Probabilistic Reasoning, Risk Assessment & Expected Value

Dependent Sampling Without Replacement, Odds & Expected Value

When trials are dependent, the outcome of an initial selection alters the sample space and shifts the probability distribution for subsequent events. In Decision Making, candidates frequently encounter dependent sampling, odds translations, and expected value evaluations.

1. Conditional Probability & Dependent Events

When two events are dependent, the probability that event $B$ occurs depends on whether event $A$ has already transpired.

Formal Definition: The conditional probability of $B$ given $A$ is expressed as:
$$P(B|A) = \frac{P(A \cap B)}{P(A)}, \quad \text{where } P(A) > 0$$
The General Multiplication Rule: For any two events, dependent or independent:
$$P(A \cap B) = P(A) \times P(B|A)$$

2. Sampling Without Replacement (The Decrement Rule)

In finite populations sampled without replacement, each consecutive selection removes an item from the pool.

The Decrement Rule: When calculating the probability of consecutive favorable draws without replacement, decrement both numerator and denominator by 1 on each successive stage:
$$P(A_1 \cap A_2) = \frac{k}{N} \times \frac{k - 1}{N - 1}$$
Contrast this with sampling with replacement, where the sample space resets to $N$ and probabilities remain static across every draw.

Worked Clinical Problem: Pathology Tray Biopsy Vials

Scenario: A histology laboratory tray holds 12 unlabeled biopsy vials collected from a suspicious lymph node resection:
4 vials contain malignant tissue ($M$).
8 vials contain benign tissue ($B$).
A pathologist selects 2 vials consecutively at random without replacement.
Question: What is the probability that both selected vials contain malignant tissue?
Draw 1: The probability that the first vial is malignant:
$$P(M_1) = \frac{4}{12} = \frac{1}{3}$$
Initial favorable selection probability is 4/12.
Draw 2: Because the first vial was malignant and is not replaced, the tray now contains 11 vials total, of which exactly 3 are malignant:
$$P(M_2 | M_1) = \frac{3}{11}$$
Updated conditional probability is 3/11.
Joint Probability: Apply the General Multiplication Rule:
$$P(M_1 \cap M_2) = \frac{4}{12} \times \frac{3}{11} = \frac{12}{132} = \frac{1}{11} \approx 0.0909$$
The probability that both vials are malignant is 1/11 (or 9.09%).
Examiner Trap Contrast: If a candidate forgets to decrement and assumes replacement, they compute $(4/12) \times (4/12) = 16/144 = 1/9 \approx 0.1111$, selecting the 1/9 (11.11%) distractor!

3. Odds vs Probability (Linguistic & Mathematical Conversion)

Decision Making questions often present statistical odds rather than raw probabilities. Candidates must convert between these two distinct representations instantly:

Odds in Favor: Compares the number of favorable outcomes ($a$) directly to the number of unfavorable outcomes ($b$):
Odds in favor = a : b
$$\text{Odds in favor} = a : b$$
Conversion from Odds to Probability: Probability compares favorable outcomes to the total sample space ($a + b$):
P(E) = a / (a + b)
$$P(E) = \frac{a}{a + b}$$
Odds Against: Compares unfavorable outcomes ($b$) to favorable outcomes ($a$):
Odds against = b : a
$$\text{Odds against} = b : a \implies P(E) = \frac{a}{a + b}$$
Conversion from Probability to Odds: If the probability of an outcome is $P$, the odds in favor are:
$$\text{Odds in favor} = \frac{P}{1 - P} = P : (1 - P)$$
Numerical Example: If the odds in favor of a surgical complication are stated as 1 to 4 (or $1:4$), the probability of the complication is:
$$P = \frac{1}{1 + 4} = \frac{1}{5} = 0.20 \quad (20\%)$$
The true complication probability is 0.20 (20%). Candidates who mistakenly calculate $1/4 =$ 0.25 (25%) fall into the classic denominator trap.

