Evaluate Precision-Recall Tradeoff in Screening

Context

MediScan Health built a binary classification model to flag patients as high risk for diabetic retinopathy so they can be referred for specialist review. The current model is being considered for rollout across primary care clinics, but clinical leaders are concerned about whether it is missing too many true cases or sending too many healthy patients to follow-up.

Current Performance

Metric	Model A (Current)	Model B (Alternative)
Precision	0.91	0.72
Recall	0.58	0.86
F1 Score	0.71	0.78
Accuracy	0.94	0.89
AUC-ROC	0.88	0.90
Positive prediction rate	6.4%	13.9%

Test set size: 20,000 patients. Disease prevalence in the test set is 8.0% (1,600 actual positive cases).

The Problem

The product team wants a clear recommendation on the difference between precision and recall, which metric should matter more in this deployment, and whether Model A or Model B is better aligned with the business and clinical objective.

Requirements

Explain the difference between precision and recall using the numbers above.
Compare Model A and Model B, including the tradeoff each model makes.
Use the prevalence and metrics to estimate the practical impact on patients and clinic operations.
Recommend which model to deploy and justify when you would prioritize precision vs. recall.
Suggest how threshold tuning could change the decision.

Constraints

Specialist review capacity is limited to 2,800 referrals per month.
Missing a true case can delay treatment and increase risk of vision loss.
Unnecessary referrals increase cost and patient anxiety, but are less severe than missed disease.

Problem

Context

Current Performance

Metric	Model A (Current)	Model B (Alternative)
Precision	0.91	0.72
Recall	0.58	0.86
F1 Score	0.71	0.78
Accuracy	0.94	0.89
AUC-ROC	0.88	0.90
Positive prediction rate	6.4%	13.9%

Test set size: 20,000 patients. Disease prevalence in the test set is 8.0% (1,600 actual positive cases).

The Problem

Requirements

Explain the difference between precision and recall using the numbers above.
Compare Model A and Model B, including the tradeoff each model makes.
Use the prevalence and metrics to estimate the practical impact on patients and clinic operations.
Recommend which model to deploy and justify when you would prioritize precision vs. recall.
Suggest how threshold tuning could change the decision.

Constraints

Specialist review capacity is limited to 2,800 referrals per month.
Missing a true case can delay treatment and increase risk of vision loss.
Unnecessary referrals increase cost and patient anxiety, but are less severe than missed disease.

Problem

Context

Current Performance

Metric	Model A (Current)	Model B (Alternative)
Precision	0.91	0.72
Recall	0.58	0.86
F1 Score	0.71	0.78
Accuracy	0.94	0.89
AUC-ROC	0.88	0.90
Positive prediction rate	6.4%	13.9%

Test set size: 20,000 patients. Disease prevalence in the test set is 8.0% (1,600 actual positive cases).

The Problem

Requirements

Explain the difference between precision and recall using the numbers above.
Compare Model A and Model B, including the tradeoff each model makes.
Use the prevalence and metrics to estimate the practical impact on patients and clinic operations.
Recommend which model to deploy and justify when you would prioritize precision vs. recall.
Suggest how threshold tuning could change the decision.

Constraints

Specialist review capacity is limited to 2,800 referrals per month.
Missing a true case can delay treatment and increase risk of vision loss.
Unnecessary referrals increase cost and patient anxiety, but are less severe than missed disease.

Problem

Context

Current Performance

Metric	Model A (Current)	Model B (Alternative)
Precision	0.91	0.72
Recall	0.58	0.86
F1 Score	0.71	0.78
Accuracy	0.94	0.89
AUC-ROC	0.88	0.90
Positive prediction rate	6.4%	13.9%

Test set size: 20,000 patients. Disease prevalence in the test set is 8.0% (1,600 actual positive cases).

The Problem

Requirements

Explain the difference between precision and recall using the numbers above.
Compare Model A and Model B, including the tradeoff each model makes.
Use the prevalence and metrics to estimate the practical impact on patients and clinic operations.
Recommend which model to deploy and justify when you would prioritize precision vs. recall.
Suggest how threshold tuning could change the decision.

Constraints

Specialist review capacity is limited to 2,800 referrals per month.
Missing a true case can delay treatment and increase risk of vision loss.
Unnecessary referrals increase cost and patient anxiety, but are less severe than missed disease.