Types of DS2 Predicates

DS2 defines three classes of deterministic predicates, which can be used alone, in combination with each other, or in combination with probabilistic predicates:

• Level 1 deterministic predicates are the least complex predicates. They are written to fire in the presence of obvious concepts – concepts that are known to be equivalent to or closely related to the target condition. For example, if “HIV” or any clinical fact known to indicate or treat HIV is present, then an HIV predicate would return “True.”
• Level 2 deterministic predicates are predicates of moderate complexity. They are written to fire in the presence of correlated concepts, such as comorbidities or co-occurrences, but only if certain conditions are met. For example, an HIV predicate might be configured to return “True” when HIV comorbidities are present, but only if the record had a “Level 1” concept to begin with.
• Level 3 deterministic predicates are predicates of significant complexity based on specific clinical rules. For example, if “HIV” is target, then a level 3 predicate may fire when two or more indirectly related concepts that suggest HIV is (in fact likely to be) present. Or it may fire when a comorbidity is present and is consistent with an HIV-related laboratory result such as a CD4 count within a particular range.

The challenge with deterministic predicates is in striking a balance between redacting too much and too little.

A probabilistic approach can help with this challenge by applying real-world probabilities and machine learning techniques to develop predicates that learn to understand the density of ties between networked concepts.  A Predicate/Reducer system based on such approaches can be optimized to redact the fewest number of clinical facts while still successfully preventing the inference of the targeted condition.