Efficient use of resources for workload management in analytical laboratories is the need of the hour. Irrespective of their size, laboratories are working to reduce their operational costs and shorten turnaround times to the delight of their customers. In the current economic situation, it is harder to have a dedicated analyst for a single instrument. We have a system of sharing the workload in our laboratory; for example, an analyst whose primary responsibility is HPLC has secondary responsibility for FT-IR and GC and is cross-trained in those. This way, when demand is low for HPLC, he can work on other instruments. This helps in efficient utilization of equipment resources as well.
Cross-training the analyst is essential in order to ensure that the laboratory’s workload can be carried out without any hiccups and to enable the samples to move more smoothly through our workflow. It also gives an analyst the chance to use characterization techniques that could be complementary to those he or she has experience with. Cross-training is also advantageous to the laboratory in the sense that the primary analyst can go on a vacation without worrying about the pending samples on his desk when he returns. However, the laboratory manager still has to ensure the quality of the data delivered.
The main concerns with respect to measurement variation are an analyst’s bias, training and instrument experience. Our chromatographic laboratory quality team decided to use Six Sigma statistical tools for measuring variation between analysts to effectively crosstrain spectroscopy analysts to handle chromatography. An exhaustive analysis of errors was derived from gage repeatability and reproducibility (“Gage R&R”) studies. The studies gave us an estimate of the precision to tolerance (P/T) ratio of the measurement.
Data from three analysts was used for the Gage R&R study. Analysts A and B were experienced, and analyst C was to be cross-trained in HPLC. Our customer requirements demanded that analyst-to-analyst variation be minimized. In this particular study, scope was restricted to measurement of errors arising from instrument operation and the analyst taking the measurements.
The training sets were taken from fiber additive samples in representative percentage concentrations. An initial set of results was obtained, using existing methods and conditions, with three samples of the additive prepared by experienced analyst B. Each analyst ran three samples of six injections each for a total of 18 runs, resulting in a total of 54 runs for each trial. So far, two sets of trials have been completed, which are mentioned as two phases in this article, and a Gage R&R study was conducted.
Initial results for the first phase are shown in Figure 1. Total variation is expressed as a percentage and consists of equipment variation, analyst variation, part-to-part variation and Gage R&R. Repeatability is the variation in measurements obtained with one gage in the same measuring environment when used several times by one analyst while measuring the identical characteristic on the same part. It is also commonly known as equipment variation. Reproducibility is the variation in the average of measurements made by different analysts using the same gage when measuring the identical characteristic on the same part. It is commonly known as appraiser variation.
Figure 1. Initial Gage R&R (ANOVA) studies (first phase) generated after injecting additive sample 18 times in HPLC by each analyst.
Initial Gage studies showed repeatability to be 82 percent, reproducibility among the analysts to be 31 percent and total gage to be a whopping 88 percent. The acceptable range for all three values should be <30 percent. These values for repeatability and reproducibility showed that total variation was due to the equipment/ method and, to some extent, the analyst (analyst C needs to be trained according to the SOP). Also, part-to-part variation was found to be 47 percent; ideally, part variation should be high to represent the full variation of the process. A rule of thumb is that total Gage R&R should be <10 percent to be acceptable; however, 10–30 percent may be acceptable based upon the risk of the application, cost of measurement device, cost of repair, etc.
The effective resolution of the measurement system can be determined by its ability to distinguish data within several data classifications or distinct categories from the observed process for a given application. The number of distinct categories in this initial phase of the Gage R&R study was found to be one, but it needs to be at least four or more for it to be a good gage. In Minitab 15™ (a software application used for conducting Six Sigma studies), 6.00 is the default for the study variation for a Gage R&R study. This is the Z value range that calculates a 99.73 percent potential study variation based on the calculated standard deviation of the variation seen in the parts chosen for the study.
In Gage R&R, the R-chart is a special range chart that tells us whether the measurement is within the statistical control and usually gives us an idea about repeatability. Here, operator/analyst C has an outlier, and this could be attributed to insufficient operator training. Operator/ analyst B is within acceptable range due to his experience with the instrument/method. The same information is also obtained from the box plot on the right, where measurements by operator are represented. The measurement variation of the second analyst is less in comparison with the third analyst. The X-bar chart by analyst gives the reproducibility portion of the Gage R&R. It helps us with the removal of special-cause variation. In the X-bar chart, only two points fall outside the limits, and so the measurement process is not adequate to detect part-topart variations; hence, the gage is not good.
Operator-sample interaction in the Gage R&R study also helps us understand reproducibility among the operators. Compared with the other operators, analyst C has some interaction with the samples. Analyst C is biased positively with the measurements and measures the samples higher than the other analysts, as seen from the operator-sample interactions.
Figure 2 shows a cause-and-effect diagram, otherwise known as a “fishbone” diagram, which allowed the HPLC quality team to systematically identify and display all the possible causes related to the bottlenecks in the quality of the data and turnaround time of the HPLC tests. The four main categories identified were machinery/equipment, people, methods and materials. Primary causes were shown to be the operator, method and environment. The detailed SOP included protocols for syringe cleaning and bubble elimination to improve reproducibility. Noise elimination due to the change in temperature was also addressed to reduce measurement variation.
Figure 2. “Fishbone” diagram with the four main categories identified included for tests in HPLC; machinery/equipment, people, methods and materials.
After the training of analyst C was completed, the data from the second-phase Gage R&R studies showed tremendous improvement compared to the first-phase data. Figure 3 shows Gage studies conducted for the second trial. Higher part-to-part variation (97 percent) was an indication that the samples are representative of full variation of the process. Improvement in precision was observed, as repeatability of the gage was found to be 15 percent. Exhaustive training imparted to analysts also had improved the reproducibility to 15 percent. Total Gage R&R was 22 percent, which is within the acceptable range. The number of distinct categories was found to be six, showing that this was a good gage. Overall, our gage was found to be good, with high part-to-part variation and acceptable repeatability and reproducibility. Improvement in gage can primarily be attributed to training of the operators using the detailed SOP created after the fishbone diagram was developed.
Figure 3. Second phase Gage R&R (ANOVA) studies generated after injecting additive sample 18 times in HPLC by each analyst after training in detailed SOP.
When we interpret the R-chart from the Gage R&R data, all the points are within limits (Upper Control Limit (UCL) and Lower Control Limit (LCL)), and so the measurement process is under control and is said to be consistent. In the X-bar chart, more than 50 percent of the points lie outside the UCL and LCL, hence the gage is good. Our analysts showed considerable improvement at the end of the study, as inferred from operatorsample interaction in the Gage R&R study. Very low operator-to-sample interaction confirms that analyst C is not biased while making the measurements as compared to first-phase measurements. This shows that analyst C has been trained well using the detailed SOP prepared after incorporating the syringe cleaning procedure and noise reduction precautions. Analyst C’s discussions with his manager and with experienced analyst B helped him achieve the cross-training required to maintain the quality of the data.
Thanks to Ravi Kelkar and the HPLC team in analytical science and technology at Aditya Birla Science & Technology Limited.