Precision in healthcare is possible only if changes in patient state, and response to intervention, can be confidently predicted.
Parity’s AI-enabled healthcare analytics platform provides high-accuracy solutions for actionable clinical analytics, alerting, and patient response prediction.
Parity enables reliable multivariate biomarker discovery and accurate patient response prediction. On a public oncology dataset used to benchmark drug response prediction, Parity achieves over 50% higher specificity than the best published alternative, at comparably high sensitivity levels.
For many drugs, prediction is complex, and there is no traditional univariate biomarker. Multivariate models must incorporate large, diverse datasets, including patient genome, tumor mutations, gene expression, and clinical variables.
A key barrier to success with these models is the danger of overfitting – clinical trials have many potential features (data points) per each of a small number of examples (patients in the trial). This leads to a high risk of the model not generalizing beyond the trial.
Parity addresses the overfitting challenge with a proprietary ensemble of machine learning combined with knowledge automatically derived from the entirety of the research literature, delivering:
- Higher accuracy from the same data, compared to competing techniques
- Faster pharmaceutical trials: improved recruitment and higher response rates
- Simple predictive decision support for clinical users
Example: STRING DB, the leading curated database of protein networks, associates RPL5 with other ribosomal proteins, which is useful for a protein-centric view, but not necessarily informative for cancer patient response prediction.
But Parity’s Literature Based Hypothesis Discovery (LBHD) automatically mines the entirety of the biomedical research literature and finds putative associations of RPL5 to cancer-related genes, enabling more precise feature selection in machine learned response models.
Additional applications of Parity's literature mining include automated knowledge bases associating human gene variants with phenotypes, and prediction of novel gene/phenotype/PGx relationships.
Clinical Analytics and Alerting
Parity’s healthcare analytics platform enables real-time predictive applications to support clinicians in the early detection of critical disease states. Early detection enables rapid response, driving better patient outcomes and lower healthcare costs.
The industry leading accuracy of Parity’s platform is derived from unique applications of artificial intelligence:
- Machine Learning to discover subtle predictive patterns in clinical variables
- Natural Language Processing to accurately mine deep, disease-specific information
- Intelligent physiological signal processing from raw device data streams
- Machine trained models intelligently filter artifacts and noise
- Semantic interpretation of all data to infer context and meaning
Applications of the Parity platform include CV:Sepsis™, a cloud-based solution that automatically monitors hospital patients and brings clinician attention to those at risk of sepsis.
On a combined measure of alert timeliness and specificity, CV:Sepsis outperforms competitive systems by >3x.
Clinical Natural Language Processing (NLP)
The Parity platform enables deep mining of medical records, including domain-specific NLP and semantic analysis across all data types.
In addition to being a key component of CV:Sepsis, applications of Parity’s Clinical NLP include extraction of concepts and relationships to semi-automate patient record abstraction, and fully automated calculation of patient acuity and risk scores.
Clinical NLP is a separately licensable component of Parity’s platform.