Sensor-derived heel pressure metrics capture reversible gait dysfunction beyond conventional gait measures in normal pressure hydrocephalus

Abstract

Background

The response to the cerebrospinal fluid (CSF) tap test helps determine the surgical eligibility of individuals with idiopathic normal pressure hydrocephalus (iNPH); however, conventional gait assessments have limited specificity and low negative predictive value. We investigated whether commercially available, low-cost sensor-embedded insoles could offer sensitive gait biomarkers for detecting changes in gait following the CSF tap test.

Methods

Seventy-seven individuals with iNPH were prospectively enrolled. Participants performed conventional gait assessments, including the 10-m walking test (10MWT), wearing insoles with pressure sensors and an accelerometer before and after the CSF tap test. Spatiotemporal and plantar pressure parameters were compared between individuals with iNPH and control participants without iNPH and between responders and non-responders to the CSF tap test, defined by subjective and objective symptom improvements. Decision curve analysis evaluated clinical utility.

Results

Individuals with iNPH exhibited significantly impaired gait and balance, reduced heel and toe pressures, and increased pressure variability compared with control participants. Responders (n = 48) demonstrated a significant post-tap increase in right heel pressure and decreased variability (p < 0.001 and p = 0.002, respectively), with sustained improvements at the 3-month follow-up in 19 participants who subsequently underwent surgical intervention. Decision curve analysis showed that models incorporating plantar pressure metrics in combination with the 10MWT yielded higher net benefit than the 10MWT model alone, with the greatest average Δnet benefit observed when multiple plantar pressure metrics were combined (+ 0.093).

Conclusion

Sensor-derived heel pressure metrics represent a digitally quantifiable biomarker that enhances decision-analytic evaluation of CSF tap test response and may improve early treatment stratification in iNPH.

Trial registration The project was retrospectively registered at ClinicalTrials.gov (NCT06722768) on December 9, 2024.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12984-026-02087-3.

Introduction

Idiopathic normal pressure hydrocephalus (iNPH) is one of the few potentially reversible neurodegenerative conditions in older adults, characterized by gait disturbance, cognitive decline, and urinary incontinence [1]. Shunt interventions are effective regardless of the specific surgical treatment [2–4]. However, heterogeneous treatment responses and procedure-related risks [1, 5–7], underscore the need for accurate preoperative stratification of surgical candidates. Quantitative tools to support such decision-making remain limited.

Symptom improvement following a cerebrospinal fluid (CSF) tap test is commonly used as a predictive indicator for shunt outcomes [8, 9]. Among the classic symptom triad, gait disturbance, characterized by short strides, shuffling, and instability during walking, is the earliest manifestation to respond to the CSF tap test [3]. Several quantitative measures, including the Timed Up and Go Test (TUGT) and the 10-m walking test (10MWT), are commonly used to evaluate gait improvement after CSF removal [10–12]. However, threshold values for clinically significant change vary across studies [3, 11–14], with a sensitivity of 58% (range: 26%–87%) and a specificity of 75% (33%–100%) [15]. Moreover, even patients with subjective improvement but no measurable change on conventional assessments after the CSF tap test demonstrated clinical benefit following shunt surgery [14, 16]. The negative predictive value has been reported to be as low as 20% in some cases [11], suggesting that current assessment methods may lack sufficient sensitivity to detect subtle changes. Recent retrospective work has suggested that simplified balance-based assessments may be more sensitive to change following a CSF tap test in iNPH [17]. However, some tasks proved challenging for approximately 20% of patients, thereby limiting the feasibility of such assessments in routine clinical practice.

Recent advances in wearable sensing technologies enable continuous and high-resolution quantification of human movement in real-world settings. Devices such as inertial measurement units and instrumented insoles allow for granular analysis of spatiotemporal parameters and plantar pressure dynamics [18], which may capture subtle features that are not detectable through conventional clinical tests [14, 19]. In our previous work, we demonstrated that affordable and easily deployable sensor-embedded insoles can capture distinguishable gait characteristics during standard clinical tests in people with gait disorders of diverse etiologies, including those with Parkinson’s disease and iNPH [20]. Furthermore, variability in plantar pressure has been associated with impaired motor automaticity and postural control, even in cases where conventional spatiotemporal parameters remain unchanged [21]. These findings suggest that sensor-derived metrics may serve as sensitive digital biomarkers, providing incremental value beyond traditional clinical assessments in identifying individuals with iNPH who are likely to benefit from CSF diversion procedures.

Here, we hypothesized that sensor-derived measures from commercial instrumented insoles can capture gait alterations and enhance identification of individuals with iNPH likely to benefit from CSF diversion. First, we characterized differences in spatiotemporal and plantar pressure features between individuals with iNPH and control participants without iNPH. Second, we identified the sensor-derived metrics that showed differential changes in response to the CSF tap test between responders and nonresponders and whether the observed changes were maintained after shunt surgery. Finally, we evaluated whether combining multiple sensor-derived metrics could refine identification of responders versus non-responders after a CSF tap test.

