Using Decision Tables To Verify The Logical Consistency And Completeness Of Clinical Guidelines

Using Decision Tables To Verify The Logical Consistency And Completeness Of Clinical Guidelines: Fever Without Source In Children Under Three.
Robert L Wears, MD, MS; Phyllis H Stenklyft, MD; Robert C Luten, MD
University of Florida Health Science Center Jacksonville, Department of Surgery, Division of Emergency Medicine Jacksonville, FL 32209

Objective: To demonstrate formal evaluation of clinical guidelines for consistency and completeness using the technique of decision tables.

Methods: Published and locally developed clinical guidelines for management of fever without source in children < 3 years old were verified by decision tables. Decision tables are a formal methodology developed in computer science to verify the completeness and logical consistency of large, complex rule sets commonly found in expert systems. This process has six steps:

1. Display conditions to be tested (eg,WBC>15,000?) in tabular form in an exhaustive and mutually exclusive manner.
2. Eliminate logically impossible combinations of conditions.
3. Translate guideline statements into rules of the form "if <conditions> then <actions>."
4. Assign actions to the remaining combinations of conditions based on these rules.
5. Simplify the table by removing irrelevant conditions and redundant actions.
6. Identify omissions (actions for which there is no condition set) and contradictions (identical conditions sets leading to one or more different action sets).
7. Subtables are frequently used to simplify the decision table formulation.

Results: The locally developed guideline had 192 possible unique condition sets; after elimination of impossibilities and redundancies, this was reduced to 14. One technical inconsistency and one significant area of incompleteness were discovered. The published guideline had 23,040 possible condition sets; this was greatly simplified by using subtables. (Eg, the subtable for infants 28 - 90 days of age collapsed from 288 to 34 condition sets). Again, several areas of technical inconsistency and incompleteness were identified.

Conclusions: Both published and locally developed clinical guidelines were incomplete and contained technical inconsistencies. Verification of internal consistency and comprehensiveness of guidelines by simple inspection is difficult in all but the simplest situations. Therefore, formal verification methods such as these should be used to ensure the logical integrity of clinical guidelines before they are recommended for general use. While formal methods cannot address the validity of guidelines, ensuring their logical coherence is a necessary (albeit insufficient) condition for a clinical guideline to be valid.