Transcript Biological Variation

Biological Variation
Dr WA Bartlett
Biochemical Medicine
Ninewells Hospital & Medical School
Dundee
Scotland
Objectives

Identification the nature of biological
variation.
 Appreciation of the significance of
biological variation in clinical
measurements.
 Attain insight into the determination and
application of indices of biological
variation.
Identification the nature of
biological variation.
What is meant by the term
biological variation in the context
of clinical biochemistry?
A component of the variance in
biochemical measurements
determined by the physiology of the
subjects observed.
Components of Variance in
Clinical Chemistry
Measurements

Analytical variance.

Within Subject biological variance.

Between Subject biological variance.
Biological Variation

All clinical chemistry measurements
change with time.
 Knowledge of temporal changes useful in
diagnosis and interpretation.
 Rate of change may be useful in prognosis.
 Understanding of the sources of biological
variation in non-diseased subjects is
fundamental to the development of
reference data.
Sources of Biological Variation

Biological Rhythms (time)
 Homeostasis
 Age
 Sex
 Ethnicity
 Pathology
 Stimuli
Practical significance of
biological variation.

What is the significance of this result?
 Is the performance of the analytical
method appropriate (imprecision,
accuracy)?
 When should I measure it again?
 Has this result changed significantly over
time?
 Changes in variability be used as a tool?
Models of Biological Variation

Assume values represent random
fluctuation around a homeostatic setting
point.
 More general model allows correlation
between successive results. (Time series
and non-decayed biological variation)
Quantifying Biological
Variation
How are you going to quantify biological
variation?
You have to dissect out the components
of variance: s2total = s2Analytical + s2Individual + s2Group
Quantifying Biological
Variation
s2Analytical =
s2Individual =
s2Group =
Average variance of replicate assays
within run analytical variance
Average biological within subject
variance.
Average Variance around the
homeostatic setting point
Variance of true means among subjects.
Variance in homeostatic setting points
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
Analytical
Variance
Subject 1
Within Subject
Variance
Subject 2
Between Subject
Variance
Subject 3
Quantifying Biological
Variation
How do you do the experiment?
 Subjects
How many?
 Collect specimens Number? Frequency?
 Analyse
specimens Minimise s2Analytical ?
 Analyse
data Outliers? Statistics?
 Apply
results of analysis.
Quantifying Biological
Variation
Estimates of biological variation are
similar regardless of: Number of subjects
Time scale of study (Short v Long?)
Geography
A lot of information can be obtained
from small studies.
Within Subject Variation (CVI,%) for Serum Sodium and Urea
No. of
Time
Sexb status Na+ Urea
subjects
11
11
62
11
10
14
111
37
274
15
9
15
16
0.5 h
8h
1d
2 weeks
4 weeks
8 weeks
15 weeks
22 weeks
6 months
40 weeks
2d 6 weeks
8 weeks
m
m
m
m
F
m
m
F
m
H
H
H
H
H
H
H
H
H
H
RF
HP
DM
0.6
0.5
0.6
0.7
0.9
0.5
0.6
0.5
0.5
0.7
0.8
0.8
0.8
2.2
6.0
4.8
12.3
14.3
11.3
15.7
11.1
11.2
13.9
6.5
14.5
13.0
Collection of Specimens.

Conditions should minimise pre-analytical
variables.
Healthy subjects.
Usual life styles.
No drugs (alcohol, smoking?).
Phlebotomy by same person.
Same time of day at regular intervals.
Set protocol for sample transport, processing &
storage.
Analysis of Specimens

Need to minimise analytical imprecision.
 Ideal : Single lots of reagents and calibrants.
Single analyst and analytical system.
Single or very small number of
batches.
Preferred Protocol: Cotlove et al

Healthy subjects.
 Specimens taken at set time intervals.
 Specimens processed & stored frozen.
 When ALL specimens are available: Analysis of all samples in a single run.
Simultaneous replicate analysis.
Quality control to monitor drift
Preferred Protocol: Cotlove et al
Advantage: Minimisation of s2Analytical
Disadvantages: Limits the number of specimens and subjects
that can be studied.
Analyte must be stable on storage.
Other Protocols: Costongs et al
 Collection and
 Singleton
storage as before.
assay of all samples in a single
run.
 Duplicate
assay of QC or patient pool to
estimate s2Analytical
Other Protocols: Costongs et al
Disadvantages: True estimate of s2Analytical ?
Integrity of QC materials
 Viral infections of pools
Vial to vial variability in QC
Other Protocols: Costongs/Moses et al

Samples assayed once or in duplicate on
the day of collection
Disadvantage: s2individual confounded by between batch
variance.
Advantage: Useful if analyte is unstable.
Analysis of Data
2
Stages
– Identification of outliers
– Nested analysis of variance
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
Analytical
Variance
Subject 1
Within Subject
Variance
Subject 2
Between Subject
Variance
Subject 3
Applications of BV Data

Setting of analytical goals.
 Evaluating the significance of change in
serial results.
 Assessing the utility of reference
intervals.
 Assessing number of specimens required
to estimate homeostatic set points.
Applications of BV Data

Assessment of reporting strategies.
 Selecting the best specimen.
 Comparing utility of available tests.
Setting of analytical goals.