Methods

Participants

Participants diagnosed with possible iNPH were prospectively recruited at Yongin Severance Hospital, Korea, between June 2021 and October 2025. All processes for examining, evaluating, and classifying participants were performed by a multidisciplinary team consisting of a physiatrist, a neurologist, and a neurosurgeon, each with > 10 years of clinical experience in their respective fields. The diagnostic criteria for possible iNPH were based on the Japanese Guidelines for Management of Idiopathic Normal Pressure Hydrocephalus [3] and were as follows: (1) individuals with gait disturbances such as shuffling, magnetic, and wide-based walking with or without cognitive impairment and urinary incontinence; (2) other neurological or nonneurological diseases could not completely explain the aforementioned symptoms; and (3) no evidence of preceding conditions known to cause ventricular dilation, such as subarachnoid hemorrhage, meningitis, head injury, congenital/developmental hydrocephalus, and aqueductal stenosis. All participants underwent brain magnetic resonance imaging, and ventricular enlargement was confirmed by an Evans ratio of > 0.3. To ensure a valid gait assessment, we excluded participants who were unable to perform the 10MWT without assistance.

Sample size estimation was guided by our prior work on heel pressure change under dual-task conditions in older adults and individuals with Parkinson’s disease [21]. Based on observed effect sizes (Cohen’s d ≈ 0.64), 38 participants per group would be required to achieve 80% power at a two-sided alpha level of 0.05.

Gait, balance, cognition, and bladder symptoms were evaluated in all participants. After an initial evaluation, participants underwent a spinal CSF tap test (30 mL), and assessments were repeated 4–6 h later. Participants were classified into responders and non-responders by consensus of the multidisciplinary team based on integrated clinical assessment, including objective improvements in gait and balance as well as subjective functional improvements reported by participants and their caregivers after the test. Objective assessments, subjective evaluations, and consensus classification were all performed independently of the insole-derived gait measurements.

Surgical treatment was considered based on participant preference and comorbidity-related surgical risks, through a shared decision-making process involving participants, caregivers, and clinicians. Surgical treatment consisted of either lumboperitoneal (LP) shunt placement or endoscopic third ventriculostomy (ETV), as determined by the treating neurosurgeon. The LP shunt utilized the Strata™ NSC LP Shunt (Medtronic PS Medical, Inc., Minneapolis, MN, USA), which does not incorporate an anti-siphon mechanism. The initial valve setting was 2.0, and valve adjustments were made when clinically indicated. Gait and balance were reassessed within 24–48 h after surgery, and a follow-up evaluation was performed 3 months postoperatively.

This study was initially approved by the Institutional Review Board and Ethics Committee of the Yongin Severance Hospital in 2021 and subsequently re-approved with a revised protocol on August 8, 2024 (IRB No. 9-2024-0130). Informed consent was obtained from all the participants. The study protocol was conducted in accordance with the tenets of the Declaration of Helsinki.

Assessment

Cognitive and urinary symptoms were assessed at baseline using the Mini-Mental State Examination and International Prostate Symptom Score (IPSS), respectively. The IPSS was administered to both male and female participants as a validated measure of lower urinary tract symptoms [22, 23]. Gait and balance were evaluated using the 10MWT, TUGT, and Berg Balance Score (BBS). Both the 10MWT and TUGT were video-recorded for detailed review. For the 10MWT, participants started from a standing position and walked at their self-selected comfortable speed over a 10-m walkway [24]. Walking speed was calculated using the time required to traverse the full 10-m distance. Routine assistive devices were permitted when necessary. The TUGT was performed using a standard protocol in which participants stood up from a chair, walked 3 m at a comfortable pace, turned around a marked point, returned to the chair, and sat down [25]. Use of routine walking aids and arm support for standing was permitted, but no physical assistance was provided. A single trial was performed for both the 10MWT and TUGT at each assessment. Balance was assessed using the BBS, which comprises 14 functional tasks related to balance, each rated on a scale of 0 to 4, with a maximum total score of 56. The presence and frequency of clinical features during the tests, such as slowness, abnormal postural adjustments, freezing, abnormal stance, disequilibrium, stiff trunk/legs, leg apraxia, improvement with cues, frontal signs, short stride, shuffling, narrow base, festination, start hesitation, turn en bloc, and Parkinsonian signs [26], were assessed by consensus.

During the 10MWT, participants wore commercial instrumented insoles (GDCA-MD, Gilon, Republic of Korea) containing four force-sensitive resistors (FSRs) and a three-axis accelerometer. The reliability and validity of the GDCA-MD were previously established [21, 27]. The insoles transmitted data via a Bluetooth low-energy 4.2 module to a mobile device with sensor data sampled at 40 Hz and stored in the mobile device. Fourteen data streams were recorded for each participant during the test and used in subsequent analyses.

Sensor-derived gait and plantar pressure metrics

All insole-derived gait parameters were analyzed by an independent investigator after completion of data collection. The investigator was blinded to participants’ responder/non-responder classification and other clinical outcomes during data processing and analysis. The spatiotemporal gait parameters, including step count, cadence, stride length, stride time, and swing ratio, were calculated using a proprietary algorithm that was validated against a 3D opto-electronic gait analysis system [21]. Stride regularity and symmetry were computed using an unbiased autocorrelation function [20, 28]. We also investigated the changes in plantar pressure distribution during walking by analyzing the raw data streams from the pressure sensors. The FSR sensor data were recorded as relative values () and converted into pressure units (kPa) using a regression equation established in a previous study [27]:

The local maxima corresponding to the peak pressure during the gait cycles were extracted for each of the four plantar regions (big toe, medial forefoot, lateral forefoot, and heel). These peak values were subsequently averaged across the cycles to obtain the mean peak pressure for each sensor per participant. The coefficients of variation (CV) of the peak pressure values were calculated to quantify within-participant variability. The first and last peaks were excluded to ensure that the analysis reflected a stable gait. The gait assessment workflow is illustrated in Fig. 1.

