Prospective gait analysis in patients from the French registry of glycogen storage disease type III: implications for clinical trials Authors: Jean-Yves Hogrel, Frédéric Fer, Isabelle Ledoux, François Petit, Martha Darce-Bello, Philippe Labrune, Karim Wahbi, Dalila Habes, Antoine Gardin, Marion Masingue, Pascal Laforêt, Valérie Decostre Published: 2026 ABSTRACT AbstractBackground and objectivesGlycogen storage disease type III (GSD-III) is a genetic metabolic disorder affecting liver and muscle, frequently leading to walking difficulties in adults. This study aimed to quantify walking performance and its progression in GSD-III, identify relevant outcome measures for future clinical trials, and determine predictors of gait decline.MethodsParticipants enrolled in the French GSD-III registry underwent an annual assessment of their 6-min walk distance (6MWD) combined with 3D trunk accelerometry. Muscle strength for flexion and extension of the knee and ankle, as well as Motor Function Measure (MFM) scores, were also collected.ResultsAmong registry participants, 93% were ambulant without assistance. The 6MWD of participants with GSD-III (n = 46, aged 10–49 years) was lower than that of controls (n = 53) (adjusted p < 0.001), due to reduced stride frequency and stride length/height ratio (adjusted p < 0.001). Accelerometry revealed no abnormalities beyond those related to slower walking speed. 6MWD correlated with muscle strength and accelerometry variables. In adults, total MFM score declined with age (adjusted p < 0.001), whereas 6MWD remained stable overall, despite a gradual decline in a few patients. Preliminary models suggest that lower baseline MFM sub-scores and specific trunk accelerometry variables may predict 6MWD deterioration.DiscussionThe walking impairment remains stable in most patients. MFM sub-scores and accelerometry variables may help identify the rare individuals at risk of gait decline. Unlike 6MWD, total MFM score worsened in adulthood and appears to be the most relevant muscle outcome measure for future clinical trials in GSD-III.French GSD-III Registry registration numberNCT06616545 was retrospectively registered on 27 SEP 2024.Supplementary InformationThe online version contains supplementary material available at 10.1007/s00415-026-13793-2.Keywords: Glycogen storage disease type III, Gait, Trunk accelerometry, Muscle strength Introduction Glycogen storage diseases (GSDs) are caused by pathogenic variants in genes encoding enzymes or transporters involved in glycogen metabolism. Of the 20 known types of GSDs, glycogen storage disease type III (GSD-III, Cori/Forbes disease) accounts for ~ 24% of all GSD cases, and shows an incidence of 1 case per 83,000–100,000 live births. GSD-III results from autosomal recessive mutations in the AGL gene (85 kb), encoding the glycogen debranching enzyme (GDE). Accumulation of abnormal glycogen affects both liver and striated muscle in GSD-IIIa, or only liver in GSD-IIIb. In childhood, GSD-III presents with hepatomegaly, elevated liver enzymes, short stature, and hypoglycaemia, with little skeletal muscle involvement even in GSD-IIIa [1]. Muscle weakness progressively develops in adolescence and adulthood [1–4]. Symptomatic cardiomyopathy, often left ventricular hypertrophy (LVH), occurs in about half of patients [1]. Currently, treatment is limited to individualized dietary management [1, 4, 5]. Research is ongoing into enzyme replacement [6], pharmacological [7], gene [4, 8–10], and combined pharmacological–gene therapies [11]. Identifying sensitive outcome measures of skeletal muscle involvement is essential for future clinical trials. Whole-body muscle magnetic resonance imaging (MRI) reveals age-related fatty replacement of muscle, mainly in thoracic extensors, glutei, hamstrings, and calves [12]. Myopathy worsens with age and may appear in childhood [13–15]. In adults, it manifests as generalized, symmetrical weakness, often predominant in hands and lower limbs [3, 16–18]. Lower-limb weakness and postural misalignment likely impair gait in GSD-III. The 6-min walk test (6MWT) measures the maximum possible distance (6MWD) walked on a flat surface in 6 min. It allows self-paced effort with short breaks [19, 20] and is widely used in myopathy studies and trials [21, 22]. Retrospective studies in GSD-III report impaired 6MWD [12, 15] but no prospective follow-up combining 6MWD and accelerometry has been published. Accelerometers offer a simple, non-invasive means to quantify gait [23]. The French GSD-III registry prospectively collects demographic, muscular, metabolic, cardiac, dietary, and therapeutic data from 21 centres. Accelerometry measurement was offered to patients assessed for muscle function at the Institute of Myology, Pitié-Salpêtrière Hospital, Paris. The objectives of the present study were to (i) quantify the proportions of ambulant and non-ambulant patients recorded in the French GSD-III registry, (ii) compare 6MWD, gait accelerometry, and strength between unassisted ambulant patients and healthy controls, (iii) identify accelerometric, strength, and MFM variables correlated with 6MWD, (iv) analyse age-related changes in 6MWD, gait, strength, and MFM to identify sensitive outcome measures, (v) determine baseline predictors of 6MWD decline over time. Methods Study design and participants All patients enrolled in the French GSD-III registry (NCT06616545) and followed at the Institute of Myology (Paris, France) underwent prospective annual assessments from the age of 10, including a 6MWT with trunk accelerometry and measurements of muscle strength and function. Follow-up measurements were collected between 2013 and 2023 using the same type of instruments and generally by the same physiotherapist. Patient results were compared with healthy controls assessed under the same conditions in the Accelerix (NCT01522690) and NatHis DM1 (NCT07362316) studies conducted between 2010 and 2015. All protocols were approved by the ethics committee CPP–Ile de France VI–Groupe hospitalier Pitié Salpêtrière, in accordance with the ethical standards laid down in the 1964 Declaration of Helsinki and its later amendments. All participants provided written