Many factors in the SEND preparation process contribute to inconsistency with the authoritative and audited Study Report. But a persistent issue is the lack of standard terminology and consistent parsing of qualitative data such as in MI – Microscopic, MA-Macroscopic and CL-Clinical Observations that will improve quality and reduce costs.
This paper describes a continuously improving process using machine learning algorithms driven by a digital representation of the Study Report to provide recommendations automatically for parsing observations to STRESC, Modifiers and Severity.
The recommendation engine semantically recombines the SEND components to match the findings as reported in the Study Report allowing the automated comparator tool to check the consistency of the qualitative incidence counts and the quantitative data in SEND against the PDF Report.
Business Problem
In general Pathologist and veterinarians either enter their original findings in Clinical Observations or Pathology data entry systems which is then mapped to ORRES in SEND data generation These ORRES are further split by certain automation rules and patterns, and converted to Base Pathological Process, Modifiers and Severity. Modifiers that are commonly used include organ-specific topography, distribution, character of the change and duration (Frame and Mann, 20083).
The Base Pathological terms and modifiers that are created either by the rule-based approach or by Experts curating each unique terms. Often these methods are not reliable and inconsistent within or across studies. The FDA requisite of submitting Base Pathological Terms, Severity and modifiers along with ORRES aids dynamic way of performing incidence count at different levels. As certain base pathological terms appear in more than an organ, the toxicity pattern of a drug in one or more targets organs will be exposed out easily when taking Base terms separately along with severity for group summary incidence tabulation. The splitting of MI/MA lesions to the granularity as expect by FDA requires very scrupulous attention as it may easily compromise the ability to detect a test-article effect or may lead to the appearance of a test-article effect when none is actually present.
Read Full Paper: Ensuring consistency of SEND Datasets with Study Reports using Machine Learning Algorithms
Many factors in the SEND preparation process contribute to inconsistency with the authoritative and audited Study Report. But a persistent issue is the lack of standard terminology and consistent parsing of qualitative data such as in MI – Microscopic, MA-Macroscopic and CL-Clinical Observations that will improve quality and reduce costs.
This paper describes a continuously improving process using machine learning algorithms driven by a digital representation of the Study Report to provide recommendations automatically for parsing observations to STRESC, Modifiers and Severity.
The recommendation engine semantically recombines the SEND components to match the findings as reported in the Study Report allowing the automated comparator tool to check the consistency of the qualitative incidence counts and the quantitative data in SEND against the PDF Report.
Business Problem
In general Pathologist and veterinarians either enter their original findings in Clinical Observations or Pathology data entry systems which is then mapped to ORRES in SEND data generation These ORRES are further split by certain automation rules and patterns, and converted to Base Pathological Process, Modifiers and Severity. Modifiers that are commonly used include organ-specific topography, distribution, character of the change and duration (Frame and Mann, 20083).
The Base Pathological terms and modifiers that are created either by the rule-based approach or by Experts curating each unique terms. Often these methods are not reliable and inconsistent within or across studies. The FDA requisite of submitting Base Pathological Terms, Severity and modifiers along with ORRES aids dynamic way of performing incidence count at different levels. As certain base pathological terms appear in more than an organ, the toxicity pattern of a drug in one or more targets organs will be exposed out easily when taking Base terms separately along with severity for group summary incidence tabulation. The splitting of MI/MA lesions to the granularity as expect by FDA requires very scrupulous attention as it may easily compromise the ability to detect a test-article effect or may lead to the appearance of a test-article effect when none is actually present.
Read Full Paper: Ensuring consistency of SEND Datasets with Study Reports using Machine Learning Algorithms