Robust decision procedures

Belton and Stewart (2002, p. 239) criticize that with the ELECTRE methods it is not possible to perform sensitivity or robustness analyses on the weights and the thresholds in an automated or interactive way. Also, they point out that weights and thresholds have no clear interpretation in terms of the decision maker s preferences and thus are often modified in an ad-hoc manner to obtain adequate outranking relations (Roy (1990b, pp. 176-179) discusses how to assign numerical values to these parameters). Additionally, Belton and Stewart (2002, pp. 246-247) criticize that the aggregation procedure of ELECTRE II and III is difficult to communicate and that with ELECTRE III counterintuitive results and rank reversal may occur if an alternative is added or removed. 

Method development to provide robust analyte-specific procedures from complex biomatrices has received significant attention. Modular on-line and off-line approaches have been reported that focus on a systematic and integrated approach to select optimal extraction and separation chemistries [27]. These approaches have automated data analysis, whereby the feedback loop from experimentation to decision can be significantly reduced [18]. Wells has comprehensively discussed the use of automated strategies for the rapid determination of optimal extraction chemistries to maximize analyte recovery and selectivity [28]. The book includes up-to-date information on available robotic platforms for automated sample preparation and comprehensive information on labware. 

Assay validation characterizes the assay performance so that the significance of the measured assay values obtained is readily understood [75]. Test methods should be validated when important decisions are to be based on the data generated [30]. Thus, the extent of method validation depends on the stage of clinical supply manufacture. The key elements of assay validation for a method are to establish reliability, the intra- and interlaboratory test variation, and relevance, the meaning of the results for a specific purpose [19]. The robustness of an analytical procedure (according to the ICH-Validation of Analytical Methods, 1993) is its measured capacity to be unaffected by small variations in controlled parameters and reliability under normal usage [35]. 

A sensitivity analysis can be carried on with the DSS (Lopes et al., 2009), which allows the evaluation of robustness of the model based on data and parameters appUed. The results shown that the procedure was robust enough to ensure the decision maker with the recommendation proposed. 

In the MOO (multiobjective optimization) method expected cost and robustness measure are simultaneously optimised. The robust model is based on the stochastic model (Eq. 10, with a = 0) having an additional objective of controlling the variability of performances of individual scenarios. The worst-case scenario is taken as the objective variable for the robustness measure and a decision-making procedure is applied to choose the best robust design alternative for the case study. 

Nevertheless, there are also examples of long linear syntheses, such as the 13-step synthesis of JNZ092, that are viable because they take advantage of the availability of well-elaborated, robust research/development procedures. The decision on a final strategy therefore needs to balance the value of already existing information against the gain and the effort required to develop alternative syntheses. 

---