informed consent. Six-minute walk test The 6MWT was typically conducted about 1.5 h after a meal (breakfast or lunch). Participants were instructed to walk as far as possible for 6 min along a dedicated 25 m corridor, turning around cones, with pauses allowed if needed. Participants wore their usual low-heeled shoes. The 6MWD was expressed as a percentage of the predicted distance based on gender, age, height, and, when applicable, weight. Predicted 6MWD values were derived from published models for ages 3–18 years [24], 20–50 years [25] and 50–85 years [26]. For participants aged 19 or 50, the model used for most of their follow-up was applied. Walking speed was calculated by dividing the 6MWD by the duration of the test (6 min). Trunk accelerometry A portable triaxial accelerometer (Locometrix®, Centaure Metrix, Evry, France) was placed at the lumbar level using a belt to record trunk accelerations during the 6MWT [27, 28]. Participants enrolled in the registry were permitted to use a walking aid if necessary, during the 6MWT. However, in the present study, only unassisted walk tests conducted with the accelerometer were analysed. Using custom software in LabView (National Instruments), the following gait parameters were extracted as described previously [28] from a 10-s steady-state walking segment showing maximal interstep regularity: (1) Average magnitude of trunk accelerationsThe acceleration root mean square (RMS) was calculated for each axis: anteroposterior (AP), mediolateral (ML) and vertical (V). It reflects the average magnitude of trunk accelerations during walking in each axis [29]. (1) Average magnitude of trunk accelerations The acceleration root mean square (RMS) was calculated for each axis: anteroposterior (AP), mediolateral (ML) and vertical (V). It reflects the average magnitude of trunk accelerations during walking in each axis [29]. (2) Gait regularity and symmetryAn unbiased autocorrelation procedure was performed to quantify the walk regularity and symmetry on the AP and V axes [30]. The autocorrelation of the ML acceleration signal was not analysed due to its unstable form [31]. (2) Gait regularity and symmetry An unbiased autocorrelation procedure was performed to quantify the walk regularity and symmetry on the AP and V axes [30]. The autocorrelation of the ML acceleration signal was not analysed due to its unstable form [31]. (3) From the power spectral density (PSD) of the acceleration signal along each axis, the following variables were calculated:(i) Stride frequency was determined from the AP accelerometer signal. The step frequency was sought by locating the frequency corresponding to the highest peak in the PSD zone between 0 Hz and the median frequency, where the step frequency is expected. As muscle impairment [3, 12] and gait (cf Results) are symmetrical in GSD-III, stride frequency was obtained by dividing the step frequency by 2.(ii) Stride length was calculated from the average walking speed multiplied by the average stride duration (= 1/average stride frequency).(iii) The mechanical power (W/kg) was calculated from the integral of the PSD of the acceleration signal along each axis. The total power is the sum of the power along each axis.(iv) Shock waves during walking.The shocks caused by the impact of feet on the ground are reflected in the ratio of high-frequency shock wave energy to low-frequency step energy (HF:LF). The cutoff frequency separating low frequencies from high frequencies was determined by multiplying the DSP’s lowest peak frequency by 2.5 [28].(v) Entropy is a physical measure of how disordered or unpredictable something is. The disorder or complexity of the acceleration signal in each direction was evaluated using Shannon spectral entropy [32]. (3) From the power spectral density (PSD) of the acceleration signal along each axis, the following variables were calculated:(i) Stride frequency was determined from the AP accelerometer signal. The step frequency was sought by locating the frequency corresponding to the highest peak in the PSD zone between 0 Hz and the median frequency, where the step frequency is expected. As muscle impairment [3, 12] and gait (cf Results) are symmetrical in GSD-III, stride frequency was obtained by dividing the step frequency by 2.(ii) Stride length was calculated from the average walking speed multiplied by the average stride duration (= 1/average stride frequency).(iii) The mechanical power (W/kg) was calculated from the integral of the PSD of the acceleration signal along each axis. The total power is the sum of the power along each axis.(iv) Shock waves during walking.The shocks caused by the impact of feet on the ground are reflected in the ratio of high-frequency shock wave energy to low-frequency step energy (HF:LF). The cutoff frequency separating low frequencies from high frequencies was determined by multiplying the DSP’s lowest peak frequency by 2.5 [28].(v) Entropy is a physical measure of how disordered or unpredictable something is. The disorder or complexity of the acceleration signal in each direction was evaluated using Shannon spectral entropy [32]. (i) Stride frequency was determined from the AP accelerometer signal. The step frequency was sought by locating the frequency corresponding to the highest peak in the PSD zone between 0 Hz and the median frequency, where the step frequency is expected. As muscle impairment [3, 12] and gait (cf Results) are symmetrical in GSD-III, stride frequency was obtained by dividing the step frequency by 2. (i) Stride frequency was determined from the AP accelerometer signal. The step frequency was sought by locating the frequency corresponding to the highest peak in the PSD zone between 0 Hz and the median frequency, where the step frequency is expected. As muscle impairment [3, 12] and gait (cf Results) are symmetrical in GSD-III, stride frequency was obtained by dividing the step frequency by 2. (ii) Stride length was calculated from the average walking speed multiplied by the average stride duration (= 1/average stride frequency). (ii) Stride length was calculated from the average walking speed multiplied by the average stride duration (= 1/average stride