Six fresh-frozen cadaver foot specimens obtained with a mid-tibial amputation from one female and five male donors (who had a mean age [and standard deviation] of 63 ± 18 years [range, thirty-two to seventy-eight years] at the time of death and a mean body weight of 74.6 ± 16.8 kg) with no evidence of previous surgery, deformity of the first ray, or preexisting foot deformity (pes cavus or pes planus) were approved for use by the institutional review board for this repeated-measures design. Dissections were performed to expose nine extrinsic tendons of the foot above the superior extensor retinaculum. The tendons were the extensor hallucis longus, extensor digitorum longus, tibialis anterior, tibialis posterior, flexor hallucis longus, flexor digitorum longus, peroneus brevis, peroneus longus, and Achilles. All specimens used in this study were right feet.
Metal fixtures were used to simulate fusion of the first metatarsophalangeal joint in the cadaver specimens. A series of modifications made to the first metatarsal and proximal phalanx allowed attachment of the modeling apparatus. Initially, the dorsomedial surfaces of the first metatarsal and proximal phalanx were exposed and the soft tissue was removed. The dorsal surfaces of the metatarsal head and the base of the phalanx were shaped flat with use of a bone saw (Stryker, Kalamazoo, Michigan), and the extensor hallucis brevis tendon was sacrificed in this process. The extensor hallucis longus tendon was preserved and retracted laterally. Steel hanger bolts (Hanger Bolt and Stud, Greenfield, Indiana) were permanently fixed into polymethylmethacrylate-filled cavities that opened on the superior surface of the bones. The bolts interfaced with a set of custom, interchangeable aluminum plates that were secured with threaded nuts (Fig. 1). The array of testing plates represented incremental changes in dorsiflexion angle and accommodated changes in the center of rotation of the joint21,22 such that the reduced bone surfaces lay perpendicular to each plate.
The testing plate was designed to be similar to the low-profile contoured plate currently available for use in arthrodesis procedures13; however, its rigidity was intended to represent a fully healed joint. Finite element analysis was performed with use of ALGOR DesignCheck (Alibre, Richardson, Texas) to verify the robustness of the testing plates during gait. A linear elastic model was employed, and a solid mesh was created of bricks, wedges, pyramids, and tetrahedra (7296 nodes and 7448 elements). The material selected was aluminum 6061-T6 alloy (an elastic modulus of 68.9 GPa and a shear modulus of 26.0 GPa). The plate did not yield when subjected to a 1334-N load applied to a fixed construct (0.021 mm of maximum displacement, with a minimum safety factor of 6.6).
Given the unique cut angles of the bone surfaces across the foot specimens, mediolateral radiographs were made to associate each plate condition with an anatomical dorsiflexion angle. A custom spring-loaded apparatus was used to apply a compressive load (25% of the donor's reported body weight) along the longitudinal axis of the tibia (Fig. 2). Radiographs were made for each foot-plate combination. Photoshop CS software (Adobe, San Jose, California) was used to measure the dorsiflexion angles with use of the landmarks described previously8. A group of testing plates was then selected such that the anatomical range extended from approximately 10° to 30° of dorsiflexion. The repeatability of controlling the dorsiflexion angle with plate fixation was confirmed through pilot studies of the dorsiflexion angle (±0.7° across plate changes and ±0.6° across user measurements).
Each plate was fabricated to produce 15° of valgus alignment, but the valgus angle of each toe was not controlled. Instead, each ray was fixed in a position similar to its natural or preferred valgus angle. The anatomical valgus angle resulting in each foot was measured through anteroposterior radiographs with use of a procedure similar to that for the dorsiflexion measurement. Radiolucent testing plates, structurally equivalent to the aluminum plates described previously, were substituted to allow for visibility of the joint.
The stance phase of the gait cycle was simulated by means of the robotic gait simulator23, consisting of a Rotopod (R 2000; Parallel Robotic Systems, Hampton, New Hampshire) and linear electromechanical actuators (Exlar, Chanhassen, Minnesota) (Figs. 3 and 4). Each foot was mounted in the simulator, and retroreflective markers were placed on the lateral malleolus, medial malleolus, and two points on the foot mounting frame representing the superior-inferior axis of the tibia. A six-camera motion analysis system (Vicon, Lake Forest, California) was used to register the position of the foot relative to the robotic gait simulator so that the appropriate robot trajectory could be determined. Custom aluminum clamps were used to attach each extrinsic tendon to its respective actuator; the Achilles tendon was coupled with use of a custom liquid nitrogen freeze clamp to accommodate its comparatively higher loading.