4. Expected Value & Long-Run Mathematical Expectation

The Expected Value $E(X)$ represents the probability-weighted average outcome across repeated trials:

$$E(X) = \sum_{i=1}^{k} x_i \times P(X = x_i)$$

Long-Run Interpretation: The expected value is rarely an outcome possible on a single trial (for example, the expected roll of a six-sided die is 3.5). Instead, it defines the long-run average per iteration over hundreds of repetitions.
Expected Frequency: To calculate the expected number of times an event will occur across $n$ independent trials, multiply the trial count by the event probability:
E(count) = n * P(E)
$$E(\text{count}) = n \times P(E)$$

5. Probability Trees (Sequential Branching Rules)

When evaluating multi-stage clinical decisions, probability trees visualize complex sequential dependencies. Candidates must enforce three governing laws:

The Node Conservation Law: The sum of probabilities along all branches radiating from any single decision node must equal exactly 1.0:
$$\sum P_{\text{branches}} = 1.0$$
The Trajectory Product Law: The joint probability of traversing any specific sequence from the root to a terminal branch tip is the continuous product of the probabilities along that trajectory:
$$P(\text{Trajectory}) = P(E_1) \times P(E_2 | E_1) \times P(E_3 | E_1 \cap E_2)$$
The Leaf Summation Law: If an outcome can be reached via multiple distinct mutually exclusive pathways, the total probability is the arithmetic sum of the individual leaf probabilities:
$$P(\text{Outcome}) = \sum P(\text{Leaf}_k)$$

Worked Clinical Problem: Expected Value in Outpatient Triage Pathways

Scenario: An acute care triage physician models patient observation duration under two diagnostic pathways for suspected biliary colic:
Pathway Alpha (Direct Ultrasonography):
70% probability of clear ultrasound: discharged after 2.0 hours.
30% probability of gallstones requiring admission: admitted for 20.0 hours.
Expected Duration: $E(X_\alpha) = (0.70 \times 2.0) + (0.30 \times 20.0) = 1.4 + 6.0 =$ 7.4 hours.
Pathway Beta (Laboratory Panel First, Selective Ultrasonography):
50% probability of normal laboratory values: discharged after 1.0 hour.
50% probability of abnormal laboratory values triggering urgent ultrasound:
Of this cohort, 40% require 20.0 hours admission, and 60% are discharged after 3.0 hours.
Branch product for delayed discharge: $0.50 \times 0.60 = 0.30$ at 3.0 hours.
Branch product for urgent admission: $0.50 \times 0.40 = 0.20$ at 20.0 hours.
Expected Duration: $E(X_\beta) = (0.50 \times 1.0) + (0.30 \times 3.0) + (0.20 \times 20.0) = 0.5 + 0.9 + 4.0 =$ 5.4 hours.
Deductive Conclusion: Pathway Beta reduces expected stay duration by 2.0 hours per patient (7.4h vs 5.4h), demonstrating how tree branch products aggregate into expected value.

Crucial Conceptual Boundary

If the odds against a medication causing side effects are 7 to 3, does the probability of experiencing side effects equal 7/10?

No! Stated odds against are 7 to 3, meaning 7 unfavorable outcomes (no side effects) for every 3 favorable outcomes (side effects). The probability of experiencing side effects is 3 / (7 + 3) = 3/10 = 0.30 (30%).

Clinical Risk Measures & The False Positive Paradox

Medical decision making frequently requires interpreting clinical trial endpoints and evaluating diagnostic screening protocols. Candidates must master 2x2 clinical contingency tables, distinguish absolute from relative risk, and understand why high-accuracy diagnostic tests produce high false-positive rates in low-prevalence screening.

1. 2x2 Clinical Trial Contingency Tables & Epidemiological Event Rates

In clinical trials, patient outcomes are organized into a 2x2 contingency table comparing an experimental intervention to a control or placebo group:

Control Event Rate (CER): The proportion of patients in the control cohort who experience the adverse clinical outcome:
$$\text{CER} = \frac{\text{Adverse Events in Control}}{\text{Total Patients in Control}}$$
Experimental Event Rate (EER): The proportion of patients in the experimental cohort who experience the adverse clinical outcome:
$$\text{EER} = \frac{\text{Adverse Events in Experimental}}{\text{Total Patients in Experimental}}$$

Trial Cohort	Adverse Outcome Occurred	Adverse Outcome Prevented	Total Evaluated	Baseline Event Rate
Control Group	Events ($A$)	Non-Events ($B$)	$A + B = N_C$	$\text{CER} = \frac{A}{A + B}$
Experimental Group	Events ($C$)	Non-Events ($D$)	$C + D = N_E$	$\text{EER} = \frac{C}{C + D}$

