Study Notes: Chapter 2–5 on Risk Measures

Chapter 2: Risk Of Lung

  • Context from transcript: first compute risks for two groups and compare them to assess how smoking relates to lung cancer.
  • Key data (as stated):
    • Risk of lung cancer among smokers: 0.028(=2.8%)0.028\quad(=2.8\%)
    • Risk of lung cancer among nonsmokers: 0.002(=0.2%)0.002\quad(=0.2\%)
  • Major concepts introduced:
    • Risk (incidence) in each group: the probability (proportion) of developing lung cancer within the group.
    • Risk Ratio (relative risk, RR): compares risk in two groups.
    • Odds and Odds Ratio (OR): compares odds of exposure between cases and non-cases (or exposure groups in a 2x2 table).
    • Risk Difference (absolute difference in risk between groups).
  • Calculations and formulas:
    • Risk Ratio (RR) between smokers (group 1) and nonsmokers (group 0):
    • RR=p<em>1p</em>0=0.0280.002=14.RR = \frac{p<em>1}{p</em>0} = \frac{0.028}{0.002} = 14.
    • Interpretation: Smokers have 14 times the risk of developing lung cancer compared with nonsmokers.
    • Notes on interpretation:
    • RR > 1 indicates higher risk in the exposed group (smokers).
    • RR is unitless and independent of absolute risk magnitude.
    • 2x2 table setup (notation used in the transcript):
    • Let a = number of smokers with lung cancer.
    • Let b = number of smokers without lung cancer.
    • Let c = number of nonsmokers with lung cancer.
    • Let d = number of nonsmokers without lung cancer.
    • Then the table looks like:
      • Smokers with cancer: a
      • Smokers without cancer: b
      • Nonsmokers with cancer: c
      • Nonsmokers without cancer: d
    • Odds Ratio (OR) formula and construction:
    • OR for exposure in cases vs non-cases (using the 2x2 table):
      • OR=a⋅db⋅c.OR = \frac{a \cdot d}{b \cdot c}.
    • Transcript-specific (numbers attempted): a = 84, b = 11{,}037, c = 104, d = 1{,}996 (based on the spoken values).
      • Then the computed value would be:
      • OR=84⋅199611037⋅104≈0.146.OR = \frac{84 \cdot 1996}{11037 \cdot 104} \approx 0.146.
    • Important caveat: The transcript’s specific a/b/c/d pairing appears inconsistent with the direction you’d expect if smoking increases cancer risk. In standard 2x2 wiring, a/b/c/d should be ordered so that OR > 1 when exposure increases disease risk. This example shows a common pitfall: mislabeling cells or reversing exposure/outcome can yield an OR < 1 even when the exposure is a risk factor. Always verify which cells are “exposed with disease” (a), “exposed without disease” (b), “unexposed with disease” (c), and “unexposed without disease” (d).
    • Risk Difference (RD):
    • RD=p<em>1−p</em>0=0.028−0.002=0.026.RD = p<em>1 - p</em>0 = 0.028 - 0.002 = 0.026.
    • In percentage points: RD=2.6%.RD = 2.6\%.
    • Interpretation: The absolute excess risk attributable to smoking is 2.6 percentage points in this example.
  • Summary of Chapter 2 takeaways:
    • Smoking is associated with a higher risk of lung cancer in this data (RR = 14 in the transcript’s numbers).
    • OR can be computed from the 2x2 counts via OR=adbcOR = \frac{a d}{b c}, but ensure the 2x2 labeling is correct to avoid misinterpretation.
    • RD provides the absolute difference in risk between smokers and nonsmokers: RD=0.028−0.002=0.026.RD = 0.028 - 0.002 = 0.026.
  • Key formulas to remember:
    • RR=p<em>1p</em>0RR = \frac{p<em>1}{p</em>0}
    • OR=adbcOR = \frac{a d}{b c}
    • RD=p<em>1−p</em>0RD = p<em>1 - p</em>0

Chapter 3: Risk Difference

  • Central idea: how to interpret the difference in risk between two groups when the direction and labeling of groups can affect the sign of the RD.
  • Transcript context: discussion around risk difference in a heart attack/aspirin scenario and the orientation of groups (placebo vs treatment).
  • Important conceptual points:
    • RD depends on which group is named as the numerator (treatment) and which is the denominator (control) in the difference.
    • If you compute RD as ptreatment − pcontrol, a beneficial treatment yields a negative RD (since ptreatment < pcontrol).
    • Alternatively, some texts report RD as control − treatment to yield a positive number for beneficial treatments. Always state the convention you are using.
  • Example considerations from transcript:
    • They discuss A and B in a setup involving heart attack, aspirin, and placebo, and the sign of the RD.
    • They considered the question: should RD be computed as (risk with placebo) − (risk with aspirin) or the reverse? The transcript notes suggest the sign can flip based on convention, and they acknowledged the resulting RD could be negative (e.g., −0.77) in that framing.
  • General formulas (clear convention):
    • If you define $p{ ext{treat}}$ as the risk in the treatment group and $p{ ext{control}}$ as the risk in the control group, then:
    • RD=p<em>exttreat−p</em>extcontrol.RD = p<em>{ ext{treat}} - p</em>{ ext{control}}.
    • If the treatment lowers risk, RD < 0.
    • The Number Needed to Treat (NNT) is defined as NNT=1∣RD∣NNT = \frac{1}{|RD|} when RD is negative (use the absolute value for the magnitude).
  • Conceptual reminder:
    • The sign and magnitude of RD are essential for interpreting whether a treatment is beneficial and for calculating NNT/NNH.
  • Connection to Chapter 2 concepts:
    • RD complements RR and OR by giving an absolute, intuitive measure of risk change that is anchored to the baseline risk.

