THE “LINKED EVIDENCE APPROACH” TO ASSESS MEDICAL TESTS: A CRITICAL ANALYSIS

Abstract

OBJECTIVES: A linked evidence approach (LEA) is the synthesis of systematically acquired evidence on the accuracy of a medical test, its impact on clinical decision making and the effectiveness of consequent treatment options. We aimed to assess the practical utility of this methodology and to develop a decision framework to guide its use. METHODS: As Australia has lengthy experience with LEA, we reviewed health technology assessment (HTA) reports informing reimbursement decisions by the Medical Services Advisory Committee (August 2005 to March 2012). Eligibility was determined according to predetermined criteria and data were extracted on test characteristics, evaluation methodologies, and reported difficulties. Fifty percent of the evidence-base was independently analyzed by a second reviewer. RESULTS: Evaluations of medical tests for diagnostic (62 percent), staging (27 percent), and screening (6 percent) purposes were available for eighty-nine different clinical indications. Ninety-six percent of the evaluations used either the full LEA methodology or an abridged version (where evidence is linked through to management changes but not patient outcomes). Sixty-one percent had the full evidence linkage. Twenty-five percent of test evaluations were considered problematic; all involving LEA (n = 22). Problems included: determining test accuracy with an imperfect reference standard (41 percent); assessing likely treatment effectiveness in test positive patients when the new test is more accurate than the comparator (18 percent); and determining probable health benefits in those symptomatic patients ruled out using the test (13 percent). A decision framework was formulated to address these problems. CONCLUSIONS: LEA is useful for evaluating medical tests but a stepped approach should be followed to determine what evidence is required for the synthesis.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 56 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.