2. Clinical Risk Metrics: ARR, RR, RRR & NNT

From CER and EER, four essential clinical risk metrics are derived:

Absolute Risk Reduction (ARR): The arithmetic difference in event rates between control and treated cohorts:
ARR = CER - EER
$$\text{ARR} = \text{CER} - \text{EER}$$
Relative Risk (RR): The ratio of risk in the experimental group compared to the control group:
RR = EER / CER
$$\text{RR} = \frac{\text{EER}}{\text{CER}}$$
Relative Risk Reduction (RRR): The proportional reduction in event rates relative to the baseline control risk:
RRR = ARR / CER = 1 - RR
$$\text{RRR} = \frac{\text{ARR}}{\text{CER}} = \frac{\text{CER} - \text{EER}}{\text{CER}} = 1 - \text{RR}$$
Number Needed to Treat (NNT): The number of patients who must receive the experimental treatment to prevent one additional adverse outcome:
NNT = 1 / ARR
$$\text{NNT} = \frac{1}{\text{ARR}}$$

Worked Clinical Trial Problem: Novel Anticoagulant Study

Scenario: A randomized clinical trial evaluates a novel oral anticoagulant (Drug X) versus standard therapy for deep vein thrombosis (DVT) prevention over 12 months, with 2,000 patients enrolled in each cohort:
Control Cohort ($N_C = 2000$): 160 patients develop DVT.
$$\text{CER} = \frac{160}{2000} = 0.080 \quad (8.0\%)$$
Baseline control risk is 0.080 (8.0%).
Drug X Cohort ($N_E = 2000$): 40 patients develop DVT.
$$\text{EER} = \frac{40}{2000} = 0.020 \quad (2.0\%)$$
Treated experimental risk is 0.020 (2.0%).
Step 1: Calculate Absolute Risk Reduction (ARR):
$$\text{ARR} = \text{CER} - \text{EER} = 0.080 - 0.020 = 0.060 \quad (6.0\%)$$
The absolute risk reduction achieved is 0.060 (6.0%).
Step 2: Calculate Relative Risk (RR):
$$\text{RR} = \frac{\text{EER}}{\text{CER}} = \frac{0.020}{0.080} = 0.25$$
The relative risk ratio is 0.25.
Step 3: Calculate Relative Risk Reduction (RRR):
$$\text{RRR} = \frac{\text{ARR}}{\text{CER}} = \frac{0.060}{0.080} = 0.75 \quad (75\%)$$
The proportional risk reduction relative to baseline is 75%.
Step 4: Calculate Number Needed to Treat (NNT):
$$\text{NNT} = \frac{1}{\text{ARR}} = \frac{1}{0.060} = 16.666\dots \approx 16.67$$
The Clinical Rounding Law: In clinical practice, you cannot treat a fraction of a patient. Treating 16 patients is insufficient to avert one complete DVT event. Therefore, NNT must ALWAYS round UP to the nearest integer:
$$\lceil 16.67 \rceil = 17 \text{ patients}$$
Candidates must treat 17 patients with Drug X to prevent one additional DVT.

3. Diagnostic Screening Parameters & 2x2 Test Cross-Tabulation

Evaluating medical screening tests requires cross-tabulating test outcomes against true disease status:

Sensitivity (True Positive Rate): Probability that the test is positive given the patient has the disease:
$$\text{Sensitivity} = \frac{\text{TP}}{\text{TP} + \text{FN}}$$
Specificity (True Negative Rate): Probability that the test is negative given the patient is healthy:
$$\text{Specificity} = \frac{\text{TN}}{\text{TN} + \text{FP}}$$
False Positive Rate (FPR): Probability that a healthy patient tests positive ($1 - \text{Specificity}$):
$$\text{FPR} = \frac{\text{FP}}{\text{TN} + \text{FP}} = 1 - \text{Specificity}$$
Positive Predictive Value (PPV): Probability that a patient who tests positive genuinely has the disease:
$$\text{PPV} = \frac{\text{TP}}{\text{TP} + \text{FP}}$$
Negative Predictive Value (NPV): Probability that a patient who tests negative is genuinely healthy:
$$\text{NPV} = \frac{\text{TN}}{\text{TN} + \text{FN}}$$

