The quality assessment of GA pattern studies is represented in Table 1. Among the case–control studies, only two studies, namely Adair et al. (40) and Cimolin et al. (34), did not reach a sufficient quality score because of a lack of adequate matching groups and identifying confounding factors. In addition, some concerns were resolved about statistical analysis methods in one of these studies (40), such as the use of discrete variable analysis for continuous variables and the lack of any correction method. The other studies achieved a Good (5, 34, 37–39, 41, 42, 44–47) or Fair (6, 7, 9, 35, 43, 70) quality judgment, associated with a low risk of bias, but presented some limitations. Bonnefoy et al. (39) presented groups different in age, Klebe et al. (6) had some incomplete data (no SIAS marker in 6 patients-50% of the sample, for safety device use), and Wolf et al. (37) declared that comparison of more homogeneous subgroups might be possible but with limited statistical power and the risk of additional bias effects. Some authors (6, 35, 39, 43, 46) did not provide a clear description of the subject characteristics, comparability, matching, or recruitment. Regarding the identification and the management of confounding factors, the authors mostly instructed the control subject to walk at a low comfortable speed (38, 42, 44, 46), to avoid any potential bias due to speed differences between groups and to ensure that the general characteristics of gait could be compared. The subject match was also often done based on age. Some authors (40, 41, 44) did not allow the use of aids to perform gait, resulting in a restriction of the sample size.

The RoB of GA studies was assessed by means of the ROBINS-I tool, as represented in Figure 2. A RoB in D1 resulted whenever the authors did not make appropriate matching between patients and control groups, in particular, considering confounding factors such as age, anthropometric data, or walking speed. For example, Bonnefoy et al. (39) presented the patient's group data including the results of different gait analyses performed by the same group of patients over the years. In the study by Adair et al. (40), the healthy control group's data were derived from another study and the information about comparability was not reported. In case reports and case series studies, no information was provided regarding D1. Only one author performed a blind matching among groups (37). A serious risk in D2 was identified in case reports (47) and case series (6, 9): because of the nature of the studies, the results might affect the recruitment. The risk in D2 was also serious if the patients were retrospectively recruited or excluded (37, 39) or recruited depending on the ability to perform gait analysis (40, 43). Studies based on a retrospective design presented a serious risk in D3. D4 was considered inappropriate in GA pattern studies, because no therapeutic intervention was considered, except gait analysis. All the studies achieved low risk in D5 because they did not report a high relevant percentage of missing data. An overall moderate risk in D6 was evidenced because the methods of gait data analysis were comparable across groups and the outcome measure was probably not influenced by the knowledge of the investigators, nonetheless, no one used any type of blinding method for the analysis. Only the study by Armand et al. (9) presented a serious risk in this domain because the authors compared GA performed with different instruments over the years. An overall moderate RoB was evidenced also in D7 because, in all studies, the outcome measurements and analyses were consistent with the a priori plan, though a pre-registered protocol was never reported. Overall, according to the ROBINS guide flowchart, eight studies presented a serious risk and 11 a moderate risk of bias.

The quality of intervention studies is represented in Table 2. NHI scale for pre-post non-controlled studies was used for retrospective observational cohort studies (54, 57) and a randomized pre-post-intervention study (53). JBI case–control scale was used to assess the study by Marsden et al. (71) in which the authors compared the effect of different Functional Electric Stimulation (FES) stimulating patterns between HSP patients and controls providing an inter- and intragroup analysis. Denton et al. (53) randomized the order of presentation of two types of intervention (cooling or warming) in a crossover design and analyzed the pre-post effect, including comparison with controls. Therefore, to avoid bias in reporting results, this study was assessed either with pre-post or with RCT tools for quality assessment.

In the NHI scale, the study questions were clearly stated in the title, abstract, or text. Among pre-post studies, the main limitations were inadequate representation of the population of interest, whenever only pure HSP phenotype was included, small sample size with the lack of power calculation, and no-blind design. Item 12 was considered non-applicable for this type of study population. In the final quality judgment, three studies were Good, six were Fair, and one was Poor.

