4.3 Statistical Analysis & Risk Quantification

[!NOTE] Curriculum Focus: Risk calculations and statistical test selection are among the highest-yield topics in the MRCP(UK) Part 1 examination. Candidates must be fluent in calculating Absolute Risk Reduction (ARR), Number Needed to Treat (NNT), Number Needed to Harm (NNH), and Odds Ratios from 2x2 data tables, as well as interpreting 95% confidence intervals and selecting correct parametric versus non-parametric statistical tests.

Evidence-based clinical decision-making requires accurate quantification of therapeutic benefit and harm. A physician must be able to decipher whether a reported drug benefit reflects a transformative absolute risk reduction or merely a modest relative improvement inflated by clinical reporting.

---

Measures of Disease Frequency

Epidemiology measures how frequently disease occurs in populations using three primary metrics:

1. Prevalence

• Definition: The proportion of individuals in a specified population who have the disease at a defined point in time: Prevalence = Existing Cases (Old + New) / Total Population at Risk
• Characteristics: A dimensionless proportion (ranging from 0 to 1, or 0% to 100%). It reflects a single cross-sectional snapshot.
• Prevalence Relationship: Prevalence is a function of both incidence and disease duration: Prevalence ≈ Incidence x Average Disease Duration A drug that improves survival in a chronic disease without curing it (e.g., insulin in type 1 diabetes) increases disease prevalence by prolonging duration.

2. Cumulative Incidence (Incidence Proportion / Risk)

• Definition: The proportion of initially disease-free individuals who develop the disease over a specified time horizon: Cumulative Incidence = New Cases Developing During Follow-up / Population at Risk at Baseline
• Limitation: Assumes all participants are observed for the entire study interval; cannot account for individuals who enter late or drop out.

3. Incidence Rate (Incidence Density)

• Definition: The rate at which new events arise per unit of person-time at risk: Incidence Rate = New Cases Developing During Follow-up / Total Person-Time at Risk (e.g., person-years)
• Advantage: Accommodates varying lengths of patient follow-up, staggered trial enrollment, and censoring (death, loss to follow-up).

---

Risk Quantification Formulas

In a clinical trial or prospective cohort study, subjects are classified into experimental and control arms, yielding a standard 2x2 contingency table:

Core Mathematical Formulations

1. Experimental Event Rate (EER) and Control Event Rate (CER)

• Control Event Rate: CER = c / (c + d)
• Experimental Event Rate: EER = a / (a + b)

2. Relative Risk (Risk Ratio, RR)

• Formula: Relative Risk = EER / CER = [a / (a + b)] / [c / (c + d)]
• Application: Cohort studies and randomized controlled trials (where true incidence can be measured directly).
• Interpretation:
  • RR = 1.0: No difference in risk between groups.
  • RR < 1.0: Intervention reduces risk (protective effect).
  • RR > 1.0: Intervention or exposure increases risk.

3. Odds Ratio (OR)

• Formula: Odds Ratio = (Odds of Exposure in Cases) / (Odds of Exposure in Controls) = (a / c) / (b / d) = (a x d) / (b x c)
• Application: Case-control studies (where background population denominators are unknown).
• The Rare Disease Assumption: When the baseline disease prevalence is low (< 5%–10%), a is negligible relative to b, and c is negligible relative to d. Under these conditions, the Odds Ratio closely approximates the Relative Risk (OR ≈ RR). If the condition is common (> 10%), the OR significantly exaggerates the relative risk.

4. Absolute Risk Reduction (ARR) and Relative Risk Reduction (RRR)

• Absolute Risk Reduction (Risk Difference): ARR = CER - EER
• Relative Risk Reduction: RRR = ARR / CER = (CER - EER) / CER = 1 - RR
• Absolute Risk Increase (ARI): When the experimental arm produces higher adverse events: ARI = EER - CER

[!IMPORTANT] The RRR Deception: Pharmaceutical literature frequently highlights RRR because it yields impressive numbers. If a drug reduces event risk from 2% to 1%, the RRR is 50%, but the ARR is only 1%. Clinical impact is determined by ARR, not RRR.

---

Number Needed to Treat (NNT) and Harm (NNH)

The Number Needed to Treat (NNT) represents the average number of patients who must receive the specific intervention over a defined duration to prevent one additional adverse clinical event:

NNT = 1 / Absolute Risk Reduction (ARR) = 1 / (CER - EER)

NNH = 1 / Absolute Risk Increase (ARI) = 1 / (EER - CER)

The Three Golden Rules of NNT for MRCP(UK)

1. Always express ARR as a decimal before inverting:
   • If ARR = 4%, write as 0.04. NNT = 1 / 0.04 = 25.
   • Never divide 1 by 4 (which yields 0.25)!
2. Always round UP to the next whole integer:
   • If a calculation yields an NNT of 14.2, the answer is 15, not 14. Treating 14 patients leaves the 15th partially unprevented; to ensure one full event is averted, 15 patients must be treated.
   • Conversely, for NNH, round DOWN (e.g., 18.8 -> 18) to provide a conservative, prudent estimate of harm.
3. Always link NNT to a time frame:
   • An NNT of 25 over 6 months represents a vastly different clinical commitment than an NNT of 25 over 10 years.

