Cognitive Load

A user interface design pattern that sequences complex clinical information and advanced controls to reduce cognitive load, revealing details only as they become contextually necessary.

• Mechanism: Initially presents only high-level AI findings (e.g., '3 medications extracted'). Detailed evidence, confidence scores, and correction tools appear on demand or hover.
• Clinical Application: A reviewer sees a summary card for a prior authorization; clicking expands the specific FHIR resources and source text snippets.
• Benefit: Prevents information overwhelm, allowing the reviewer to focus on a single decision at a time rather than parsing a dense, fully-populated screen.

A visual comparison interface that highlights the specific textual, structural, or coding differences between an AI-generated output and a human-corrected version to accelerate validation.

• Inline Highlighting: Uses color-coded strikethroughs (red for deletions) and underlines (green for insertions) directly on the clinical text.
• Field-Level Comparison: For structured data, a side-by-side table shows the AI-extracted value versus the corrected value for each attribute (e.g., dosage, frequency).
• Cognitive Impact: Eliminates the need for the reviewer to mentally compare two separate documents, reducing extraneous cognitive load and cutting correction time by up to 40%.

A feature that directly links an AI-generated clinical statement or code to the exact sentence or paragraph in the original medical record, enabling rapid evidence verification.

• Implementation: Clicking an extracted entity (e.g., 'Type 2 Diabetes Mellitus') scrolls the source document viewer to the originating text and applies a temporary highlight.
• Bidirectional Linking: Hovering over source text can also indicate which AI extractions were derived from that specific passage.
• Load Reduction: Removes the high-effort task of manually scanning a lengthy clinical note to confirm an AI assertion, transforming verification from a search task into a simple confirmation task.

The automated prioritization and categorization of review queue items based on urgency, clinical severity, or model uncertainty to ensure the most critical cases are handled first.

• Uncertainty-Based Queue: Items where the model's confidence threshold is low (e.g., < 0.85) are surfaced to the top, as they are most likely to contain errors.
• Clinical Severity Sorting: Documents mentioning high-acuity conditions or critical medications are flagged for immediate review.
• Cognitive Benefit: Prevents the alert fatigue associated with reviewing a random mix of easy and difficult cases. Reviewers can maintain a high straight-through processing (STP) rate on confident items while focusing mental effort on ambiguous edge cases.

A granular annotation task where a human reviewer adjusts the start and end character offsets of a highlighted medical entity in unstructured text to fix extraction boundary errors.

• Direct Manipulation: The reviewer drags the boundary handles of a highlighted span (e.g., 'chest pain') to include a missed modifier (e.g., 'severe chest pain radiating to left arm').
• Alternative to Re-typing: Avoids the high cognitive cost of manually re-entering structured data by allowing a quick, mouse-driven correction of the AI's initial selection.
• Downstream Impact: Precise span correction feeds directly into an active learning loop, providing high-quality token-level feedback to improve the underlying medical named entity recognition model.

A mechanism that automatically applies a single human correction to identical or semantically similar errors across a batch or downstream dataset to maintain consistency.

• Batch Application: If a reviewer corrects a misspelled drug name in one document, the system can optionally apply that same fix to all other instances in the current review queue.
• Semantic Propagation: Using dense vector embeddings, the system identifies and flags semantically identical errors (e.g., a consistently mislabeled lab value) for one-click bulk correction.
• Efficiency Gain: Directly attacks the review burden metric by preventing a human from performing the same repetitive correction dozens of times, preserving mental stamina for novel, complex clinical judgments.

A user interface design pattern that sequences complex clinical information and advanced controls to reduce cognitive load, revealing details only as they become contextually necessary. Core mechanism: Information is initially hidden and displayed on demand, preventing the reviewer from being overwhelmed by the full complexity of a patient record.

• Primary content: Shows only the AI-extracted entity and its immediate source context
• Secondary content: Expands to reveal full document, confidence scores, and related ontologies
• Clinical impact: Reduces time-to-decision by 30-40% in radiology report review interfaces
• Implementation: Uses accordion panels, hover tooltips, and progressive form wizards

A visual comparison interface that highlights the specific textual, structural, or coding differences between an AI-generated output and a human-corrected version to accelerate validation. Core mechanism: Inline diffs with color-coded additions (green) and deletions (red) allow the reviewer to instantly perceive what changed without mentally comparing two documents.

• Granularity levels: Character-level for span corrections, field-level for structured data, document-level for summaries
• Cognitive benefit: Eliminates the need for working memory to hold two versions simultaneously
• Common pattern: Side-by-side panels with synchronized scrolling and unified diff views
• Error prevention: Reduces missed corrections by making every delta visually salient

A feature that directly links an AI-generated clinical statement or code to the exact sentence or paragraph in the original medical record, enabling rapid evidence verification. Core mechanism: Bidirectional highlighting between the extracted structured field and its provenance in the unstructured source text.

• Implementation: Click on an extracted diagnosis to scroll and highlight the source sentence in the original note
• Cognitive load reduction: Eliminates the need for reviewers to manually search lengthy documents
• Trust calibration: Builds appropriate reliance by making the model's evidence transparent
• Audit efficiency: Reduces source verification time from minutes to seconds per entity

The automated prioritization and categorization of review queue items based on urgency, clinical severity, or model uncertainty to ensure the most critical cases are handled first. Core mechanism: A scoring function that assigns each item a priority score derived from confidence thresholds, clinical risk factors, and service-level agreements.

• Uncertainty-based routing: Low-confidence predictions appear first, when reviewer attention is freshest
• Severity weighting: Critical findings (e.g., malignancy mentions) are escalated regardless of confidence
• Batch optimization: Groups similar task types to reduce context-switching costs
• Cognitive benefit: Prevents decision fatigue by ensuring high-cognitive-load tasks are distributed across a shift

A state of desensitization caused by an excessive volume of low-value or false-positive notifications, leading clinicians to ignore or override critical AI-generated warnings. Core mechanism: When the signal-to-noise ratio of alerts drops below a tolerable threshold, the human operator's attention system adaptively filters out all alerts—including genuine ones.

• Root cause: Overly sensitive models with poor precision that flag every minor deviation
• Clinical consequence: A 49-96% override rate for drug interaction alerts in EHR systems
• Mitigation: Tiered alerting with severity levels, suppression of duplicate alerts, and contextual relevance filtering
• Design principle: Every alert must earn its interruption cost by providing actionable, high-stakes information

A specialized interface component designed to visually align and compare two conflicting data sets, such as an AI-derived medication list and an existing EHR record, for manual merging. Core mechanism: A unified table or card layout that places conflicting entries in adjacent rows with clear visual indicators of match, mismatch, or absence.

• Visual encoding: Green for matches, amber for partial matches, red for conflicts, grey for items present in only one source
• Action affordances: One-click accept, reject, or modify for each discrepancy
• Cognitive load reduction: Transforms a complex comparison task into a simple visual scan
• Error rate: Structured reconciliation UIs reduce medication reconciliation errors by up to 50% compared to free-text review