Among case reports, only two did not reach a sufficient score (59, 60). The most frequent limitation was the lack of declaring the presence or the absence of collateral effects (58, 59, 61, 62, 64); Dan et al. (59) and Klebe et al. (60) did not provide a clear description of the patient's characteristics.

Among RCT (crossover design), one study reached an Excellent rating (10/11) (69) and four studies had a Good quality rating (7/11). Only one study (66) presented a baseline comparison between the crossover groups but did not implement any type of blind procedure. In most of these studies, the operator who administered the treatment was not blinded, whereas the patients and the outcome assessors were blind. In 50% of the studies, the management of missing data was addressed.

Limited to crossover RCTs, the RoB was assessed by means of the ROB2 tool (see Figure 3).

Relative to D1, Denton et al. (66) gave no information about concealed allocation, Ardolino et al. (68) used an alternate allocation design which is considered an incorrect randomization process (72). Some concerns were attributed to the study by Ardolino et al. (68) relative to the D1b domain because the authors did not specify whether the number of participants allocated to each of the two groups corresponded or not, even if they took into account sufficient time to carryover the effects. Denton et al. (66) and Antczak et al. (67) presented RoB in D2 because no information was given regarding the possible influence of unblinding treatment providers on the outcome, even though patients and evaluators were blind. Some concerns in D4 emerged in the study by Antczak et al. (67) because the evaluation of the datasets was not blindly executed. Regarding D5, all the authors referred the study to a pre-specified trial protocol, except for Denton et al. (66). Finally, only Diniz de Lima et al. (69) presented a low RoB.

All patients fulfilled the diagnostic clinical criteria for HSP according to defined criteria (2, 13, 20). The number of included HSP patients ranged across studies from 6 to 50 for the case–control studies and from 1 to 6 for the case series studies. A total of 341 HSP patients were included across all the 19 studies. No explicit differentiation was done among HSP patients with early or late onset—with onset predominantly above or below 35 years according to Harding (13). Nonetheless, nine studies focused on children and young subjects (named median age <18 years old). Most studies considered pure forms of HSP. Very few patients presented complicated forms, including urinary disturbances (6), intellectual deficit, ataxia, and peripheral neuropathy (7).

Characteristics of included studies are represented in Supplementary Table 2.

Since the included studies considered either minors (mostly compared to spastic diplegic subjects) or adults (compared to stroke patients) as samples, the results are presented separately.

The results of the included studies are represented in Table 3. With sample sizes ranging from 1 to 29, a total population of 111 patients was included in these studies.

The oldest study by Klebe et al. (6) found a typical gait pattern of patients with sporadic or HSP, compared to healthy subjects, consisting of reduced speed, cadence, and step length; increased step width and increased variation of stride length; reduced sagittal knee range of motion (ROM), with increased minimal knee angle; reduced step height with reduced maximum hip angle; increased maximum ankle angle due to equinovarus feet; and circumduction but no significant variation of foot progression angle.

Van Beusichem et al. (7) applied the Rodda (73) gait classification system for cerebral palsy, to describe the pattern of four subjects affected by a complicated form of HSP due to de novo KIF1A mutations. All four gait classes were represented with a progression from classes I and II to III and IV at the last evaluation at 10–18 years.

Pulido-Valdeolivas et al. (5) identified six gait patterns in a group of 26 HSP subjects, aged 4–17 years; the authors compared gait analysis data among patients and healthy subjects, by means of Dynamic Time Warping (5). Pattern I, in the early phase of HSP, was “close to normal” with slightly increased stance time and double support, hip and knee flexion at initial contact (IC), and delayed peak knee flexion in the swing phase. Pattern II presented overall increased anterior pelvic tilt and hip flexion, and increased knee flexion at IC. Pattern III was characterized by knee recurvatum, with reduced and delayed peak knee flexion in the swing phase. Crouch gait corresponded to pattern IV, while constant and severe anterior pelvic tilt, with recurvatum and equinus, distinguished pattern V. Pattern VI was similar to “jump knee” pattern in CP patients (74). Spatiotemporal parameters were relatively spared for patterns I, II, and VI; while they were impaired in patterns III, IV, and V. Asymmetry was described in 27% of HSP subjects, with different patterns in right and left limbs. The authors also found a correlation between GMFCS stages and increased knee flexion at IC, pelvic rotation, and obliquity. They concluded that knee flexion and non-sagittal pelvic movements were relevant indicators of HSP progression. Overlapped polyneuropathy determined an increased range of pelvic rotation in terminal swing and increased time to peak knee flexion, and was most reported in patterns I, II, and III. Abnormal visual evoked potentials (VEPs) were more frequent in subjects classified in pattern III (knee recurvatum).

