What is Sensitivity?
Sensitivity (Sn):
- correctly identify as diseased those actually with disease.
- Positive in Disease
- Is 1 – false neg rate.
- Denominator = condition present
Use a high Sensitivity to rule out
- Snout: If 100% Sensitive, all (-) are True Negatives
What is Specificity?
Specificity (Sp):
- correctly identify as non-diseased those actually without the disease
- Negative in Health
- 1 – false pos rate
- Denominator = condition absent
Use a high Specificity to Rule in
- Spin: If 100% Specific, all (+) are True Positives
What is Predictive Value (+)?
Predictive Value (+):
- proportion true + of those screening as +
- Denominator = test positive
Probability that person actually has disease, given a pos test result
What is Predictive Value (-)?
Predictive Value (-):
- proportion true - of those screening as –
- Denominator = test negative
Probability that person actually not have disease, given neg test result
What are the 3 ways to measure Risk?
- Absolute Risk
- Incidence of disease in group initially free of condition
- new cases ratio
- new cases/group
- Attributable Risk
- Incidence of disease attributable to exposure
- subration
- exp - non-exp
- Relative Risk
- How many times more likely exposed develop disease relative to non-exposure
- division
- exposed/non-exposed
Number Needed to Treat (NNT) vs. Number Needed to Harm (NNH)
Number Needed to Treat (NNT)
- NNT = 1 / Absolute Risk reduction
- # needed to tx to prevent 1 case
Number Needed to Harm (NNH)
- 1 / Absolute Risk increase
- # needed to tx to cause 1 extra case
When should we use Relative Risk (RR)?
Cohort studies
- 2 groups to start:
- exposed & not exposed
- cases/conditions develop or appear over time
RR determines how much more likely exposed develop disease compared to non-exposed
————————
How to calculate:
RR = (a/a+b)/(c/c+d)
When should we use an Odds Ratio (OR)?
Case Control studies
- 2 groups to start:
- cases & not cases (controls)
OR determines relative freq of exposure among cases and non-cases
- Odds that a case is exposed divided by odds that a control is exposed
————————
How to calculate:
OR = ad/bc
- If OR >1 = freq of exp higher among cases thus increased risk
- If OR <1 = freq of exp lower among cases, thus decreased risk or ‘protection’
The stronger the association, the higher the OR
Continuous vs. Categorical Variables
Continuous = T-test (group 1 mean vs. group 2)
- Appropriate for quantification
- Infinite scale
- body weight
- height
- BP in mm Hg
- If limited to integers —> Discrete
- # cigs/day
- # children
- visits
- days
Categorical = Chi Squared (group A vs group B)
- Not appropriate for quantification
- Classified, categories, bins
- BMI as ↑ or ↓
- Height as tall or short
- BP as hypo or hypertensive
- Race, Gender, Eye Color
- If 2 possible values → Dichotomous
- dead or alive
- If categories not ordered → Nominal
- blood type
- If categories ordered → Ordinal
- mild, moderate, or severe
