Affiliations: [a] Department of Neurology, Besançon University Hospital, Besançon, France | [b] Department of Clinical Psychiatry, Besançon University Hospital, Besançon, France | [c] Rapid-Fr network (Regional Network for Diagnostic Aid and Management of Patients with Cognitive Impairment – Department of Franche-Comté), Besançon, France | [d] Clinical Methodology Center, Besançon University Hospital, Besançon. France | [e] EA 481 Laboratory of Neurosciences, University of Franche-Comté, Besançon, France | [f] CIC-IT Inserm, Besançon University Hospital, Besançon, France | [g] Chrono-environnement Laboratory, UMR6249 CNRS, University of Franche-Comté, Besançon, France
Correspondence:
[*]
Correspondence to: Gilles Chopard, PhD, Centre Hospitalier Universitaire, consultation de psychiatrie, MEMCP, niveau 1, boulevard Fleming, 25030 Besançon Cedex, France. Tel.: +33 03 81 21 90 19; Fax: +33 03 81 21 88 17; [email protected]
Abstract:
Background: A single cutoff is widely used to screen amnestic mild cognitive impairment (aMCI). However, results of screening test performance are never adjusted for spectrum effect and spectrum bias.
Objectives: To assess the potential impact of spectrum effect and spectrum bias on screening test performance and clinical decision.
Methods: The ability of the combination of Memory Impairment Screen (MIS), the Isaacs Set Test (IST), and the Mini-Mental State Examination (MMSE) to distinguish aMCI (n = 3,330) from patients with subjective cognitive complaints (SCC) (n = 1,522) was investigated across a wide range of age and educational backgrounds. The spectrum effect was defined as the variation of the sensitivity and/or the specificity across different subgroups. A spectrum bias was highlighted if the likelihood ratio (LR) observed in a subgroup of subjects statistically differed from the LR observed in the overall sample.
Results: For the MIS-IST pairing, the overall sensitivity and specificity were equal to 72.5% and 75.2% , the positive LR (LR+) and the negative LR (LR–) were equal to 2.91 and 0.37, respectively. Across the different age-education subgroups, the sensitivities ranged from 43.7% to 92.5% and specificities from 39.3% to 95.2%. LR+ and LR– ranged from 1.51 to 9.10 and 0.13 to 0.59, respectively. A statistically significant spectrum bias was found in some subgroups and may result in differences between the post-test probabilities. Similar results were also found for the MMSE.
Conclusion: These findings could potentially affect the clinician’s decision with a possible greater impact in elderly patients with a lower educational level.
