Construct Validity Indicators from the ARMADA STUDY for the Tablet Computer‐Based NIH Toolbox Cognition Battery

AbstractBackgroundThe Advancing Reliable Measurement in Alzheimer’s Disease and Cognitive Aging (ARMADA) study was designed to validate and further develop the full tablet‐based NIH Toolbox Assessment of Neurological and Behavioral Function within an older, racially diverse sample. This ARMADA cohort study of the Toolbox Cognition Battery (NIHTB‐CB) reports construct validity, assessed through its two components: convergent (high correlations to related constructs) and discriminant (low correlations with dissimilar constructs) validity.MethodsParticipants (age range 65–99 years, 142 aged 85+) were consensus conference‐diagnosed with normal cognition (N = 367), amnestic mild cognitive impairment (N = 136), and Alzheimer’s disease (N = 68). NIHTB‐CB assesses domains of Fluid Executive, including Set Shifting (Dimensional Card Sort/DCCS), Inhibitory Control and Attention, (Flanker) and Working Memory (List Sort Working Memory/LSWM); Fluid Processing Speed (Pattern Comparison/PCPS), Fluid Episodic Memory (Picture Sequence Memory/PSM) and Crystalized Language (Oral Reading/ORR, Vocabulary/TPVT). Participants were also given the National Alzheimer’s Disease Coordinating Center Unified Data Set (UDS3) cognitive measures: Memory Composite (Benson Figure and Craft Stories Delayed Recall); Executive Composite (Trail Making, Semantic Fluency), Auditory Attention (Digit Span Forward), Figure Copy (Benson), and Language (Multi‐lingual Naming Test). Unadjusted, (uncorrected) standard scores were created based on the full Toolbox and UDS populations.ResultBased on the large sample size, most all correlations between NIHTB‐CD and UDS3 were significant, though patterns were clear. NIHTB‐CB/Fluid Executive tests had highest correlations with the related UDS3 Executive factor composite (r = 0.53‐0.60), approaching threshold for a “good” designation (>0.6), and “low” designation correlations (r<0.3) for Auditory Attention. PSM correlated highest with UDS Episodic Memory (r = 0.55), and lowest with Auditory Attention (r = 0.23). PCPS was highest with the Executive Composite (r = 0.53) and lowest with Language (r = 0.26). Crystalized NIHTB‐CB correlated highest with UDS Language (r = 0.41‐0.50) versus ORR with Memory (r = 0.22) and TPVT with Copy (r = 0.26).ConclusionResults demonstrated that the highest correlations between NIHTB‐CD and UDS3 measures were between related constructs and lowest correlations were between dissimilar constructs. Thus, convergent/discriminant expectations were supported across NIHTB‐CB measures, reinforcing NIHTB‐CB construct validity. Consistent results across diagnoses and other analytic approaches are also presented, supporting these findings.