Towards Protecting Vital Healthcare Programs by Extracting Actionable Knowledge from Policy

Abstract

In challenging economic times, obtaining value for money by ensuring financial integrity and fairer distribution of services are among the top priorities for social and health-care systems globally. However, healthcare billing policies are complex and identifying non-compliance is often narrow-scope, manual and expensive. Maintaining 'integrity' is a challengeensuring that scarce resources get to those in need and are not lost to fraud and waste. Our approach fuses recent advances in dependency parsing with a policy ontology to convert the content of regulatory healthcare policy into human-friendly policy rules, that are amenable to machineexecution, with human oversight. We describe the ontology-guided transformation of textual patterns into a semantically-meaningful knowledge graph of rules, outline our experiments and evaluate results against policy rules obtained from professional investigators. The aim is to make a policy-compliance 'landscape' visible to healthcare programs -helping them identify Fraud, Waste or Abuse.

Collaborative artificial intelligence system for investigation of healthcare claims compliance

“…For the first two tasks (extraction and execution of rules) we built a ground truth 24,25,29 with the help of domain experts who manually defined rules corresponding to policy documents for dental providers for the states of Iowa 30 and Colorado 31 (US)-all domain experts who helped design and test Clais are professional FWA investigators. The resulting 141 ground truth rules are based on the same ontology that guides Clais automatic extraction of rules from text, and are formalized into knowledge graphs; there are 90 rules from the Iowa policy document (a knowledge graph with 1977 vertices and 2447 edges), and 51 rules from the Colorado policy document (a knowledge graph with 1651 vertices and 2044 edges); the ontology and the ground truth rules are publicly available 32 .…”

“…We build upon recent natural language processing (NLP) techniques 24,25 to automatically identify dependencies between relevant entities and relations described in a fragment of policy text, and to assemble them into a rule. Clais uses a configurable NLP extraction pipeline, where each component can be replaced or complemented by others with similar functionalities.…”

“…The core of the NLP extraction pipeline transforms textual patterns between ontological annotations into sets of RDF triples, which constitute a partial rule knowledge graph; it uses two domain-agnostic approaches: semantic role labeling 25 and semantic reasoning 24 . Semantic role labeling is based on a declarative information extraction system 58 that can identify actions and roles in a sentence (who is the agent, what is the theme or context of the action, if there are any conditionals, the polarity of the action, temporal information, etc.).…”

“…Dependency trees capture fine-grained and relevant distant (not necessarily consecutive in the text) syntactic connections by following the shortest path between previously annotated ontological entities. For each linguistically connected predicate-arguments tuple in the sentence, the dependency tree searches the ontology definitions for non-ambiguous paths that could connect the tuple elements in a semantically correct way; the search is based on parametrized templates 24 .…”

“…1) analyzes policy documents, identifies paragraphs of text that may define compliance or non-compliance definitions of patients' benefits and translates them into rules that are both human interpretable and machine executable. We build on our previous work [24][25][26] leveraging different natural language processing techniques and a rich domain ontology (co-created with domain experts), which captures repeatable templates for the translation of policy text into rules, even in the absence of labelled data. The rules, formalized as knowledge graphs, consist of a set of logical conditions having precise semantics (defined in the ontology).…”