International Business Machines Corporation

Merative US L.p.

Eric W Will

Richard Joseph Stevens

Richard D Dettinger

Frederick Allyn Kulack

Adam Clark

Jeffrey Wayne Tenner

Thomas John Eggebraaten

Shannon Everett Wenzel

Fernando Jose Suarez Saiz

David Gerald Herbeck

Mark Gregory Megerian

William Craig Rapp

Joel C Dubbels

Peter John Johnson

Kimberly Diane Kenna

Patrick Michael Wildt

Ikhua Tong

Christopher Chute

James Buntrock

Serguei V Pakhomov

Janice R Glowacki

Brian Edward Olson

Lisa Wellman

Techniques for identifying missing evidence are provided. A plurality of documents, each comprising digitally encoded natural language text data, is received. The plurality of documents is processed to determine a plurality of pair-wise comparisons between a plurality of therapies, where each of the plurality of pair-wise comparisons indicate a relative efficacy of at least one therapy in the plurality of therapies, as compared to at least one other therapy in the plurality of therapies. A knowledge graph is generated based at least in part on aggregating the plurality of pair-wise comparisons, and the knowledge graph is analyzed to identify one or more knowledge gaps within the knowledge graph. Finally, at least an indication of the identified one or more knowledge gaps is output.

Identifying knowledge gaps utilizing cognitive network meta-analysis

Method and apparatus for performing feature engineering using negative inferences are provided. One example method generally includes identifying a plurality of concepts and analyzing a corpus of documents to determine a first co-occurrence rate for a first concept and a second concept in the plurality of concepts. The method further includes analyzing the corpus of documents to determine a second co-occurrence rate for the second concept and at least a third concept of a set of concepts related to the first concept and determining an inverse relationship between the second concept and the third concept. The method further includes generating test data for training a machine learning model including a negative inference between the second concept and the third concept and training the machine learning model using the test data.

Negative inferences in machine learning treatment selection

Methods and apparatuses for performing clinical trial editing using machine learning are provided. One example method generally includes receiving information of a first clinical trial that is being drafted and identifying, in a corpus of literature, a plurality of documents that are relevant to the first clinical trial, based on the title of the first clinical trial. The method further includes providing the plurality of documents and the plurality of criteria to a machine learning model configured to output for each respective criterion of the plurality of criteria, a confidence value for the respective criterion, receiving as output from the machine learning model the confidence value for each respective criterion and, upon determining that a first criterion has a first confidence value below a predefined threshold, prompting a user to verify the first criterion.

Clinical trial editing using machine learning

A method and apparatus for providing curated criteria to identify one or more candidates for a clinical trial is disclosed. A computer processor identifies a first input criterion for the clinical trial. The processor employs a trained first recurrent neural network (RNN) configured as an encoder to encode the first input criterion. The encoder extracts key features of the medical condition of the patient. The processor employs a trained second RNN configured as a decoder to generate a curated output criterion by processing the encoded first input criterion based on the derived key features. The processor employs a machine learning model to ingest the curated output criterion to identify the one or more candidates for the clinical trial.

Recurrent neural network to decode trial criteria

Techniques for unbounded therapy evaluation are provided. A request to suggest a potential therapy based on a patient profile is received. A plurality of accepted therapies is determined based on the patient profile, where the plurality of accepted therapies is based on stored definitions obtained from one or more subject matter experts. Next, a plurality of therapy components is identified based on the plurality of accepted therapies. A plurality of potential therapies is then identified based on the plurality of therapy components, where none of the plurality of potential therapies are included in the plurality of accepted therapies. A score is generated for a potential therapy of the plurality of potential therapies based on analyzing a knowledge graph, where the score indicates a suitability of the potential therapy for a patient associated with the patient profile. Finally, the potential therapy is provided, along with an indication of the score.