Fig. 1.

Overview of the gait assessment and sensor-based analysis workflow. Participants underwent conventional gait assessments and sensor-based gait analysis before and after the CSF tap test. The instrumented insoles collected data at 40 Hz from four plantar pressure sensors (toe, medial forefoot, lateral forefoot, heel) and three-axis inertial sensors (ax, ay, az) in each insole, generating a total of 14 data streams per participant. Signals were processed using a proprietary algorithm to derive spatiotemporal gait parameters. Plantar pressure metrics, including mean peak pressure and its coefficient of variation (CV), were additionally calculated

Participants underwent conventional gait assessments and sensor-based gait analysis before and after the CSF tap test. The instrumented insoles collected data at 40 Hz from four plantar pressure sensors (toe, medial forefoot, lateral forefoot, heel) and three-axis inertial sensors (a_x, a_y, a_z) in each insole, generating a total of 14 data streams per participant. Signals were processed using a proprietary algorithm to derive spatiotemporal gait parameters. Plantar pressure metrics, including mean peak pressure and its coefficient of variation (CV), were additionally calculated.

Statistical analysis

First, we compared the gait characteristics of individuals with iNPH with those of older adults without iNPH (N = 30; mean age, 75.03 ± 6.44 years; 15 males) from our previous study [20]. Between-group differences were assessed using the chi-square test for categorical variables (sex) and the two-sample t-test for continuous variables. Effect sizes were calculated as Cohen’s d. We also examined the association between plantar pressure metrics and conventional assessments to investigate the clinical relevance of plantar pressure measurements. Pearson correlation analyses were conducted separately within the iNPH and control groups to assess group-specific association patterns between plantar pressure metrics and conventional assessments.

Next, we analyzed the differences between responders and non-responders in the CSF tap test among the individuals with iNPH. Clinical characteristics were compared between the groups using the chi-square test for categorical variable (sex) and two-sample t-tests for continuous variables. Normality of continuous variables was assessed using the Shapiro–Wilk test. Within-group changes in the gait indices after the test were assessed using paired t-tests and presented as mean differences with 95% confidence intervals (CIs). Cohen f², accompanied by 95% CIs, was calculated to estimate the magnitude of the observed differences.

Additionally, we examined the proportion of responders who exceeded the minimal clinically important difference (MCID) thresholds established in previous studies, which are defined as follows: 0.1 m/s for gait speed, 4 points for the BBS, and 4 s for the TUGT in individuals with iNPH [29, 30].

Among those who underwent surgery, we analyzed longitudinal changes in gait parameters that had significantly differed between CSF tap test responders and non-responders. Owing to non-normality (Shapiro–Wilk, p < 0.05) and small sample size (N = 19), we used the nonparametric Friedman test to assess differences across four time points: pre-tap, post-tap, immediate postoperative, and 3-month postoperative periods. Post-hoc pairwise comparisons were conducted using the Wilcoxon signed-rank test.

Given the exploratory nature of this study and its relatively small sample size, formal correction for multiple comparisons was not applied across the analyses to avoid overlooking potentially meaningful findings. Effect sizes and 95% confidence intervals were reported to facilitate interpretation of the results.

Finally, we evaluated whether adding sensor-derived heel pressure metrics to conventional gait tests improves discrimination and clinical decision-making for identifying responders to the CSF tap test. Receiver operating characteristic (ROC) curve analysis assessed the ability of conventional gait tests and heel pressure metrics to distinguish responders from non-responders. Percent changes from pre- to post–tap test were used as predictors in logistic regression models adjusted for age and sex. An additional model combining the 10MWT and heel pressure metrics was constructed to assess the incremental value of the sensor-derived metrics. Predicted probabilities from these models were used to generate ROC curves, and differences in area under the curve (AUC) were compared using the DeLong test [31].

Clinical utility was further evaluated using decision curve analysis (DCA) by calculating net benefit across threshold probabilities [32]. Net benefit was evaluated for the age- and sex-adjusted 10MWT model and for models incorporating heel pressure AVR, heel pressure CVR, or both metrics in combination with the 10MWT. Given the limited sample size (N = 77; 48 responders), internal validation was performed using bootstrap resampling with 10,000 iterations to estimate mean values and 95% confidence intervals.

All statistical analyses were conducted using Python (version 3.12.4; Python Software Foundation, Wilmington, DE, USA), with packages including pandas, NumPy, Matplotlib, Seaborn, SciPy, StatsModels, and scikit-learn. Statistical significance was set at p < 0.05.