Chapter 4: Absolute Risk Reduction

  • Core distinction: ARR, RR, and OR are related but convey different information.
  • Transcript cues and standard definitions:
    • Relative Risk (RR) and Relative Risk Reduction (RRR):
    • If the relative risk is reduced by half, RR = 0.5, and the Relative Risk Reduction is RRR=1−RR=1−0.5=0.5RRR = 1 - RR = 1 - 0.5 = 0.5, i.e., a 50% reduction relative to the baseline risk.
    • Absolute Risk Reduction (ARR):
    • ARR is the absolute difference in risk between control and treatment: ARR=p<em>control−p</em>treatmentARR = p<em>{control} - p</em>{treatment} (positive if treatment lowers risk).
    • Number Needed to Treat (NNT):
    • NNT=1ARRNNT = \frac{1}{ARR} (for ARR expressed as a proportion, i.e., not percentage points).
  • Worked conceptual example (based on the transcript’s discussion):
    • If placebo risk of a heart attack is R0 and aspirin treatment lowers risk to R1, then:
    • RR = R1R0\frac{R1}{R0}
    • RRR = 1−RR=1−R1R01 - RR = 1 - \frac{R1}{R0}
    • ARR = R0−R1R0 - R1
    • NNT = 1ARR\frac{1}{ARR}
    • A qualitative takeaway from the transcript: the statement that aspirin makes you “half as likely” to have a heart attack aligns with an RR of about 0.5 in some contexts, which corresponds to an RRR of about 0.5 (50%).
  • Worked numerical illustration (illustrative, not pulled from transcript):
    • Suppose baseline risk (control) p<em>0=0.02p<em>0 = 0.02 and treatment risk p</em>1=0.01p</em>1 = 0.01.
    • Then:
    • RR=p<em>1p</em>0=0.010.02=0.5RR = \frac{p<em>1}{p</em>0} = \frac{0.01}{0.02} = 0.5
    • RRR=1−RR=0.5RRR = 1 - RR = 0.5
    • ARR=p<em>0−p</em>1=0.02−0.01=0.01ARR = p<em>0 - p</em>1 = 0.02 - 0.01 = 0.01
    • NNT=1ARR=10.01=100NNT = \frac{1}{ARR} = \frac{1}{0.01} = 100
  • Practical implications:
    • ARR depends on the baseline risk; RR can look impressive even when ARR is small if baseline risk is low.
    • OR (odds ratio) can diverge from RR as events become more common, so choose the effect measure that best communicates risk in a given context.
  • Summary of key relationships:
    • RR=p<em>1p</em>0RR = \frac{p<em>1}{p</em>0}
    • RRR=1−RR=1−p<em>1p</em>0RRR = 1 - RR = 1 - \frac{p<em>1}{p</em>0}
    • ARR=p<em>0−p</em>1ARR = p<em>0 - p</em>1
    • NNT=1ARRNNT = \frac{1}{ARR} (for ARR in proportion)
  • Contextual note on the transcript:
    • The speaker attempted to relate relative and absolute risk reductions but had some confusion about sign and interpretation. The clarified framework above helps resolve orientation issues and provides a consistent way to compute NNT from ARR.

Chapter 5: Conclusion

  • Consolidated ideas from prior chapters:
    • Different measures answer different questions:
    • RR/OR tell how many times more/less likely an outcome is in one group vs another.
    • RD/RR/ARR provide absolute differences or proportional reductions, which matter for practical decision-making and patient counseling.
    • The direction and orientation of groups matter for interpreting RD and NNT. Always state which group is treated as the numerator and which is the denominator.
    • OR can approximate RR when outcomes are rare, but with more common outcomes it can diverge and be harder to interpret.
  • Practical takeaways for exam prep:
    • Be able to compute and interpret RR, OR, RD, ARR, and NNT from a 2x2 table.
    • Be explicit about which group is p0 (control) and which is p1 (treatment).
    • Remember the formulas and the relationships among the measures:
    • RR=p<em>1p</em>0RR = \frac{p<em>1}{p</em>0}
    • OR=adbcOR = \frac{a d}{b c} (with a, b, c, d defined in the 2x2 table)
    • RD=p<em>1−p</em>0RD = p<em>1 - p</em>0
    • ARR=p<em>0−p</em>1ARR = p<em>0 - p</em>1 (commonly used to express treatment benefit as a positive value)
    • RRR=1−RRRRR = 1 - RR
    • NNT=1ARRNNT = \frac{1}{ARR} (for ARR as a proportion; use absolute value for RD-based NNT when RD is negative)
  • Final reflection on the transcript's content:
    • Some numerical details in the transcript were inconsistent or ambiguously labeled (e.g., the exact a/b/c/d counts for the OR calculation). In your study notes, prioritize the standard definitions and consistent table labeling, and verify numbers against your lecture slides or problem sets.
    • The overarching theme is to understand how these measures differ, how to interpret them, and how to communicate risk clearly in real-world contexts (clinical decisions, patient counseling, public health messaging).