> For the complete documentation index, see [llms.txt](https://www.parapathology.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://www.parapathology.com/statistics-and-bioinformatics/sensitivity-specificity-predictive-values-and-decision-making.md).

# Sensitivity, Specificity, Predictive Values & Decision Making

Diagnostic test accuracy (DTA) and medical decision analysis quantify how effectively a diagnostic test, biomarker, or screening tool discriminates between patients with and without a target condition.

## Core Diagnostic Performance Metrics

For a binary diagnostic test evaluated against an established reference standard, cases are categorized in a $2 \times 2$ contingency table:

| Test Outcome              | Disease Present ($D^+$)    | Disease Absent ($D^-$)         | Total                           |
| ------------------------- | -------------------------- | ------------------------------ | ------------------------------- |
| **Test Positive ($T^+$)** | True Positive ($TP$)       | False Positive ($FP$)          | Total Test Positive ($TP + FP$) |
| **Test Negative ($T^-$)** | False Negative ($FN$)      | True Negative ($TN$)           | Total Test Negative ($FN + TN$) |
| **Total**                 | Total Diseased ($TP + FN$) | Total Non-Diseased ($FP + TN$) | Total Cohort ($N$)              |

### Fundamental Definitions

1. **Sensitivity (True Positive Rate):** $$\text{Sensitivity} = \frac{TP}{TP + FN} = P(T^+ \mid D^+)$$ The probability that a diseased individual tests positive. High sensitivity is crucial for **screening tests** to "rule out" disease (SnNOut: high **S**e**n**sitivity, **N**egative test rules **Out**).
2. **Specificity (True Negative Rate):** $$\text{Specificity} = \frac{TN}{TN + FP} = P(T^- \mid D^-)$$ The probability that a disease-free individual tests negative. High specificity is critical for **confirmatory tests** to "rule in" disease before invasive or toxic interventions (SpPIn: high **Sp**ecificity, **P**ositive test rules **In**).
3. **Positive Predictive Value (PPV):** $$\text{PPV} = \frac{TP}{TP + FP} = P(D^+ \mid T^+)$$ The proportion of test-positive individuals who truly have the condition. **PPV depends heavily on disease prevalence** in the tested cohort.
4. **Negative Predictive Value (NPV):** $$\text{NPV} = \frac{TN}{TN + FN} = P(D^- \mid T^-)$$ The proportion of test-negative individuals who are truly disease-free. Like PPV, **NPV varies with pre-test disease prevalence**.
5. **Likelihood Ratios ($LR^+$ and $LR^-$):** Likelihood ratios combine sensitivity and specificity into prevalence-independent metrics:
   * **Positive Likelihood Ratio ($LR^+$):** $$LR^+ = \frac{\text{Sensitivity}}{1 - \text{Specificity}} = \frac{P(T^+ \mid D^+)}{P(T^+ \mid D^-)}$$ ($LR^+ > 10$ indicates strong evidence to rule in disease).
   * **Negative Likelihood Ratio ($LR^-$):** $$LR^- = \frac{1 - \text{Sensitivity}}{\text{Specificity}} = \frac{P(T^- \mid D^+)}{P(T^- \mid D^-)}$$ ($LR^- < 0.1$ indicates strong evidence to rule out disease).
6. **Youden's $J$ Index:** $$J = \text{Sensitivity} + \text{Specificity} - 1$$ Captures overall discriminative ability ($0 \le J \le 1$) and serves as a standard optimization metric for cutoff selection in [ROC analysis](/statistics-and-bioinformatics/roc-analysis.md).

***

## Bayesian Updating & Fagan Nomograms

Diagnostic testing updates the clinical probability of disease via Bayes' rule:

$$\text{Pre-test Odds} = \frac{\text{Pre-test Probability}}{1 - \text{Pre-test Probability}}$$ $$\text{Post-test Odds} = \text{Pre-test Odds} \times \text{Likelihood Ratio}$$ $$\text{Post-test Probability} = \frac{\text{Post-test Odds}}{1 + \text{Post-test Odds}}$$

A **Fagan nomogram** graphically aligns pre-test probability, the calculated likelihood ratio, and post-test probability on three logarithmic axes, enabling clinicians to intuitively appreciate how test results alter diagnostic certainty.

***

## Tooling & Practical Implementation

In the [jamovi](/statistics-and-bioinformatics/jamovi.md) and R ecosystem, [**meddecide**](/statistics-and-bioinformatics/jamovi/meddecide.md) provides dedicated interactive and programmatic tools for medical decision analyses:

* **`decision`**: Computes all accuracy metrics with confidence intervals from patient raw data.
* **`decisioncalculator`**: Instant $2 \times 2$ calculator using summary cell counts ($TP$, $FP$, $TN$, $FN$) without needing raw patient records.
* **`decisioncompare`**: Compares multiple tests against a gold standard using McNemar's test and multi-axis radar plots.
* **`decisioncombine`**: Evaluates parallel, serial, and majority test combination rules.
* **`cotest` & `sequentialtests`**: Models co-testing and staged diagnostic algorithms.
* **`nogoldstandard`**: Fits latent class models (LCA) when the reference standard is imperfect.

***

## See Also

* [**meddecide**](/statistics-and-bioinformatics/jamovi/meddecide.md)
* [**ROC analysis**](/statistics-and-bioinformatics/roc-analysis.md)
* [**Kappa**](/statistics-and-bioinformatics/kappa.md)
* [**Power Analysis**](/statistics-and-bioinformatics/power-analysis.md)
* [**Jamovi**](/statistics-and-bioinformatics/jamovi.md)
* [**Statistics and Bioinformatics**](/statistics-and-bioinformatics/statistics-and-bioinformatics.md)