A SIMPLE, GRAPHICAL-METHOD TO EVALUATE LABORATORY ASSAYS

Evaluation methods of laboratory assays often fail to predict the larg e, infrequent errors that are a major source of clinician complaints. We present a simple, graphical method to evaluate laboratory assays, w hich focuses on detecting large, infrequent errors. Our method, the fo lded empirical cumulative distribution plot or, more simply, mountain plot, is prepared by computing a percentile for each ranked difference between the new and reference method. To get a folded plot, one perfo rms the following subtraction for all percentiles over 50: percentile = 100 - percentile. Percentiles (y axis) are then plotted against diff erences or percent differences (x axis). The calculations and plots ar e simple enough to perform in a spreadsheet We also offer Windows base d software to perform all calculations and plots. The mountain plot co mpared to the difference plot focuses attention on two features of the data: the center and the tails. We prefer the mountain plot over othe r graphical techniques because: 1. It is easier to find the central 95 % of the data. 2. It is easier to estimate percentiles for large diffe rences (e.g., percentiles greater than 95%). 3. Unlike a histogram, th e plot shape is not a function of the intervals. 4. Comparing differen t distributions is easier. 5. The plot is easier to interpret than a s tandard empirical cumulative distribution plot. Difference and moutain plots each provide complementary perspectives on the data. We recomme nd both plots. This method can also be used with data from a wide vari ety of other applications, such as clinical trials and quality control .