07agm PT ccil barry loescher - cala.ca · Findings •CAEAL Old –Scores of 5,4,2 or 0 were...
Transcript of 07agm PT ccil barry loescher - cala.ca · Findings •CAEAL Old –Scores of 5,4,2 or 0 were...
Background
• CCIL strongly endorses the requirement foracceptable PT performance to maintain accreditation
Background
• ISO / IEC 17025 and the accreditors, SCC,CAEAL andMDDEP require PT participation
• No consensus Internationally on frequency
• Majority of laboratories
– Once every 4 years (minimum)
– 2 X per year
Background
• Under active consideration internationally
• APLAC has a committee addressing the issue
– Leaning toward a 1 X 2 sample approach
– SCC and CAEAL attend all meetings
Background
• PT providers must be accredited to ISO Guide 43(Can-P-1594:2001)
• CAEAL provides the lion’s share of Canadian PTsamples
Background
• Historically CAEAL has distributed 2 sets of four PTsamples / yr for the parameters they supply.
• Require a minimum of two samples / yr for otheraccredited parameters.
Background
• PT costs are the major component of accreditationcosts
$4,400$2.900Small Lab
$30,000$5,500Large Lab
PTAssessment
Background
• PT samples are an essential part of a Lab QualityProgram
• Can we reduce the number of samples and maintainthe integrity of the program?
• CCIL decided to evaluate the effectiveness of thecurrent program
The Study
• Member labs were ask to submit 10 sets of failing ormarginal data (Score 70% to 80%) plus the data forthe same parameter(s) in the subsequent study
• 232 data sets were submitted from 21 labs
The Study
• In response to customer feedback and workshopinput, CAEAL recently developed and implemented anew scoring system
– New vs Old was evaluated
– CAEAL New vs SCC protocol was also evaluated
The Study
• Items evaluated
– Range of PT sample sets
– How often do all four samples trend in the samedirection?
The Study
• Items evaluated
– Effect of reduced sample sets
• Recalculated scores using samples
– 1 & 3, 2 &4
– 1, 3
Disclaimer
• The conclusions discussed below are based on alimited data set
• Analysis of larger data sets may lead to differentconclusions.
Findings
• CAEAL Old– Scores of 5,4,2 or 0 were assigned to each sample
on the following basis:– < 1 sd from the mean 5– >1 < 2 sd from the mean 4– >2 < 3 sd from the mean 2– >3 sd from the mean 0– Scores from 4 samples summed X 5– Scores ≥ 70 were acceptable
Findings
• CAEAL New– Calculate Z score = │(xi – X) │/ sd– Where– xi is the reported result– X is the assigned value, usually the adjusted
mean– sd is the standard deviation of the adjusted mean
Findings
• CAEAL New– Z Score modified to account for low level data– Z score = │(xi – X) │/ √ ( sd2 + (RDL/3)2)– Z > 6.6 assign Z = 6.6– “<“ reported, RDL > assigned value• Z=2 (some exceptions)
– “<“ reported, RDL < assigned value• Use RDL value to calculate Z
Findings
• Reporting Low Level Data– 2 Approaches RDL = MDL, RDL >> MDL– Result < 5X MDL and other techniques more
sensitive• Greatly increased chance of failure
Findings
• CAEAL New– Calculate average Z Score– PT score (%) = 100 + (-15* avgZ).– PT Score ≥ 70% is acceptable
Findings
• CAEAL New vs Old
– 77% of the results improved
– Average 9.5 points
– 39 of 180 results went from fail to pass
• CAEAL New vs SCC
– CAEAL 102 failures
– SCC 79 failures
Findings
• Sample Ranges
– The average concentration range covered by thefour samples all parameters was 9
– Organics 6
– Wet Chemistry 11
– Metals 12
– 4 of 149 sets had a range > 50
Findings
• Sample Ranges
– Most current analytical techniques cover 2 ½ to4 ½ orders of magnitude
– In general the CAEAL PT samples do not coverthe calibration range
Findings
• Trends
– 79% of the data sets had all four samples on thesame side of the assigned value
– Confirmed by CAEAL analysis of a full years data
• PT > 80 48%
• PT 70 – 80 79%
• PT < 70 73%
Findings
• Reasons
– Samples normally analyzed together in a singlerun. Samples are homogeneous, relatively closein concentration
– Most sources of lab error produce consistent bias
• Inaccurate standards
• Calibration bias
• Drift
Findings
• Recurrent Problems
– 11% of data sets had problems on the subsequent set
– CAEAL 3 full studies
• Consecutive failures 23%
– i.e. a significant number of labs were unable to correct theirproblems
– Argues for at least a 2X / yr frequency
Findings CAEAL Protocol
8%5%Pass to Fail
7%13%Fail to Pass
3%2%Change > 20%
18%16%Change > 10%
Samples 2 & 4Samples 1 & 3Score ChangeCompared to4 Sample Set
Findings CAEAL Protocol
19%18%Pass to Fail
11%9%Fail to Pass
16%11%Change > 20%
43%33%Change > 10%
Sample 4Sample 2Score ChangeCompared to4 Sample Set
Findings SCC Protocol
79
4 Samples
12%12%
2727Fail to PassPass to Fail
5252Failures
Samples 2 & 4Samples 1 & 3
Findings SCC Protocol
79
4 Samples
21%20%
4845Fail to PassPass to Fail
3134Failures
Sample 4Sample 2
Findings
• Two sample sets correlate well with 4 sample sets
– 17% Pass to Fail or Fail to Pass
• These data represent about 15% of the total
– ≈ 10% failure ≈ 5% marginal
– Other 85% all pass
Findings
• Extrapolating to the full data set
– 2 Samples vs 4
CAEAL Protocol
– 97.5 % of labs, same result
SCC Protocol
– > 98% of labs, same result
Conclusions:
• Frequency should not be decreased
– Substantial %age of repeat problems
• Two sample sets provide adequate problem detection
– Similar pass / fail rates
Conclusions:
• Four sample sets do not provide enough additionalinformation to warrant the additional effort
– All 4 samples trend the same direction 79% of the time
– Average range < one order of magnitude
Current Status:
• Report sent to CAEAL and SCC
• Prompt response from both
• Additional database analysis by CAEAL
– Indicated a risk based approach might be viable
• Willing to entertain changes
– Provided the program isn’t weakened