Diagnosis of Pancreatic Cystic Neoplasms - PowerPoint PPT Presentation

1 / 23
About This Presentation
Title:

Diagnosis of Pancreatic Cystic Neoplasms

Description:

Set equal to 0. dy/dx 1 = 0. dy/dx = 1. Hence, maximum sum value of ... ROC curves plotted (data not shown) and numerical algorithms used to integrate AUC: ... – PowerPoint PPT presentation

Number of Views:1276
Avg rating:3.0/5.0
Slides: 24
Provided by: att78
Category:

less

Transcript and Presenter's Notes

Title: Diagnosis of Pancreatic Cystic Neoplasms


1
Diagnosis of Pancreatic Cystic Neoplasms
  • Peter Attia
  • May 16, 2006

2
Background
  • Neoplastic nature of pancreatic cysts recognized
    for more than a century
  • Past several decades have great progress in the
    classification and characterization of such
    lesions
  • Most pancreatic cysts are pseudocysts,
    complicating acute and/or chronic pancreatitis
  • However, 10 15 are neoplastic

3
Differential Diagnosis
Pancreatic Cyst
Inflammatory
Infectious
Congenital
Neoplastic
  • Pseudocyst
  • Echinococcal
  • cyst
  • Simple
  • Polycystic dz
  • - CF
  • - VHL
  • - PDK/L
  • Enterogenous
  • Serous cystadenoma
  • Mucinous
  • IPMN
  • Solid pseudopapillary
  • Cystic endocrine
  • Ductal adeno with
  • cystic degeneration
  • Acinar cell
  • cystadenocarcinoma

4
Distribution of Cystic Neoplasms
5
  • PURPOSE
  • Determine the most accurate test for
    differentiating mucinous from nonmucinous cystic
    lesions of the pancreas

6
Materials and Methods
  • Multicenter trial initiated in 1999 (12 centers)
  • Patients (with or without symptoms) found to have
    pancreatic cysts gt10 mm on US or CT
  • Excluded if PT/PTT, platelets inappropriate, AP,
    presence of pancreatic abscess
  • 341 patients enrolled over 2 years and underwent
    EUSFNA
  • Cystic lesions aspirated under EUS guidance using
    one pass of 19- or 22-guage needle
  • Morphologic findings recorded
  • 1. adjacent mass (? mucinous)
  • 2. macrocystic septations (? mucinous)
  • 3. honeycombed septations (? nonmucinous)
  • 4. diffusely thickened wall (? nonmucinous)

7
Materials and Methods (cont)
  • Cytologic findings recorded
  • 1. Mucinous epithelium (clusters of glandular
    cells with cytoplasmic mucin)
  • 2. Nonmucinous epithelium (flat monolayers of
    small cuboidal cells or inflammatory cells)
  • Histologic classification
  • 1. Mucinous cystic neoplasm (benign, borderline,
    malignant)
  • 2. Nonmucinous cystic neoplasm
  • Cystic lesions arising from IPMN were considered
    mucinous
  • Tumor markers CEA, CA 72-4, CA 125, CA 19-9, and
    CA 15-3
  • Separate ROC curves were plotted for each tumor
    marker and the AUC was used to quantify
    predictive power
  • The cutoff value was selected to optimize both
    specificity and sensitivity

8
Seemingly major digression
True Condition
Negative
Positive
Test Result
Total
Positive
FP
TP
T
Negative
TN
FN
T-
Total
D-
D
Sensitivity TP/D True Positive / Total
Disease True Positive Rate (TPR)
Specificity TN/D- True Negative / Total
Disease -
1 Specificity 1 TN/D- (D- TN)/D-
FP/D-
False Positive / Total Disease -
False Positive Rate (FPR)
9
But what if the test is not binary?
  • Sometimes the results of a test fall into one of
    two obviously defined categories ? hence one
    sensitivity/specificity pair
  • What if the test is more complicated?
  • e.g. CT characteristics of a lung nodule
  • Five-point scale of evaluation
  • benign
  • probably benign
  • possibly malignant
  • probably malignant
  • malignant

10
Cutoff Level
  • In the previous example there are 4
    specificity/sensitivity pairs
  • How are they related?
  • As the cutoff decreases ?

