What is TQM? What does it consist of (w/ brief explanation of each)?
- Total Quality Management: management process for continuous quality improvement
- Q planning/processes: processes are designed (or selected) to be error "minimal" ("preventative")
- Q control: On-going monitoring of testing process for accuracy ("right now")
- Q assessment: review of system to determine system problems ("overall")
- Q improvement: re-design of processes using findings of Q assessment ("reaction/prevent")
Highest error frequency times for clinical care? Lowest?
- Highest: test choice, result use
- Lowest: analytical
What is method validation and what does it consist of?
- Verification/Establishment of the analytical performance of a method
- consists of - Accuracy, Precision, Reportable range, Reference intervals
What must be completed before reporting patient results for an FDA-cleared test system? An uncleared test system?
- Cleared: demonstrate that it is comparable to mfgr established specs (Accuracy, Precision, Reportable range)
- Verify that the mfgr reference values are appropriate for your pt pop
- Uncleared: determine Accuracy, Precision, Analytical sensitivity, Analytical specificity, Reportable range, Reference intervals, Any other characteristic required for test performance
What were the key issues that CLIA-88 introduced?
- Test complexity (mod, high, waived, PPM)
- Personnel Standards
What is "bias"?
- Another term for systematic error
- Could refer to proportional or constant variation
How can one assess the accuracy of a test method?
- Calibrators: referenced to the definitive method
- External Standards: NIST, CAP, pure materials
- Recovery experiment: assess matrix effects
What are the three method comparisons? Give details about the method most used for CLS.
- Definitive method: "gold standard"
- Reference method: traceable to definitive method
- Comparable method: compare to another lab (or another assay within your own lab)
- Analyze both methods over 5-20 days and plot on scattergram (X=reference, Y=candidate)
- Visually evaluate data
- Calculate linear regression, r, and Sy.x
- *note - proportional error affects slope, constant error affects intercept
Define correlation coefficient, standard error of estimate, coefficient of variation,
- Correlation coefficient (r): is there a relationship? (1=perfect)
- Std Error... (Sy.x): basically a stddev around a regression line (95% of points within ±2 Sy.x)
- Coefficient of var (C.V.): relative standard deviation ()
- puts the stddev raw # into context
- *all are used during method comparison
What is a Gaussian Distribution of data?
- "bell shaped curve"
- ~2/3 data within ± 1SD
- ~95% data within ± 2SD
- 99.7% data within ± 3SD
Repeatability vs reproducability
- Both are measurements of precision
- Repeatability: successive results with no changes (reagent, calibration, analyst, etc)
- Reproducability: similar results after changes are made
What is the F-test and how do we use it?
- Statistical test to compare variance (var=stdev2)
- F = var1/var2 where var1>var2 (F is always >1)
- Compare to a table to determine if variances are unequal
- Used during precision testing (method comparison)
Define analytical measuring range
AMR: range that can be directly reported by instrument without dilution
How do you determine LOD/LOQ? What is functional sensitivity?
- run sample w/o analyte 10-20 times, calc mean and SD
- Limit of detection (LOD): mean + 3sd
- Limit of quantitation (LOQ): mean + 10sd
- *results below LOD/LOQ reported as <nn
- Functional sensitivity: lowest concentration that can be measured with C.V. of 20% (or 10%)
Potential interferents w/ types of interference and if they typically result in pos or neg results. How is this tested?
- Hemolysis: spectral+, chemical-, addition+(artificial increase)
- Turbidity: light scatter+/-, volume displacement-
- Bilirubin: spectral+, chemical-
- Interference testing is similar to accuracy recovery experiment (compare same sample w/ and w/o interferent)
Determining reference interval CLSI C28
- Very involved
- 120 subjects per partition (age, sex, etc when appropriate)
- includes questionnaire
- Transference more common (verify validity of existing range to your assay and population)
- 20-60 pts
Bayesian Statistics - Sensitivity, Specificity, Predictive value+, Predictive value-, Efficiency, Prevalence
- Sensitivity: fraction of pts w/ disease detected
- Specificity: fraction of pts w/o disease that test neg
- PV+: proportion of pos tests that represent disease
- PV-: proportion of neg tests that represent lack of disease
- Efficiency: proportion of tests that correctly classify pt
- Prevalence: proportion of tested popn with disease (TP+FN/TP+FP+TN+FN)
What are ROC plots?
