Approaches to Modeling

The multimarker strategies described in the preceding sections have been based entirely on either dichotomized or categorical (quartiles) interpretation of biomarker results. For biomarkers with graded linear or even nonlinear relationships with risk, consideration of the absolute concentration is likely to offer incremental information compared to a simple categorization as "positive" or "negative." However, interpretation of multiple biomarker results together as continuous variables substantially increases the complexity of reporting to the clinician and may fail to integrate the information in a manner that provides a useful overall assessment. For example, should the biomarker results of a patient with a BNP of 240 pg/mL, cTnI of 0.5 ng/mL, and hsCRP of 0.6 mg/L be reported exactly as such, or as BNP positive/cTnI positive/CRP negative, or as a multimarker score equal to 2? More sophisticated modeling approaches, such as based on neural network analysis, may be employed to transform the absolute concentrations into a single composite score. Such an approach has the advantage of being simple for the clinician to implement but

Fig. 10. Risk of death or MI at 6 mo stratified by cTnT (threshold: 0.01 pg/L), sCD40L (threshold:

5 pg/L), and MPO (threshold: 350 pg/L) among patients enrolled in c7E3 Anti-Platelet Therapy in

Unstable Refractory Angina Trial and allocated to placebo (n = 547). (Data are from ref. 40.)

Fig. 10. Risk of death or MI at 6 mo stratified by cTnT (threshold: 0.01 pg/L), sCD40L (threshold:

5 pg/L), and MPO (threshold: 350 pg/L) among patients enrolled in c7E3 Anti-Platelet Therapy in

Unstable Refractory Angina Trial and allocated to placebo (n = 547). (Data are from ref. 40.)

Table 2

Challenges in Developing a Multimarker Strategy

• It requires adequate clinical and analytic validation of each individual biomarker, including determination of appropriate sample handling, analytic performance, biological variability, and clinical decision limits.

• Relationship between the biomarker and the risk of specific clinical events may differ; that is, not all biomarkers are associated with the same clinical events with equal strength.

• Dichotomization of biomarker results may be overly simple and even misleading with respect to the magnitude of risk.

• Complex modeling is likely to hinder clinical application, whereas an overly simplified formulation may limit the validity and discriminatory capacity of the strategy.

• Evidence regarding the implications for therapy is critical to guiding clinical implementation.

may not be sufficiently flexible to provide the clinician with the information necessary to make decisions regarding therapy (see the next section).

Was this article helpful?

0 0

Post a comment