LABORATORY STATISTICS

Accuracy & Precision

Accuracy Precision
The nearness or The ability of an
closeness of the assayed analytical method to give
value to the true or target repeated results on the
value same sample that agree
with one another.
Descriptive Statistics
• Descriptive Statistics is the foundation for monitoring
performance or quality control.
• Laboratory data patterns are described by their center,
spread and shape.
• Assessment of data dispersion or spread assess the
predictability in laboratory measurement.
Measures of Center
• Mean, Median and Mode -It is the most common used descriptors of
the center of a data set.
Measures of Center
Mean
• is calculated by adding the values of all the individual data
points and dividing that sum by the total number of data
points.
• Most common
Median
• “middle”
• Used when the data are skewed so its calculation will not
be affected by outliers.
Mode
• rarely used; most frequent observation
• it is used to describe data with two centers (bimodal)
Examples (Find the Mean)
1. 5,4,6,5,3,7,5
2. 5,4,6,8,9,7
Examples (Find the Median)
1. 5,4,6,5,3,7,5
2. 5,4,6,8,9,7
Examples (Find the Mode)
1. 3,4,5,5,5,6,7
2. 4,5,5,5,6,7,8,9,9,9
Measures of Spread
• Assessment of data dispersion or spread allows
laboratorians to assess the predictability in the
laboratory test.
• It represents the the relationship of all the data points
to the mean.
• Range, Standard Deviation, Coefficient of Variation
and Standard Deviation Index
Measures of Spread
Range
• simplest expression of spread or distribution
• it is the difference of highest and lowest score in a data
Standard Deviation
• It is a measure of dispersion of values from the mean.
• Helps describe the normal curve. A measure of
distribution range.
Variance - measure of variability; it determines significant
difference between groups of data.
Measures of Spread
Coefficient of Variation
• a percentile expression of the mean.
• An index of precision.
• Compares SDs
SD Index
• the difference between the value of a data point and the mean
value divided by the group’s SD
• SDI = 0 (ACCURATE)
• SDI = 3 (INACCURACY)
Variance
SDI
CV

SD
Measures of Shape
•Gaussian distribution curve
• Normal Distribution Curve/Bell-shaped curve
• Johann Karl F. Gauss
• It describes many continuous laboratory variables and
shares several unique characteristics.
Measures of Shape
• Total area under the
Gaussian curve is 1.0
• ±1SD – 68.3%
• ± 2SD – 95.5 %
• ± 3SD – 99.7%
x - mean (x – mean)2
1. Find the mean.
2. Find the difference of the individual values & mean.
3. Find the square of difference of the individual values & mean.
4. Find the SD.
5. Identify values of 1SD. (1)(mean)(sd)
6. Identify values of 2 SD.
7. Identify values of 3 SD.
1. Find the mean. 1. 18
2. Find the difference of the 2. 0, 0.2, 0.1, 0, 0.2, 0.1, 0, 0.2, 0.1
individual values & mean. 3. 0.16
3. Find the square of difference of 4. 0.13
the individual values & mean.
5. 17.9 – 18.1
4. Find the SD.
6. 17.7 – 18.3
5. Identify values of 1SD.
7. 17.6 – 18.4
6. Identify values of 2 SD.
7. Identify values of 3 SD.
MEASUREMENT OF PRECISION
• Within-run (repeatability) - the same laboratorian
performs 20 replicate analyses in a single run on a
single day.

• Within-laboratory (across-run or between-run) -

Similarly, within-laboratory precision is estimated by
measuring a sample 20 times over multiple days.
SOURCES OF ANALYTIC VARIABILITY
• Operator technique
• Instrument differences
• Test accessories
• Contamination
• Environmental conditions (e.g., temperature, humidity)
• Reagents
• Power surges
• Matrix effects (hemolysis, lipemia, icterus)
Descriptive Statistics of Groups of Paired
Observations
• It is used in Comparison of Method (COM) studies
• Compares a REFERENCE METHOD and the NEW METHOD.
• Reference Method is plotted on the x-axis
• Test Method is plotted on the y-axis
Descriptive Statistics of Groups of Paired
Observations
• Represented by a LINEAR REGRESSION
• Linear regression is a method of determining the
relationship between a dependent variable (y) with one or
more independent variables (x).
• Slope, y-intercept and correlation coefficient = r
• Correlation Coefficient
• Measures the strength of the relationship between the two
methods.
• m = slope
• B = y-intercept
•x=0
•y=?
• r = correlation coefficient
• +: both variables increase and
decrease together
• -: one variable increases, other
variable decreases
• 1.0 = perfect relationship
Anscombe
quartet
slope, y-intercept, and
correlation coefficients
are all identical, yet
visual inspection
reveals that the
underlying data are
completely different.
• The difference between the reference and test method is called
ERROR.

