Biostatistics Case Studies Peter D. Christenson Biostatistician Session 2: Diagnostic Classification.

Slides:



Advertisements
Similar presentations
Logistic Regression I Outline Introduction to maximum likelihood estimation (MLE) Introduction to Generalized Linear Models The simplest logistic regression.
Advertisements

Logistic Regression.
Evaluating Diagnostic Accuracy of Prostate Cancer Using Bayesian Analysis Part of an Undergraduate Research course Chantal D. Larose.
1 Using Biostatistics to Evaluate Vaccines and Medical Tests Holly Janes Fred Hutchinson Cancer Research Center.
April 25 Exam April 27 (bring calculator with exp) Cox-Regression
Journal Club Alcohol and Health: Current Evidence March-April 2005.
Chapter 11 Survival Analysis Part 2. 2 Survival Analysis and Regression Combine lots of information Combine lots of information Look at several variables.
Prevention Strategies Rajesh G. Laungani MD Director, Robotic Urology Chairman, Prostate Cancer Center Saint Joseph’s Hospital, Atlanta.
Notes on Logistic Regression STAT 4330/8330. Introduction Previously, you learned about odds ratios (OR’s). We now transition and begin discussion of.
Statistics for Health Care
Basic Statistical Concepts Donald E. Mercante, Ph.D. Biostatistics School of Public Health L S U - H S C.
Assessing Survival: Cox Proportional Hazards Model Peter T. Donnan Professor of Epidemiology and Biostatistics Statistics for Health Research.
STAT E-150 Statistical Methods
Prostate Cancer Screening 2012 Paul L. Crispen, MD Department of Surgery University of Kentucky.
Prostate Cancer Screening Assistant Professor Charles Chabert Men’s health Seminar Ballina April 2011 prostates.com.au.
M Ravanbod Medical oncologist Bushehr – 11/91 A 50 y/o white man comes for check up and wants to discuss about prostate cancer. Negative family history.
Marshall University School of Medicine Department of Biochemistry and Microbiology BMS 617 Lecture 12: Multiple and Logistic Regression Marshall University.
Logistic Regression II Simple 2x2 Table (courtesy Hosmer and Lemeshow) Exposure=1Exposure=0 Disease = 1 Disease = 0.
Multiple Choice Questions for discussion
Logistic Regression III: Advanced topics Conditional Logistic Regression for Matched Data Conditional Logistic Regression for Matched Data.
Prostate Screening in 2009: New Findings and New Questions Durado Brooks, MD, MPH Director, Prostate and Colorectal Cancer.
Prostate Cancer James B. Benton,M.D.. Prostate Cancer Significant of the clinical problem Early detection/screening Prevention/Management.
Biostatistics Case Studies 2005 Peter D. Christenson Biostatistician Session 4: Taking Risks and Playing the Odds: OR vs.
SoftPAP® A Novel Collection Device for Cervical Cytology.
Surrogate End point for Prostate Cancer- Specific Mortality After RP or EBRT A D’Amico J Nat Ca Inst 95,
Biostatistics Case Studies Peter D. Christenson Biostatistician Session 5: Analysis Issues in Large Observational Studies.
Prostate Cancer Screening in 2013: Reports of its Death Are Greatly Exaggerated Norm D. Smith, M.D. Associate Professor Co-Director Urologic Oncology University.
April 11 Logistic Regression –Modeling interactions –Analysis of case-control studies –Data presentation.
Statistics for Health Care Biostatistics. Phases of a Full Clinical Trial Phase I – the trial takes place after the development of a therapy and is designed.
Assessing Survival: Cox Proportional Hazards Model
EDRN Approaches to Biomarker Validation DMCC Statisticians Fred Hutchinson Cancer Research Center Margaret Pepe Ziding Feng, Mark Thornquist, Yingye Zheng,
Reliability of Screening Tests RELIABILITY: The extent to which the screening test will produce the same or very similar results each time it is administered.
How do we know whether a marker or model is any good? A discussion of some simple decision analytic methods Carrie Bennette (on behalf of Andrew Vickers)
“The African American Prostate Cancer Crisis in Numbers”
PCa Screening New Areas of Research Francesco Montorsi Milan.
Prevention with Finasteride Ian M. Thompson, MD October, 2009.
April 6 Logistic Regression –Estimating probability based on logistic model –Testing differences among multiple groups –Assumptions for model.
April 4 Logistic Regression –Lee Chapter 9 –Cody and Smith 9:F.
Logistic and Nonlinear Regression Logistic Regression - Dichotomous Response variable and numeric and/or categorical explanatory variable(s) –Goal: Model.
Assessing Binary Outcomes: Logistic Regression Peter T. Donnan Professor of Epidemiology and Biostatistics Statistics for Health Research.
MBP1010 – Lecture 8: March 1, Odds Ratio/Relative Risk Logistic Regression Survival Analysis Reading: papers on OR and survival analysis (Resources)
Biostatistics in Practice Peter D. Christenson Biostatistician LABioMed.org /Biostat Session 4: Study Size and Power.
Biostatistics in Practice Peter D. Christenson Biostatistician Session 4: Study Size and Power.
Jennifer Lewis Priestley Presentation of “Assessment of Evaluation Methods for Prediction and Classification of Consumer Risk in the Credit Industry” co-authored.
Biostatistics in Practice Peter D. Christenson Biostatistician Session 6: Case Study.
A.N.N.C.R.I.P.S The Artificial Neural Networks for Cancer Research in Prediction & Survival A CSI – VESIT PRESENTATION Presented By Karan Kamdar Amit.
1 Lecture 6: Descriptive follow-up studies Natural history of disease and prognosis Survival analysis: Kaplan-Meier survival curves Cox proportional hazards.
Evaluating Screening Programs Dr. Jørn Olsen Epi 200B January 19, 2010.
Biostatistics in Practice Peter D. Christenson Biostatistician Session 4: Study Size for Precision or Power.
Biostatistics Case Studies 2007 Peter D. Christenson Biostatistician Session 1: The Logic Behind Statistical Adjustment.
Unit 15: Screening. Unit 15 Learning Objectives: 1.Understand the role of screening in the secondary prevention of disease. 2.Recognize the characteristics.
Organization of statistical research. The role of Biostatisticians Biostatisticians play essential roles in designing studies, analyzing data and.
Sample Size Determination
Biostatistics Case Studies 2007 Peter D. Christenson Biostatistician Session 2: Aging and Survival.
BIOSTATISTICS Lecture 2. The role of Biostatisticians Biostatisticians play essential roles in designing studies, analyzing data and creating methods.
Biostatistics Case Studies Peter D. Christenson Biostatistician Session 3: Missing Data in Longitudinal Studies.
Biostatistics Case Studies 2006 Peter D. Christenson Biostatistician Session 1: Demonstrating Equivalence of Active Treatments:
Diagnostic Likelihood Ratio Presented by Juan Wang.
Logistic Regression Logistic Regression - Binary Response variable and numeric and/or categorical explanatory variable(s) –Goal: Model the probability.
Radical Prostatectomy versus Watchful Waiting in Early Prostate Cancer Anna Bill-Axelson, M.D., Lars Holmberg, M.D., Ph.D., Mirja Ruutu, M.D., Ph.D., Michael.
Prostate cancer update Suresh GANTA Consultant urological surgeon Manor Hospital.
PSA screening Cost Conscious Project Kristopher Huston January 2016.
© 2010 Jones and Bartlett Publishers, LLC. Chapter 12 Clinical Epidemiology.
What are the Chances Dr? Nick Pendleton. Can I have a Prostate Check? ?
Screening Tests: A Review. Learning Objectives: 1.Understand the role of screening in the secondary prevention of disease. 2.Recognize the characteristics.
Clinical Epidemiology
April 18 Intro to survival analysis Le 11.1 – 11.2
How do we delay disease progress once it has started?
Prostate Cancer Screening- Update
Narrative Reviews Limitations: Subjectivity inherent:
Presentation transcript:

