Title: Using Biostatistics to Evaluate Vaccines and Medical Tests
1Using Biostatistics to Evaluate Vaccines and
Medical Tests
- Holly Janes
- Fred Hutchinson Cancer Research Center
2Two projects
- Evaluating a candidate HIV vaccine The Step
Study - Statistical methods for evaluating medical tests
PSA screening test for prostate cancer
3The Step Study
- To evaluate a candidate HIV vaccine aimed at
- Preventing HIV infection
- Delaying disease progression in those who become
HIV infected - 2004 to 2007
- North America, South America, Caribbean,
Australia - 3000 HIV negative participants randomized to
vaccine or placebo - Tested approximately every 6 months for HIV
infection
4Vaccine was ineffective at preventing infection
- Estimated annual rate of HIV acquisition
- 3.1 (2.1 to 4.3) for placebos
- 4.6 (3.4 to 6.1) for vaccinees
5Evaluating vaccine effects on disease progression
- In the subset of participants who became HIV
infected - As of October, 2007 81 male infections
- Not enough female infections to study
- Did the vaccine recipients who became infected
have slower disease progression than the placebos
who became infected?
6Measures of HIV disease progression
- Time to initiation of antiretroviral therapy
(ART) - HIV viral load repeated measures over time
- CD4 cell count repeated measures over time
7Demographic Characteristics of HIV Infected
Participants
8No Vaccine Effect on Time to ART Initiation
9Vaccine effects on viral load and CD4 cell count
- Repeated measures over time on each subject
- Set values to missing after ART initiation
- Lots of missing data, due to
- ART initiation
- Patient dropout
- Missed visits
- Missing values are informative!!
10Sample Individual Viral Load Trajectories
11Population Trends in Viral Load
12Analysis of Viral Load and CD4 Cell Count
- Statistical methods
- Longitudinal data methods allow for repeated
measures over time on the same subjects - Missing data methods incorporate information
about missing data - Imputation
- Inverse probability weighting
- Findings
- No evidence that vaccine and placebo groups have
different levels or trends in viral load or CD4
cell count
13Evaluating Medical Tests
14Cancer Screening Tests
- Aimed at finding disease before it causes
symptoms - Early-stage disease usually easier to treat
- Commonly used screening tests
- Mammography, for breast cancer
- Pap test, for cervical cancer
- PSA test, for prostate cancer
15Evaluating cancer screening tests
- How accurate is the test?
- How often is cancer found? (true positive rate)
- How often are healthy individuals told they have
cancer? (false-positive rate) - Screening tests must have very low false positive
rates - The test is applied in the general population
- The vast majority of subjects do not have cancer
- A positive test result leads to invasive
follow-up procedures (eg biopsy), unnecessary
cost and stress - If false positive rate is 5, 5,000 unnecessary
biopsies for every 100,000 people screened
16PSA test for prostate cancer
- Commonly used screening test for prostate cancer
in men over 50 - Utility is hotly debated
- Test measures amount of prostate-specific antigen
(PSA) in the blood - High value suggests cancer
- What is high?
- Positive test result prompts biopsy
17Quantifying test accuracy
- The true positive rate (TPR)
- Proportion of subjects with cancer who test
positive - The false positive rate (FPR)
- Proportion of healthy subjects who test positive
- How to define test positive for a quantitative
test?
18How to define test positive?
19TPR 0.98 FPR 0.75
20TPR 0.75 FPR 0.25
21TPR 0.25 FPR 0.02
22The ROC Curve
TPR vs. FPR as the test-positive threshold is
varied
23Quantifying the accuracy of the PSA test
- The age of the man matters
- PSA increases with age, in the absence of cancer
- Age is a strong risk factor for cancer
- If we ignore age, PSA performance will look
artificially high - Men with cancer are older on average
- Older men tend to have higher PSA
- Confounding
24An Age-Adjusted ROC Curve
- TPR vs. FPR among men of the same age
- This allows the test-positive threshold to
depend on age
25The Age-Adjusted ROC Curve for PSA
When FPR 0.025, TPR 0.17 (0.13 to
0.21) When FPR 0.05, TPR 0.27 (0.21 to
0.33)
26Summary
- Evaluating the efficacy of a candidate HIV
vaccine - The Step trial
- Vaccine effects on time to ART, viral load, CD4
- Statistical methods that accommodate longitudinal
data, missing data - Statistical methods for evaluating medical tests
- Eg PSA for prostate cancer screening
- The tradeoff between TPR and FPR
- Statistical method to adjust for covariates