Armand et al. (9) examined the gait evolution in several subjects affected by HSP (mutations in SPG3A) from the same family. The Gait Deviation Index (75) differed among subjects, but it showed an overall tendency to amelioration from childhood to adolescence and deterioration from adolescence to adulthood.

Adair et al. (40) interestingly analyzed trunk and pelvis kinematics and found increased ROM of the trunk and pelvis in the sagittal plane, with increased posterior trunk lean and anterior pelvic tilt and increased trunk obliquity during the swing phase.

The other four studies included spastic DCP subjects in comparison with HSP and healthy controls (34, 36, 37, 39). HSP and DCP showed similar patterns, and, in both groups, sagittal kinematics could be categorized according to the classification by Sutherland and Davids (74). The principle noticeable difference was that HSP subjects presented more often and longer knee hyperextension during midstance compared to DCP (34, 36, 37). Based on findings by Bonnefoy-Mazure et al. (39), HSP presented significantly reduced gait speed compared to both controls and DCP. Another peculiar characteristic of HSP subjects was increased trunk ROM and peak angular velocity in the sagittal plane during the swing phase (37, 39). HSP presented upper limb patterns similar to healthy subjects, while DCP kept their arms symmetrically elevated, with shoulders abducted and elbows flexed (39).

The results of the included studies are represented in Table 4. The sample size ranged from 6 to 50 subjects. A total of 230 subjects were included in 10 studies; the mean age at the time of GA was 47 years (SD 2.8 years). The mean disease duration among six of these studies (6, 41–44, 46) (in the others no data were available) was 19.7 years, with a prevalence of early-onset forms compared to late-onset forms.

Serrao et al. (41) and Martino et al. (43) identified three kinematic patterns as distinctive of HSP, compared to healthy subjects: increased ROM at the hip, with normal values at knee and ankle; reduced knee and ankle ROM, with normal hip ROM; and reduced ROM at hip, knee, and ankle. A reduced ROM at the knee and ankle was described also in other studies (38, 42, 44, 46), with a decreased foot lift (46), compared to healthy subjects. An increased and premature calf muscle activity was observed both in HSP and stroke subjects, compared to controls, but the contribution of the stretch reflex was excluded (35). Marsden et al. (38) demonstrated that the shorter latency stretch-evoked plantar flexor activity correlated with the increased passive stiffness found at the gastrosoleus in HSP patients, compared to controls. Conversely, no significant difference in knee extensor stiffness was recorded, by comparing HSP and controls. A significant reduction of strength was described, in particular, at the plantar flexors and knee extensors (38). Patterns of coactivation at the electromyography (EMG) were described at dorsi-plantar flexors (41–43) and extensors–flexors of the knee (41, 42). Furthermore, mapping the motor neuron activation in the lumbosacral enlargement of HSP subjects, the activity timings in lumbar and sacral segments tended to be quasi-synchronous because of a progressive widening of the activity involving the sacral segments (43). Conversely, healthy subjects showed distinct loci of activation of sacral and lumbar segments during late and early stance, respectively. Coactivation resulted to correlate with higher energy consumption during gait, based on center of mass (COM) displacements during the gait cycle (42).

Van Vugt et al. (70) analyzed the dynamic postural instability of HSP subjects starting from the distance between the center of pressure and the center of mass (COP-COM separation) to the margin of stability (61) (MOS). The authors found a significantly lower MOS in medio-lateral direction at heel strike and midstance, and in antero-posterior direction at midstance, compared to healthy subjects. van Lith et al. (45) enquired about the anticipatory postural adjustments (APAs) at gait initiation in HSP and controls by studying the StartReact effect. Delayed APAs were observed in HSP subjects, though a starling acoustic stimulus (45) (SAS) positively affected their response, by reducing the activation delay of tibialis anterior (TA) and rectus femoris (RF), close to controls' values. Conversely, the soleus (SO) inhibition was not accelerated upon administration of the SAS.