Biostatistics Case Studies Peter D. Christenson Biostatistician Session 2: Diagnostic Classification

Case Study

PSA Background PSA secreted by prostatic epithelial cells; test developed in the late 1970s. Currently, usually screen PSA>4.0 ng/ml. Why 4.0? Unclear; early studies of CaP patients showed many PSA > 4.0? Hesitant to unnecessarily take biopsies with lower PSA. A few studies did suggest substantial CaP with lower PSA. Prostate Cancer Prevention Trial specified biopsy at study end, regardless of PSA.

Issues and Goals Over-diagnosis in men over 50: Microscopic evidence in 33% of autopsy / cystoprostatectomy specimens. 9-16% currently diagnosed with CaP. 3% CaP mortality. Screening (early diagnosis): Maximize curable CaP detection and exclude as many men as possible from unnecessary biopsies. Prognosis (predicting outcome): Predict mortality using pre- and post-biopsy info.

Catalona, et al (1991) Men ≥ 50 years of age with PSA ≥ 4.0. Biopsy those with abnormal DRE or ultrasound. N=1653 N=1516 N=107 (6.5%)N=30 (1.8%) N=85N=27 N=19/85=22%N=18/27=67% PSA<4.0PSA≥10 4.0≤PSA<10 ? Abnormal DRE / US CaP in biopsy

Krumholtz, at al (2002) Prostate screening program. Recommend biopsy if high PSA and/or abnormal DRE. PSA cutoff changed from 4.0 to 2.6 mid-study. CaP in 156/601=26% with 2.6≤PSA≤4.0 and 97/309=31% with 4.0<PSA≤10.0. Report on 94 with embedded prostatectomy specimens. PSA≤4.0  more organ-confined; not greater over- detection.