frequency). (iii) The mechanical power (W/kg) was calculated from the integral of the PSD of the acceleration signal along each axis. The total power is the sum of the power along each axis. (iii) The mechanical power (W/kg) was calculated from the integral of the PSD of the acceleration signal along each axis. The total power is the sum of the power along each axis. (iv) Shock waves during walking.The shocks caused by the impact of feet on the ground are reflected in the ratio of high-frequency shock wave energy to low-frequency step energy (HF:LF). The cutoff frequency separating low frequencies from high frequencies was determined by multiplying the DSP’s lowest peak frequency by 2.5 [28]. (iv) Shock waves during walking. The shocks caused by the impact of feet on the ground are reflected in the ratio of high-frequency shock wave energy to low-frequency step energy (HF:LF). The cutoff frequency separating low frequencies from high frequencies was determined by multiplying the DSP’s lowest peak frequency by 2.5 [28]. (v) Entropy is a physical measure of how disordered or unpredictable something is. The disorder or complexity of the acceleration signal in each direction was evaluated using Shannon spectral entropy [32]. (v) Entropy is a physical measure of how disordered or unpredictable something is. The disorder or complexity of the acceleration signal in each direction was evaluated using Shannon spectral entropy [32]. RMS and power variables were normalized to walking speed, which is known to influence lower trunk accelerations [33, 34]. Strength measurements Maximal isometric flexion and extension strength was measured in a seated position with knees flexed at 90°, bilaterally, using a specific dynamometer for the ankle (MyoAnkle [35]) and knee (Biodex 3®, Biodex Medical Systems Inc., Shirley, NY, United States). Ankle muscle strength measurements were only introduced in mid-2016, which explains missing data. The highest result from at least three trials was used for analysis and expressed as a percentage of the predicted value for gender, age, height and/or weight based on published models for ankle muscle strength [35] and knee strength for children up to 17 years of age [36] and adults between 20 and 80 years of age [37]. For participants aged 18 and 19, the model usable for most of their follow-up was applied. Motor Function Measure The MFM (Motor Function Measure) is a standardized scale used to assess motor function in individuals with neuromuscular disorders. The MFM-32 version includes 32 tasks (items) grouped into three functional domains: D1 (13 items related to standing and transfers), D2 (12 items concerning axial and proximal motor function), and D3 (7 items concerning distal motor function). The items, detailed in the MFM manual (https://mfm-nmd.org/?lang=en) [38], reflect daily activities [39]. Each item is scored on a four-point ordinal scale ranging from 0 (unable to initiate the task) to 3 (performs the task normally). The score for each domain (D1, D2, D3), expressed as a percentage, corresponds to the sum of the scores obtained by the participant on all items in that domain, divided by the maximum possible score for that domain (maximum score = 39 for D1, 36 for D2, 21 for D3) and multiplied by 100. For the MFM-32, the total score corresponds to the sum of the scores obtained on the 32 items divided by 96 and multiplied by 100. Higher percentages reflect better motor performance. Statistical analysis Given the rarity of GSD-III, no power analysis was conducted. Data processing and statistical analyses were carried out using R software (version R.4.4.0) and particularly its glmnet, pROC and ggplot2 packages. Statistical significance was set at p < 0.05. Comparison of baseline variables in ambulant GSD-III and healthy participants The baseline variables between participants with GSD-III and their healthy peers were compared using a Mann–Whitney U test for continuous variables and a chi-square test for categorical variables or proportions. To adjust for multiple comparisons and control the familywise error rate, p values were adjusted using the Bonferroni correction method. Correlations We assessed correlations among various clinical and accelerometric variables, for all visits combined, using Spearman’s rank correlation coefficients. Given the large number of tests performed in the correlation matrix, we adjusted the significance thresholds to reduce the likelihood of type I errors. Age dependence of gait, accelerometry and strength variables in GSD-III Participants were included in analyses if they had at least one complete evaluation visit with accelerometric data. Loss to follow-up was addressed by including all available data from each participant in the longitudinal analyses, regardless of the number of visits completed. Participants who discontinued follow-up were not excluded from analyses. To analyse changes in the 6MWD, accelerometric parameters, and muscle strength in relation to patient age, we employed multiple linear regression models. Adjusted p values were calculated using the Benjamini–Hochberg procedure to control for false discovery rates, ensuring that the significance of results was not due to multiple comparisons. Interaction terms (age × sex × group) were tested in the regression models to assess whether the effect of disease status varied according to age or sex. Interactions were considered significant at p values < 0.05. Search for predictors of the 6MWD decline Given that the most bothersome impacts for adults with GSD-III include on average difficulty walking, the use of walking aids or devices, and feeling different from their peers [2], we attempted to identify the baseline predictive factors of 6MWD decline. Participants with GSD-III and accelerometric recordings were divided a posteriori into 2 groups: (i) the group with a 6MWD (in %pred) decline over time (i.e., 6MWD loss of at least 1.5%pred/year of follow-up and/or 6MWD < 70%pred at the last visit, choice of cutoff values is based on our dataset), and (ii) the group with a stable 6MWD (in %pred) over time made up of the remaining participants. Continuous variables at baseline were compared between the two