Multirule procedure

The multirule procedure developed by Westgard and associates uses a series of control rules for interpreting control data. The probability for false rejections is kept low by selecting only those rules whose individual probabilities for false rejection are very low (0.01 or less). The probability for error detection is improved by selecting those rules that are particularly sensitive to random and systematic errors. The procedure requires a chart having lines for control limits drawn at the mean 1 s, 2 s, and 3 s, and is adapted to existing Levey-Jennings charts by the addition of one or two sets of control limits. 

The use of the multirule procedure is similar to the use of a Levey-Jennings chart, but the data interpretation is more structured. To use the multirule procedure, follow these steps ... 

An example application of the multirule procedure is shown in Figure 19-14, where the top chart is for a high-concentration control material and the bottom chart is for a low-concentration material. Table 19-2 summarizes the interpretation of the charted data, providing the run number, the accept and/or reject decision, control rules violated, and the type of error suspected based on the rule violations. It is important to note that the rule is applied only within a run, so that between-run systematic errors are not wrongly interpreted as random errors. However, the rule may be applied across materials, meaning that one of the observations can be on the low material and the other on the high material, as long as they are within the same run. On the other hand, note that the Izs, 4,j, and 10 rules can be applied across runs and materials. This effectively increases n and improves the error detection capabilities of the procedure. 

Comparison of the probability for error detection between the multirule procedure and the Levey-Jennings chart having 3s limits shows improved error detection for the multirule procedure. The rule improves the detection of random error and the 22s, 4is, and 10 rules improve the detection of systematic error. Elimination of the lO f rule does not cause much loss in error detection but does considerably reduce the amount of control data that must be inspected thus the simplification may malce the multirule procedure easier to use. The 4]s rule could possibly be elim-... 

When compared with the Levey-Jennings chart with 3s limits, the cusum procedure provides better detection of systematic errors but less sensitivity for random errors. Because of the low sensitivity to random errors, cusum should not be used alone, but rather should be included with a Levey-Jennings procedure on a combined chart or used as a separate chart along with a Levey-Jennings chart. Performance is similar to the multirule procedure for low ns, at least as far as detecting systematic errors. However, the multirule procedure is more sensitive to random errors because of the added rule. 

For this example where the observed imprecision is 2.0% and the observed inaccuracy is 1.0%, the four lines above tire operating point correspond to the top four QC procedures listed in the key on the right. All four will provide at least 90% detection of critical systematic errors, but then false rejection rates wiU vary from 0.03 to 0.18, or 3% to 18%. Appropriate choices would be the I2.5S single rule with n-4 or the multirule with n = 4 to keep false rejections low. 

---