Aided Intelligence for Healthcare

ayfie can artificially derive conceptual meaning from even small datasets, such as patient notes. This text is analyzed by using a combination of Machine Learning and Linguistics. With robust and customizable dictionaries and grammars, meaning can be analyzed on a few sentences, a few paragraphs, an entire document. This can then be processed against a rule set for compliance, measured to detect costly cloning of information or tied to a structure ontology for Computer Assisted Coding (CAC).

The unique and powerful ability ayfie holds to understand language and transform it into predicate structures, which are in turn accessible to mathematical methods, also powers previously unattainable insights on unstructured patient notes. The combination of human-guided intelligence with statistical analysis and machine learning has given healthcare providers new tools to manage quality care initiatives. For example, tying insight into workflow alerting allows for intervening care for patients with a high-risk of readmission.

Alerts are flexible and intuitive. Social resources don't need to rely on data systems experts to adjust their alerting rules; they can easily define new pattern-matching and correlation strategies that are informed by their domain expertise. Detection strategies should evolve as new risks are identified. Now they can.

ayfie Local Grammars

The only way to correctly and efficiently extract the right information.

In contrast to popular approaches to syntax and parsing, ayfie is strongly committed to the use of extremely large local grammar systems which can analyze substantial portions of natural language with a very high degree of accuracy.

Because we are still a far cry from a comprehensive and semantically realistic treatment of large fragments of any natural language, ayfie aims at describing well-understood semantic subsets in a detailed way. This holds for the description of both argument and predicates. The latter express relations between arguments (ayfie calls them “propositional forms”) and they come in a number of very different forms that are rarely distinguished in other approaches to syntax.

Local grammars for propositional forms have an advantage, in that, they identify the nature of the predicates by separate grammars for the different ways of expressing propositional forms (thereby also capturing the many variants of the same underlying propositional form). On the other hand, there are also very detailed grammars for the different types of argument structures involving such notions as persons, dates, organizations, locations and many others, each which requires a grammar specification of its own.

General, rule-based grammars that do not make such semantic distinctions from the beginning cannot capture the structure of utterances in a reliable way.

Best of Breed in Text Analytics in Healthcare

Similar conceptually to "spell check," the software uses natural language processing in order to highlight key terms and phrases for ICD-9 CM, ICD-10 CM. The aided intelligence of ayfie allows for easier, less labor-intensive coding.

The hospital compliance risk could increase faster than before with the Civil Monetary Penalties (CMPs) for violating the False Claims Act (FCA) increasing substantially under an interim final rule that the Department of Justice (DOJ) published June 30 (81 Fed. Reg. 42491). The interim final rule, which is effective August 1, raises per claim penalties for FCA violations from the current per-claim range of $5,500 to $11,000 to a new minimum-maximum range of $10,781 and $21,562.

ayfie correlates result data to concept extractions from unstructured information that appears in medical records for desired care and business result. This is discussed in the case study below for the problem of readmissions.

Use Case: Determining High Risk Patients for Re-admittance Through Physician Notes

In the last 10 years, it’s become far more common for physicians to keep records electronically. Those records could contain a wealth of medically useful data:

  • hidden correlations between symptoms, treatments and outcomes
  • indications that patients are promising candidates for trials of new drugs

Much of that data, however, is buried in physicians’ freeform notes. To date, no one has effectively been able to extract meaningful data from such unstructured text due to word-sense disambiguation. In a physician’s notes, the word “discharge,” for instance, could refer to a bodily secretion or to release from a hospital.

Our technology can infer words’ intended meanings, which allows us to build never before seen insights into physician notes. We recently partnered with a top 50 healthcare provider in the United States, to analyze over 30 million patient progress notes and ultimately develop an application that can both detect and notify administrators when new patients are at high risk for re-admittance, in real time while the patient is still being treated at the hospital. Accordingly, this application is highly scalable as it can easily be adapted to other health care institutions. From an industry perspective, the value of this application is significant.

Recently, in the U.S., the Department of Health and Human Services mandated a transition from a fee-for-service delivery system to a value-based system, whereby healthcare providers would be reimbursed for Medicare and Medicaid patients - not based on services performed, but for the overall care provided for a particular health matter. Accordingly, the need for a product that provides tools to reduce readmissions and improve both the quality and efficiency with respect to patient care has never been greater. This process can aide in mitigating readmissions and the potential for severe peanlties.

At a Glance

  • Primary decision driver: Aided intelligence, information extraction and visualization
  • Value Proposition: Create predictive alerts for Medicare patients with high risk of re-admissions at time of care based on free form notes in medical documentation
  • Pain-point: No ability to analyze largely unstructured information collected by medical professionals. Hospitals are subject to claw backs of insurance payouts by the government in the event of high re-admittance
  • Delivery: create structure from unstructured data that (a) correlate notes with common traits of high risk patients to (b) create an accurate detection system to notify hospital administrators when a patient currently in care is at high risk, psychosocial factors in re-admissions, correlated information with patient interviews, re-admissions

Leveraging Machine Learning, Linguistic Analysis and years of experience solving complicated problems.

Contact us