A gait trajectory was simulated on the basis of kinematic and kinetic in vivo gait data recorded from ten healthy subjects (six female and four male individuals with a mean age of 52 ± 7 years [range, forty-one to sixty-three years] and a mean body weight of 70.2 ± 10.2 kg) who performed four to five gait trials each. A force plate (Kistler Instrument, Amherst, New York) mounted perpendicular to the simulator platform moved in order to replicate the relative in vivo trajectory of the tibia with respect to the ground. Simultaneously, the linear electromechanical actuators applied variable tension to the nine extrinsic foot tendons. The tendon force trajectories were estimated from the physiologic cross-sectional area (cm2), the maximum specific isometric tension (N/cm2), and the electromyographic activity (percentage of maximum voluntary contraction) of the muscle24-28.
Gait was simulated on the basis of 50% of the mean body weight (365.9 N) of the cadaver specimens and approximately one-fifteenth of the self-selected gait velocity of the in vivo subjects (a ten-second stance phase of the gait cycle). The ground reaction force and plantar pressure data were recorded with use of the force plate and the Pedar pressure-measuring insole (Novel, Munich, Germany), respectively. The insole was positioned to remain in contact with the medial aspect of the forefoot throughout the stance phase. Load cells (DSM; Transducer Techniques, Temecula, California) placed in series with the actuators recorded the tendon forces, and video-capture devices monitored the gait trials from both sagittal and dorsal perspectives. Saline solution was periodically applied to the tendons for tissue preservation. During the course of the study, each foot specimen underwent one to two freeze-thaw cycles. All gait trials for a given specimen were completed during one thaw period.
The fidelity of each gait trial was determined through analysis of the tendon force control and ground reaction force behavior. Video footage and manual inspection were used to assess tendon attachment, evidence of bone fracture, and rigidity of plate fixation. Through a manual, iterative process, the following simulator parameters were tuned to increase fidelity relative to the target trajectories: the gain on the Achilles tendon force trajectory (20% to 33% of estimated physiologic force), a robot trajectory offset (?x) along the superior-inferior tibial axis, and an ankle dorsiflexion angle at the point of heel strike.
A trial was considered successful if it satisfied the following criteria: the Achilles tendon force trajectory visually tracked its target force (a post hoc analysis demonstrated that the root mean square error was <3% of peak force), non-Achilles tendon force trajectories performed with <10 N root mean square error, the vertical ground-reaction force values at the first and second peak exhibited less than ±10% error, and the arthrodesis model exhibited no signs of laxity or fracture.
Peak vertical ground-reaction-force error was calculated as the percent error between the in vivo and in vitro peak magnitudes. Tendon force error was determined through a root mean square calculation between the in vivo and in vitro force data points. Three successful gait trials were obtained for each foot-plate combination for a total of ninety-five trials, and the testing sequence descended sequentially through the range of dorsiflexion (i.e., 30°, 25°, 20°, 15°, and 10°). The sample size was selected on the basis of the availability of specimens and testing resources.
The boundaries of contact between the plantar surface of the foot and the pressure-measuring insole were constant throughout all gait trials for a given specimen; this was confirmed through video review. The contact regions of the first metatarsal head, the second metatarsal head, and the hallux were mapped, and quadrilateral masks were created to define these regions in Novel Database software. Given variations in foot size, the mask size was unique to each foot. Video footage was also reviewed to verify that motion at the second metatarsal head was unrestricted by the adjacent fusion apparatus at the first metatarsophalangeal joint.
Two variables of interest were calculated within each mask region. Peak pressure was defined as the highest magnitude recorded by any sensor circumscribed by a mask. The pressure-time integral was calculated by summing the peak pressure values across all frames in a trial (50-Hz sampling rate).
Statistical Methods
From a statistical standpoint, the design of this study was multilevel: variability in the pressure measures (peak pressure and pressure-time integral) could arise because of variability between feet or variability in repeated measures within feet. Because of this complexity in design, linear mixed-effects models were used to estimate regression lines characterizing the relationship between pressure (peak pressure or pressure-time integral, under the hallux or the metatarsal head) and dorsiflexion angle and to test for a significant association between pressure measures (the dependent variables) and dorsiflexion angle (the independent variable). These models not only yield an estimate of the slope of change in pressure per unit increase in dorsiflexion angle, but they also model variability in pressure across feet separate from the variability in pressure within feet. They also account for the variability in the estimate of slope across feet. The dorsiflexion angle that minimizes pressure was estimated by calculating the intersection of the hallux and metatarsal head regression lines estimated in the mixed-effects models described above and with use of a bootstrap to estimate 95% confidence intervals.