groups using the Mann–Whitney U test. Categorical variables were compared using Fisher’s exact test. Receiver Operating Characteristic (ROC) analysis was used to evaluate the discriminatory ability of each variable to identify participants with 6MWD decline (in %pred) vs those with stable 6MWD. The area under the ROC curve (AUC) was extracted for each variable. All variables showing a p value < 0.05 in that univariate analysis were entered into multivariable logistic regression models: the full model (M1), the clinical-only model (M2), the MFM-only model (M3) and the accelerometric-only model (M4), including all variables, clinical variables, MFM sub-scores and accelerometric variables in the modelling, respectively. Stepwise selection (“both” mode, based on Akaike Information Criterion) was applied to retain the most parsimonious combination of predictors. Model performance was assessed using the AUC, AUC-Leave-One-Out (AUC–LOO), which measures the performance of the model in cross-validation (performance on new data), sensitivity and specificity. DeLong’s test compared the AUCs between two models. Statistical significance was set at p < 0.05. Missing data Missing data were handled using complete case analysis. Participants with missing values for specific variables were excluded from analyses involving those variables, while remaining included in analyses, where data were available. Results Participants Figure 1 shows that among the 68 participants included up to October 2023 in the French GSD-III registry, 61 (90%) had their walking ability entered. Of these 61 participants (median age 24 years, range 4–66 years), 57 (93%) were able to walk without support, 1 (< 2%) had been walking with assistance in daily life since the age of 53 and 3 (5%) had been non-ambulatory since the ages of 36, 55 and 60. Among ambulatory participants without support, 46 completed the 6MWT with an accelerometer. Diagram of the data available in the French GSD-III registry in October 2023 for the analysis of 6MWD with accelerometry. N number of participants, FSHD Facioscapulohumeral muscular dystrophy Their baseline demographic and clinical data are presented in Table 1 (which also includes, for informational purposes, data not analysed from non-ambulatory or supported-walking patients). The group of participants aged 10–49 included 18 adolescents under the age of 18 and 28 adults aged 18 or over, with a balanced gender ratio (57% female). The distribution of age, height, weight, and BMI is provided in Supplementary Fig. 1. Nearly half (46%) carried homozygous mutations, and 4% had a GSD-IIIb-related variant (c.18_19del in patients 8 and 9 in Table 1). Symptom onset occurred between 1 month and 16 years (median 9 months). BMI ranged from 16–39 kg/m2, lower in teenagers (20 ± 3 kg/m2) than adults (25 ± 4 kg/m2). Left ventricular hypertrophy was present in 48%, more common in teenagers (61%) than adults (39%). Fifty-seven per cent followed a high-protein diet and 9% followed a ketogenic diet, with a higher proportion of teenagers on a diet. Approximately half of the participants regularly experienced hypoglycaemia. No cirrhosis or liver transplant was reported, though some adults had steatosis or hepatic adenoma. Muscle weakness affected the lower limbs (~ 40% of participants) more than the upper limbs (~ 10–20%) or the axial/trunk area (~ 15%). Achilles tendon contractures were more frequent in adults (25%) than in children (6%). Table 1Demographic and clinical data at baseline for participants with GSD-IIIOnly the data from the 46 ambulant participants who did not use a walking aid were analysed in this study. For informational purposes, the unanalysed data regarding non-ambulant participants or those using a walking aid are also presented in this table. All gene variants are based on NM_000642.3. The participant with ID 45 underwent enzymatic diagnostics. The participant with ID 46 was able to perform the 6MWT without assistance and was counted as ambulant without assistance but used a cane in everyday life. The patient numbers highlighted in grey indicate the 17 participants whose 6MWD (in %pred) declined on the basis of an average decrease of at least 1.5%pred/year and/or a 6MWD < 70%pred at their last visit as described in §3.5 on the predictive variables for progression of 6MWD.ND not determined. 0: absent, 1: present. BMI Body mass index, F Female, M Male, LVH Left ventricular hypertrophy, LL Lower limb, Several participants were siblings: 31 and 33; 36 and 37; 26 and 27; as well as 38, 39 and 40. Participants 18 and 20 are father and son, respectively. The high-protein diet consisted of 2.5 to 3g of protein/kg/day, while the ketogenic diet was a modified Atkins diet Demographic and clinical data at baseline for participants with GSD-III Only the data from the 46 ambulant participants who did not use a walking aid were analysed in this study. For informational purposes, the unanalysed data regarding non-ambulant participants or those using a walking aid are also presented in this table. All gene variants are based on NM_000642.3. The participant with ID 45 underwent enzymatic diagnostics. The participant with ID 46 was able to perform the 6MWT without assistance and was counted as ambulant without assistance but used a cane in everyday life. The patient numbers highlighted in grey indicate the 17 participants whose 6MWD (in %pred) declined on the basis of an average decrease of at least 1.5%pred/year and/or a 6MWD < 70%pred at their last visit as described in §3.5 on the predictive variables for progression of 6MWD. ND not determined. 