Work

Gaussian Versus GAMLSS Norms for the Trail Making Test: Implications for Cognitive Impairment Classification

Year: 2026
Type: article
Study design: Observational Study
Abstract: Objective. To determine whether distributional assumptions in Trail Making Test (TMT) norms affect percentile calibration, impairment classification, and discrimination between cognitively normal adults and individuals with subjective cognitive decline (SCD) or mild cognitive impairment (MCI). Method: The sample comprised 587 adults aged 55 years or older: 334 controls, 149 with SCD, and 104 with amnestic MCI. Gaussian regression and generalized additive models for location, scale, and shape (GAMLSS) were compared for TMT-A, TMT-B, and their difference (B-A). Norms were estimated exclusively in controls. Repeated five-fold cross-validation assessed predictive performance, calibration, and discrimination; bootstrap analyses evaluated percentile stability and paired area-under-the-curve (AUC) differences. Right-censoring was incorporated for TMT-B and B-A. Results: Variable-dispersion log-normal modeling was selected for TMT-A, whereas Box–Cox t (BCT) models were selected for TMT-B and B-A. Compared with Gaussian norms, GAMLSS reduced TMT-A P95 impairment classifications in SCD (21.5% to 12.1%) and increased TMT-B P90 classifications in MCI (63.5% to 73.1%). BCT improved TMT-B discrimination between controls and SCD (AUC .670 vs. .600; adjusted bootstrap ΔAUC confidence interval [.004, .145]). B-A discriminated controls from MCI less effectively than TMT-B (AUC .808 vs. .869; adjusted interval for ΔAUC [-.114, -.024]). Conclusions: Flexible norms improved distributional fit and changed impairment classifications, but discrimination gains were measure- and group-dependent. B-A offered no discriminative advantage over TMT-B in this sample.
Language: English
Field: Medicine
Open Access status: green
OpenAlex ID: W7236587328
FWCI:
Cites:
Cited by:
Related to: 10