4.2 Diagnostic Testing, Screening & Predictive Values

[!NOTE] Curriculum Focus: The MRCP(UK) Part 1 frequently tests mathematical calculations based on 2x2 contingency tables. Candidates are expected to rapidly compute sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and positive/negative likelihood ratios (LR+ and LR-), as well as interpret the clinical consequences of altering diagnostic cut-off points on ROC curves.

Diagnostic tests do not provide absolute diagnostic certainty; instead, they alter the clinical probability that a condition is present. Evidence-based diagnostic decision-making relies on applying Bayesian principles to convert pre-test probability into post-test probability.

---

The Core 2x2 Contingency Table

All fundamental diagnostic metrics are derived from a standard 2x2 table comparing the index test result against the reference ('gold standard') diagnosis:

Intrinsic Test Parameters: Sensitivity and Specificity

Sensitivity and specificity describe the performance of a test against an established reference standard. They are intrinsic biological properties of the test; assuming the diagnostic threshold and clinical spectrum remain constant, they do not change when the disease prevalence changes.

1. Sensitivity (True Positive Rate)

• Formula: Sensitivity = True Positives / Total Diseased = a / (a + c)
• Clinical Meaning: The probability that a patient with the disease tests positive.
• Clinical Mnemonic: SnNOut — A test with high Sn (Sensitivity), when N (Negative), rules Out the disease. A highly sensitive test generates very few false negatives (FN). Example: D-dimer in low-probability pulmonary embolism; high sensitivity ensures a negative result safely excludes PE.
• False Negative Rate (FNR): The proportion of diseased individuals missed by the test: False Negative Rate = 1 - Sensitivity = c / (a + c)

2. Specificity (True Negative Rate)

• Formula: Specificity = True Negatives / Total Non-Diseased = d / (b + d)
• Clinical Meaning: The probability that a patient without the disease tests negative.
• Clinical Mnemonic: SpPIn — A test with high Sp (Specificity), when P (Positive), rules In the disease. A highly specific test generates very few false positives (FP). Example: Anti-dsDNA antibodies in systemic lupus erythematosus; high specificity confirms SLE.
• False Positive Rate (FPR): The proportion of healthy individuals falsely flagged as diseased: False Positive Rate = 1 - Specificity = b / (b + d)

---

Predictive Values and Prevalence Dependence

Unlike sensitivity and specificity, predictive values reflect clinical practice directly: "Given my patient's positive or negative test result, what is the probability that they actually have the condition?"

Predictive Value Formulas

• Positive Predictive Value (PPV): PPV = True Positives / Total Test Positives = a / (a + b)
• Negative Predictive Value (NPV): NPV = True Negatives / Total Test Negatives = d / (c + d)

Impact of Disease Prevalence (Bayes' Theorem in Practice)

Predictive values are heavily dependent on disease prevalence in the tested cohort:

+------------------------------------------------------------------------------------------+
|                       Impact of Changing Disease Prevalence                              |
+------------------------------------------------------------------------------------------+
| Disease Prevalence INCREASES:                                                            |
| • Positive Predictive Value (PPV) INCREASES                                              |
| • Negative Predictive Value (NPV) DECREASES                                              |
|                                                                                          |
| Disease Prevalence DECREASES (e.g., General Population Screening):                       |
| • Positive Predictive Value (PPV) PLUMMETS (Most positive tests are False Positives!)   |
| • Negative Predictive Value (NPV) APPROACHES 100%                                        |
+------------------------------------------------------------------------------------------+

Worked Example: The Screening Paradox

Consider an assay with 90% sensitivity and 95% specificity applied to two different populations of 10,000 individuals:

1. High-Prevalence Clinic Cohort (Prevalence = 20%):

   • Diseased: 2,000 patients | Healthy: 8,000 patients
   • True Positives = 2,000 x 0.90 = 1,800
   • False Positives = 8,000 x (1 - 0.95) = 400
   • PPV = 1,800 / (1,800 + 400) = 1,800 / 2,200 = 81.8%

2. Low-Prevalence Population Screening (Prevalence = 1%):

   • Diseased: 100 patients | Healthy: 9,900 patients
   • True Positives = 100 x 0.90 = 90
   • False Positives = 9,900 x 0.05 = 495
   • PPV = 90 / (90 + 495) = 90 / 585 = 15.4%

[!IMPORTANT] MRCP Exam Trap: In the low-prevalence screening setting, despite an excellent 95% specificity, nearly 85% of all positive test results are false positives. This is the fundamental reason unselected screening for rare disorders is discouraged.

