**AI for Medical Treatment Week 1 Quiz Answers**

*Measuring Treatment Effects*

Question 1) Given the following statistical information of patients for a treatment arm and a control group, which one corresponds to a correct setup of a randomized control trial?

Question 2) You are part of a medical team trying to create an alternative treatment for patients with lung cancer. Your group performs several experiments and reports results with the following p-values. Which has the most statistically significant result?

- p-value = 0.5
- p-value = 0.001
- p-value = 0.0003
**p-value=0.0001**

Question 3) Given an average risk reduction (ARR) of 0.2, on average, how many people need to receive the treatment in order to benefit one of them (NNT)?

**5**- 10
- 20
- 0.8

Question 4) You are studying the effect of a new treatment for heart attack, your job consists in looking at outcomes of the effect in patients, fill the unit level treatment effect column using the Neyman-Rubin causal model, and then calculate the average treatment effect.

*Tips:*- The event doesn't occur: 0
- The event occurs: 1
- Unit Level Treatment Effect: -1 represents a benefit, o represents no effect, 1 represents harm.

**0.375**- -0.75
- 0.75
- -0.375

Question 5) Calculate the conditional average treatment effect applying the Two-Tree Learner method, the patient has an Age=61 and BP= 130.

- -0.24
**-0.20**- 0.24
- 0.43

Question 6) Using the S-Learner, or Single Tree, method, what is the conditional average treatment effect for a 61 year-old patient with a blood pressure (BP) of 140?

- 0.02
- 0.22
- 0.10
**We can't estimate the conditional ATE using this S-Learner.**

Question 7) Which considerations are relevant to the S-Learner Method? Choose all that are correct.

- Your model is more likely to overfit your data.
- The Decision Tree might decide not to use the treatment feature.
**This model might produce a treatment effect estimate of o for everyone.**- Since the two models are using each half of the data, there are fewer samples available to learn the relationships between the features.

Question 8) Which considerations are relevant to the T-Learner Method? Choose all that are correct.

- Your model is more likely to overfit your data.
- The Decision Tree might decide not to use the treatment feature.
- This model might produce a treatment effect estimate of o for everyone.
**Since the two models are using each half of the data, there are fewer samples available to learn the relationships between the features.**

**AI for Medical Treatment Week 2 Quiz Answers**

*Information Extraction with NLP*

Question 1) Which of the following is not true about BERT's inner word representations?

**Each unique word can have exactly one vector representation**- The representation of a word depends on the words around it
- Words which are similar in meaning are typically close as vector
- None of the above

Question 2) True or False: the start and end vectors are fixed throughout training

- True
**False**

Question 3) Which of the following is a difference between BERT and LSTM models?

- BERT is trained using backpropagation while LSTMs are not
- BERT can be trained on multiple languages, while LSTMs cannot
**BERT takes entire sequences as input, while LSTM models process words one by one**- BERT uses regular word vectors, while LSTMs use contextualized word vectors

Question 4) Given the following word vectors and start and end vectors, determine the start and end of the sequence of interest.

- start: The, end: cancer
- start with, end: gene
- start: gene, end: associated
**start: breast, end. Cancer**

Question 5) You find that a radiology report mentions "edema". Which of the following can you immediately conclude?

- The x-ray contains edema
- The x-ray contains pneumonia
- The x-ray does not contain edema
**None of the above**

Question 6) Use the following entry in SNOMED CT to help determine the positive labels for this x-ray report.

- common cold: 0, lesion: 0
- common cold: 0, lesion: 1
- common cold: 1, lesion: 1
**common cold: 1, lesion: 0**

Question 7) Let's see why F1 is used instead of the regular mean of precision and recall. Let's say the mean of precision and recall is at least 0.75. Which of the following could be the true value of the precision?

- 0.75
- 0.5
**Both**- Neither

Question 8) Now let's say F1 score is at least 0.75. Now which of the following values of precision are possible?

**0.75**- 0.5
- Both
- Neither

Question 9) Compute the F1 score for pneumonia and mass separately based on the following retrieved labels and ground truth:

- (0.5, 0.83)
**(0.5, 0.8)**- (0.75, 0.8)
- None of the above

Question 10) Now compute the F1 score for all labels jointly:

- 1.35
- 0.61
**0.66**- None of the above

**AI for Medical Treatment Week 3 Quiz Answers**

*ML Interpretation*

Question 1) You train the random forest pictured below and it gets a c-index of 0.90. After shuffling the values for x, your dataset is the following. What is the variable importance for x?

- 0.05
- 0.1
- 0.5
**0.65**

Question 2) Say you have trained a decision tree which never splits on a variable X. What will be the variable importance for X using the permutation method?

- 0.5
- A random number between 0 and 1
**0**- There is too little information to say

Question 3) We have the following table the output of a model fon an example using subsets of the variable. What is the Shapley value for s_BP?

- 0.0
- 0.02
- 0.05
**0.125**

Question 4) We have the following table the output of a model fon an example using subsets of the variable. What is the sum of the Shapley value for S_BP and d_BP?

- 0.0
- 0.2
- 0.05
**0.15**

Question 5) Could the following table of outputs be given by a linear model with no interactions (assume not including a feature means setting it to 0)?

- Yes
**No**

Question 6) Now assume we add Age as a variable. What is the new Shapley value for s_BP?

- 0.0
**0.09**- 0.125
- 0.20

## Post a Comment