Source of Funding
The Department of Veterans Affairs Rehabilitation Research and Development Service and the University of Washington Medical Student Research and Training Program played no role in this investigation.
Data from six feet demonstrated that peak pressure and pressure-time integral under the hallux were negatively correlated with dorsiflexion angle, while peak pressure and pressure-time integral under the metatarsal head were positively correlated with dorsiflexion angle (p < 0.004 in all cases) (Table I; Figs. 5-A through 6-B). The angle at which the regression lines intersected was 24.7° (95% confidence interval, 19.3° to 30.5°) for peak pressure and 21.3° (95% confidence interval, 17.2° to 27.0°) for pressure-time integral. The anatomical valgus angle averaged 14.0° ± 4.7° across the specimens.
To assess load transfer to the lesser metatarsals, an analysis of pressure data for the head of the second metatarsal was performed. For the number of cadaver feet that we tested, there was no relationship between peak pressure under the second metatarsal head compared with the dorsiflexion angle (p = 0.17) or for pressure-time integral under the second metatarsal head compared with the dorsiflexion angle (p = 0.067). Pressure data for the heads of the third through fifth metatarsals were not obtained.
The peak in vitro vertical ground-reaction forces had a mean error of 5.9% ± 4.3% and 3.2% ± 2.3%, compared with the first and second in vivo peaks, respectively (Fig. 7). All non-Achilles tendon forces satisfied the fidelity criterion of <10 N root mean square error (Table II). Achilles tendon force tracking was assessed visually, and post hoc analysis demonstrated that the root mean square error was <3% of peak force. Thus, relative to its peak target force, the Achilles tendon performed with the greatest fidelity.
Arthrodesis of the first metatarsophalangeal joint is indicated for arthropathy of the joint4 or for patients requiring operative revision of a failed hallux valgus correction5. In our study, minimal peak and integrated plantar pressures were achieved at 24.7° and 21.3°, respectively. These values fall within the generally suggested dorsiflexion range4,5,9-17 of 20° to 25°.
Pressure under the first metatarsal head increased with dorsiflexion, and the pressure under the hallux was inversely related. We are unaware of previous studies that have examined the angle-pressure relationship in postoperative gait. Discrete values have been reported, however, from in vivo measurements. Gibson and Thomson9 reported a mean dorsiflexion angle of 26° ± 7° and a mean valgus angle of 9° ± 6° at one year following arthrodesis. Pressure was reported at two years following surgery, averaging 23.0 ± 10.6 N/cm2 under the first metatarsal head. When a dorsiflexion angle of 26° was input into our linear mixed-effects model for peak pressure, we calculated 24.6 N/cm2 under the first metatarsal head (a difference of 1.6 N/cm2). It is important to note that, in our study, foot specimens experienced 50% body weight, unlike the study by Gibson and Thomson in which subjects were allowed to load their feet to full body weight (normal walking). However, the relationship between body weight and plantar pressure has been shown to be variable29, and we may not be able to attribute the similarity in the measurements to the fidelity of our model.
DeFrino et al.19 reported a mean dorsiflexion angle of 15.7° ± 6.9° and normalized pressures of 1.13 and 0.51 N/cm2·kg under the hallux and first metatarsal head, respectively. Using the mean 50% body weight (365.9 N) in our simulation, we normalized our regression equations to control for body weight and perform a comparison. Inputting 15.7° into the normalized linear mixed-effects model produces peak pressure values of 0.85 N/cm2·kg (a difference of 0.28 N/cm2·kg) and 0.36 N/cm2·kg (a difference of 0.15 N/cm2·kg) under the hallux and first metatarsal head, respectively. Although such literature values are discrete, their similarity to our calculated outputs may support the applicability of our simulation to clinical practice. Again, we acknowledge the variability between body weight and peak plantar pressure29.
Other studies of dorsiflexion angle and plantar pressure do not readily allow for comparison with our findings, as they examined qualitative changes10,11,18. Namely, three follow-up studies of patients undergoing metatarsophalangeal joint arthrodesis utilized a Harris mat to characterize pressure distribution in feet with mean dorsiflexion angles ranging from 22° to 29°.
