Rule-Based Models Remain a Reliable Tool for Detecting Medical Overtreatment, Despite the Rise of AI

Is there still a place for rule-based models in the age of AI? The answer is a resounding yes when it comes to detecting medical overtreatment. While AI-based approaches have made significant strides in many areas of healthcare, insurance companies continue to use rule-based models for detecting potential fraud and overbilling.

The primary reason for this is the lack of interpretability of AI model results. How can you persuade the clinic that a particular treatment is considered overtreatment if your only proof is that the model thinks with a 90% probability it is overtreatment? In contrast, rule-based models can provide links on particular clinical recommendations, and treatment protocols. Its transparent and easy to understand, making it easier for insurers to negotiate with clinics and hospitals.

But creating effective rules-based model takes a lot of efforts. To create effective rules, insurance companies must have a deep understanding of their Tables of Benefits (ToB), clinical recommendations, and treatment protocols. They must also have technical expertise in coding rules and optimizing their performance. A good rule-based model should contain millions of rules, taking into account the vast number of existing diagnoses and services. Many small insurance companies lack the competencies and resources to create and maintain such complex models, making it economically infeasible for them.

In contrast, larger insurance companies are already leveraging the benefits of rule-based models. However, they must ensure the relevance and impartiality of the rules, which can be challenging. Some specialized players in this space focus too heavily on technical competencies, neglecting the critical medical expertise required to create effective rules. As a result, their clients may have to take on the work of creating rules, which defeats the purpose of outsourcing the function of detecting overbilling.

Furthermore, the situation is more complicated in certain markets, such as Mexico where the majority of clinics follow their own services classifications despite existence of a national services classifier (CUPS). This requires additional efforts from insurance companies to match those different names to a standard one, adding extra time for medical experts to validate the claim.

Another specific example of the benefits of rule-based models in medical overtreatment detection is Saudi Arabia, where medical service coding standards (NPHIES) were implemented to replace the previously inconsistent naming of services. This allowed insurance companies to better analyse clinical bills and to increase rejection rates from 20% to 35%. Clinics were not prepared for this and started to overcharge even more, leading to an increase in disputed payments. To resolve such disputes, evidence in the form of links to clinical recommendations and medical articles from authoritative sources is required, providing a rules-based approach.

In conclusion, while AI-based approaches may offer some benefits in healthcare, rule-based models remain a reliable and effective way for insurance companies to detect potential fraud and overbilling, particularly when it comes to identifying cases of overtreatment. The key is to ensure that the rules are continually updated, relevant, and impartial, and that insurers have access to the right competencies to create and maintain these models.