Overview of eIQ Review

Permissions: eIQ Review: Allows licensed users to access eIQ Review in Data Central. Default access is not granted. Access to eIQ Review must be manually enabled in User Management.

eIQ Review can be enabled for studies in Data Central. eIQ Review provides users with a way to conduct an assisted data review. Users can see a list of data points that need human review with pre-set visualizations that support their review. The output can be used to improve data quality and increase efficiency in the data review process. They can drill down to underlying subject data to better understand and visualize the atypical data, ask further questions about the data, and finally take review actions, all without leaving the eIQ Review panel.

Available eIQ Review Models

Central Statistical Monitoring applied to Labs and Vitals - CSM Labs and Vitals
A subject-level univariate (1-dimensional) anomaly detection model that runs on SDTM mapped data and identifies subjects that have atypical labs and / or vitals. Users receive a list of subjects with an anomalous lab or vital sign.
Anomalous AE Durations - AE Durations
A Machine Learning (ML) model trained on historic clinical data. The model predicts whether reported AEs have anomalous durations. Users receive a list of records with anomalous AE durations. This model requires SDTM mapped data.
Anomalous Shift in Labs and Vitals - Shift in Labs and Vitals
A Machine Learning (ML) model trained on historical data. The model predicts data discrepancies that have an atypical shift in lab values and vital signs within both normal and out-of-range values. Users get a list of records that have an anomalous shift in labs and vital signs. This model requires using SDTM mapped data.
General Data Review - Univariate Outlier Detection
A study-agnostic outlier detection model that runs on raw data and identifies atypical values in several domains (Labs, Safety, CM, MH, etc.). Output is based on selected variables within configuration. Users get a list of values that are atypical according to patterns identified in the current study.
General Data Review - Incorrect Item Type
A study-agnostic item type detection model that runs on raw data and identifies incorrect item types in several domains (Labs, Safety, CM, MH, etc.). Output is based on selected variables within configuration. Users get a list of values that have the incorrect type according to patterns identified in the current study (e.g., non-numeric value in a numeric field).
Anomalous CM Durations - CM Durations
A Machine Learning (ML) model trained on historic data. The model predicts if reported Concomitant Medications (CMs) have an anomalous duration by reviewing records with an end date or that are ongoing. This model requires using SDTM mapped data.
Domain Classification - MH and AE - Domain Classification
A classification predictive model that predicts the correct domain based on coded and free text fields. The model uses natural language processing (NLP) and large language models (LLM), using supervised learning, and is trained on historical data with the target ground truth labelled by data review experts. The model identifies records and data that may have been entered in the incorrect form or domain and requires correction from the site to ensure data quality. It automates the following checks:
- Medical History condition / event entry corresponds to a Condition / Event
- Adverse Event entry corresponds to a Condition / Event
Concomitant Medication (CM) Consistency - CM Consistency
A Machine Learning (ML) model trained on historical data using semi-supervised learning with some records ground truth labelled by data review experts. It uses an anomaly detection algorithm to identify fields within CM records that are inconsistent, incorrectly entered as other or have missing data, and provides suggestions for correct data entry. The model includes predictions for dose unit in the recorded data set.
Concomitant Medication (CM) Indication - CM Indication
A Machine Learning (ML) model identifying inappropriate medication-indication pairs in the study data by cross-referencing recorded indications with the open-source dataset. It assigns a matching score between the recorded indications and the indications retrieved from the open-source dataset to curate a list of anomalous pairs that need human review. Detection of such data discrepancies helps clinical data review teams identify data quality issues more efficiently, thereby reducing the need for manual checks.
Audit Trail Review Model - Irregular Data Patterns
A model that identifies site-level irregular data patterns across multiple indicators, such as irregular amounts of data deletions, data changes per domain, per user, after verification, and per month. Each indicator has a pre-defined sheet in the Audit Trail Review Workspace that includes both visualizations supporting the review and data listings supporting issue creation.

Review, Feedback, and Issue Management

Tip: Usual Data Central functionality is available, such as filtering and sorting columns, using panel icons to export (download) a listing, creating issues, marking records as reviewed, and maximizing / restoring a panel. Users can also float additional lists and charts on an eIQ Review sheet.

Review Workflow

Records identified by eIQ Review are configurable for review. A single record or multiple records can be marked as reviewed or unreviewed, and all records within a panel can be marked as reviewed or unreviewed. Records marked as reviewed do not appear as requiring review again unless the data record changes before a subsequent import. By default, most listings have pre-programmed filters applied to show only records that are new or updated since the last review. Review statuses are tracked separately for each reviewer role. Only the reviewer roles configured to review that domain display. Additional reviewer roles can be configured based on the study's needs. Steps to mark records as reviewed / unreviewed are the same for marking records as reviewed in Data Central.

Model Feedback Option

For all eIQ Review models, except CSM Labs and Vitals and Irregular Data Patterns, users performing reviews can provide feedback within the Anomalous Records listing, indicating whether the anomalous prediction is accurate.

To provide feedback on the model prediction:

For a single record: Click the Thumbs Up icon in a row to accept the prediction, the Thumbs Down icon to mark the prediction as incorrect and mark the record as reviewed, or right-click a record and select the from the Feedback options.
For multiple records:
1. Click the checkboxes at the left of the rows.
2. Click the Thumbs Up icon in the panel toolbar.
3. Select from the drop-down: Accept prediction, Incorrect prediction, Mark as Reviewed, or Remove Feedback(s).

Issue Creation

Users may also create issues within the eIQ Review models. Issue text is auto populated based on the model, eliminating the need to manually input the issue text. Steps to create issues are the same as creating issues in Data Central.