Intra- and Inter-Session Reliability of Average- and Trial-Level Motor Evoked Potential Amplitudes

Background: Motor evoked potential (MEP) amplitude elicited by single-pulse transcranial magnetic stimulation (TMS) is widely used as an indirect index of corticospinal excitability (CSE). Averaged MEP amplitudes are commonly used to improve measurement stability, but averaging may obscure some trial-level information across the recording sequence. As trial-level and hierarchical modelling approaches become increasingly common, it remains important to determine how averaged and trial-level MEP amplitudes should be interpreted as measures of CSE. Objective: This study compared the intra-session and inter-session reliability of averaged and trial-level MEP amplitudes, with exploratory block analyses used to examine whether trial-order structure was obscured by averaging. Methods: Twelve healthy adults completed two sessions with three CSE assessments at T1, T2, and T3 (T1 and T2 20 min apart; T3 at least 48 h later). At each time point, 25 single-pulse MEPs were elicited at 120% resting motor threshold and recorded from the right first dorsal interosseous muscle. Reliability was assessed using intraclass correlation coefficients (ICC), coefficients of variation (CV), minimal detectable change at the 95% confidence level (MDC95), Bland–Altman analysis, linear mixed-effects models, exploratory sequential block analyses, and Bayesian variance decomposition. Results: Average MEP amplitudes remained stable across T1, T2, and T3. Average-level reliability was good across all timepoints (ICC = 0.899) and excellent within session (T1 vs. T2: ICC = 0.992, CV = 4.935%, MDC95 = 0.191 mV). Inter-session average-level reliability remained good (ICC = 0.844–0.852). In contrast, trial-level analyses revealed substantial within-participant variability, with low ICCs, high CVs, and large MDC95 values. Bland–Altman analyses showed minimal mean bias but substantially wider limits of agreement for trial-level data. Exploratory sequential block analyses showed that, in the late block, T1 differs significantly from T3. Bayesian modelling indicated that residual trial-level variability exceeded between-subject variability, with a posterior ICC of 0.336. Conclusions: Averaging 25 MEPs provides a reliable estimate of corticospinal excitability, but it masks variability. Trial-level and exploratory block analyses suggest that this variability may contain meaningful temporal structure, supporting trial-level modelling as a complementary approach for validating and interpreting averaged MEP estimates.