Scientists Replace Peak Pressure With Predictive Force Model for Personalized Orthoses

For decades, the design of foot orthoses and insoles has been anchored to a single number: the peak plantar pressure, the highest force-per-area reading recorded under the sole during a step. It is intuitive, it comes straight off the sensor, and it maps naturally onto tissue-level concerns such as focal overload and ulceration risk. But a new study argues that this convention has quietly become a liability. Peak pressure, researchers now show, is not the mechanical load the body imposes on the foot at all. It is a local quotient of force and contact area, and that area is set as much by the measuring device, the shoe, the sock, and the skin as by the person walking. The mechanically primary quantity, the force transmitted through the foot, is fixed by body mass and gait dynamics through Newton’s second law and is, in principle, independent of how the contact is distributed. When an orthosis is designed to meet a pressure target, the effort of personalizing it is spent matching a number that the device itself changes the moment it is worn.

The study, published in Frontiers in Bioengineering and Biotechnology by Sergio Fuentes del Toro and Josue Aranda-Ruíz of Universidad Carlos III de Madrid, set out to test whether the applied regional force is a more stable descriptor of plantar loading than peak pressure. The researchers drew on two large, open-access datasets acquired on entirely different measurement platforms: the CAD WALK healthy-control dataset, recorded barefoot on an RSscan footscan platform with rectangular sensors sampled at 500 Hz, and StepUP-P150, recorded on a Stepscan platform with square 5 mm sensors sampled at 100 Hz under both barefoot and shod conditions at four walking speeds. After quality control, 27,385 of the 30,424 recorded StepUP-P150 footsteps were retained, alongside all 2,640 pre-segmented footsteps from the 55 CAD WALK participants. The two platforms differ in sensor geometry, native units, and sampling rate, which is precisely what allowed the device-independence of the results to be tested.

Because the two systems report pressure in different native units, the team did not rely on nominal calibrations. Instead, they calibrated force against the impulse-momentum principle identically for both platforms. Over a walking cycle, the mean vertical ground reaction force equals body weight, so the vertical impulse of a single foot during stance must equal body weight times the step time. This identity yields a per-trial calibration factor in closed form, computed for each combination of participant, interface condition, and walking speed rather than once per platform. Across the 330 StepUP-P150 trials, the factor had a mean of 0.0391 with a coefficient of variation of just 7.8 percent, indicating a single, well-defined plate constant describes each system rather than a quantity that drifts from step to step. With this calibration, the whole-foot peak force reached 1.25 times body weight on average at preferred speed, squarely within the physiological range for level walking. On CAD WALK, whose data are already expressed in physical units, the same procedure acted as an independent check, recovering a stance impulse ratio of 0.968 and correcting a systematic under-reading of roughly three percent that increased with body mass.

The first major result concerns reproducibility. Within participants of StepUP-P150, across repeated barefoot footsteps at preferred speed, the body-weight-normalized applied force varied less from step to step than the robust peak pressure. The between-step coefficient of variation was 0.18 for whole-foot force and 0.20 for forefoot force, against 0.26 for peak pressure. Consistently, the number of footsteps required to bring the running mean to within five percent of its stable value was 11 for whole-foot force and 15 for forefoot force, against 28 for peak pressure, so the applied force stabilizes in roughly half the footsteps needed by the pressure reading. Footstep order within a walking pass had no systematic effect on the peak applied force, confirming that these counts reflect genuine step-to-step variability rather than initial acceleration artifacts.

The second and most striking result is the dissociation produced when the foot-support interface changed. In paired barefoot and shod trials at preferred speed across all 55 participants, the body-weight-normalized applied force moved from 1.251 to 1.260 times body weight, a change of 0.7 percent that was not statistically significant. Over the same trials, the robust peak pressure fell from 572 to 386 kPa, a reduction of 32.5 percent that was highly significant. Taken per participant, the mean absolute change induced by the interface was 3.7 percent for the applied force and 31.8 percent for the peak pressure, making the pressure reading 8.6 times more sensitive to the interface than the force. Inserting a shoe between the foot and the plate leaves the force transmitted through each region essentially unchanged while it redistributes and cushions the pressure, which is precisely the quantity a pressure-based design would read as a large effect. The applied force, being the mechanically conserved quantity, is unaffected.