Results

Baseline characteristics

Ninety-four participants with suspected iNPH were screened for eligibility (Fig. 2). Of these, 17 were excluded because of the inability to complete the 10MWT independently or because of the presence of acute medical conditions unrelated to the study (e.g., infection or fall-related injury) that could affect gait or study participation. Consequently, 77 participants were included in the final analysis, of which 48 were classified as responders. Among the responders, 20 underwent shunt surgery, while 28 did not, primarily owing to surgical concerns (n = 15) or satisfaction with post-tap symptom improvement (n = 13). One participant was lost to follow-up after surgery, resulting in 19 participants being included in the postoperative outcome analysis.

Fig. 2.

Flow diagram of participant inclusion and classification

Comparison with control participants without iNPH

Individuals with possible iNPH demonstrated significantly poorer performance on clinical gait and balance tests, including the 10MWT, TUGT, and BBS, than control participants without iNPH (Table 1). Spatiotemporal gait parameters revealed that individuals with possible iNPH had lower cadences, shorter step lengths, longer stride times, and larger swing ratios. Their gait rhythm was also more irregular and asymmetric than that of controls.

Table 1.

Baseline clinical parameters of control participants and participants with possible iNPH

			Controls (N = 30)	Possible iNPH (N = 77)	Mean Δ (95% CI)	p-value	Effect size (Cohen’s d)
Baseline demographics
Age			73.70 ± 6.82	71.36 ± 9.44	− 2.34 [ − 5.62, 0.94]	0.160	− 0.27
Sex (Male: Female)			15: 15	48: 29	−	0.244	−
Height (cm)			161.13 ± 6.11	161.10 ± 9.05	− 0.03 [ − 3.05, 3.00]	0.985	0.00
BMI (kg/m2)			23.73 ± 2.85	24.80 ± 3.21	1.06 [ − 0.21, 2.34]	0.100	0.34
Gait and balance parameters
10-m walk test (m/s)			1.29 ± 0.18	0.80 ± 0.29	− 0.49 [ − 0.58,  − 0.40]	< 0.001†	− 1.86
Step count (n)			13.90 ± 1.47	28.39 ± 17.61	14.49 [10.46, 18.52]	< 0.001†	0.97
Cadence (spm)			118.31 ± 8.39	111.92 ± 19.77	− 6.38 [ − 11.79,  − 0.98]	0.021*	− 0.37
Stride length (m)			1.28 ± 0.23	0.87 ± 0.32	− 0.41 [ − 0.52,  − 0.30]	< 0.001†	− 1.38
Stride time (s)			1.03 ± 0.07	1.10 ± 0.17	0.07 [0.02, 0.11]	0.005**	0.45
Swing ratio (%)			35.83 ± 2.03	34.74 ± 2.34	− 1.09 [ − 2.00,  − 0.18]	0.020*	− 0.48
Regularity			0.89 ± 0.13	0.78 ± 0.15	− 0.11 [ − 0.17,  − 0.05]	< 0.001†	− 0.76
Symmetry			0.91 ± 0.06	0.61 ± 0.38	− 0.30 [ − 0.39,  − 0.21]	< 0.001†	− 0.91
Timed up-and-go test (s)			10.06 ± 1.80	19.68 ± 12.04	9.63 [6.82, 12.43]	< 0.001†	0.94
Berg balance scale			55.93 ± 0.25	46.66 ± 10.30	− 9.27 [ − 11.61,  − 6.93]	< 0.001†	− 1.06
Peak pressure (kPa for mean, % for CV)
Toe	Lt	Mean	230.30 ± 54.25	195.62 ± 58.33	− 34.68 [ − 58.57,  − 10.79]	0.005**	− 0.61
		CV	8.88 ± 5.18	9.14 ± 6.82	0.27 [ − 2.18, 2.71]	0.829	0.04
	Rt	Mean	235.05 ± 46.38	202.45 ± 52.42	− 32.60 [ − 53.33,  − 11.87]	0.003**	− 0.64
		CV	7.41 ± 5.54	8.57 ± 5.86	1.16 [ − 1.27, 3.58]	0.344	0.20
Medial forefoot	Lt	Mean	213.82 ± 46.49	228.70 ± 44.55	14.88 [ − 4.97, 34.74]	0.138	0.33
		CV	13.98 ± 7.76	7.62 ± 4.61	− 6.36 [ − 9.42,  − 3.30]	< 0.001†	− 1.12
	Rt	Mean	214.83 ± 31.96	232.51 ± 36.59	17.68 [3.34, 32.03]	0.017*	0.50
		CV	13.45 ± 8.39	7.84 ± 4.22	− 5.62 [ − 8.88,  − 2.36]	0.001**	− 0.99
Lateral forefoot	Lt	Mean	130.02 ± 35.55	145.71 ± 42.33	15.69 [ − 0.47, 31.85]	0.057	0.39
		CV	30.62 ± 13.15	16.25 ± 11.05	− 14.37 [ − 19.82,  − 8.91]	< 0.001†	− 1.23
	Rt	Mean	150.34 ± 44.50	148.90 ± 43.30	− 1.44 [ − 20.52, 17.63]	0.880	− 0.03
		CV	22.57 ± 15.84	17.17 ± 12.81	− 5.41 [ − 11.93, 1.12]	0.102	− 0.39
Heel	Lt	Mean	252.88 ± 35.85	222.23 ± 45.57	− 30.65 [ − 47.33,  − 13.98]	< 0.001†	− 0.71
		CV	3.16 ± 1.71	7.23 ± 8.01	4.07 [2.15, 5.98]	< 0.001†	0.59
	Rt	Mean	265.05 ± 27.78	244.46 ± 37.98	− 20.59 [ − 33.88,  − 7.29]	0.003**	− 0.58
		CV	3.11 ± 1.85	6.55 ± 5.37	3.44 [2.05, 4.82]	< 0.001†	0.74

