Much ado about nothing a commentary on leftcensored biomarker data - PowerPoint PPT Presentation

Loading...

PPT – Much ado about nothing a commentary on leftcensored biomarker data PowerPoint presentation | free to view - id: 15bb8f-ZDc1Z



Loading


The Adobe Flash plugin is needed to view this content

Get the plugin now

View by Category
About This Presentation
Title:

Much ado about nothing a commentary on leftcensored biomarker data

Description:

Measurement processes may yield biomarker values below a predetermined lower bound ... Determination of detection limits for MIP-1 ELISA ... – PowerPoint PPT presentation

Number of Views:88
Avg rating:3.0/5.0
Slides: 20
Provided by: whit56
Category:

less

Write a Comment
User Comments (0)
Transcript and Presenter's Notes

Title: Much ado about nothing a commentary on leftcensored biomarker data


1
Much ado about nothing a commentary on
left-censored biomarker data
  • SPER Advanced Methods Workshop
  • June 20th, 2006
  • Brian W. Whitcomb

2
The issue
  • Measurement processes may yield biomarker values
    below a predetermined lower bound

3
Reporting of biomarker data below a lower
threshold
  • Reporting threshold is equal to 2.2

4
Why is there a threshold?
  • Issues related to biomarker, vs. issues related
    to noise

5
Conventional determination of the limit of
detection (LOD)
  • Defined by the IUPAC as the concentration or
    quantity …derived from the smallest measure, XL,
    that can be detected with reasonable certainty
    for a given analytical procedure
  • XL is the mean of the blanks plus k standard
    deviations of the blanks for the LOD, IUPAC
    suggests k 3

6
Determination of detection limits for MIP-1ß ELISA
Sensitivity The Minimum Detectable Dose (MDD)
was determined by adding two standard deviations
to the mean RFU of twenty zero standard
replicates and calculating the corresponding
concentration.
30 assays (each of 20 blanks) MDD 0.36 - 6.66
pg/mL Avg. MDD 2.12 pg/mL
From RD Systems, Minneapolis, MN
7
Conventional determination of the limit of
detection (LOD)an example
  • BLANK SERIES
  • 10.0
  • 5.0
  • 8.1
  • 7.1
  • 4
  • 11.3
  • 12.0
  • 8.0
  • 7.7
  • 7.0
  • mean 8.02
  • Std dev 2.53

8
Example of LOD left-censored data
Blanks
True X XN(20,10)
9
Example of LOD left-censored data
Blanks
True X
Observed X
Assumes error is Gaussian, independent of X and
homoscedastic
10
Example of LOD left-censored data
Blanks
Observed X Blanks subtracted
These observations are below the threshold and
are subject to special treatment
11
How should these data be treated?
  • Depends upon
  • Question
  • Data

12
Depends upon question…
  • Descriptive
  • Mean and variance
  • use available data and proportion missing to
    estimate distribution (Gupta 1952)
  • find expected value of missing data (Richardson
    and Ciampi 2004)
  • use non-parametric statistics (median and range)
  • Analytical
  • Linear relation with other continuous variables
    or log-linear relation with odds of binomial
    variables
  • use of expected value (Richardson and Ciampi
    2004)
  • replace by zero (Schisterman et al. 2006)

13
Depends upon the data…
  • What is the distribution of the noise in the
    measurement process?
  • and
  • Is the error is independent of the biomarker?

Calibration curve is a simple validation study
with replicate measures across a range of
biomarker levels
?
14
A simple approach to evaluate independence of
error
15
A simple approach to evaluate independence of
error
If xgtLOD then x If xltLOD then xerror
If xgtLOD then xerror If xltLOD then x
P(errorX) P(error)
LOD
LOD
LOD
16
Using relation of error with biomarker to
determine treatment of NDs
  • Error is independent of x
  • Consider issue a measurement error problem and
    treat accordingly (e.g., Spiegelman et al. 1997)
  • Error is a function of x
  • Nonlinear calibration models (Carroll et al.
    1995)

17
Using relation of error with biomarker to
determine treatment of NDs
  • Error is limited to levels below the threshold,
    or data below threshold are unavailable
  • Or
  • Distribution of X can be assumed
  • substitution methods may yield minimal bias
    (Hornung and Reed 1990 Richardson and Ciampi
    2004 Schisterman et al. 2006)

18
Conclusions
  • Treatment of data below a detection threshold
    depends upon the reasons for the threshold and
    the nature of the data above and below the
    threshold
  • Both must be understood before a decision is
    reached regarding those data

19
The end
  • Thank you
About PowerShow.com