Qc/statistical Analysis 1

Front 1

Descriptive Statistics

Back 1

Summarize a data set into a few numbers that represent or describe a group of numbers.

Mean

Front 2

Inferential Statistics

Back 2

Compare information from one set of data to another set of data to make a comparison or inference about the data

T-test or linear regression

Front 3

Population

Back 3

the entire group of items that could possibly be measured

Represented by N

Front 4

Sample

Back 4

A subset from the population. Should be randomly selected to accurately represent the population

Represented by n

Front 5

Variable

Back 5

the characteristic you want to measure

Front 6

data

Back 6

the values you obtain from a sample population

A single value is call a datum

Front 7

data set

Back 7

set of variable connected

Front 8

experiment

Back 8

a planned activity whose result yield a data set

Front 9

parameter

Back 9

a numerical value that summarizes one aspect of the data from a population

Front 10

Statistic

Back 10

a numerical value that summarizes one aspect of the data from a sample

Front 11

Qualitative

Back 11

Yes/No values

pos/neg

less than greater than

Front 12

Quantitative

Back 12

an actual number

Front 13

Accuracy

Back 13

the measurement closest to the true value

Inaccuracy: lack of closeness to the true value

Front 14

Precision

Back 14

the repeatability of the process how reproducible are your values

Front 15

Inaccuracy and Variability

Back 15

Inaccuracy must be detectable and is accepted within it's defined limits

variability: is expected and accepted within defined limits

Front 16

Variability

Back 16

We must decide what is acceptable

In the lab we test QC and use QC rules to do this

Front 17

Qualitative Data

Back 17

Gives a general impression:

• Pie charts
  • Bar graphs: vertical or horizontal categorize data, show a trend

Front 18

Ways to show quantitative data

Back 18

• Data columns
• Dot chart or line graph
• Frequency distribution
  • Histogram- smoothing out frequency trend

Front 19

Frequency distribution

Back 19

• Data can be graphed by how often they appear or occur

Front 20

Histogram

Back 20

• The frequency distribution of repeated measurement should have a bell shaped most should fall within the center

Front 21

Meanm Median, Mode

Back 21

In a perfect distribution they should be the same

Front 22

Mean

Back 22

add all the numbers divided by the number of data

Front 23

Median

Back 23

Middle number

Even set of numbers divide by 2

Odd set of number (n+1)/2

Not effected by extreme values

Front 24

Mode

Back 24

the number, or numbers, that occur the most frequently

Some data is bimodal, sometimes there is no modal number

not effected by extreme values

Front 25

Measures of variability- Gaussian distribution

Back 25

• If data is perfectly distributed, a histogram of the data appears as a bell shape
• Central tendencies: mean, median, mode
  • Variability: variance, SD, % coefficient of variation

Front 26

Measures of variability-Standard deviation and variance

Back 26

Front 27

Coefficient of Variation

Back 27

The smaller the CV the better the precision

Front 28

Shift from the mean

Back 28

systematic error

Can fix

Front 29

5 stages of QC

Back 29

1. Preparation- what QC should I use
2. Planning- what can go wrong
3. Prevention- minimize errors
4. Practice- train employees, compliance
5. Performance- testing and eval

Front 30

Why test QC material

Back 30

• So we know our testing procedure is accurate and precise
• Ensure results are reliable, accurate, correct, trustworthy
• Is the instrument always right
  • Regulatory issue: CLIA '88

Front 31

QC material- Selection and USe

Back 31

• Anything that simulates a patient sample, with known analyte values
• QCs matrix must be similar to the patient specimen
  • Caution: QC material should be handled as if it contain bio-hazardous material

Front 32

Assayed control

Back 32

QC material that has been repeatedly analyzed by the manufacturer:

Includes means and SDs

Statistical data is supplied with the QC material

Expensive

Front 33

Unassayed controls

Back 33

have not been tested for concentration or variability

Front 34

For all controlled material

Back 34

The mean and the SD must be verified or established in your lab before use in your QC program

QC with different lot # may have different analyte concentrations

Front 35

New lot #

Back 35

New mean and SD must be established

Front 36

Minimum QC

Back 36

two levels per day for each analyte
Exception: more than 2 levels may be required, more often than once per day (blood gas)