Sensitivity
Specificity
11
Receiver Operating Characteristic Curves
  • The ROC curve is defined as a plot of test
    sensitivity (true positive rate) as the
  • y-coordinate versus its false positive rate
  • (1-specificity) as the x-coordinate
  • This is a very effective method of evaluating the
    performance of a diagnostic test

12
What does this look like?
13
How to quantify? Upper and lower limits?
AUC Area Under Curve
Test A (best possible) AUC 1
Test D (chance diagonal) AUC 0.5
Hence,
Test A gt Test B gt Test C gt Test D
14
How to extract optimal sensitivity and
specificity from ROC curve?
1
  • y sensitivity and x 1 specificity
  • maximize sensitivity specificity
  • ? maximize y (1 x)
  • Differentiate with respect to x and
  • Set equal to 0
  • dy/dx 1 0
  • ? dy/dx 1
  • Hence, maximum sum value of
  • sensitivity specificity attained
  • when slope of ROC curve 1

y
0
1
x
15
How to extract optimal sensitivity and
specificity from ROC curve?
1
  • y sensitivity and x 1 specificity
  • maximize sensitivity specificity
  • ? maximize y (1 x)
  • Differentiate with respect to x and
  • Set equal to 0
  • dy/dx 1 0
  • ? dy/dx 1
  • Hence, maximum sum value of
  • sensitivity specificity attained
  • when slope of ROC curve 1

y
0
1
x
16
Final word on sensitivity and specificity
Specificity f(x)
1
If you want to maximize the sum of
specificity and sensitivity, consider the
following function h(x) f(x)g(x), where
f(x)gt0, g(x)lt0 for all x on 0,N
Sensitivity g(x)
0
0
N
Test value
The maximum value of the sum of specificity and
sensitivity i.e. h(x) f(x)g(x) occurs when
the derivative of h(x) 0 ? f(x)g(x) 0 ?
f(x) -g(x). In other words, when the slopes
of sensitivity and specificity curves are of
equal magnitude (but opposite sign). NB This
is NOT necessarily where the curves intersect!
17
Results Patient Characteristics
18
Results Histology of Mucinous Neoplasms
  • Most were mucinous cystic neoplasms (rather than
    IPMN), 52/68 76
  • About half of these were malignant, 29/52 56

19
Results Accuracy of Tumor Markers
  • Each TM cutoff value found (empirically) by
    optimizing sensitivity and specificity ? based on
    this cutoff highest sensitivity and specificity
    shown in Table 3
  • Optimal CEA found (empirically) to be 192 ng/mL,
    approximately where the curves intersected (by
    chance)
  • ROC curves plotted (data not shown) and numerical
    algorithms used to integrate AUC
  • Varied from 0.7930 (CEA) to 0.5011 (CA15-3)

20
Results Accuracy
  • EUS morphology, cytology, and CEA (with cutoff of
  • 192 ng/mL) compared for sensitivity,
    specificity, and accuracy

21
Results Combination Testing
  • Three combinations of diagnostic tests evaluated
  • The combination of morphology, cytology, or CEA
    produced the greatest sensitivity (91) of any
    single test or combination
  • However, the accuracy of CEA alone was highest

22
Conclusions
  • CEA alone, with a cutoff value of 192 ng/mL
    (based on the assay used by the authors), had
    greater accuracy in differentiating mucinous from
    nonmucinous pancreatic cystic neoplasms than any
    other test, alone or in combination
  • The sensitivity of detecting mucinous lesions
    could be increased by the addition of other
    diagnostic tests (e.g. morphology, cytology), but
    this came at a cost namely decreased specificity
    and accuracy
  • This study confirmed that CEA and CA 72-4 were
    present in much higher concentrations in mucinous
    rather than serous cysts

23
Conclusions
  • This type of study, and the analysis undertaken,
    underscores the importance of understanding when
    it makes sense to have higher sensitivity vs.
    higher specificity (is it better to jail innocent
    people or have guilty people walk free?)
Write a Comment
User Comments (0)
About PowerShow.com