Values are presented as mean ± SD; Controls: Control participants without iNPH; *p < 0.05, **p < 0.01, †p < 0.001

Bold values indicate statistically significant results (P <0.05)

In terms of plantar pressure distribution, peak pressures in both the toe and heel regions were significantly lower in the possible iNPH group than in controls bilaterally. Notably, pressure variability exhibited contrasting regional patterns. The CV was significantly reduced in the left lateral forefoot region and bilateral medial forefoot regions, whereas it was significantly increased in the heel region (left: 3.16 ± 1.71% vs. 7.23 ± 8.01%, 95% CI 2.15 to 5.98; right: 3.11 ± 1.85% vs. 6.55 ± 5.37%, 95% CI 2.05 to 4.82, both p < 0.001), indicating increased variability during initial contact.

Pearson correlation analyses revealed distinct association patterns between plantar pressure metrics and functional assessments in the two groups (Additional file 1: Fig. S1). In control participants, heel pressure metrics were significantly associated with gait speed, whereas TUGT and BBS showed no significant relationships with plantar pressure parameters. In contrast, plantar pressure metrics showed broader associations with functional assessments in individuals with possible iNPH. Significant correlations were observed not only with the 10MWT but also with TUGT and BBS, and these relationships involved both heel and toe pressure metrics. Better functional performance was associated with greater plantar loading and lower variability in both heel and toe regions.

Responders versus non-responders: post-tap and postoperative changes

In individuals with iNPH, there were no statistically significant differences between the responder and non-responder groups in terms of demographic factors or baseline functional status before the tap test (Additional file 1: Table S1). Responders showed significant improvement in gait speed (10MWT), mobility (TUGT), and balance (BBS) after the CSF tap test (all p < 0.001; Table 2). Step count decreased while stride length and gait symmetry increased. However, only 19 (39.6%), 14 (29.2%), and 13 (27.1%) participants surpassed their respective MCID thresholds.

Table 2.

Changes in conventional gait and balance parameters before and after the CSF tap test between responders and non-responders

	Responder (N = 48) Δ (95% CI)	Non-Responder (N = 29) Δ (95% CI)	Between Groups β (95% CI)
10-m walk test (m/s)	0.08 [0.04, 0.13]†	− 0.01 [ − 0.07, 0.04]	0.09 [0.03, 0.15]**
Step count (n)	− 4.10 [ − 8.19,  − 0.02]*	− 0.14 [ − 2.46, 2.18]	− 4.21 [ − 8.65, 0.24]
Cadence (spm)	0.51 [ − 2.36, 3.38]	− 5.65 [ − 16.08, 4.78]	3.70 [ − 2.35, 9.76]
Stride length (m)	0.08 [0.04, 0.12]†	0.00 [ − 0.04, 0.04]	0.08 [0.02, 0.13]**
Stride time (s)	0.00 [ − 0.03, 0.03]	0.04 [ − 0.04, 0.11]	− 0.03 [ − 0.10, 0.03]
Swing ratio (%)	− 0.16 [ − 0.81, 0.49]	− 0.72 [ − 1.40,  − 0.05]*	0.57 [ − 0.38, 1.52]
Regularity	0.02 [ − 0.01, 0.05]	− 0.02 [ − 0.07, 0.02]	0.05 [0.00, 0.09]
Symmetry	0.10 [0.01, 0.19]*	− 0.01 [ − 0.11, 0.09]	0.11 [ − 0.02, 0.23]
Timed up-and-go test (s)	− 2.56 [ − 3.77,  − 1.35]†	− 1.29 [ − 3.53, 0.95]	− 1.80 [ − 3.84, 0.24]
Berg balance scale	2.88 [1.75, 4.00]†	0.97 [ − 0.26, 2.20]	2.17 [0.54, 3.81]**

Values are presented as mean ± SD; *p < 0.05, **p < 0.01, †p < 0.001

Bold values indicate statistically significant results (P <0.05)

In the between-group analysis, linear regression analyses adjusted for age and sex revealed significant associations between the groups and changes after the CSF tap test for the 10MWT (p = 0.004, Cohen f² = 0.12) and BBS (p = 0.010, Cohen f² = 0.10). By contrast, TUGT showed a nonsignificant trend toward improvement in responders (p = 0.083, Cohen f² = 0.04). For spatiotemporal gait parameters, only the change in stride length differed significantly between responders and non-responders.