- Receiver Operating Characteristic plots
- Plot true negatives (TN) (1-specificity) vs true positives (TP) (sensitivity) for varying cutoff values
- the point closest to the upper left corner is best cutoff (maximum efficiency)
- Area under curve of 1 (no lines visible) is perfect
- Area under curve of .5 (y=x) is 50/50
Assayed vs unassayed QC
- Assayed: mfgr provides values (conc, var)
- Unassayed: user determined values
Levy-Jennings method and its flaws
- Run controls 20 times (during appropriate times)
- Calc mean and SD, create chart
- Plot controls on chart and reject if outside limit (2SD)
- Problems: False rejection (4.5% 1 ctrl, 9% 2 ctrls), No warning capability, No ID of error TYPE
How do various types of errors affect gausian distribution of a control? What are these types of errors typically related to?
- random error: widening of curve
- instrument problems (pipetting, mixing, washing, electronics, etc)
- systematic error (shift): abrupt move in running mean
- physical event (maint, rgt change, calibration, etc)
- systematic error (trend): gradual, continuous change in running mean
- degradation of component (QC material, aging lamp, reagant, etc)
- mixed errors: systematic type error that occurs randomly (leaking pipetter, sample evaporation)
Describe each of the Westgard Multi-rules (including alternative rules)
- 1-2s: one control exceeds 2SD limit.
- Warning rule, failure does not reject assay, initiates determination of other rule set
- 1-3s: one control excees 3SD limit.
- Implies random error (only one point)
- 2-2s: two simultaneous (or consecutive) values exceed the SAME limit of 2SD (both + or both -)
- Implies systematic error
- *can be across run or within run
- R-4s: range between two observations within a run exceeds 4SD
- Implies random error
- 4-1s: four consecutive observations exceed the same control limit (+1SD or -1SD)
- Implies systematic error
- *can be across run (last 4) or within run (last 2 for 2 ctls)
- 10x: ten consecutive observations fall on the same side of the mean
- Implies long-standing systematic error
- *can be across run (last 10) or within run (last 5 for 2 ctls)
- across controls = within run = both controls
- within controls = across run = one control
- alternative rules: for assays with 3 control materials
- 9x or 12x rule
- 2 of 3: 2 out of last 3 ctls exceed same 2SD limit
- can be within or across runs
What are the modified Westgard multi-rules? What are they for?
- Use 1-3s, 2-2s, and R-4s rules to reject run
- Use 4-1s and 10x as warnings for preventative maintenance (accept run)
- These tests are statistically relevant, but may not be medically relevant
- For tests with long intervals between calibrations
What are a few alternate approaches to the Westgard multi-rules?
- Westgard EZ rules: software that uses your inputted CV and quality goals to determine the best rules
- examples include 1-5s, etc (personalized)
- Multi-stage QC: different rules based on likelihood of error
- Start-up (any changes) - more sensitive rules (more likelihood of error)
- Monitoring (during run) - single limit rules (less likelihood of error)
What is the appropriate response (steps) to QC failure
- 1. Determine type of error (R or S)
- 2. Relate type of error to likely causes
- *Consider factors in common on multi-test systems
- 3. Relate causes to any recent changes/actions
- 4. Verify solution. Document remedy.
- *NOTE - repeating controls is statistically likely to bring them into QC range, even if the true reading is outside of the acceptable value (doubled SD, shifted SD)
Three types of External Quality Control w/ brief explanation
- Internal QC: daily assay of control materials
- Peer comparison: send QC results to central body (vendor?) for comparison with others
- Proficiency testing: assay of unknown samples provided by external body. Mandated.
- must meet peer-group consensus +/- abs value, percent value, or statistical range (depends on the analyte)
- 3 sets of 5 samples per year. 4/5 is passing. Must pass 2/3 consecutive.
- *note - if test isn't in CLIA you must 'verify accuracy' twice per year (usually send to reference lab)
Give 5 examples of patient-based QC w/ detail
- Serum indices: spectro est of hemolysis, icterus, and lipemia
- Test inter-relationships: T. protein always > albumin, etc
- Physiological limits: "absurd values"
- may indicate clot, bubble, or IV dilution
- Delta checks: comparison of current to previous results (absolute, relative, or rate)
- Insensitive, many false positives
- Average of normals: monitor running avg of selected patient results (in real time)
- equivalent to X bar on DxH
What are the physiological limits (absurd values) for the following analytes: Albumin, bili, creatinine, CK, cholesterol, Ca + K
- Albumin: <1.5 or >6.0 mg/dL
- Bilirulin: <0.2 mg/dL
- Creatinine: < 0.3 mg/dL
- CK: <5 U/L
- Cholesterol: <50 mg/dL
- Ca + K: Ca <2.0 mg/dL, K >10.0mmol/L (EDTA)
What are the 3 ways that IV Fluid might interfere with the values for various chemistry tests?
- Normal saline (0.9% NaCl): increased Na and Cl
- Dextrose 5% in water (D5W): Glucose 5,000mg/dL (absurd value)
- Lactated Ringer's: lactate 28mmol/L (absurd value)