Chapter 4: Data Quality and Validation

4.1 The Importance of Data Quality

Data quality determines whether research conclusions are trustworthy and findings reproducible. Poor quality data leads to incorrect conclusions, wasted resources, failed replication, and potentially harmful clinical decisions. In medical research, data quality is not merely a technical concern—it is an ethical imperative.

Quality assurance must be built into every stage of research, from study design through data collection, processing, and analysis. Detecting and correcting quality issues early prevents propagation of errors and reduces remediation costs.

4.2 Data Quality Dimensions

Data quality is multidimensional. Each dimension addresses a different aspect of fitness for use:

Core Quality Dimensions

Dimension	Definition	Measurement
Accuracy	Data correctly represents real-world values	Error rates, validation against gold standard
Completeness	Required data elements are present	Missing value percentages
Consistency	Data values are coherent across dataset	Logical checks, cross-field validation
Timeliness	Data is current and available when needed	Data age, update frequency
Validity	Data conforms to defined formats and rules	Format compliance, range checks
Uniqueness	No unintended duplicates	Duplicate detection rates

4.3 Quality Assurance During Collection

Preventing errors is more efficient than correcting them:

Collection Quality Controls

• Electronic Data Capture: Built-in validation rules, range checks, skip logic
• Standardized Instruments: Validated questionnaires, calibrated equipment
• Training: Data collectors trained on protocols, common errors
• Real-Time Validation: Immediate feedback on data entry errors
• Double Data Entry: Independent entry with reconciliation for critical data
• Source Document Verification: Compare entered data against source

Validation Rule Types

Rule Type	Description	Example
Range Check	Value within expected bounds	Age between 0-120 years
Format Check	Data matches expected pattern	Date as YYYY-MM-DD
Cross-Field	Logical relationship between fields	Death date after birth date
Referential	Foreign key exists	Patient ID in patient table
Pattern Match	Regular expression validation	Email format, phone number

4.4 Data Cleaning and Transformation

Post-collection data processing must maintain data integrity while improving quality:

Cleaning Procedures

• Missing Data Assessment: Characterize missingness patterns (MCAR, MAR, MNAR)
• Outlier Detection: Statistical methods (IQR, Z-score) and domain knowledge
• Duplicate Resolution: Identify and merge or remove duplicate records
• Standardization: Normalize formats, units, encodings
• Correction: Fix identifiable errors with documentation

4.5 Data Validation Frameworks

Systematic validation ensures consistency across the data lifecycle:

Framework	Application	Features
Great Expectations	Data pipelines	Expectation suites, data docs
CDISC Conformance	Clinical trial data	Pinnacle 21, OpenCDISC
Pandas Profiling	Exploratory analysis	Automated quality reports
OHDSI DQD	OMOP databases	Data quality dashboard

4.6 Quality Metrics and Reporting

Measuring and communicating quality enables improvement and appropriate data use:

Quality Indicators

• Completeness Rate: Percentage of non-missing values per field
• Conformance Rate: Percentage passing validation rules
• Error Rate: Errors per number of data points
• Plausibility: Percentage of values within expected distributions
• Timeliness: Data latency from source to availability

4.7 Reproducible Data Processing

Data processing must be reproducible for validation and replication:

Reproducibility Requirements

• Code Version Control: All processing scripts in Git with version tags
• Environment Specification: Document software versions, dependencies
• Pipeline Documentation: Describe processing steps and order
• Intermediate Outputs: Save checkpoint datasets for verification
• Automated Testing: Unit tests for data processing functions