After the CSF tap test, plantar pressure parameters differed significantly between responders and non-responders (Table 3). Heel peak pressure increased significantly in responders on both sides (left: Δ = 13.06 kPa, 95% CI 6.04 to 20.08; right: Δ = 14.88 kPa, 95% CI 8.75 to 21.02, both p < 0.001), whereas non-responders exhibited no change. Consequently, between-group differences in mean heel peak pressure were significant bilaterally (left: 95% CI 3.12 to 26.09, p = 0.013, f² = 0.09, right: 95% CI 7.38 to 25.90, p < 0.001, f² = 0.18). Increases in mean peak pressure were also observed in the toe and medial forefoot regions, resulting in significant between-group differences. Furthermore, the CV in the right heel was significantly decreased in responders (Δ = − 1.58, 95% CI  − 2.55 to  − 0.60, p = 0.002) but remained unchanged in non-responders (p = 0.806), leading to a significant between-group difference (95% CI − 3.36 to − 1.14, p < 0.001, f² = 0.23). The CV in the toe region also decreased in responders on both sides, but with a smaller effect size than that observed for the heel (Cohen’s f²: left = 0.09, right = 0.06). Even among the 29 responders who did not meet the MCID for the 10MWT, CV of right heel peak pressure showed a median reduction of 11.1% (IQR: − 30.2% to 10.4%). No significant within- or between-group differences were observed for the lateral forefoot region.

Table 3.

Changes in mean and CV of peak pressures before and after the CSF tap test between responders and non-responders

			Responder (N = 48) Δ (95% CI)	Non-Responder (N = 29) Δ (95% CI)	Between Groups β (95% CI)
Toe	Lt	Mean	18.79 [10.02, 27.57]†	− 4.09 [ − 14.31, 6.13]	20.91 [7.63, 34.18]**
		CV	− 1.18 [ − 2.85, 0.50]	3.20 [0.27, 6.14]*	− 3.50 [ − 6.30,  − 0.69]*
	Rt	Mean	27.73 [18.63, 36.83]†	2.89 [ − 9.04, 14.82]	21.46 [7.99, 34.93]**
		CV	− 1.66 [ − 3.23,  − 0.10]*	0.78 [ − 1.17, 2.72]	− 2.41 [ − 4.63,  − 0.18]*
Medial forefoot	Lt	Mean	10.76 [4.16, 17.36]**	− 1.14 [ − 10.28, 7.99]	11.29 [1.43, 21.16]*
		CV	0.31 [ − 0.98, 1.60]	2.11 [0.05, 4.18]*	− 1.91 [ − 4.08, 0.26]
	Rt	Mean	7.17 [0.28, 14.05]*	− 4.47 [ − 14.35, 5.41]	12.42 [1.50, 23.34]*
		CV	0.20 [ − 1.25, 1.65]	1.35 [ − 0.90, 3.60]	− 1.09 [ − 3.63, 1.45]
Lateral forefoot	Lt	Mean	5.67 [ − 2.13, 13.48]	2.93 [ − 6.43, 12.30]	2.42 [ − 9.95, 14.79]
		CV	1.13 [ − 0.72, 2.97]	2.35 [ − 1.39, 6.09]	− 0.58 [ − 3.85, 2.70]
	Rt	Mean	5.84 [ − 2.47, 14.15]	4.64 [ − 5.13, 14.41]	0.70 [ − 12.28, 13.67]
		CV	0.60 [ − 1.99, 3.18]	− 2.18 [ − 4.94, 0.58]	2.56 [ − 1.09, 6.22]
Heel	Lt	Mean	13.06 [6.04, 20.08]†	− 2.18 [ − 11.83, 7.46]	14.61 [3.12, 26.09]*
		CV	− 0.29 [ − 1.47, 0.89]	− 0.18 [ − 1.79, 1.43]	− 0.09 [ − 1.97, 1.79]
	Rt	Mean	14.89 [8.75, 21.02]†	− 0.14 [ − 6.78, 6.50]	16.64 [7.38, 25.90]†
		CV	− 1.58 [ − 2.55,  − 0.60]**	0.14 [ − 1.00, 1.27]	− 2.25 [ − 3.36,  − 1.14]†

Values are presented as mean ± SD; Units: kPa for Mean, % for CV; *p < 0.05, **p < 0.01, †p < 0.001

Bold values indicate statistically significant results (P < 0.05)

Among the 20 participants who underwent surgical treatment, 19 completed the 3-month follow-up (12 ETVs and 7 LP shunts). Improvements in gait, balance, and plantar pressure observed after the CSF tap test were maintained postoperatively (Additional file 1: Table S2; Fig. 3).

Fig. 3.

Changes in the right heel plantar pressure parameters across four time points: pre-tap, post-tap, immediate postoperative, and 3-month postoperative assessments. a Mean peak pressure. b Coefficient of variation. Dots represent individual participants; center lines indicate medians, boxes indicate interquartile ranges, and whiskers indicate the range within 1.5 × IQR. Data were obtained from participants who underwent surgical treatment and completed all follow-up evaluations (N = 19). Statistical comparisons were performed using the Friedman test followed by post-hoc Wilcoxon signed-rank tests. *Indicates p < 0.05, **indicates p < 0.01, †indicates p < 0.001

Predictive utility of plantar pressure metrics

Given the consistent post-tap changes and longitudinal improvements observed in right heel pressure metrics, we evaluated their ability to discriminate responders from non-responders using ROC curve analysis and DCA. Percentage changes in the mean peak pressure (AVR), coefficient of variation (CVR), and conventional gait measures were analyzed.