Accepted analytical goal for imprecision: CVGoal = ½ CVI
therefore: CVAnalytical = CVGoal
= ¼ of the s2Individual if achieved.
(Harris. Am J Clin Pathol 1979:72;274)
Utility of Analytical Goals

Assessment of methods and equipment.
 Should be addressed in early stages of
method development.
 Index of Fiduciality: CVAnalytical /CVGoal
If <1 analytical goal met
(Fraser Clin Chem 1988:34;995)
Evaluating the significance
of change in serial results.

Critical Difference or Reference Change value
indicates the value by which 2 serial results
must differ to be considered statistically
significant: CD = 2½ * Z * (CVA2 + CVI2)½
Probabilty = 95% Z = 1.96
Probability = 99% Z = 2.58

Only valid if the variance of s2Individual is
homogenous.
(Costongs J Clin Chem Clin Biochem 1985;23:7-16)
Multipliers for (CVA2 + CVI2) ½ to Obtain Critical
Difference at Different Levels of Probability
Multiplier
(2 ½ * Z)
Probability of
false alarm
3.64
2.77
2.33
1.81
1.47
1.19
0.95
0.01
0.05
0.10
0.20
0.30
0.40
0.50
Probability
99%
95%
90%
80%
70%
60%
50%
Significance of Change?
63 year old patient: Cholesterol 1 = 6.60 mmol/L
Cholesterol 2 = 5.82 mmol/L
Significant change ?
Cva = 1.6%
CVI = 6.0%
RCV = 2½ * Z * (CVA2 + CVI2)½
95%RCV = 1.414 * 1.96 * (1.6 ½ + 6.60 ½) ½ = 17.2%
99%RCV = 1.414 * 2.58 * (1.6 ½ + 6.60 ½) ½ = 22.6%
Actual Change = ((6.60 – 5.82)/6.60)*100= 11.8%
Dispersion =Z* (SD2A + SD2I)
Dispersion of first result = result ± 1.96 SD : 95% level 6.60 = 5.80 –7.40
99% level 6.60 = 5.54 – 7.66
Dispersion of 2 result
95% level = 5.82 = 5.11 – 6.53
99% level = 5.82 = 4.89 – 6.75
Overlap: therefore neither significantly or highly
significantly different
Can use the formula to ascertain the probability that
change is significant. Calculate Z using the (((6.65.82)/6.6)*100%) as RCV and look up in tables. 82% in
this case.
USE of RCV

Handbooks reports, 95% and 99%
probabilities that change is significant.
(> or >> * or **)
 Delta checking, exemption reporting.
– 95% auto validate, 99% refer for clinical
validation or renanalysis.
Index of Heterogeneity

Measure of the heterogeneity of variance within
the study population: ratio of the observed CV of the set of subjects
variances (SDA+I2) to the theoretical CV ( / 2/n-1)
for the set.

The ratio should =1 (1SD = 1/ /2n )

Large ratio = more heterogeneity.
(Costongs J Clin Chem Clin Biochem 1985;23:7-16)
Assessing the utility of
reference intervals.

Utility of population based reference data?
 Ratio of Within to Between subject variances.
Index of Individuality = CVI / CVG
Population Ref Intervals: Index <0.6 = Limited in Value
Index >1.4 = Applicable
Biological Variation &Utility of Reference
Intervals
Number of specimens
required to estimate
homeostatic set points.
n = ( Z. CVA+ I/D)
where: Z = number of Standard deviates for a
stated probablity (e.g. 1.96 for 95%).
D = desired % closeness homeostatic set
point.
Number of specimens required to
estimate homeostatic set points: Cholesterol testing
How many samples (n) required to
estimate set point within ±5% given: CVI = 4.9% CVA = 3% (Recommended)
Substitute equation: n = ( Z. CVA+ I/D)
n = [1.96·(32 + 4.92)½/5]2 = 5.07
RCV at 95% and Number. of Specimens Required
to Assess the Homeostatic Set Point at Different Levels of
Imprecision
CVA
CVI
RCVa Number of
(%)
(%)
(%) specimensb
2.0
3.0
4.0
5.0
6.0
7.0
8.0
9.0
10.0
15.0
20.0
aRCV
(p <0.05) = 2.77 (CV
4.7
4.7
4.7
4.7
4.7
4.7
4.7
4.7
4.7
4.7
4.7
2
14.1
15.4
17.1
19.0
21.1
23.4
25.7
28.1
30.6
43.5
56.9
4
5
6
7
9
11
13
16
19
38
65
+ CV 2)½, assuming no statistical evidence of heterogenity
Assessment of reporting
strategies

Results may be reported in different
formats
e.g. 24h Urinary creatinine output: CVI for concentration = 23.8%
CVI for output per collection = 13.0%
CD for concentration = 66.0%
CD for output = 36.2%
Selecting best Specimen.

e.g early morning urines for albumin
versus 24h collections.

Random hormone measurements versus
timed measurements.
Comparing Available Tests

Creatinine v Creatinine Clearance

FT4 v TSH in replacement situations

FT4 v Total T4
Reference Intervals
Dr WA Bartlett
Birmingham Heartlands & Solihull
NHS Trust (Teaching)