0: absent, 1: present. BMI Body mass index, F Female, M Male, LVH Left ventricular hypertrophy, LL Lower limb, Several participants were siblings: 31 and 33; 36 and 37; 26 and 27; as well as 38, 39 and 40. Participants 18 and 20 are father and son, respectively. The high-protein diet consisted of 2.5 to 3g of protein/kg/day, while the ketogenic diet was a modified Atkins diet Comparison of baseline variables in ambulant GSD-III and healthy participants Supplemental Fig. 2 shows 3D-axis accelerometric recordings of a healthy subject and a subject with GSD-III matched for age and gender. In Table 2, baseline gait and accelerometric data of ambulant GSD-III participants were compared with healthy controls. The groups were comparable in gender, age, height, weight, and BMI (Table 2, Supplementary Fig. 1). The 6MWD (in meters and %pred) and walking speed were significantly lower in patients with GSD-III, mainly due to reduced stride frequency. The mean ± standard deviation of their 6MWD was 76 ± 16%pred (Table 2). Consequently, stride length, derived from walking speed and frequency, was also significantly shorter, both in metres and normalised to height. RMS and power values on all axes were lower in GSD-III, except when normalised to walking speed or total power. Despite slower gait in GSD-III, regularity, symmetry, and entropy were similar to controls. Although HF and LF integrals of the acceleration PSD were reduced in GSD-III, the HF/BF ratio on each axis was unaffected. Correlation matrix heatmap of gait, accelerometry and strength variables in GSD-III. Due to the number of tests in the matrix, the significance thresholds have been lowered to 10–6 (***), 10–5 (**) and 10–4 (*). RMS unit is gravity (g). Strength and 6MWD are expressed in percentage of predicted value (%pred). Table 2Comparison of baseline demographic, gait and accelerometric variables in ambulant GSD-III and healthy participantsVariablesCTRL (n = 53)GSD-III (n = 46)Ratio of meansP valueAdjusted p valueProportionMeanSDMedianMinMaxProportionMeanSDMedianMinMaxGSD3/CTRLDemographicsGender (%M)57430.2711.000Patients ≤ 18 yrs (%)34390.7461.000Patients > 18 yrs (%)66610.7461.000Age (years)—All patients25122395724122310490.4701.000Height (cm)16614167130189161181681191880.3401.000Weight (kg)6018602592632265221100.5631.000BMI (kg.m−2)2142115302352316390.0511.000Gait6MWD (m)669716495388314891045182006770.73 < 0.001 < 0.0016MWD (% predicted)1041210478124761677301020.73 < 0.001 < 0.001Speed (m.s−1)1.860.201.801.492.311.360.291.440.561.880.73 < 0.001 < 0.001AccelerometryStride frequency (Hz)1.190.111.171.031.471.040.101.030.831.320.87 < 0.001 < 0.001Stride length (m)1.570.151.541.282.031.300.251.300.601.810.83 < 0.001 < 0.001Stride length:height0.950.080.950.811.150.810.150.840.341.040.86 < 0.001 < 0.001RMS_AP (g)0.400.080.370.270.540.310.070.300.180.490.77 < 0.001 < 0.001RMS_ML (g)0.360.120.340.180.750.250.080.240.100.460.69 < 0.001 < 0.001RMS_V (g)0.520.110.520.310.740.430.120.430.150.650.82 < 0.0010.011RMS_AP: Speed (g.s.m−1)0.210.040.210.150.290.230.060.230.140.480.1321.000RMS_ML: Speed (g.s.m−1)0.190.050.180.100.330.190.060.170.090.430.5351.000RMS_V: Speed (g.s.m−1)0.280.050.270.190.410.320.070.300.190.590.0080.211Interstride regularity_AP0.800.140.850.400.960.810.110.830.480.940.9361.000Interstride regularity_V0.900.100.930.590.980.920.050.930.801.000.3981.000Interstep regularity_AP0.760.110.780.490.940.730.120.760.410.900.1551.000Interstep regularity_V0.890.080.920.650.980.880.090.900.500.970.3311.000Gait symmetry_AP0.970.110.960.771.280.900.110.940.641.080.0491.000Gait symmetry_V0.990.050.990.911.200.950.080.960.601.040.0040.122Power_AP (W/kg)7.883.236.743.4915.784.862.234.541.1711.590.62 < 0.001 < 0.001Power_ML (W/kg)6.965.225.621.6227.623.382.123.020.4010.760.49 < 0.001 < 0.001Power_V (W/kg)14.115.5413.444.8025.819.895.059.121.0821.340.70 < 0.0010.011POWER_TOTAL (W/kg)28.9512.3827.6111.6362.0518.138.4317.073.0940.640.63 < 0.001 < 0.001Power_AP: Power total0.280.050.270.180.420.280.060.260.170.450.5961.000Power_ML: Power total0.230.070.210.090.450.190.080.170.070.450.0050.134Power_V: Power total0.500.080.490.350.670.530.100.550.180.720.0110.258Power_AP: Speed (W.kg−1.s.m−1)4.181.503.592.087.423.551.383.561.158.070.0741.000Power_ML: Speed (W.kg−1.s.m−1)3.632.413.050.9411.962.491.432.140.476.660.0050.135Power_V: Speed (W.kg−1.s.m−1)7.502.707.463.1214.167.073.087.071.5214.860.4261.000Power_TOTAL: Speed (W.kg−1.s.m−1)15.315.6614.046.9827.3313.115.0412.163.6428.300.0941.000Entropy_AP3.520.503.452.744.743.390.493.282.304.510.1961.000Entropy_ML3.280.673.232.144.643.540.543.542.284.510.0350.764Entropy_V2.520.462.431.683.672.350.422.321.533.600.1541.000HF:LF_AP0.930.400.870.272.080.730.350.670.221.660.0120.285HF:LF_ML1.451.511.060.238.791.491.021.190.194.660.1881.000HF:LF_V0.360.210.310.081.180.380.290.280.061.230.6661.000 Comparison of baseline demographic, gait and accelerometric variables in ambulant GSD-III and healthy participants Correlations between gait, accelerometry and strength variables in GSD-III Figure 2 shows correlations between 6MWD, accelerometric variables (191 recordings across all visits), MFM, and muscle strength. 6MWD (%pred) correlated with MFM D1, D3, and total, but not D2, and with stride frequency, stride length/height ratio, RMS, and power. The 6MWD(%pred) was positively correlated with vertical regularity between steps and between strides, and negatively correlated with vertical entropy and HF/LF ratio. Muscle strength (%pred) correlated with MFM D1, total MFM, and 6MWD, except the ankle dorsiflexors (no 6MWD link). All strength measures correlated with stride length/height ratio. For knee/ankle flexion–extension, right and left strength were highly correlated and their ratios ≈ 1 ± 0.2 are consistent with gait symmetry. Therefore, mean bilateral strength was used for further analyses. Age dependence of gait, accelerometry and strength variables in GSD-III The median follow-up duration was 5.4 years (range: 0–9.5 years). Participants completed a median of 5 evaluation visits over the study period. The total cumulative follow-up time was 228.2 patient-years. To identify variables most affected by disease progression, the mean change percentage and age dependence of each variable are shown in Table 3. The mean change percentage represents the average annual variation with age, expressed as a percentage of the overall mean. No interaction terms were tested significant. Subgroup analyses were performed descriptively by age and, therefore, disease progression. Table 3Age dependence of gait, accelerometry and strength variables in non-assisted ambulant participants with GSD-IIITeenagers (n = 20)Adults (n = 29)All (n = 46)Mean change percentageP valueAdjusted P valueMean change PercentageP valueAdjusted P valueMean