Human Information Processing, Health Information Technology, and Medical Outcomes

To the Editor: In their Commentary, Drs Duncan and Evens¹ are correct that before the wholesale adoption of health information technology, a profound re-evaluation of the designs of current systems is needed. However, their emphasis on explicit algorithms warrants refinement, and their dismissal of subjective input deserves reconsideration.

Medical research has been overwhelmingly about pathophysiology and has generated few answers to practice-based questions. Most practical decisions currently have little or no evidence to support an algorithm or guideline. Even if all of the research to sustain the necessary guidelines were affordable, however, there would be a more profound problem. Explicit algorithms presuppose that a singular target point exists. Yet optimal outcomes for individual patients require the reconciliation of multiple targets, including statistical evidence that may or may not apply; patient and system resources; patient, family, and professional preferences; and the individual's intersecting disease states. The balance of competing forces creates entire landscapes of possibilities, in which optimization is relative to a field of solutions rather than the linear migration toward a single global minimum.

Additionally, nuanced observations and subjective hunches are the very impetus that drives improvement of the target. Guidelines pull extreme behaviors toward a mean and compress the bell curve of performance. Although that improves poor functioning, it also stifles the innovation that yields high performance. It is not random behavior that places superior achievement ahead of the bell but the constant needling at discerned, if often mistaken, causality. In 1954, landmark work by Meehl² revealed the limitations of subjective judgment and clinical experience, but a meta-analysis of more recent studies has found that clinical experience can improve decision-making accuracy by nearly 13%, and the effect is consistent across venues.³

The formal mathematics of multi-objective optimization actually requires information from user judgment to isolate a final solution from the set of optimized possibilities.⁴ This need for input from outside the system is representative of Gödel's incompleteness, a very real limitation that any comprehensive system of explicit algorithms will always confront.

Health information technology can ultimately support the complexity of providing patient care, but only with a substantial departure from current reductionist approaches⁵ and a recognition of the complementary contributions of evidence and clinical judgment. Optimal care of the individual patient will indeed require new precision in information capture and pattern extraction, but even if the evidence were pursued to an N-of-1 scale, its application would remain incomplete without human judgment.