Worked Example: Statin Primary Prevention

In a 5-year randomized trial of hypercholesterolaemic patients:

• Control Group 5-year MI incidence: 8.0% (CER = 0.08)
• Statin Group 5-year MI incidence: 5.2% (EER = 0.052)
• Absolute Risk Reduction: ARR = 0.08 - 0.052 = 0.028 (2.8%)
• Relative Risk Reduction: RRR = 0.028 / 0.08 = 35%
• Number Needed to Treat: NNT = 1 / 0.028 = 35.71 -> 36 patients treated for 5 years

---

Hypothesis Testing, Inferential Errors & Power

Statistical hypothesis testing evaluates whether observed clinical differences reflect genuine biological effects or random sampling variation.

The Hypotheses

• Null Hypothesis (H0): There is no genuine difference between study groups; any observed disparity is due to chance alone.
• Alternative Hypothesis (H1): There is a true difference between study groups.

Type I and Type II Errors

• Type I Error (alpha): Rejecting the null hypothesis when it is actually true. The threshold is conventionally set at alpha = 0.05 (5%). A p-value < 0.05 indicates that if the null hypothesis were true, there is less than a 5% probability of observing an effect this large purely by chance.
• Type II Error (beta): Failing to reject the null hypothesis when a true difference actually exists. Conventionally set at beta = 0.10 to 0.20 (10%–20%).
• Statistical Power (1 - beta): The probability of correctly detecting a true difference if one exists. High-quality trials target a power of 80% or 90% (0.80 or 0.90).

Determinants of Statistical Power

Power increases when:

1. Sample size increases (n increases).
2. True effect size increases (larger ARR or difference between means).
3. Measurement variability / standard deviation decreases (more homogeneous population).
4. Significance threshold is relaxed (increasing alpha from 0.01 to 0.05).

---

P-Values versus 95% Confidence Intervals

P-Values

A p-value quantifies the probability of obtaining data at least as extreme as observed, assuming the null hypothesis is correct. It provides a continuous measure of evidence against H0, but does not indicate effect magnitude, precision, or clinical relevance.

95% Confidence Intervals (CI)

A 95% confidence interval provides a range within which the true population parameter lies with 95% certainty across repeated sampling experiments.

The Critical Rules of Confidence Interval Interpretation

+------------------------------------------------------------------------------------------+
|                       Confidence Interval Significance Rules                             |
+------------------------------------------------------------------------------------------+
| 1. For DIFFERENCES (Mean Difference, Risk Difference, ARR):                              |
|    • Null value = 0 (Zero)                                                               |
|    • If 95% CI DOES NOT span 0 --> Statistically Significant (p < 0.05)                  |
|    • If 95% CI SPANS 0 (e.g., -1.2 to +4.5) --> NOT Statistically Significant (p >= 0.05)|
|                                                                                          |
| 2. For RATIOS (Relative Risk, Odds Ratio, Hazard Ratio):                                 |
|    • Null value = 1.0 (One)                                                              |
|    • If 95% CI DOES NOT span 1.0 --> Statistically Significant (p < 0.05)                |
|    • If 95% CI SPANS 1.0 (e.g., 0.82 to 1.45) --> NOT Statistically Significant (p >= 0.05|
+------------------------------------------------------------------------------------------+

Statistical vs Clinical Significance

• Large Sample Size Trap: An overpowered trial (n = 40,000) may demonstrate a statistically significant reduction in systolic blood pressure of 0.4 mmHg (95% CI 0.1 to 0.7 mmHg; p = 0.008). While statistically significant, it is clinically irrelevant.
• Small Sample Size Trap (Underpowered): A small trial (n = 30) evaluating mortality reduction may report an ARR of 15% (95% CI -3% to 33%; p = 0.08). Although statistically non-significant, it cannot exclude a transformative clinical benefit.

---

Parametric versus Non-Parametric Statistical Tests

Selecting the correct statistical test depends on:

1. Data Type: Continuous (interval/ratio), Ordinal (ranked/Likert scale), or Categorical (nominal).
2. Distribution: Normal (Gaussian bell curve) versus Non-Normal (skewed).
3. Number of Groups: Two groups versus three or more groups.
4. Sample Pairing: Independent (unpaired) versus Matched/Repeated measures (paired).

The Master Statistical Selection Matrix