Constant
Systematic
Error Proportional
Random
Random Error
• It is present in all measurements and can be either positive or
negative, typically a combination of both positive and negative errors
on both sides of the assigned target value.
• Random error can be a result of many factors including instrument,
operator, reagent, and environmental variation.
• Random error is calculated as the SD of the points about the
regression line (Sy/x).
• Ideal: 0.00
Systematic Error
• It influences observations consistently in one direction (higher
or lower).
• The measures of slope and y-intercept provide estimates of
the systematic error.
• Systematic error can be further broken down into constant
error and proportional error.
Constant Systematic Error
It exists when there is a continual difference between the test
method and the comparative method values, regardless of the
analyte concentration.
This constant difference, reflected in the y-intercept, is called
constant systematic error.
Ideal value: 0.00
Proportional Systematic Error
• It exists when the differences between the test method and the
comparative method values are proportional to the analyte
concentration.
• The error is proportional, because it will increase with the
analyte concentration.
• Proportional error is present when the slope is NOT 1.
MEASUREMENT OF BIAS
• This examines patient samples by the method being
evaluated (test method) with a reference method. It is used
primarily to estimate the systematic error in actual patient
samples, and it may offer the type of systematic error
present (proportional vs. constant).
• Ideally, the test method is compared with a standardized
reference method (gold standard), a method with
acceptable accuracy compared to its imprecision.
STATISTICAL ANALYSIS OF BIAS
• Linear regression analysis yields the slope (b), the y-intercept
(a), the SD of the points about the regression line (Sy/x), and
the correlation coefficient (r); regression also yields the
coefficient of determination (r2)
• r2 indicates the proportion of variation explained by one
variable to predict another and ranges from 0 to 1.
INFERENTIAL STATISTICS
Nonparametric method:
• Statistical test that makes no
specific assumption about the
distribution of data.
• Nonparametric methods rank
the reference data in order of
increasing size.
• Nonparametric tests are the
recommended analysis for most
reference range intervals.
Parametric method:
Statistical test that assumes
the observed values, or
some mathematical
transformation of those
values, follow a (normal)
Gaussian distribution.
Inferential Statistics
Nonparametric method:
1. Mann-Whiteney test
2. Wilcoxon-Signed Rank test
3. Kruskal-Wallis test
4. Friedman test
5. Chi-square test
6. Spearman Rank order
correlation
Parametric Tests
t-test
• used to determine whether there is statistically significant difference
between the means of two groups of data

F-test
• used to determine whether there is statistically significant difference
between the standard deviations of two groups of data

ANOVA
• ANOVA tests whether several different groups (three or more) all have
the same mean.
Method Evaluation
QUALITY MANAGEMENT
SYSTEM
Traditional Total Quality
• Acceptable quality
• Error-free quality
• Department focused
• Organization-focused
• Quality as expense • Quality as means to lower cost
• Defects by workers • Defects by system
• Management-controlled worker • Empowered worker
• Status quo • Continuous quality improvement
• Manage by intuition • Manage by fact
• Intangible quality • Quality defined
• We vs. they relationship • Us relationship
• End-process focus • System process
• Reactive system • Proactive system
Quality Management System
All of laboratory’s policies, processes, procedures and resources
needed to achieve quality testing.

Quality Assurance/Assessment
Process by which laboratory ensures quality results by close monitoring
preanalytical, analytical, and postanalytical stages of testing.

Quality Control
Part of the analytical phase of Quality Assurance (QA); process of monitoring
results from control samples to verify accuracy of patient results.
QUALITY CONTROL
• It is part of the process management component of the quality system
that integrates good laboratory practices to ensure correct patient
results.
• It is the process of ensuring that analytical results are correct by testing
known samples that resemble patient’s sample and comparing their
determined values with their expected values (control values).
Quality Assurance Quality Control
• Process-oriented • Product-oriented
• Plan to avoid defect • Try to find defects and correct
in the first place them while making the product
• QA is all about • QC is all about detection
prevention • QC is used to verify the quality of
• QA involves the product
processes managing
quality
OBJECTIVES OF QUALITY CONTROL
• To check the stability of the
machine.
• To check the quality of
reagents.
• To check technical errors.