Front 37

Frequency of QC testing

Back 37

Instrument Manufacturer sets

Experience dictates frequency of QC

Criticality of the analyte dictates frequency

Front 38

QC levels

Back 38

Normal: Ref range

Below: the normal/ref range

Above: the normal/ref range

Front 39

QC level 1

Back 39

usually in the "normal/ref range"

Front 40

QC level 2

Back 40

usually in the abnormal range ( some high some low)

Front 41

QC level 3

Back 41

Low abnormal, normal, and high abnormal- used in hematology for drug testing

Front 42

Evaluation of QC data

Back 42

Based on Gaussian Curve:

• Test QC material and evaluation QC data
  • QC rules are based on a normal Gaussian distribution of QC data: Mean=Median=Mode, presence of random error give the curve its shape

Front 43

Levey-Jennings Chart

Back 43

• Daily QC data is recorded on a Levey-Jennings chart
• L-J chart plots QC data over time
  • First the mean and SD are determined by collecting a minimum of 20 QC data for an analyte
    • The Gaussian distribution is calculated using the means and multiples of the SD

Front 44

Creating a Levey-Jennings Chart

Back 44

• turn gaussian curve on its side
• extend X and SD lines to the right
• Label the X-axis (days of the month) and Label the y-axis (units of concentration)
  • Label chart with: qc level, lot #, data, analyte name

Front 45

Analytic Errors

Back 45

• Proper distribution of QC data: 95% of the data within 2SD
• Random errors: creates curve
  • Systematic error: affects all results predictably (shift, trend)
    • RE+SE=TE

Front 46

Random Errors

Back 46

• Imprecision- occurs by chance
• Related to precision/imprecision
• Random error defines the SD and %CV
  • Routine maintenance and calibration will not reduce the occurrence of random QC erro

Front 47

Systemic Error

Back 47

• Represents a real-time shift or bias
• Affects all samples
  • affects accuracy
    • Routine maintenance and calibration can minimize and correct systemic error
    • Shift and trend

Front 48

Gaussian distribution

Back 48

• 1SD: 68.2%
• 2SD: 95.5%
• 3SD: 99.7

Front 49

Shift

Back 49

• A bias exist that causes all data to be higher or lower than is correct
  • Abrupt change- away from the mean; is immediately measurable

Front 50

Trend

Back 50

• The bias that exists becomes greater and greater, but slowly over time
  • Slow progression: data moving slowly away from the mean, either positive or negative direction; it takes time for its effect to be measurable

Front 51

Shift/Bias

Back 51

The QC data shifts/move abruptly away from the mean for number of days

Front 52

Trend

Back 52

QC data either decreases or increases consistently over a period of 7 days

Front 53

Westgard QC multi-rules

Back 53

IDs the presence of:

• True QC failures: true positive
• To reject false alarms: false positive
  • Direct the technologist to the correct response in resolving the Failed QC

Front 54

One (1) QC warning rule

Back 54

gets your attention

Front 55

Five (5) QC failure rules

Back 55

used to reject qc data: 5 rules are applied if 2 QC levels are used, other rules exist if 3 qc levels are used

Front 56

Why repeat QC with a 13s QC error

Back 56

• 13s error is RE
• Probability that the number is a true value
• If this number is truly a random error, where will the next fall
  • Repeat QC is easiest and is recommended

Front 57

Why troubleshoot QC with a 22s error

Back 57

• Repeating is a waste of time and money
• What is the probability that these 2 numbers represent the true value of QC
  • Probability proves these two do not represent the true value. A bias exists

Front 58

Troubleshoot QC Systemic error

Back 58

• Inspect L-J control chart and failed QC to determine type of error
• Relate type of error to possible cause
• consider factors in common w/ QC and on multi-test systems
• relate causes to recent change
  • verify the solution and document the remedy

Front 59

41s error

Back 59

• Four concurrent occurrences >1SD
• systemic error detector
• QC has failed
  • Repeating QC is a waste of time; troubleshoot

Front 60

10x failure rule

Back 60

• 10 consecutive control values lie on the same side of the mean
• systematic error detector
  • QC has failed; troubleshoot

Front 61

Look at QC rules when

Back 61

QC violates the 1-2s warning rule, look at QC data continually

Front 62

Common mistakes in use of QC

Back 62

• No reviewing all the QC results and QC rules before testing patient results
• Don't just repeat and repeat the same failed QC material hoping for a better result. Have a plan
  • Don't mix up QC rules with real results