Finally, lower velocity (35, 42, 43, 70), lower cadence (38, 70), longer double support phase (70), increased step width (41, 42), and increased lateral flexion of the trunk were reported (70). Contrasting data emerged regarding step length and stance duration, being reduced or similar to controls (41–43).

Included intervention studies focused on adult subjects and no study was found including minors. Population characteristics are summarized in Table 5. Most studies considered pure forms of HSP. Very few patients presented complicated forms, including ataxia, peripheral neuropathy, retinopathy, and epilepsy (53). The methods and results of these studies are reported in Table 6.

Seven studies researched botulinum toxin injections to reduce spasticity (49, 52, 54–57, 69). Four were cohort prospective studies (49, 52, 55, 56), one was a double-blind randomized crossover study (69), and two were retrospective studies (54, 57). Xeomin (54, 56, 57), Prosigne (69), Dysport (52, 55, 57), and Botox (49, 57) were used (whenever indicated, dilution was 2 to 5 ml). The most frequently injected muscles were gastrocnemius, soleus, adductors magnus and longus, and gracilis (52, 54, 55, 69). One study included tibialis posterior (49). Two studies extended injections to other targets: hamstrings and quadriceps (56, 57); quadratus lumbi, tibialis anterior, flexor digitorum and hallucis, and extensor longus hallucis (56). The sample size ranged from 15 to 55 subjects. A total of 170 subjects were included in the seven studies, of which 98 were male subjects. An overall synthesis of age range was not feasible because data were differently reported as mean, median, or range values, but all patients were over 18 years old. After the injections, self-administered daily stretching (10 min for 2–3 times) (52, 56) or physiotherapy (49, 54, 57) was prescribed. The follow-up ranged from 8 weeks to 5 months (49, 52, 54, 56, 57, 69), only the study by Servelhere et al. (55) did not provide any follow-up assessment. Studies reported a transient reduction of spasticity according to the Modified Ashworth Scale (MAS) and of the injected muscles' strength according to the Medical Research Council (MRC) scale, in the short term. Both receded at 4–5 months follow-up assessment. Furthermore, an increase in range of movement (ROM) within 3 months after injection was reported as increased dorsiflexion, knee flexion, and hip abduction depending on the targeted muscle. Short-term improvement in gait velocity was reported by all studies (49, 52, 54, 56), except Servelhere et al. (55) and De Lima et al. (69). No significant differences were demonstrated at functional tests, except by Paparella et al. (57). This study reported significant improvements at the following tests, 3 months after botulinum and intensive physiotherapy, in 18 subjects: Spastic Paraplegia Rating Scale (SPRS), Walking Handicap Scale (WHS), 10-m walking test (10MWT), 2-min walking test (2MWT), Timed UP and Go test (TUG), the visual analogical scale (VAS), and numeric rating scale (NRS) which assessed the perceived quality of life and pain.

Transient side effects were reported in 19 subjects: muscle strength reduction (52, 55, 69), bruise, transient pain, paresthesia in the site of the injection (69), impairing gait quality (55, 69), sleepiness, and blurred vision in one subject (55). Paparella et al. (57) denied adverse effects.

Four studies researched intrathecal baclofen to reduce spasticity (59–61, 63). Three were case reports (59, 61, 63) and one was a retrospective cohort study (60). Gait analysis at a self-chosen comfortable speed was recorded before and after intrathecal bolus testing (59, 60) or before and after pump implantation. Increased gait velocity and step length were reported by all authors. Klebe et al. (60) described improvement in 5 patients over 10; among them, 2 subjects refused pump implantation because they experienced weakness and unsteadiness. Dan et al. (59) showed that ITB normalized the planar covariation of elevation angles of the thigh, shank, and foot over the gait cycle, thus improving the coordination of the lower limb and reducing mechanical energy expenditure. Heetla et al. (63) reported the reduction of spasticity using MAS, without strength loss and improvement at TUG, which lasted 6 months after implantation. Molteni et al. (61) observed a reduction in the slope of the moment–angle curve of the ankles, which lasted 2 years after pump implantation. The overall baclofen dose range was 25–108 μg, and the overall number of patients involved was 13.