The third result establishes predictability. Across the steady-state speeds of StepUP-P150, spanning 0.43 to 1.94 meters per second in the barefoot condition, the applied force followed a power law of body mass and walking speed. Fitted across the full barefoot cohort and speed range, the total applied force obeyed F = 17.96 m^0.910 v^0.184, with mass in kilograms and speed in meters per second, reproducing the measured force with a coefficient of determination of 0.96. Under leave-one-subject-out cross-validation, the cross-validated coefficient was 0.955, essentially identical to the in-sample value, indicating the model does not overfit. The mass exponent, significantly below unity, indicates a slightly sub-proportional, allometric scaling of force with body mass. Walking speed, which does not enter the impulse calibration, contributed a significant incremental effect, raising the mass-only fit from 0.917 to 0.965. By contrast, the identical model form applied to peak pressure explained only 0.33 of its variance. Even with mass and speed known, the peak pressure remains largely unpredictable, a quantitative separation that disqualifies it as a governable design input.

The spatial dimension proved equally universal. Every footstep was registered onto a common, length-normalized anatomical frame, resampled to an isotropic 2-mm grid, aligned using the temporal sequence of stance rather than its instantaneous shape, and scaled by heel-to-hallux length. On this canonical frame, the per-cell load map was highly consistent across participants, with a between-participant correlation of 0.892. The map recovered the expected anatomy: a loaded rearfoot, a low-load medial arch, an oblique metatarsal ridge, and a single medial hallux. Crucially, the CAD WALK and StepUP-P150 maps agreed with a cross-system Pearson correlation of 0.95, and their marginal load profiles coincided after length normalization. The highest-load cells, the rearfoot and the metatarsal heads, were also the most stable across participants, so the regions that matter most for design are the best determined. This coherence across subjects and devices licenses averaging the two systems, with equal weight per platform, into a single universal load map released with the study.

Together, the scaling law and the universal map combine into a closed-form model that reconstructs the plantar force field of an arbitrary individual from three inputs alone: body mass, walking speed, and foot length. The total applied force follows the power law, and the force at any cell of the foot follows by distributing that total over the universal map. The released regional force targets, 0.86 times body weight at the rearfoot and 1.10 at the forefoot, fix the demand each region must carry. As a worked example, for a 70-kilogram person walking at 1.30 meters per second, the model gives a total applied force of approximately 900 Newtons, or 1.31 times body weight, which is then multiplied by the released map to obtain the force carried by each cell. The framework yields a distinct, individualized absolute-force field for each person rather than a single averaged shape, with predicted totals rising from 701 Newtons for the lightest, slowest individual to 1,294 Newtons for the heaviest, fastest.

The practical implication is that an orthosis or insole is better specified from a device-independent map of regional applied force than from a device-specific field of peak pressure. The researchers are explicit that where a direct plantar-loading measurement is available, it remains the preferred input, and the role of the model and universal map is to supply a device-independent estimate when direct measurement is unavailable. The advantage is portability and independence from the acquisition device, not superior accuracy relative to subject-specific measurement. Much of the current insole-design literature optimizes against peak pressure or its redistribution, and in silico prescription pipelines increasingly personalize insole geometry from such inputs. Grounding these pipelines in the applied force, expressed as a body-weight fraction and distributed over the universal map, would make the resulting prescriptions more transferable between acquisition devices and footwear conditions because the input is then defined by the conserved mechanical demand rather than by the interface through which it happened to be measured.

Several limitations bound these conclusions. The analysis was restricted to healthy adults walking at self-selected and controlled speeds on level ground; running, stair and slope locomotion, and clinical populations such as the diabetic foot, in which peak pressure is itself the clinical target, were not examined. The predictive model was fitted and evaluated within StepUP-P150 barefoot walking, so its coefficients call for external validation on independent cohorts, and the model predicts the population-average distribution rather than an individual’s departures from it. The universal map was built from barefoot data, and the shod condition was acquired on the platform rather than through a true in-shoe sensor, so the interface effect reported is that of the shoe as an interposed layer, not of the full in-shoe environment. The regional partition followed the 40/30/30 convention of the source of the design targets, adopted from the literature rather than independently validated. Future work follows directly: extending the framework to running and clinical gait, validating the shod condition with genuine in-shoe measurement, and, most directly, the mechanical design of a personalized insole in which each region of the map is assigned a working stiffness matched to its local applied force, a design step that is the subject of a companion study.

Subject of Research: Biomechanics

Article Title: From pressure to force: a device-independent, predictive model of plantar loading for personalized orthoses

Article References: Fuentes del Toro, S., & Aranda-Ruíz, J. (2026). From pressure to force: a device-independent, predictive model of plantar loading for personalized orthoses. Frontiers in Bioengineering and Biotechnology, 14, Article 1950679. https://doi.org/10.3389/fbioe.2026.1950679

Image Credits: AI Generated

DOI: 10.3389/fbioe.2026.1950679

Keywords: biomechanics, plantar pressure, foot orthoses, gait analysis, personalized medicine, predictive model, orthotics, walking, pressure, force, device-independent, predictive