Stress contributes to decreased physical and emotional wellbeing if not managed at an appropriate level. However, when it comes to clinical diagnostics, stress testing may be necessary to reveal the performance of your assays in today’s high throughput, fast-paced clinical testing environment. One example is the HgbA1c (HgbA1c) assay.

HgbA1c is recommended as a marker for diabetes diagnosis and monitoring (1). Compared to fasting glucose and glucose tolerance tests, HgbA1c has the advantage of convenience and freedom from acute perturbations (2). With the current focus on wellness and disease prevention, HgbA1c level is frequently a clinical performance measure and is incorporated in the physician quality reporting system. Therefore, it is a frequently ordered test in both primary and specialty care settings.

Through the efforts of NGSP and IFCC, analytical methods for HgbA1c are generally well standardized and characterized. Vendors certify their methods through NGSP to document traceability to the DCCT reference method. This is usually achieved by running result comparison of a low number (usually 10-20) of fresh blood samples. No significant bias against DCCT method supports traceability. Downstream in a clinical laboratory, before assay go-live, typically 20-30 samples are used to perform comparison studies as part of assay verification. Does this approach give us enough confidence in the real clinical performance of the assay?

Recently, we had the opportunity to conduct wellness testing, including plasma lipid panel and whole blood HgbA1c for an employer. In a period of 2 months, 200-700 samples/day were analyzed for the above tests. Because of the high volume, samples were batched and run during low clinical volume hours. We had a medium throughput HPLC instrument for routine clinical testing before the endeavor. In order to accommodate the large volume, we validated the HgbA1c immunoturbidimetric assay on our high throughput automation line. Besides other validation studies, method comparison was done by running 25 samples using the immunoassay and the HPLC method. Good correlation and no apparent bias was seen in this comparison study. Once wellness testing commenced, we started to get spotted feedback that the HgbA1c level was running higher than what it should be. We carried out another correlation study between the immunoassay and the HPLC during this period, and still did not observe statistically significant bias.

To troubleshoot, we pulled all the data and plotted the overall population distribution as well as daily means and medians. We observed a shift of HgbA1c to the higher values compared to the NHANES population distribution. We also identified “peaks and valleys” of daily means/medians and QCs that corresponded to maintenance change of cuvettes on the automation line. This suggested absorbance errors. We did a stress testing of the immunoassay by running 200 batched HgbA1c samples, and repeating the first 20 samples after the 200 batch. A higher bias of 15-20% was observed from the stress testing. This bias was not eliminated by implementing cuvette washing program after each HgbA1c sample, for the specific generation of the immunoassay. The bias was evident after running a batch of as low as 50 samples. We hypothesized this bias are caused by build-up of blood cell fragments in the cuvettes when continuous whole blood samples are run through the system. Cell fragment deposition and coating of the cuvettes caused the turbidity measurement to go up artificially. Feedback was given to the vendor, who subsequently released a second generation of the assay that eliminated most bias upon stress testing.

Lessons learned: always try to simulate your real clinical testing scenario and drill/pilot/stress test your assay before going live. After going live, monitor your QC and patient moving average or daily mean/median closely.

References:

  1. International Expert Committee. International Expert Committee report on the role of the A1C assay in the diagnosis of diabetes. Diabetes Care 2009; 32: 1327–1334.
  2. Bonora E. and Tuomilehto J. The Pros and Cons of Diagnosing Diabetes with A1c. Diabetes Care 2011; 34 Supplement 2: S184-S190