Functional electrical stimulation (71) was enquired by two studies. One case report by Pease et al. (58) reported improvements in gait velocity and knee extension in the stance phase, after FES on the quadriceps and anterior compartment of the leg. Marsden et al. (71) examined a cohort of 11 long-term users of FES (at least 6 months) with and without stimulation and compared them with matched controls. With stimulation (mainly at dorsiflexion, and in some cases also at hip abductors and extensors), an increase in gait velocity and dorsiflexion torque was reported. Long-term follow-up was missing.

One study (65) explored the effect of Electrical Twitch Obtaining Intramuscular Stimulation over the low back and gluteal area, in a mixed population, including one HSP adult. The patient experienced increased speed and reduction of falls and back pain.

Ardolino et al. (68) presented a double-blind, randomized, crossover, and sham-controlled study about anodal transcutaneous spinal Direct Current Stimulation delivered over the thoracic spinal cords (T10–T12). Eleven HSP subjects were involved. They maintained their usual pharmacological treatment but no other intervention (i.e., physiotherapy) was performed during the trial. A significant reduction of spasticity was observed at the Ashworth scale, in particular, at knee extensors and hip flexors, 2 months after treatment. No other functional outcome was improved.

De Souza et al. (8) reported a subject who underwent chronic spinal cord stimulation (SCS) implantation in the posterior epidural space of T11–T12. Alternating ON/OFF phases allowed studying the effect of SCS: improvements in muscle strength and spasticity and at SPRS were reported in ON phases.

Antczak et al. (67) researched the effect of repetitive transcranial magnetic stimulation (rTMS) by means of a blinded, randomized, crossover, and sham-controlled study. Fifteen patients were enrolled, with one dropping out due to a seizure that occurred during a stimulating session. Other adverse effects were headache (several subjects) and sleeplessness (one subject). Usual physiotherapy and oral drugs (59) were provided during the trial. The strength of the proximal and distal muscles of the lower limbs increased, and the spasticity of the proximal muscles decreased. Nonetheless, no functional improvements were observed at the TUG and the 10-m walk test (10MWT).

Two studies (51, 64) researched robot-assisted gait training with partial body weight support: one case report (64) and a cohort study (51) involving 13 pure HSP patients. The treatment lasted 6 weeks. The case report (64) included physiotherapy and overground walking, while the study by Bertolucci et al. (51) provided a gradual reduction of the robotic guidance force and increased workload. An overall improvement in functional tests was observed, with non-significant change in strength, spasticity, and pattern of gait. In the cohort study (51), the improvement was maintained at a 2-month follow-up.

The outcome of an 8-week intensive physiotherapy program including stretching, strengthening, and functional exercise, in two HSP subjects, was described by Samuel et al. (76). The authors reported an improvement in all tests after completing the intervention period: TUG, Functional Reach Test (FRT), 10mWT, and 2mWT.

Zhang et al. (50) researched gait analysis changes after a 10-week hydrotherapy program in 11 HSP subjects. A significant improvement in gait velocity was reported. A significant decrease in the transverse plane rotation of hip, knee, and ankle and an increase in hip and knee peak extension moment were reported.

Klebe et al. (48) performed an open-label study with a longitudinal follow-up at 6 months, in 22 patients treated with 60 mg of methylphenidate per day. Non-significant improvement was observed at gait analysis, MAS, or MRC, at the last assessment. Nausea and sleep disturbances were reported as collateral effects, but only one dropout was recorded, based on worsening of pre-existing urinary disturbances.

Two studies by Denton et al. (53) (randomized treatment with healthy controls) and in Denton et al. (66) (randomized crossover) researched the role of lower limb superficial heating in a total of 43 HSP subjects and 19 controls. The authors demonstrated that heating reduced spasticity, increased dorsiflexor rate of force generation and nerve conduction velocity, and slightly improved gait speed while cooling (53) induced the opposite effects. The application of heating or cooling wrap lasted 30 min.