Diagnostic Screening Status	Truly Diseased ($D^+$)	Truly Healthy ($D^-$)	Marginal Row Totals	Predictive Value Metric
Screen Positive ($T^+$)	True Positives (TP)	False Positives (FP)	$\text{TP} + \text{FP}$	$\text{PPV} = \frac{\text{TP}}{\text{TP} + \text{FP}}$
Screen Negative ($T^-$)	False Negatives (FN)	True Negatives (TN)	$\text{FN} + \text{TN}$	$\text{NPV} = \frac{\text{TN}}{\text{TN} + \text{FN}}$
Marginal Column Totals	$\text{TP} + \text{FN}$	$\text{FP} + \text{TN}$	Total Population ($N$)	Screening Baseline
Diagnostic Metric	$\text{Sensitivity} = \frac{\text{TP}}{\text{TP} + \text{FN}}$	$\text{Specificity} = \frac{\text{TN}}{\text{TN} + \text{FP}}$		$\text{FPR} = \frac{\text{FP}}{\text{TN} + \text{FP}}$

4. The False Positive Paradox (Low Baseline Prevalence Screening)

When screening an asymptomatic population for a rare medical condition, low baseline prevalence causes healthy individuals to vastly outnumber diseased individuals. Even an extraordinarily accurate diagnostic test produces more false positives than true positives.

Worked 10,000-Cohort Diagnostic Screening Model

Scenario: A nationwide screening program evaluates 10,000 asymptomatic individuals for a rare endocrine disorder:
Baseline Disease Prevalence: 0.2% ($0.002$).
Screening Test Sensitivity: 95% ($0.95$).
Screening Test Specificity: 90% ($0.90$).
Population Stratification:
1. Truly Diseased Cohort:
$$\text{Diseased} = 10000 \times 0.002 = 20 \text{ individuals}$$
True Positives (TP) $= 20 \times 0.95 =$ 19 individuals.
False Negatives (FN) $= 20 - 19 =$ 1 individual.
2. Truly Healthy Cohort:
$$\text{Healthy} = 10000 - 20 = 9980 \text{ individuals}$$
False Positive Rate $= 1 - 0.90 = 0.10$ ($10\%$).
False Positives (FP) $= 9980 \times 0.10 =$ 998 individuals.
True Negatives (TN) $= 9980 - 998 = 8982 \text{ individuals}$.
Positive Predictive Value (PPV) Calculation:
Total positive test results $= \text{TP} + \text{FP} = 19 + 998 = 1017$.
Positive Predictive Value:
$$\text{PPV} = \frac{\text{TP}}{\text{TP} + \text{FP}} = \frac{19}{1017} \approx 0.01868 \quad (1.87\%)$$
The Counter-Intuitive Reality: Despite 95% sensitivity and 90% specificity, a patient testing positive has only a 1.87% probability of actually having the disease! Over 98% of all positive test results are false positives.

5. Cognitive Biases in Statistical Decision Making

Examiners construct Decision Making items around three universal probabilistic cognitive fallacies:

The Gambler's Fallacy (Belief in the Law of Small Numbers): The irrational belief that independent past outcomes alter the likelihood of future trials. For example, assuming that after 5 consecutive coin tosses land heads, tails is "due". In clinical triage, assuming that because the previous four patients had rare appendicitis, the fifth patient presenting with abdominal pain cannot have appendicitis.
The Base Rate Fallacy: Neglecting the prior background prevalence of a condition when evaluating diagnostic outcomes. Candidates benchmark prematurely on high test sensitivity (e.g. 95%) and assume a positive test means a 95% likelihood of disease, completely ignoring that the condition occurs in only 1 in 500 people.
The Conjunction Fallacy: Believing that a specific compound event ($A \text{ and } B$) is more probable than a single broader event ($A$). Because the joint probability is bounded by $P(A \cap B) \le P(A)$, a compound outcome can never be more likely than either constituent outcome alone.

Crucial Conceptual Boundary

If a pharmaceutical company announces that a new drug reduces stroke risk by 50%, does that mean half of the patients taking the drug are cured?

No. That 50% figure represents Relative Risk Reduction (RRR), not Absolute Risk Reduction (ARR). If the baseline stroke risk was 2% (CER = 0.02) and dropped to 1% (EER = 0.01), the absolute benefit is only 1 percentage point (ARR = 0.01), requiring 100 patients to be treated (NNT = 100) to avert one stroke.