Understanding Precision and Bias in Clinical Testing
Precision measures how consistently a test produces the same result when repeated under similar conditions, while bias (measurement trueness) reflects how accurately the test's average result matches the true value—precision addresses random error, whereas bias addresses systematic error. 1
Core Definitions
Precision
Precision is the closeness of agreement of replicate test results under stipulated conditions 1. This concept has several important subcategories:
- Repeatability: Precision when measurements are taken under identical conditions within a single run
- Intermediate precision: Precision when some conditions vary (different days, operators, reagent lots) while others remain constant
- Reproducibility: Precision when testing occurs across different laboratories 1
Precision reflects random error in the measurement process. A precise test will give you nearly the same result each time you measure the same sample, even if that result is systematically wrong 2.
Bias (Measurement Trueness)
Measurement trueness is the closeness of agreement between the average of an infinite number of replicate test results and the reference value—trueness is usually expressed numerically by the statistical measure bias 1.
Bias represents systematic error—a consistent deviation from the true value in one direction 2. For example, if a glucose meter consistently reads 10 mg/dL higher than the actual value, this represents positive bias.
Measurement Accuracy: The Complete Picture
Measurement accuracy is the closeness of agreement between a measurement result and an accepted reference value; it is an aggregate of both trueness (bias) and precision 1. A test can be:
- Precise but biased (consistently wrong in the same direction)
- Accurate on average but imprecise (scattered around the true value)
- Both precise and unbiased (ideal state)
Clinical Implications and Sources of Error
Origins of Bias and Imprecision
In clinical practice, bias and imprecision arise from multiple sources 2:
Technical factors:
- PET/CT and display hardware
- Quantitative corrections (scatter, attenuation)
- Acquisition protocols
- Calibration errors
Biological and operational factors:
- Operator errors (particularly in region of interest definition)
- Partial volume errors in imaging
- Specimen handling variations
- Pre-analytical variables 3
The most significant origins of bias and variance are biological and operational rather than purely technical 2. Under controlled conditions, modern equipment can achieve bias and precision approaching 1%, but real-world performance is substantially worse due to these human and biological factors 2.
Practical Assessment and Management
Determining "Truth"
'Truth' regarding bias can rarely be determined in human studies 2. The primary methods to assure quantitative accuracy include:
- Careful calibration with phantoms of known concentration
- Protocol definition and monitoring
- Comparison to reference standards when available 2
Clinical Significance Thresholds
The clinical significance of imprecision varies by context. For example, in FDG-PET imaging:
- Single-center studies: Changes >20% in SUVmax are likely significant
- Multi-center studies: Changes >30% are required to confidently detect true progression or regression in individual patients 2
These thresholds reflect the combined impact of technical precision, biological variation, and operational factors.
Key Pitfalls to Avoid
Confusing precision with accuracy: A test can be highly precise (reproducible) but systematically biased, making all results consistently wrong 4
Ignoring intermediate precision: Testing only within-run precision misses important sources of variation from day-to-day operations, different operators, and reagent lot changes 1, 3
Assuming fixed test characteristics: Diagnostic accuracy measures (sensitivity, specificity) vary across clinical contexts, populations, and disease severity—they are not fixed properties of a test 5
Neglecting performance around cutoffs: Measurement accuracy near clinical decision thresholds is particularly critical, as small variations can change categorical interpretations (positive/negative, normal/abnormal) 1
Overlooking biological variation: The observed variation in serial measurements includes both analytical imprecision and true biological variation within the patient 6
Quality Control Implications
Routine quality control samples typically monitor only precision (reproducibility/random error) and not trueness/bias (systematic error) 7. To comprehensively assess test performance:
- Use precision controls to monitor random error over time
- Use accuracy controls (trueness controls) to assess both systematic and random error
- Implement both within-run and between-run monitoring 3, 7
Acceptable performance standards generally require: