Scribd
Upload
10 views2 pages

Understanding Bayes Theorem in Biostatistics

The document discusses Bayes' theorem and how it can be used to update probabilities based on new information or test results. It provides examples of applying Bayes' theorem to determine the probability of a restaurant being good based on a review, and the probability a person has COVID given a positive test result. It also discusses computing probabilities more efficiently using Bayes factors and odds ratios.

Uploaded by

kingngaichow
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
10 views2 pages

Understanding Bayes Theorem in Biostatistics

The document discusses Bayes' theorem and how it can be used to update probabilities based on new information or test results. It provides examples of applying Bayes' theorem to determine the probability of a restaurant being good based on a review, and the probability a person has COVID given a positive test result. It also discusses computing probabilities more efficiently using Bayes factors and odds ratios.

Uploaded by

kingngaichow
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

Biostatistics course

- Hypothesis testing
- t-test, ANOVA
- Regression
- Risk of bias?

Bayes theorem
- Prior: P(A)
- i.e. the left rectangle
- Given B
- i.e. bottom left and bottom right rectangle
- Posterior: P(A|B)
- i.e. bottom left / (bottom left + bottom right)
- P(A|B) = P(A)P(B|A) / (P(A)P(B|A) + P(A’)P(B|A’))
- Sometimes expressed as: P(A|B) = P(A)P(B|A) / P(B)
- But often computes P(A)P(B|A) and P(A’)P(B|A’) separately
- e.g. restaurant quality
- Assume 60% chance that a restaurant is good
- P(A) = 0.6, P(A’) = 0.4
- Given a good review in openrice, what is the updated probability?
- Find P(A|B)
- If restaurant is good, 80% chance it will get a good review
- P(B|A) = 0.8
- If restaurant is bad, 20% chance it will get a good review
- P(B|A’) = 0.2
- P(A|B) = 0.6*0.8 / (0.6*0.8 + 0.4*0.2) = 0.857 (higher than prior)
- e.g. Covid test
- Assume 30% people have Covid
- P(A) = 0.3, P(A’) = 0.7
- Given a positive test, what is the updated probability?
- Sensitivity = 80% (real positive detected as positive)
- P(B|A) = 0.8
- Specificity = 90% (real negative detected as negative)
- P(B|A’) = 0.1
- P(A|B) = 0.3*0.8 / (0.3*0.8 + 0.7*0.1) = 0.774 (higher than prior)

Bayes factor
- Test result indicates disease
- Test result tells the probability of disease
- Test result updates the probability of disease
- Computation
- Remember the rectangle
- P(D|+) = P(D)P(+|D) / (P(D)P(+|D) + P(ND)P(+|ND))
- P(D|+) = Prior * sensitivity / (Prior * sensitivity + (1-prior) * false positive rate)
- How to compute faster?
- If Prior << 1
- Approximate 1-prior = 1
- Approximate prior * sensitivity << (1-prior) * false positive rate
- P(D|+) = Prior * sensitivity / false positive rate
- Define Bayes factor = sensitivity / false positive rate
- P(D|+) = Prior * Bayes factor
- But limited to prior << 1
- Can we do even better?
- Remember odds? Odds = P / (1-P)
- O(D|+) = P(D|+) / (1-P(D|+))
- = Prior * sensitivity / (1-prior) * false positive rate
- = O(D) * Bayes factor
- Comment: formula is neat, good for computation
- Very nice thought from 3B1B
- The term ‘false positive rate’ is often misinterpreted
- False positive rate = P(+|ND) = 1-P(-|ND) (i.e. 1-specificity)
- If specificity is 90%, false positive rate is 10%
- In clinical situation, suppose test result is positive
- People misinterpret this as ‘10% chance that the test result is wrong’
- In probability, it is often more intuition to think 50% probability = in 100 people,
50 people will…
- e.g. typhoon: in 100 typhoon with this track forecast, 90 typhoons will
fall in the region
- e.g. CI: in 100 experiments, 95 CI will contain the true mean
- e.g. false positive rate: in 100 people that get a positive test result, 10
people are actually healthy

Probability distribution 1, 2

You might also like