Objective:

To predict whether gait parameters from markerless video-based gait analysis, align with clinician-determined drain-trial response in individuals being worked up for iNPH

Background:

iNPH often shows gait improvement after a temporary CSF drain trial used to select shunt candidates. The 10-meter walk (10MWT) is practical and captures meaningful change, yet the trial’s negative predictive value is limited. In routine care, clinicians time the 10MWT but rarely leverage richer quantitative features (e.g., stride length/width, stride-time variability/asymmetry). We hypothesized that advanced gait metrics would better align with a clinician-determined drain response and help connect objective trial gains with post-operative outcomes. 

Design/Methods:

Single-center cohort of 28 iNPH patients with pre- and post-drain 10MWT videos were analyzed using an automated markerless pose-estimation pipeline. Primary endpoint: gait speed (m/s); responders specified as ≥20% relative increase. We derived step count, cadence, step width, and stride/step variability/asymmetry. MRIs were scored via iNPH Radscale. Within-subject change used paired tests and standardized response mean. A baseline-speed threshold was estimated by Youden’s index. The comparator was a multidisciplinary committee-determined clinical drain response. 

Results:

Mean gait speed increased from 0.544 to 0.666 m/s; mean within-subject change +0.121 m/s (+32.8%) (p<0.003). Responders: 13/28 (46.4%) by Δ≥0.10 m/s and 11/28 (39.3%) by ≥20%. Baseline pre-drain speed alone showed rule-in utility (high specificity, limited sensitivity). Advanced gait metrics changed in the expected direction (e.g., longer strides, higher cadence), stride-time variability/asymmetry effects were heterogeneous. Imaging markers and MoCA showed non-significant associations with percent speed change.   

Conclusions:

Video based 10MWT detected clinically relevant post-drain gains, with ~38% meeting the ≥20% responder criterion. Neither baseline speed, MoCA category, nor Radscale components reliably identified responders. Findings support feasibility of automated video analytics and the ≥20% benchmark, while highlighting the need for larger, prospective studies to refine simple prediction rules, select differentiating gait variables, and link drain trial changes to patient-centered post-shunt outcomes.