change PercentageP valueAdjusted P valueMFM D1 (%)−0.20.2531.000−1.6 < 0.0010.002−1.0 < 0.001 < 0.001MFM D2 (%)0.10.5181.000−0.10.5221.000−0.10.5421.000MFM D3 (%)0.50.0080.298−0.30.0090.381 < 0.10.9371.000MFM total (%)0.10.3521.000−0.6 < 0.0010.011−0.30.0020.0786MWD (%pred)0.10.7641.000−1.40.0060.257−0.80.0220.635Speed (m.s−1)0.90.0741.000−1.70.0010.053−0.60.0621.000Stride.frequency (Hz)−1.5 < 0.001 < 0.001−0.50.0160.641−0.9 < 0.001 < 0.001Stride length (m)2.2 < 0.001 < 0.001−0.70.0321.0000.40.1071.000Stride length:height−0.70.0240.792−0.70.0281.000−0.7 < 0.0010.028interstride regularity AP1.670.0010.054 > -0.10.9781.0000.60.0801.000interstride regularity V0.50.0631.0000.10.2861.0000.30.0290.722Symmetry AP−0.20.6531.000−0.4280.3081.000−0.40.1911.000Symmetry V−0.20.3661.0000.1480.2471.000 > −0.10.8221.000RMS AP (g)−2.10.0421.000−2.320.0291.000−2.20.0020.090RMS ML (g)−2.70.0321.000−2.170.0030.156−2.6 < 0.0010.010RMS V (g)−2.80.0080.298−1.20.2771.000−1.90.0100.329RMS:Speed AP (g.s.m−1)−3.2 < 0.0010.015-0.90.4641.000−1.70.0260.668RMS:Speed ML (g.s.m−1)−3.70.0020.063-0.30.8291.000−1.80.0220.635RMS:Speed V (g.s.m−1)−4.0 < 0.001 < 0.0010.10.9571.000−1.50.0471.000Power AP (W/kg)−3.40.1241.000−3.20.0971.000−3.60.0110.339Power ML (W/kg)−4.10.1251.000−5.00.0060.257−4.70.0030.115Power V (W/kg)−4.90.0130.445−1.70.1651.000−3.50.0020.062Power total (W/kg)−4.40.0250.816-2.70.0331.000−3.8 < 0.0010.032Power AP:Power total1.10.0971.000−1.80.0291.000−0.40.4311.000Power ML:Power total−0.70.6591.000−1.00.3901.000−0.80.3731.000Power V:Power total−0.40.5491.0001.20.0230.8380.50.2271.000Power AP:Speed (W.kg−1.m−1.s)-5.10.0160.534−2.00.2651.000−3.60.0060.224Power ML:Speed (W.kg−1.m−1.s)-5.40.0411.000−2.80.1271.000−4.10.0070.233Power V:Speed (W.kg−1.m−1.s)-6.9 < 0.0010.009−0.80.4351.000−3.8 < 0.0010.009Power total:Speed (W kg−1.m−1 s)-6.1 < 0.0010.034−1.50.1741.000−3.8 < 0.0010.009HF:LF AP-3.50.0931.000−0.20.9281.000−2.20.1501.000HF:LF ML-9.1 < 0.0010.0341.40.7261.000−4.30.0521.000HF:LF V2.60.3961.0002.50.2971.0001.20.4961.000Entropy AP-2.0 < 0.0010.006−0.70.2031.000−1.4 < 0.0010.008Entropy ML-2.6 < 0.001 < 0.001-0.30.4261.000−1.3 < 0.001 < 0.001Entropy V-0.60.3721.0000.10.8181.000−0.40.3181.000Knee flexion strength (%pred)0.30.7091.000-0.80.2641.000−0.10.8201.000Knee extension strength (%pred)-0.10.8861.0000.40.5921.0000.30.4871.000Ankle dorsiflexion strength (%pred)0.60.4821.0003.96 < 0.0010.0091.80.0060.224Ankle plantarflexion strength (%pred)-0.90.4861.0000.70.5051.000−0.20.7911.000The mean change percentage is defined as the average annual change in each variable as a function of age, expressed as a percentage of the overall mean value of that variable. The Teenagers group is aged less than 21. Three participants from the Teenagers group are also in the Adults group, as they were followed up before and after their 21st birthday. Significant changes with age are highlighted. Among the 46 participants at baseline, ankle strength measurements were only available for the 11 participants included in the registry after these measurements were added by amendment (see §2.3). All other variables were complete. Across all 190 longitudinal observations from the 46 participants (including baseline), MFM scores were complete. Ankle strength had the highest rate of missing data (57/190, 30%). Accelerometry variables were missing for 3 observations (1.6%), corresponding to visits, where the 6MWT was not performed. Knee extension strength was missing for 2 observations (1.1%), and 6MWD/knee flexion strength for 1 observation (0.5%). Age dependence of gait, accelerometry and strength variables in non-assisted ambulant participants with GSD-III The mean change percentage is defined as the average annual change in each variable as a function of age, expressed as a percentage of the overall mean value of that variable. The Teenagers group is aged less than 21. Three participants from the Teenagers group are also in the Adults group, as they were followed up before and after their 21st birthday. Significant changes with age are highlighted. Among the 46 participants at baseline, ankle strength measurements were only available for the 11 participants included in the registry after these measurements were added by amendment (see §2.3). All other variables were complete. Across all 190 longitudinal observations from the 46 participants (including baseline), MFM scores were complete. Ankle strength had the highest rate of missing data (57/190, 30%). Accelerometry variables were missing for 3 observations (1.6%), corresponding to visits, where the 6MWT was not performed. Knee extension strength was missing for 2 observations (1.1%), and 6MWD/knee flexion strength for 1 observation (0.5%). Age-related changes observed in growing teenagers (Table 3) were (i) decreased stride frequency (Fig. 3b) with increased stride length; (ii) decreased RMS/speed ratio along AP and V axes, and reduced total power, power V/speed and total power/speed ratios; (iii) a sharp decline in HF/LF ML ratio; and (iv) reduced AP and ML entropy. Age dependence of MFM D1 score (a), 6MWD (c), stride frequency (b) and stride length normalised to height (d) The dashed lines indicate the expected value (red in panels a and c) or the mean value (black in panels b and d) for healthy participants. The blue area in panels b and d represents the standard deviation for healthy participants. In adults, the MFM total score declined with age due to decreased MFM D1, while D2 and D3 remained stable. Ankle dorsiflexion strength increased significantly with age when considering intra-individual repetitions but showed a non-significant decrease (not shown) when intra-individual repetitions were excluded. Combining all participants (teenagers and adults) revealed significant age effects on MFM D1, stride frequency, stride length/height ratio (Fig. 3d), vertical and total power/speed ratios, and AP and ML entropy. RMS ML and total power also decreased significantly in the whole group, though not within subgroups alone. Figure 3a shows MFM D1 decline from the third decade onward, and earlier in some patients. Although impaired in GSD-III (Table 2), 6MWD (%pred) and speed did not worsen significantly