Background Extinction-based exposure is effective but relapse remains common, partly because timing recruits distinct defensive (≤ 400 ms) and expectancy (≥ 1000 ms) systems. Immersive Virtual-Reality Exposure Therapy (VRET) allows precise timing control, yet it is unclear whether timing effects observed in VR can inform personalized exposure. Methods Eighty‑one healthy adults completed acquisition and returned 24 hours later for extinction in a head‑mounted immersive VR paradigm. We manipulated CS–probe lead intervals (100, 200, 1000, 4000 ms) and predictability (predictable vs. unpredictable) during acquisition; extinction used the same leads without shocks. Eyeblink startle (EMG) from a final sample of N = 81 acquisition and N = 67 extinction files was analyzed using linear mixed‑effects models to test timing × predictability × phase interactions. Exploratory segmented fits probed nonlinearity. Post hoc, we used nested cross‑validated elastic‑net and a shallow decision tree to derive interpretable rules; clinical utility was assessed via decision‑curve analysis using out‑of‑fold predictions. Sensitivity analyses varied outcome scale, random‑effects structure, inclusion of 100‑ms trials, and phase completeness. Results In the predictable group, CS+ > CS− discrimination was present at 1000 ms (estimate = 0. 212, Holm‑adjusted p =. 0057) and 4000 ms (estimate = 1. 227, p <. 0001), but not at 100 or 200 ms. Cross‑phase extinction (formerCS+ vs. acquisition CS+) showed maximal reductions at 4000 ms (EXT − ACQ ≈ −196. 24 arbitrary units; Holm‑adjusted p <. 0001), with earlier leads non‑significant. In the unpredictable group, phase‑related attenuation was prominent at 200 ms and minimal at ≥ 1000 ms, consistent with non‑associative effects. Exploratory fits suggested a breakpoint below 1000 ms (≈ 300–400 ms), with sensitivity estimates spanning 500–1800 ms. A simple scorecard combining long‑lead discrimination, habituation slope, and short‑lead reactivity stratified extinction outcomes. Conclusions Immersive VR reproduces the dual‑window timing signature of fear learning. Extinction benefits from probe placement in expectancy‑supported windows (≥ ~1000 ms), whereas early windows remain relatively resistant. An interpretable rule that combines long‑lead discrimination, baseline habituation, and short‑lead reactivity offers a prototype for timing‑personalized exposure therapy.
Laws et al. (Sat,) studied this question.