Computations of the Eight Basic Measures in Diagnostic Testing

Diagnostic testing concerning categorical or dichotomized variables is ubiquitous in many fields including, in
particular, the field of clinical or epidemiological testing. Typically, results are aggregated in two-by-two
contingency-table format, from which a surprisingly huge number of indicators or measures are obtained. In this
chapter, we study the eight most prominent such measures, using their medical context. Each of these measures
is given as a conditional probability as well as a quotient of certain natural frequencies. Despite its fundamental
theoretical importance, the conditional-probability interpretation does not seem to be appealing to medical
students and practitioners. This paper attempts a partial remedy of this situation by visually representing
conditional probability formulas first in terms of two-variable Karnaugh maps and later in terms of simplified
acyclic (Mason) Signal Flow Graph (SFGs), resembling those used in digital communications or DNA
replication. These graphs can be used, among other things, as parallels to trinomial graphs that function as a
generative model for the ternary problems of conditional probabilities, which were earlier envisioned by Pedro
Huerta and coworkers. The arithmetic or algebraic reading or solving of a typical conditional-probability
problem is facilitated and guided by embedding the problem on the SFG that parallels a trinomial graph.
Potential extensions of this work include utilization of more powerful features of SFGs, interrelations with
Bayesian Networks, and reformulation via Boolean-based probability methods.