Analyzing knowledge graphs with unbounded insight generation

Method and apparatus for predicting clinical trial eligibility for patients. Embodiments include determining a current value for each of a plurality of attributes of a first patient. Embodiments include identifying a cohort of patients that are clinically similar to the first patient, based on the plurality of attributes. Embodiments include analyzing data associated with the cohort of patients to determine an attribute trend for at least a first attribute the plurality of attributes. Embodiments include generating a predicted value for the first attribute, based on the current value of the first attribute and the attribute trend for the first attribute. Embodiments include identifying a plurality of clinical trials based on the first attribute. Embodiments include generating a probability that the first patient will be eligible for each of the plurality of clinical trials at a future time, based on the predicted value for the first attribute.

Predicting clinical trial eligibility based on cohort trends

Method and apparatus for predicting beneficial clinical trials for patients. Embodiments include receiving one or more attributes of a first patient. Embodiments include selecting a potential clinical trial for the first patient, based on the one or more attributes. Embodiments include identifying a cohort of patients that are clinically similar to the first patient, based on the one or more attributes, wherein each patient in the cohort has undergone a respective trial that is either (i) the potential clinical trial, or (ii) a clinically similar clinical trial. Embodiments include determining, for each respective patient in the cohort, a respective outcome of the respective trial. Embodiments include generating a predicted outcome for the first patient, based on the respective outcomes for each patient in the cohort. Embodiments include refraining from recommending the potential clinical trial for the first patient, based on the predicted outcome.

Predicting patients who can benefit from a clinical trial

A method and apparatus for identifying implied criteria for a clinical trial is disclosed. An example method generally includes generating a training data set from a corpus of clinical trial specifications. The training data set may include at least a first sample corresponding to a first trial. The first sample may include a first feature based on one or more explicitly stated trial criteria, a second feature based on metadata describing the first trial, and a third feature based on patient data of patients associated with the first trial. A machine learning model is trained, using a supervised learning approach, based on the training data set. A system processes a second trial as an input to the trained machine learning model to determine one or more implied criteria that are not explicitly enumerated in a specification for the second trial.

Identifying implied criteria in clinical trials using machine learning techniques

Techniques for cognitive analysis of documents are provided. A document corresponding to a clinical trial is received. Criteria are identified for the clinical trial based on processing the electronic document using natural language processing (NLP) techniques, where the criteria specify attributes of individuals that are eligible for the trial. Further, the criteria are analyzed using NLP techniques to determine a respective confidence level for each respective criterion, where the confidence level indicates a degree of certainty that the criterion is applicable to the trial. A knowledge graph is generated based at least in part on the clinical trial, where the confidence level assigned to each criterion is included in the knowledge graph for subsequent use. Finally, one or more therapies in the knowledge graph are selected to be recommended, based on a patient profile and the confidence level assigned to each of the criteria.

Cognitive analysis of criteria when ingesting data to build a knowledge graph

A method and apparatus for filtering clinical trials using machine learning techniques is disclosed. An example method generally includes receiving a first set of filters that were applied to a first plurality of clinical trials with respect to a first patient. A system determines one or more attributes of the first patient and trains a machine learning (ML) model based on the first set of filters and the one or more attributes of the first patient. The system receives a selection of a second patient, determines one or more attributes of the second patient, and generates a second set of filters by processing the one or more attributes of the second patient using the trained ML model.

Methods and systems for recommending filters to apply to clinical trial search results using machine learning techniques

Embodiments of the invention may be used to improve enforcement and compliance with publishing rules in an automated and provable manner. Prior to publication, documents may be processed using publishing rules (workflows) based on conceptual analysis of document content. Additionally, embodiments of the invention include a content creation system configured to provide prompt feedback on content coverage. Such a system enables the creator of information to better understand what approval requirements apply to content they create and intend to publish, as the content is being created.

Content creation system configured to provide feedback on content coverage

Generating document processing workflows configured to route documents based on document conceptual understanding

Processing publishing rules by routing documents based on document conceptual understanding

Method, system, and computer program product to analyze a plurality of candidate answers identified as responsive to a question presented to a deep question answering system, by computing a first feature score for a first feature of an item of evidence, of a plurality of items of evidence, the first feature score being based on at least one attribute of the first feature, the item of evidence relating to a first candidate answer, of the plurality of candidate answers, and computing a merged feature score for the first candidate answer by applying the first feature score to a second feature score for a second feature of the item of evidence.

Growing community of inventors

Oronoco, MN, United States of America

Eric W Will