Among conventional measures (10MWT, TUGT, and BBS), the 10MWT showed the highest discrimination (AUROC = 0.663, 95% CI 0.533 to 0.783; Additional file 1: Fig. S2a). Among plantar pressure models, the combined heel AVR and CVR model showed the highest discrimination (AUROC = 0.772, 95% CI 0.651 to 0.879; Additional file 1: Fig. S2b). Adding heel AVR and CVR to the 10MWT model yielded the highest AUROC (0.784, 95% CI 0.670 to 0.886; Additional file 1: Fig. S2c). However, DeLong tests showed no statistically significant differences in AUROC between the models.

DCA demonstrated that adding plantar pressure metrics to the clinical model increased net benefit compared with the 10MWT-alone model and treat-all strategy across threshold probabilities (Fig. 4). In DCA, the threshold probability reflects the predicted probability of treatment responsiveness at which additional clinical consideration or intervention may be warranted. The model incorporating both heel AVR and heel CVR showed the greatest clinical utility, with the highest average Δnet benefit (+ 0.093) and the largest improvement at a threshold probability of 0.60 (Δnet benefit + 0.136). Higher net benefit within the intermediate threshold range (0.40–0.70) suggests that incorporating heel pressure metrics may provide additional support when considering further treatment options such as shunt surgery in individuals showing equivocal or modest improvement in conventional gait measures following the CSF tap test.

Fig. 4.

Decision curve analysis comparing the age- and sex-adjusted 10MWT model with models incorporating plantar pressure metrics for identifying responders to the CSF tap test. Percentage changes in right heel mean peak pressure (AVR) and its coefficient of variation (CVR) were evaluated. The model incorporating both AVR and CVR showed the greatest net benefit across clinically relevant threshold probabilities, suggesting added value in identifying treatment responsiveness beyond conventional gait measures alone.”

Discussion

In this study, we demonstrate the clinical utility of a low-cost sensor-embedded insole system for gait assessment in individuals with iNPH. Our findings showed that clinically meaningful gait parameters relevant to both diagnosis and prognosis can be effectively captured using a simple instrumented insole with a relatively low sampling rate. Furthermore, combining changes in the 10MWT with peak heel pressure following a CSF tap test enhanced the net clinical benefit in decision-making, suggesting that plantar pressure measurements provide incremental value for identifying surgical candidates.

In line with prior studies, several clinical parameters indicated distinct gait and balance impairments in individuals with iNPH compared with those in control participants without iNPH [3]. Individuals with iNPH exhibited slower gait speed, shorter stride length, increased step count, and irregular gait patterns with reduced regularity and symmetry. We also observed an altered plantar pressure distribution and increased variability in individuals with iNPH. Lower pressures at the toe and heel indicate decreased loading at both the initial contact and push-off phases of gait, which are critical for stability and forward propulsion during walking [33]. Additionally, region-specific variability in plantar pressure suggests more rigid forefoot loading and unstable heel contact during walking. The correlation between lower heel-toe pressure and slower 10MWT in our dataset suggests that altered plantar loading is associated with impaired gait performance in individuals with iNPH. These findings are consistent with reduced heel strike and toe-off angles, as well as decreased activation of both the pretibial and calf muscles, in individuals with iNPH [34, 35]. Furthermore, we found that both the mean peak pressures and their variability (CV) at the heel and toe were linked to gait speed and balance, whereas only the mean heel peak pressure was significantly associated with gait speed in control participants, suggesting a more widespread disruption of plantar pressure control and its functional consequences in individuals with iNPH. Although the “shuffling gait” and related kinematic/kinetic features of iNPH are well-established [35], our findings highlight that both rhythm disruption and plantar loading characteristics can be quantitatively assessed in a cost-effective and portable manner.

A recent scoping review suggested the potential benefits of exercise and physical therapy interventions for gait and balance function in individuals with iNPH, although objective longitudinal monitoring methods remain limited [36]. In this context, quantitative assessment of multiple gait domains including gait rhythmicity, stability, and plantar loading characteristics may support rehabilitation goal setting, provide complementary information for individualized exercise planning, and facilitate objective evaluation of treatment-related functional changes in individuals with iNPH. In addition, sensor-based approaches may enable repeated functional monitoring in real-world environments, including home and community settings with varying surfaces and walking conditions.

Combining data obtained from instrumented insoles may provide complementary information regarding treatment-related gait changes following the CSF tap test in individuals with iNPH. Previous studies using wearable inertial sensors have demonstrated that quantitative gait parameters, including cadence, stride length, double support duration, and gait coordination indices, sensitively reflect gait apraxia and responsiveness to CSF drainage or shunt surgery in patients with iNPH [37–39]. These studies also suggested that instrumental gait metrics may provide greater sensitivity for detecting clinically meaningful changes than conventional clinical rating scales alone. Consistent with these previous findings, our study further demonstrated the responsiveness of region-specific plantar loading patterns to CSF drainage. Changes in plantar pressure variability were observed even in some individuals without prominent improvements in conventional spatiotemporal gait parameters. Although responders showed significant improvements in gait speed and stride length at the group level, approximately 60% of responders did not exceed the established MCID threshold for gait speed at the individual level. This may have led to a reduced NPV, as previously reported [11]. However, decreases in heel pressure CV were more uniformly observed across responders, including those without significant gait speed gains. While AUROC did not significantly change, the observed increase in net benefit suggests that incorporating heel pressure metrics may provide incremental clinical value by capturing subtle treatment-related gait changes that are not fully reflected by conventional gait measures alone. This pattern was further supported by the results of 19 participants who underwent surgical intervention. Previous studies have reported that maximum gait improvement typically occurs 24–48 h after CSF drainage [40], requiring prolonged inpatient observation, which increases the logistical and financial burdens on both patients and healthcare systems. In contrast, our approach allowed assessment as early as 4 h post-CSF tap, providing timely decision support while reducing patient burden and improving clinical feasibility.