with age (Table 3), despite progressive decreases in some individuals (Fig. 3c). Consistent with stable 6MWD, no age-related decline was observed in knee or ankle flexion/extension strength (%pred) for all participants. Search for predictors of the 6MWD decline Potential predictive factors explaining the decrease in 6MWD in rare patients were investigated: Genotype Except for two mutations linked to the GSD-IIIb phenotype without muscle deficiency [40], no clear genotype–phenotype relationship was found in GSD-IIIa. Despite allelic heterogeneity, patients with identical genotypes displayed variable severity, suggesting additional genetic or environmental influences [4, 41]. Missense mutations may retain residual enzymatic activity, but their low frequency limited analysis reliability: in our cohort (Table 1), 9/50 (18%) participants had a missense mutation on at least one allele, and 2/50 (4%) were homozygous. Left ventricular hypertrophy (LVH) No previous study has quantified the effect of LVH on walking distance, though evidence indicates muscle activity benefits cardiac remodelling [42]. Among the 22 participants with LVH (Table 1), half maintained stable 6MWD and half showed a decline, suggesting LVH is not necessarily a predictor for 6MWD decline. Clinical and biomechanical variables Although 6MWD (%pred) and walking speed were reduced in GSD-III (Table 2), they did not significantly worsen with age (Table 3), except in a few participants (Fig. 3c). Among the 46 ambulant GSD-III participants, 17 were classified a posteriori as having a declining 6MWD trajectory (loss ≥ 1.5%pred/year and/or last 6MWD < 70%pred), while the remaining 29 were classified as stable (Table 1). In Table 4A, although both groups were equivalent in age, gender, height and weight (not shown), univariate analysis identified significant baseline differences in MFM D1, D2, D3, knee flexion strength (%pred), and 8 accelerometric variables. To evaluate the added predictive value of accelerometric measures, we compared four multivariable logistic regression models (Table 4B). The full model (M1), including all variables significantly different between declining and stable groups, retained stride length, MFM D2, HF:LF_V, Power V, and RMS V after stepwise selection. The clinical-only model (M2), restricted to MFM scores and knee flexion strength (%pred), the MFM-only model (M3), and the accelerometric-only model (M4) also displayed satisfactory AUC, sensitivity and specificity. Adding accelerometric variables to clinical variables (M1) improved model performance in terms of AUC, sensitivity and specificity. However, the improvement in AUC between M1 and the other models, while consistent across validation methods, did not reach statistical significance (DeLong test p > 0.05). The lower AIC for M1 (40.9) nevertheless indicates better model fit. Table 46MWD (%pred) decline predictors by ROC analysisA. Univariate analysisa) Significant clinical variablesVariablesNo decline (n = 29)Decline (n = 17)p valueAUCMFM D1 (%)93.64 ± 7.3180.85 ± 14.220.0020.78MFM D2 (%)98.66 ± 2.7495.59 ± 5.380.0090.71Knee flexion strength (%pred)73.86 ± 16.8757.98 ± 21.370.0250.70MFM D3 (%)98.04 ± 2.9895.52 ± 4.280.0270.68b) Significant accelerometric variablesVariablesNo decline (n = 29)Decline (n = 17)p valueAUCStride length (m)1.41 ± 0.201.15 ± 0.230.0010.81Stride length:height0.87 ± 0.100.73 ± 0.170.0010.79HF:LF_V0.28 ± 0.190.52 ± 0.340.0120.72Entropy_V2.23 ± 0.372.46 ± 0.380.0330.69POWER_TOTAL (W/kg)19.49 ± 6.8815.47 ± 10.140.0410.68RMS_V (g)0.45 ± 0.090.38 ± 0.140.0440.68Power_AP (W/kg)5.18 ± 1.864.17 ± 2.640.0440.68Power_V (W/kg)10.73 ± 4.018.32 ± 6.060.0440.68B. Comparison of predictive modelsModelsInput variablesSelected variablesAUCAUC-LOOSensitivitySpecificityAICM1All significant variablesStride_length, MFM D2, HF:LF_V, Power_V, RMS_V0.910.8275%100%40.9M2Clinical-only variablesMFM D1, MFM D2, knee flexion strength0.850.7871%86%46.8M3MFM-only variablesMFM D1, MFM D2, MFM D30.830.7571%90%50.7M4Accelerometric-only variablesStride length, HF:LF_V0.880.8076%93%44.2AUC Area under the curve, AUC-LOO AUC-leave-one-out, AIC Akaike Information Criterion. 6MWD (%pred) decline predictors by ROC analysis AUC Area under the curve, AUC-LOO AUC-leave-one-out, AIC Akaike Information Criterion. Practically, the probability of 6MWD decline can be expressed as where For M1 model: z = 43.4 – 5.8 × Stride length (m) – 0.27 × MFM D2 (%) + 6.0 × HF:LF V (ratio) + 1.4 × Power V (W/kg) – 60.1 × RMS V (g). Two probability thresholds can be applied depending on the clinical objective: a threshold of 0.68 maximizes specificity (100%) with 75% sensitivity, ensuring no false positives; alternatively, a threshold of 0.38 (Youden index) offers a better balance with 90% specificity and 82% sensitivity. For M2 and M4 models, the equations are provided in the Online Resource. If no accelerometer (for M1 and M4) or dynamometer (for M2) is available, model M3 can be used, which is based solely on the MFM sub-scores. The probability threshold for the M3 model can be 0.75 (specificity 100%, sensitivity 41%) or 0.45 (Youden index) (specificity 90%, sensitivity 71%). Discussion A recent study reported that 83% of adults with GSD-III experienced walking difficulties and 48% considered muscle strength and function as a key therapeutic target [2]. The present study quantified deficits in 6MWD, accelerometric, strength, and MFM variables, examined their age-related changes, and identified the most relevant muscle outcome measures for future clinical trials. Predictors of 6MWD progression in GSD-III are also proposed. Comparison of baseline variables in ambulant GSD-III and healthy participants Our findings quantitatively confirm the walking difficulties of ambulant GSD-III patients and their variability. Their mean 6MWD ± SD, reduced to 76 ± 16%pred (median: 77%pred), was consistent with the results of a previous study [15], which also found no difference between the 6MWD (in %pred) of children (median age 11.5, range 3.7–17.7 years) and adults. Furthermore, accelerometry showed that in GSD-III (i) the reduction in walking speed compared with healthy controls was likely due to a decrease in step frequency and in the step length/height ratio, (ii) the decrease in triaxial RMS and power values was associated with the reduction in walking speed. These observations may reflect insufficient muscle energy production [43, 44] associated with muscle weakness and muscle fatigue affecting gait quality [2, 45]. Correlations between gait, accelerometry and strength variables in GSD-III GSD-III shows symmetrical lower-limb weakness [3, 12, 47] quantified here for the first time by dynamometry (right:left ratio≈1). Lower-limb strength correlated with standing and transfers (MFM D1, total MFM) and walking (6MWD). The strength of knee/ankle muscle groups affected stride frequency and stride length/height ratio. 