices therefore remains a difficult task. Manual errors can often be identified by comparing the physician’s past invoices with the newly submitted receipts.

Our partner ICO-LUX offers a solution that allows for documents to be examined forensically for this purpose. ICO.Fraud checks incoming documents for various indications of fraud (e.g., line spacing, fonts, the chronology of invoice numbers) by comparing them with “genuine” documents from the same healthcare professional. This comparison results in a “fraud score” that can be taken into account by benefit billing systems such as in|sure Health Claims during automated processing. If a document is suspected to have been tampered with, the automated payment is halted and the process is attributed to a fraud specialist.

 The “fraud score” can also be used positively: provided that ICO.Fraud recognizes, beyond a doubt, that documents are “genuine,” then the documents can increasingly be processed automatically by machines. With this approach, fraud detection takes on another dimension: it mitigates the loss of control described above, and helps insurers have greater confidence in their automated processing systems.

If you would like to receive further information and advice, please contact our expert Martin Lorenz, Senior Business Developer at adesso insurance solutions.