Additionally, our findings suggest asymmetric motor control and treatment responses in individuals with iNPH. The temporal symmetry of steps was significantly lower in individuals with iNPH than in controls. Functional assessments involving postural transitions, such as the TUG and BBS, showed stronger associations with plantar pressure metrics of the right foot than with those of the left foot in individuals with iNPH. Moreover, greater changes in the right-sided plantar pressure metrics were observed among responders following the tap test. Asymmetric motor features, well documented in neurodegenerative disorders such as Alzheimer’s and Parkinson’s disease [41], may share pathophysiological mechanisms with iNPH [42]. Disruption in motor cortical circuits and white matter tracts has been linked to impaired motor control and coordination [43, 44], and their restoration has been linked to improvements in motor function [45–47]. Although some studies have noted motor asymmetry in iNPH [19, 48, 49], the extent to which this asymmetry reflects distinct underlying mechanisms and its relevance to treatment responsiveness warrants further investigation.

Limitations

This study has several limitations. First, the inclusion of participants capable of walking at least 10 m without assistance may limit the generalizability of our findings to those with more severe motor deficits. In addition, the distribution of responders and non-responders observed in this study may not be representative of the broader iNPH population and may have introduced bias in estimating group differences. Second, although the sensor-embedded insoles were designed for use within regular footwear and all assessments were performed under standardized conditions, the possibility that the insoles may have subtly influenced gait patterns or shoe fit cannot be completely excluded. Third, the small sample size and relatively short follow-up duration may not fully capture postoperative gait changes after surgical intervention. Notably, we observed significant improvements in plantar pressure metrics, stride length, and gait speed. However, rhythm-related features showed mixed changes: temporal symmetry remained unchanged, whereas gait regularity improved significantly at the 3-month follow-up. These findings are consistent with those of previous studies [50, 51], suggesting that distinct mechanisms may underlie the motor control and rhythm regulation of gait. Alternatively, rhythm-related improvements may emerge beyond the 3-month follow-up period. Fourth, because this study focused solely on gait changes following CSF drainage, changes in cognitive or urinary symptoms and their possible associations with gait were not addressed. Fifth, classification performance was assessed within a single cohort; therefore, external validation using an independent dataset is warranted. Although bootstrapping was applied to estimate the ROC CIs, it does not replace external validation and may still be limited by the small and imbalanced dataset. Finally, direct comparisons with other gait assessment platforms commonly used in individuals with iNPH, such as instrumented walkways or laboratory-based gait analysis systems, were not performed. Future studies are warranted to evaluate the agreement and complementary value of sensor-based insole measurements across different gait assessment technologies. Further studies with external validation, larger and more diverse samples, and broader clinical assessments are required to confirm and generalize our findings.

Conclusions

A simple instrumented insole may provide complementary information regarding treatment-related gait changes in individuals with iNPH, including subtle changes that may not be fully reflected by conventional gait measures alone. Sensor-based gait assessment may support functional evaluation and clinical decision-making following the CSF tap test.

Abbreviations

10MWT

10-M walking test

AUC

Area under the curve

BBS

Berg balance score

CI

Confidence intervals

CSF

Cerebrospinal fluid

CV

Coefficients of variation

DCA

Decision curve analysis

FSR

Force-sensitive resistor

INPH

Idiopathic normal pressure hydrocephalus

IQR

Interquartile range

MCID

Minimal clinically important difference

ROC

Receiver operating characteristic

SD

Standard deviation

TUGT

Timed up and go test

Author contributions

SMH: Data curation, Formal analysis, Methodology, Software, Validation, Visualisation, Writing—Original draft, Review & editing; WHL, SIC and HWY: Data curation, Investigation, Writing—Review & editing; SJC: Formal analysis, Investigation, Validation, Writing—Review & editing; JKH and NYK: Conceptualization, Formal analysis, Investigation, Methodology, Project administration, Resources, Writing—Review & editing.

Funding

This work was supported by a grant from the Korea Health Technology R&D Project through the Korea Health Industry Development Institute, funded by the Ministry of Health and Welfare, Republic of Korea (grant number: RS-2023-00265489).

Data availability

The datasets generated and/or analyzed during the current study are not openly available to protect study participant privacy and are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

This study was initially approved by the Institutional Review Board and Ethics Committee of the Yongin Severance Hospital in 2021 and subsequently re-approved with a revised protocol on August 8, 2024 (IRB No. 9-2024-0130). Informed consent was obtained from all the participants. The study protocol was conducted in accordance with the tenets of the Declaration of Helsinki.

Consent for publication

Not applicable.

Competing interests

N.Y.K. received stock options from GILON in 2025. The company had no role in the study design, data collection, analysis, interpretation of the data, or preparation of the manuscript. The authors declare no other competing interests.