---

Likelihood Ratios and the Fagan Nomogram

Likelihood ratios (LRs) combine sensitivity and specificity into a single metric that quantifies how much a given test result raises or lowers the odds of disease. Crucially, likelihood ratios are independent of disease prevalence.

Mathematical Definitions

• Positive Likelihood Ratio (LR+): LR+ = Sensitivity / (1 - Specificity) = True Positive Rate / False Positive Rate
  • Describes the odds of a positive test in a diseased patient compared to a non-diseased patient.
• Negative Likelihood Ratio (LR-): LR- = (1 - Sensitivity) / Specificity = False Negative Rate / True Negative Rate
  • Describes the odds of a negative test in a diseased patient compared to a non-diseased patient.

Clinical Interpretation Benchmarks

Bayesian Odds Calculations and Fagan Nomogram

To convert probabilities using likelihood ratios:

1. Convert Pre-test Probability (P) to Pre-test Odds: Odds = P / (1 - P)
2. Calculate Post-test Odds: Post-test Odds = Pre-test Odds x Likelihood Ratio
3. Convert Post-test Odds back to Post-test Probability: Probability = Odds / (Odds + 1)

In the clinic, a Fagan Nomogram bypasses calculation: aligning a straight ruler connecting Pre-test Probability on the left vertical axis through the calculated Likelihood Ratio on the central vertical axis allows direct reading of Post-test Probability on the right axis.

---

Receiver Operating Characteristic (ROC) Curves

When a diagnostic test yields a continuous numerical variable (e.g., serum troponin, PSA, ferritin), clinicians must select a cut-off threshold to designate results as 'positive' or 'negative'. An ROC curve visualizes the trade-off across all potential cut-off values.

   True Positive Rate (Sensitivity)
   1.0 +---------------------------------*
       |                             *   |
       |                         *       |
   0.8 |                      *          |
       |                   *             |
   0.6 |                *                |
       |             *                   |
   0.4 |          *                      |  Diagonal line: AUC = 0.5
       |       *                         |  (Chance discrimination)
   0.2 |    *                            |
       | *                               |
   0.0 +---------------------------------+
      0.0   0.2   0.4   0.6   0.8   1.0
            False Positive Rate (1 - Specificity)

Core Properties of the ROC Curve

• Axes: The vertical y-axis plots Sensitivity (True Positive Rate). The horizontal x-axis plots 1 - Specificity (False Positive Rate).
• Area Under the Curve (AUC / c-statistic):
  • AUC = 1.0: Perfect discrimination (100% sensitivity with 100% specificity).
  • AUC = 0.5: Discrimination no better than random chance (represented by the 45-degree diagonal reference line).
  • AUC 0.70 – 0.79: Acceptable discriminatory accuracy.
  • AUC 0.80 – 0.89: Excellent discriminatory accuracy.
  • AUC >= 0.90: Outstanding clinical accuracy.
• Youden's Index (J): The mathematical point on the ROC curve maximizing overall classification efficiency: J = Sensitivity + Specificity - 1 Graphically, it represents the point on the ROC curve with the greatest vertical distance from the diagonal line of chance.

Clinical Consequences of Adjusting the Cut-off Threshold

---

Wilson and Jungner Screening Criteria

Screening identifies asymptomatic disease in a population. The UK National Screening Committee (UK NSC) bases national screening policy (e.g., breast cancer mammography, diabetic retinopathy, bowel cancer FIT testing, abdominal aortic aneurysm ultrasound) on the classic Wilson and Jungner (WHO) criteria:

1. The Condition: Must be an important health problem with a recognized latent or early asymptomatic phase, and its natural history must be well understood.
2. The Test: Must be simple, safe, validated, reliable, acceptable to the target population, with an established and agreed cut-off level.
3. The Treatment: There must be an accepted, effective intervention available for patients identified with the disease, and initiating therapy during the latent phase must yield better clinical outcomes than waiting for symptoms to emerge.
4. The Healthcare System: Adequate facilities for definitive diagnosis and treatment must be available; there must be an agreed policy on whom to treat; and the financial cost of case-finding must be economically balanced relative to total medical expenditure.