Healthcare Data Science Actual Questions and Correct
Answers
Q1
What are the 6 steps in the Predictive Modeling Pipeline?
Answer: Prediction Target Cohort Construction Feature Construction Feature Selection
Predictive Model Performance Evaluation
Q2
What are the two criteria for determining if a Prediction Target is viable?
Answer: Interesting - Talk to domain experts, read publications, check for high
cost/long time/bad quality metrics Possible - Consider human performance, past
experience from similar projects, and results from prior publications
Q3
Difference between Prospective and Retrospective studies
Answer: Prospective: Identify cohort first, then start collecting data going forward in
time Retrospective: Identify study population, then trace back to retrieve historical
data Retrospective is typically less expensive and faster but may have more noise in
the data
Q4
What is a Cohort Study?
Answer: Select a group of patients who are exposed to the risk. Define the right
inclusion/exclusion criteria for all relevant patients. Example: All HF patients
discharged from hospital for studying readmission
Q5
What is a Case-Control Study?
Answer: Study that compares cases (patients with positive outcome/have the disease)
against controls (patients with negative outcome/healthy but otherwise similar). Key
is matching criteria between cases and controls (e.g., age, gender, clinic)
, Q6
What are Observation Window and Prediction Window in Feature Construction?
Answer: Observation Window: The period of time before the index date used to collect
patient data and construct features Prediction Window: The time period after the
index date during which we predict the outcome Index Date: Reference point
separating observation from prediction
Q7
What are common Feature Types in Feature Selection?
Answer: Demographics (age, sex, race) Diagnosis Lab results Symptoms Medications
Vitals (blood pressure, etc.)
Q8
What are the two main categories of Predictive Models?
Answer: Regression - Target y is continuous (metrics: MAE, MSE, R²) Classification -
Target y is categorical/binary (metrics: True/False positive rate, PPV, F1, AUC-ROC)
Popular methods include Linear Regression, Logistic Regression, SVM, Decision Trees,
Random Forest
Q9
Define the four elements of a Confusion Matrix
Answer: True Positive (TP): Correctly predicted positive False Positive (FP): Incorrectly
predicted positive (Type I error) True Negative (TN): Correctly predicted negative
False Negative (FN): Incorrectly predicted negative (Type II error)
Q10
Define Sensitivity (True Positive Rate) and Specificity (True Negative Rate)
Answer: Sensitivity (Recall/TPR) = TP / (TP + FN) Measures the proportion of actual
positives correctly identified Specificity (TNR) = TN / (TN + FP) Measures the
proportion of actual negatives correctly identified
Q11
Define Positive Predictive Value (Precision) and F1 Score
Answer: Positive Predictive Value (Precision) = TP / (TP + FP) Proportion of positive
predictions that are actually correct F1 Score = 2 (Precision Recall) / (Precision +
Recall) Harmonic mean of precision and recall, balances both metrics
Answers
Q1
What are the 6 steps in the Predictive Modeling Pipeline?
Answer: Prediction Target Cohort Construction Feature Construction Feature Selection
Predictive Model Performance Evaluation
Q2
What are the two criteria for determining if a Prediction Target is viable?
Answer: Interesting - Talk to domain experts, read publications, check for high
cost/long time/bad quality metrics Possible - Consider human performance, past
experience from similar projects, and results from prior publications
Q3
Difference between Prospective and Retrospective studies
Answer: Prospective: Identify cohort first, then start collecting data going forward in
time Retrospective: Identify study population, then trace back to retrieve historical
data Retrospective is typically less expensive and faster but may have more noise in
the data
Q4
What is a Cohort Study?
Answer: Select a group of patients who are exposed to the risk. Define the right
inclusion/exclusion criteria for all relevant patients. Example: All HF patients
discharged from hospital for studying readmission
Q5
What is a Case-Control Study?
Answer: Study that compares cases (patients with positive outcome/have the disease)
against controls (patients with negative outcome/healthy but otherwise similar). Key
is matching criteria between cases and controls (e.g., age, gender, clinic)
, Q6
What are Observation Window and Prediction Window in Feature Construction?
Answer: Observation Window: The period of time before the index date used to collect
patient data and construct features Prediction Window: The time period after the
index date during which we predict the outcome Index Date: Reference point
separating observation from prediction
Q7
What are common Feature Types in Feature Selection?
Answer: Demographics (age, sex, race) Diagnosis Lab results Symptoms Medications
Vitals (blood pressure, etc.)
Q8
What are the two main categories of Predictive Models?
Answer: Regression - Target y is continuous (metrics: MAE, MSE, R²) Classification -
Target y is categorical/binary (metrics: True/False positive rate, PPV, F1, AUC-ROC)
Popular methods include Linear Regression, Logistic Regression, SVM, Decision Trees,
Random Forest
Q9
Define the four elements of a Confusion Matrix
Answer: True Positive (TP): Correctly predicted positive False Positive (FP): Incorrectly
predicted positive (Type I error) True Negative (TN): Correctly predicted negative
False Negative (FN): Incorrectly predicted negative (Type II error)
Q10
Define Sensitivity (True Positive Rate) and Specificity (True Negative Rate)
Answer: Sensitivity (Recall/TPR) = TP / (TP + FN) Measures the proportion of actual
positives correctly identified Specificity (TNR) = TN / (TN + FP) Measures the
proportion of actual negatives correctly identified
Q11
Define Positive Predictive Value (Precision) and F1 Score
Answer: Positive Predictive Value (Precision) = TP / (TP + FP) Proportion of positive
predictions that are actually correct F1 Score = 2 (Precision Recall) / (Precision +
Recall) Harmonic mean of precision and recall, balances both metrics