The fidelity of the tendon actuation system is vital to recreating in vivo leg muscle activity. Gait simulation was performed at 50% of body weight; thus, the target ground reaction force and non-Achilles tendon forces were scaled to one-half of those derived from our in vivo source (Fig. 7). Because the Achilles tendon largely determines the peak vertical ground-reaction force in the latter part of the stance phase, it was tuned to 20% to 33% of its physiologic loading to achieve an in vitro second vertical ground-reaction force peak within 10% of that of the in vivo value. This methodology assumes that determining the appropriate Achilles tendon force gain by means of achieving an accurate vertical ground reaction force is more robust than an analytical approach based on cadaver anthropomorphic data. We did not attempt to control the temporal characteristics of the vertical ground-reaction force curve.
There are several potential limitations to our study, some of which are inherent to our cadaver model. The intrinsic muscles of the foot specimens were not active; such muscles only contributed to osseous motion through passive force-length properties. Accordingly, the sacrifice of the extensor hallucis brevis tendon was not believed to introduce variability in our gait simulations. Another potential source of variability in our study was the lack of valgus angle control between feet. We did not study valgus angle as it relates to plantar pressure changes within feet, so we did not include valgus angle as a variable in our statistical model. In a finite element model of metatarsophalangeal joint arthrodesis20, plantar pressure increased by 446 kPa over a valgus angle range of 10° to 30°. With regard to load transfer to the lesser metatarsals, the small number of feet in our study limits the interpretations of the plantar pressure results under the second metatarsal head. In particular, the lack of association between the pressure-time integral under the second metatarsal head compared with the dorsiflexion angle (p = 0.067) may be due to a type-II error. Another limitation lies in the fact that the motion analysis system was not used to measure the kinematics of the metatarsophalangeal joint during gait. Thus, despite finite element analysis of the arthrodesis model and inspection of the joint during and after gait trials, the possibility remains that osseous deformation occurred during gait.
Concerning our gait parameters, the simulation at one-fifteenth of physiologic velocity is suboptimal for observing rate-dependent changes in plantar pressure. The discrepancy between simulated body weight during gait (50%) and that for radiographic imaging (25%) could have resulted in a nondifferential misclassification of dorsiflexion angle measurements. In addition, the simulator was not driven by kinematics and kinetics from postoperative gait in patients after arthrodesis, which could differ from those of normal subjects. Brodsky et al. quantified gait behavior in subject limbs before and after metatarsophalangeal joint arthrodesis, noting a decrease in step width postoperatively30. DeFrino et al. compared the gait of the involved limb in patients who had an arthrodesis with that of the contralateral, normal limb and with the gait of healthy control subjects19. The limbs that had been treated operatively showed a decrease in step length after surgery, although the authors found no significant change in hip or knee kinematics among all limbs. The range of motion at the ankle showed no significant difference between healthy and treated limbs in either study.
Our recommendation for a dorsiflexion fusion angle requires several assumptions. We quantified the mid-diaphyseal dorsiflexion angle to account for interindividual differences in arch height. If intraoperative imaging is not used to quantify the dorsiflexion angle, the alternative may be to measure the declination angle with use of surface landmarks. Thus, the applicability of our findings may be limited to practitioners who quantify the mid-diaphyseal dorsiflexion angle at the onset. In addition, our model represented a fully healed joint. If yielding of the fixation hardware is expected postoperatively (i.e., due to the patient walking), a more conservative dorsiflexion angle may be warranted. Such yielding may explain the discrepancy between recommended dorsiflexion angle values as stated in the methodology of some studies and the angles reported postoperatively in those studies10,12. Alternatively, without angle measurements at the onset of surgery, discrepancies may occur intraoperatively. In addition, the simulator was not driven by postoperative inputs, and it is possible that the kinematics of the gait of patients after arthrodesis could differ from those of normal subjects. However, a previous motion analysis study of patients before and after first metatarsophalangeal arthrodesis found no significant difference in the hip, knee, or ankle kinematics after surgery19. Finally, it should be noted that the selection of an optimum dorsiflexion angle may be more nuanced than selecting a dorsiflexion angle at which cumulative pressures are reduced. Considerations should be made for individual patient characteristics (i.e., a previous toe deformity) that influence plantar pressure distribution during the stance phase.
Our findings support the hypothesis that an angle-pressure relationship exists and that it is inversely related for the hallux and the metatarsal head. Utilizing a cadaver model of metatarsophalangeal joint arthrodesis and dynamic gait simulation, we determined that peak pressure and pressure-time integral under the forefoot were minimized at 24.7° and 21.3°, respectively. Accordingly, a target range of 20° to 25° for a dorsiflexion fusion angle may be appropriate for patients undergoing arthrodesis of the metatarsophalangeal joint.