The meta-analysis (Figures 5A–F) included three studies (52, 56, 69) that presented an overall heterogeneity (Q = 9.35, p = 0.009 in t₀; Q = 10.40, p = 0.005 in t₁), so a random effect model was used.

The estimated means were 1.217 in t₀ (n = 94; CI [1.019; 1.416]), 1.224 in t₁ (n = 89; CI [1.015; 1.432]), and 1.346 (n = 37; CI [1.219; 1.473]) in t₂.

The change of the maximum gait velocity was not significant from t₀ to t₁ (SMD = 0.003; CI [−0.287, +0.293], p = 0.983) nor from t₀ to t₂ (SMD = 0.073; CI [−0.374, +0.520], p = 0.749). Therefore, the change from t₁ to t₂ was not significant (SMD = 0.046; CI [−0.409, +0.517], p = 0.843).

This non-significance may depend on the great variability of the individual data, but also on the peculiarities of Diniz De Lima et al. (69), in which the average of the maximum gait velocity was considerably lower than that of the other two studies, both in t₀ and in t₁. The exclusion of Diniz De Lima et al. (69) from the meta-analysis made the other studies homogeneous (Q = 0.05, p = 0.817), but the changes in the maximum gait velocity remained non-significant.

For the comparison of the comfortable gait velocity, four studies were included (49, 52, 56, 69) in the meta-analysis (Figures 6A–F). A random effects model was used, due to the substantial heterogeneity among the studies (Q = 13.26, p = 0.004 in t₀; Q = 23.63, p < 0.001 in t₁; Q = 21.98, p < 0.001 in t₂). The estimated means were 0.837 in t₀ (n = 109; CI [0.721; 0.953]), 0.876 in t₁ (n = 103; CI [0.721; 1.030]), and 0.922 (n = 50; CI [0.688; 1.157]) in t₂. The mean change was not significant from t₀ to t₁ (SMD = 0.108; CI [−0.163; 0.378]; p = 0.435), weakly significant in the comparison between the baseline and t₂ (SMD = 0.335; CI [−0.052; 0.723]; p = 0.089), and again not significant t₁ to t₂ (SMD = 0.025; CI [−0.365; 0.416]; p = 0.898). Also, in this case, Diniz De Lima (69) was peculiar compared to the other studies, because it presented a negative change of the comfortable gait velocity from t₀ to t₁ (SMD = −0.08; CI [−0.46; 0.30]). Meta-analysis was then repeated excluding this study. The new results showed a positive estimate of the mean change from t₀ to t₁, with a noticeably lower p-value, but in any case, non-significant at 95% (SMD = 0.297; CI [−0.086; 0.680]; p = 0.129).

Only two studies (55, 69) were included in the meta-analysis (Figures 7A, B), and in both cases, no follow-up data were present. Data presented heterogeneity in t₀ (Q = 4.66, p = 0.0308) and also in t₁ (Q = 504, p = 0.0248), so a random effects model was performed. The estimated means of SPRS were 18.948 in t₀ (n = 76; CI [14, 269; 23.626]) and 18.643 in t₁ (n = 76; CI [13.769; 23.517]), showing a substantial stability of this parameter (SMD = −0.041; CI [−0.359; 0.277]; p = 0.801).

The meta-analysis included two studies (52, 56) (Figures 8A–F) for this parameter that resulted quite homogeneous in the three time points (Q = 0.04, p = 0.849 in t₀; Q = 0.03, p = 0.852 in t₁; Q = 0.13, p = 0.721 in t₂). The estimated means were 10.495 in t₀ (n = 40; CI [9.468; 11.512]), 10.561 in t₁ (n = 37; CI [9.591; 11.531]), and 10.765 (n = 37; CI [9.733; 11.802]) in t₂. Due to the large variability of the data, the mean change was not significant from t₀ to t₁ (SMD = 0.0300; CI [−0.417; 0.477]; p = 0.896), from t₀ to t₂ (SMD = 0.056; CI [−0.391; 0.504]; p = 0.804), and from t₁ to t₂ (SMD = 0.038; CI [−0.418; 0.494]; p = 0.870).