Prostate Cancer Prevention Trial (PCPT) 18,882 men randomized to finasteride or placebo; up to 7 years follow-up. Annual DRE and PSA. Biopsy recommended if PSA>4.0 or abnormal DRE. End of study biopsy planned for all men without during-study diagnosis of CaP. Primary outcome = biopsy CaP positive or negative. Main result: finasteride had 25% efficacy; 18% CaP in finasteride vs. 24.4% CaP in placebo.

Current Paper: Thompson et al, 2004.

Main Results Overall, 449/2950 = 15.2% with CaP detected in biopsy.

Conclusions and Issues There is substantial CaP with low PSA values, and the rate appears to have a dose-response relationship with PSA. Is it a good screening tool? Does it have prognostic ability, at least for CaP in biopsy, since mortality was not studied. What do we make of the reported sensitivity and specificity? We first examine several “what if” scenarios with artificial data.

Scenario 1: PSA is useless

Scenario 2: PSA is a perfect test

Scenario 3: Almost perfect association

PSA Prognostic Ability in Scenario 3 Predict P(CaP) = Probability(CaP) using PSA: P(CaP if PSA = 3) = 33 ± 2 % (CI) P(CaP if PSA = 8) = 88 ± 2 % (CI) Since ±2% is very small, this study is very precise at measuring the prevalence of biopsy-evident CaP according to PSA intervals. PSA would be a decent prognostic factor for biopsy-detected CaP (but of course we actually want to predict clinical outcomes).

PSA Screening Ability Sensitivity= True positive rate = % identified among CaP +. Specificity= True negative rate = % not identified among CaP -. For screening, sensitivity is usually more important than specificity.

PSA Screening Ability in Scenario 3 using PSA>4.0 Sensitivity= 820/( ) = 65% Specificity= 2501/( ) = 87%

PSA Screening Ability in Scenario 3 using PSA>2.0 Sensitivity= 987/( ) = 78% Specificity= 1993/( ) = 55%

Conclusions from Scenarios Association ≠ screening accuracy. Good* screening needs something more like: *Few unnecessary biopsies, but detect most serious CaP.

Back to Actual Results: Overall, 449/2950 = 15.2% with CaP detected in biopsy.

Revised Table 2: Prevalence of CaP and its Precision 95% CI for CaP Prevalence PSA RangeAny GradeHigh Grade ≤ ± 2.2% % 0.6 – ± 2.1%1.0 ± 0.7% 1.1 – ± 2.4%2.0 ± 0.9% 2.1 – ± 3.9%4.6 ± 1.9% 3.1 – ± 6.4%6.7 ± 3.6% ≤ ± 1.3%2.3 ± 0.5%

Table 2: Sensitivity and Specificity Sensitivity at PSA=1.1 of 0.75 = ( )/449 Specificity at PSA=1.1of 0.33=( )/( ) Relative to only PSA≤4.0. Not useful without PSA>4.0 info.

Figure 2: Models P(CaP)=function(PSA) Risk= P(CaP)

Figure 2 Models Table 2 Data Dotted line is only to show form of model; do not extrapolate. Logistic model is not very useful with so much data, unless adjustment for other factors (e.g., family hx of CaP) is desired.

Logistic Regression Uses odds of disease = P(CaP)/[1-P(CaP)]. Log(odds) are linear in PSA  sigmoidal curve in previous graph, common for bounded outcomes. Increase in odds for a given change in PSA is proportional to PSA. For this study, log(odds)= (PSA) and P(CaP) = exp(logodds)/(1+exp(logodds) Can include other adjusting factors. Usually used for prediction (prognosis); but can define, e.g., Prob(CaP)=fixed number such as 0.22 to classify and obtain sensitivity and specificity.

Logistic Regression in Software SPSS: Select Analyze > Regression > Binary Logistic Specify CaP (1=Yes;0=N0) as dependent variable. Specify PSA as covariate. Select Options > CI for exp(B). OK SAS: proc logistic descending; class CaP; model CaP = PSA; run;

Software Output: SAS Odds Ratio Estimates Point 95% Wald Effect Estimate Confidence Limits PSA Analysis of Maximum Likelihood Estimates Parameter DF Estimate Intercept PSA

Combined Screening and Prognosis ? PSA Other Pre- Biopsy 1 Biopsy Character- istics 2 Risk of CaP Death Biopsy Moderate Low High Neg Pos 1 Such as PSA velocity 2 Such as prostate volumeSee NEJM 2004(Jul 8);351:

Summary This paper does not address PSA screening ability. Demonstrates substantial CaP even with very low PSA. Study in progress with mortality as outcome (Ref 29). Studies in progress using additional markers for early detection and of CaP (Ref 30). Possible prediction error in any study due to lab error in measuring PSA. Here, if large, say 20%, 20% error =~ 10% error in Prob(CaP).