Henry’s Clinical Diagnosis and Management, 23rd Edition, Chapter 10, p.115
QUALITY CONTROL
• Control Solutions
o Patient-like; should be tested in the same
manner as patient samples
o Liquid or lyophilized
o Composed of one or more analytes of known
concentration
• The accuracy of any assay depends on the
control solutions.
• Clinical Chemistry: 2-Levels of Control
Solutions
TYPES OF QUALITY CONTROL
• Pooled serum
o Human – more expensive but preferred
o Bovine – most common animal source
• Commercial control
o Usually lyophilized for stability
o Requires reconstitution
o Incorrect mixing = incorrect control values
A. Assayed – more expensive; company analyzed control
and established reference range
B. Unassayed – without values; get reference range
CHARACTERISTIC OF AN IDEAL QC MATERIALS
• It should resemble human sample
• Inexpensive and stable for long periods
• No communicable diseases
• No matrix effects/known matrix effects
• With known analyte concentrations
• Convenient packaging for easy dispensing and storage
TYPES OF QUALITY CONTROL
• Intralaboratory Quality Control
• Internal Quality Control
• Interlaboratory Quality Control
• External Quality Control
INTERNAL QUALITY CONTROL
• Analyses of control samples together with the patient specimens
• Daily monitoring of accuracy & precision of analytical methods
• Detection: Random and Systematic error
EXTERNAL QUALITY CONTROL
• Interlaboratory Quality Control
• It involves the proficiency testing programs that periodically provide
samples of unknown concentrations to participate in clinical
laboratories .
• It is important in maintaining long-term accuracy of the analytical
methods
• The College of American Pathologists Proficiency Program is the gold
standard for clinical laboratory external QC testing
EXTERNAL QUALITY CONTROL
• A series of unknown samples are sent to the laboratory from the
reference laboratory or authorized program provider.
• Unknown samples must be tested by the laboratorians who regularly
perform analysis of patient specimens using the same reagents and
equipment for actual patient specimens, and the results are
submitted to the program provider, preferably as soon as every
analysis is done.
• Analysis of the unknown samples should be completed within the
usual time as for the routine samples.
EXTERNAL QUALITY CONTROL
• Unknown accuracy samples should be treated like a patient specimen
to determine the true essence of the accuracy.
• Results of the proficiency testing must not be shared with their
laboratories “during the testing period”. Comparison studies can be
made after the testing cycle to identify areas for improvement.
• If there is no available proficiency testing program for certain analyte,
it is required to implement a non-proficiency test scheme.
CALIBRATION
• It is the process of verification by comparing the accuracy of the
measuring instrument against a reference standard.
REFERENCE STANDARDS
• This are substances with known concentration used.
• It is done every 6 months or more frequently if recommended by the
manufacturers.
1. Primary Standard
2. Secondary Standard
REFERENCE MATERIALS
• A primary standard reference material
• Is an ultra high purity grade compound used in analysis involving assay,
identification or purity tests.
• Secondary standards or working standards
• These are also high purity grade materials which are quantified in relation to
primary standards and put to routine use in laboratories.
• Limited validity depending on stability of the material and before expiry fresh
working standards should be prepared for future use.
LEVEY-JENNINGS CHART
• The most common intralaboratory quality control chart used in Clinical
Chemistry laboratory; considered as the Gaussian Curve on its side.
• When the observed value falls with the control limits, it can be interpreted
that the method is performed adequately.
• Points falling outside the control limits suggest that problems may be
developing.
• Multi-rules were formalized by the Western Electric Company and later
applied to the clinical laboratory by Westgard and Groth. Multi-rules
establish a criterion for judging whether an analytic process is out of
control. To simplify the various control rules, abbreviations are used to
refer to the various control rules.
ESTABLISHING STATISTICAL QC LIMITS
• Prior to implementing a new lot number of QC material,
both levels of QC material must be analyzed for 20 days.
• Shift – abrupt change in analytic processes.
• Trend – gradual change in analytic processes.
One control value
exceed 2SD from the
mean. This is used as a
warning rule to trigger
inspection of control
data by following the
rejection rules.
One control value
exceed 3SD from the
mean.
Two consecutive
values exceed the
same limit, either
+2SD or – 2SD
from the mean.
1 control
measurement
exceeds the mean
+ 2SD and another
exceeds the – 2SD
mean.
4 consecutives
values exceeds
either + 1SD and
or – 1SD mean.
The value must all
be in the same
direction.
8 consecutive
values all lie on
one side with the
mean.
12 consecutive
values all lie on
one side with the
mean.
2 out of 3 control
values exceed the
same mean plus 2s
or mean minus 2s
control limit.
3 consecutive
control values
exceed the same
mean plus 1s or
mean minus 1s
control limit.
6 consecutive
values all lie on
one side with the
mean.
9 consecutive
values all lie on
one side with the
mean.
7 control values
trend in the same
direction. i.e. get
progressively
higher or lower
RANDOM ERROR
• No trend or means of predicting it.
• Occurs once; non-repeating.
• Easy to spot
• Human error: Mislabeling, improper mixing of sample and
reagent, voltage and temperature fluctuations
• 12s, 13s, R4S
SYSTEMATIC ERROR
• Will be seen as a trend in a data
• Repeating error
• Improper calibration, deterioration of reagents, sample
instability, instrument drift, changes in standard
materials
• 22s, 41s, 10x
SHIFT
• 6 or more values on one side of the mean; maintains a
constant level.
• i.e. deterioration of standards
TREND
• Control values that increase or decrease over a period of 6
consecutive days; usually they pass through the mean.
i.e. deterioration of reagents