Interpreting Your Results
General Statistics Panel
The General Statistics panel reports the primary agreement metrics for the study. Interpret each metric as follows:
| Metric | What It Tells You |
|---|---|
| Overall Agreement (with CI) | The proportion of specimens classified identically by both methods. Review the 95% confidence interval — a wide CI indicates that more specimens are needed for a reliable estimate. |
| McNemar Test (PASS/FAIL) | PASS (p > 0.05) indicates that discordant results are distributed symmetrically — neither method systematically over-calls relative to the other. FAIL (p ≤ 0.05) indicates asymmetric discordance; investigate whether one method is biased toward positive or negative calls. |
| Positive Agreement | The fraction of reference-positive specimens also classified as positive by the test method. Low PPA indicates the test method may miss positive specimens. |
| Negative Agreement | The fraction of reference-negative specimens also classified as negative by the test method. Low NPA indicates the test method may produce false positives. |
| Test > Reference | Count and percentage of specimens where the test method assigned a higher result level than the reference. High counts suggest systematic over-calling. |
| Test < Reference | Count and percentage of specimens where the test method assigned a lower result level than the reference. High counts suggest systematic under-calling. |
| Cohen's Kappa (with CI) | Chance-corrected agreement coefficient. Values ≥ 0.80 indicate strong agreement; 0.60–0.79 moderate agreement; < 0.60 warrants investigation. Review the CI — a lower CI bound below 0.60 with limited specimens suggests insufficient precision. |
Confusion Matrix
The confusion matrix cross-tabulates reference method results (rows) against test method results (columns). Each cell shows the number of specimens assigned to that combination of classifications.
- Diagonal cells (green) — specimens where both methods assigned the same result level; these represent agreement.
- Off-diagonal cells — specimens where the two methods assigned different result levels; these represent discordance. Examine which cells are populated to determine whether the test method tends to over-call or under-call.
Bubble Chart
The bubble chart is a visual representation of the confusion matrix where bubble size reflects the number of specimens in each cell. Color coding highlights the clinical significance of each cell:
- Green bubbles — diagonal cells representing agreement between the two methods.
- Yellow bubbles — one-level discordance (adjacent off-diagonal cells). For a three-level scheme, this includes Negative vs. Indeterminate and Indeterminate vs. Positive disagreements.
- Red bubbles — two or more levels of discordance (non-adjacent off-diagonal cells). For example, a specimen called Positive by the reference method and Negative by the test method.
Ideally, all specimens fall on the diagonal (green). Large yellow bubbles warrant investigation; any red bubbles represent the most clinically significant discordances and should be reviewed individually before the method is accepted.
Configured Cutoffs Table
When custom result codes or numeric cutoffs are configured, the Visual Analysis section displays a Configured Cutoffs table summarizing the classification boundaries applied to the data. This table lists each defined level, the corresponding numeric range or categorical code for both the reference method and the test method, and the report label assigned to each level. Confirm that these boundaries match your instrument's published cutoff values and your laboratory's classification rules before accepting the study results.
Acceptance Criteria
The module does not assign an overall pass or fail determination — acceptance is the responsibility of the laboratory director. CLSI EP12 does not specify universal numeric thresholds for agreement; acceptance criteria should be defined prospectively based on the clinical application of the test and the laboratory's quality standards. As a practical reference:
- CLSI EP12 does not specify universal numeric thresholds for PPA or NPA. Acceptance criteria should be defined prospectively, consistent with the manufacturer's package insert claims and the clinical application of the test. Many laboratories require ≥90% PPA and NPA for routine qualitative tests; safety-critical applications may require ≥95% or 100% PPA.
- Confidence intervals for PPA and NPA are wide with small sample sizes. CLSI EP12 shows that with 20 total specimens (10 positive, 10 negative) and 100% agreement, the 95% CI for PPA spans 72%–100%. More specimens narrow the interval and make pass/fail determinations more defensible.
- If Cohen's Kappa is included as a supplemental metric, values ≥0.80 are generally interpreted as strong agreement in the broader literature; values below 0.60 warrant investigation. Kappa is sensitive to prevalence — a very high or low proportion of positives can produce misleadingly low Kappa even with excellent PPA and NPA. It should not be the sole basis for acceptance.
- A significant McNemar test result (p ≤ 0.05) should prompt investigation of whether the test method is systematically biased toward positive or negative calls.
- All red-bubble discordances in the bubble chart (2+ levels of disagreement) should be individually reviewed and documented before the method is accepted.