Analytic and clinical performance of two compact cholesterol-testing devices

Several relatively inexpensive compact analyzers for measuring cholesterol are available for use outside of the clinical laboratory. We evaluated the analytic and clinical performance of total cholesterol assayed with the AccuMeter (ChemTrak) and the LDX (Cholestech). Accuracy of both devices was evaluated by collecting capillary and venous whole blood from 100 subjects and assaying for total cholesterol. Results were compared with the Centers for Disease Control standardized reference laboratory method. Mean percent bias, mean absolute percent bias, and percentage of subjects with total error above +/- 8.9% were calculated and results were compared with recommendations from National Cholesterol Education Program (NCEP) for total cholesterol measurements. Precision was evaluated by assay of three pooled serum samples with both devices in duplicate in two runs/day for 20 days. The CV for each serum pool for each device was calculated and compared with NCEP recommendations for precision for total cholesterol measurements. Results with the two devices were compared. The total cholesterol mean percent bias for capillary samples was 2.1% for the LDX and -1.0% for the AccuMeter (p<0.01), and for venous samples 1.6 and -2.0%, respectively (p<0.001). The mean absolute percent bias for capillary samples was 5.4 and 5.2%, respectively (p=0.29), and for venous samples was 5.0 and 5.7% (p=0.79). Each device had an excessive number (12-22%) of individual results that exceeded NCEP recommended total error for a single cholesterol measurement (+/- 8.9%). In the precision analysis the average CV from all three serum pools was 4.0% and 5.3% for the LDX and AccuMeter, respectively (p<0.05). Thus both devices failed to meet the NCEP recommendation for precision of 3% CV. They both provided total cholesterol results that correctly classified individual patients into appropriate risk groups 95% of the time or better if values within +/- 8.9% of NCEP cut points for risk classification were ignored. Both devices met the NCEP +/- 3% requirement for total cholesterol mean percent bias but did not meet the +/- 3% requirement for CV as a measure of precision. Because of the variability in results, both devices had excessive numbers of individual subjects with total cholesterol results greater than the recommended total error limit of +/- 8.9% difference from the standardized method. Despite variability in some individual results, the rate of clinical misclassifications for coronary heart disease risk was relatively low for both devices if results near the NCEP cut points were repeated.