6MWD, stride frequency, stride length/height ratio, RMS, and triaxial power were intercorrelated. Reduced step/stride regularity was associated with higher HF/LF ratio (shock wave indicator) and entropy (acceleration signal disorder indicator), but only vertical-axis measures correlated with 6MWD, suggesting progressive weakness impairs propulsion and shock absorption. Age dependence of gait, accelerometry and strength variables in GSD-III In this prospective study, we confirmed in a larger cohort the significant decline of MFM D1 and total scores in adulthood, but not in teenagers, consistent with our earlier findings of a drop around age 30 [17, 18]. In teenagers with GSD-III, the age-related increase in stride length and concurrent decrease in stride frequency, entropy, HF/LF ratio, RMS/speed, and power/speed along some axes suggest that our accelerometric measures captured locomotor maturation, as previously described in healthy individuals [46]. Despite their early impairment, 6MWD (%pred) and walking speed did not worsen significantly regardless of age among ambulant participants. In adulthood, accelerometric variables remained stable with age, as did muscle strength (in %pred), with the exception of an increase in ankle dorsiflexion strength, probably due to initial underperformance or learning. Consequently, only MFM D1 and total scores have proved to be age- and disease-sensitive muscle outcome measures in adults and should be prioritised in clinical practice and research. Search for predictors of the 6MWD decline Despite generally stable 6MWD (%pred) impairment, some individuals declined, needed walking aids, or became non-ambulant. Our analysis identified MFM sub-scores and knee flexion strength as independent baseline predictors: lower values indicate higher risk of 6MWD decline. Adding accelerometry features to clinical variables (M1 model) markedly improved discrimination (AUC) and specificity. The perfect specificity of 100% achieved with this M1 model means that patients identified as ‘low risk’ can be confidently reassured about their future walking ability. Resources can then be focused on individuals flagged as ‘high risk’ of 6MWD decline. However, this M1 model requires a 3D accelerometer and analysis software. Alternatively, the M2 and M3 models—based on clinical predictors and MFM scores only, respectively—offer acceptable predictive performance. The accelerometry-only model (M4: stride length, HF/LF_V) performed similar to the model M3 with MFM scores alone (AUC 0.88 vs 0.83), indicating complementary prognostic value and supporting trunk accelerometry in routine assessment. However, the improvement in AUC between M1 and the other models did not reach statistical significance, likely reflecting the limited sample size inherent to rare disease research. External validation in independent cohorts remains needed. Moreover, stepwise selection may introduce instability and bias. Future studies could also explore composite accelerometric indices or machine learning approaches to derive parsimonious, clinically interpretable risk scores. Limitations First, although the Locometrix accelerometer was developed about 20 years ago, sensor technology has not fundamentally changed, and using the same device ensured consistent measurements across nearly 10 years of follow-up. Second, accelerometric variables were derived from 10-s walking segments showing maximal step regularity, excluding cone turns of the 6MWT from the analysis. While this approach may underestimate some gait defects, the abnormalities identified within the most regular segment are, therefore, robust. Third, walking ability was assessed using the 6MWT, the most used and best-validated walk test in neuromuscular trials [47]. However, it may have been exhausting for the most severely affected participants and, conversely, may have been unrepresentative of daily fatigue for milder cases. Nevertheless, it had the merit of being achievable for all ambulant participants with GSD-III. Fourth, hip muscle strength was not measured, although these muscles contribute to walking. This omission avoided overloading sessions already, including other muscle tests not presented here (e.g., upper limbs, timed tests, …). Fifth, additional biological or clinical factors may influence gait but were not investigated in this study. Sixth, although the MFM scale is a reliable, valid, sensitive (e.g. [48]), rapid, and cost-effective motor function measure across ages and severities, regulators and funders may also require patient-reported and quality-of-life outcomes to fully evaluate future GSD-III treatments. Conclusion Although 6MWD is impaired in GSD-III, it does not significantly worsen over time in most patients, though a subset shows progressive decline. MFM sub-scores remain practical clinical predictors of 6MWD progression, while accelerometric variables may provide additional predictions. Regular monitoring of these predictors and implementation of preventive measures may help preserve walking capacity and quality of life. Unlike 6MWD, MFM D1 and consequently total MFM scores deteriorate significantly with age in adults, supporting the MFM scale as a suitable clinical and outcome measure for future GSD-III trials. Finally, future studies could use wearable devices to record daily walking and fatigue, which are not captured in the 6MWT. Supplementary Information Below is the link to the electronic supplementary material. Supplementary file1 (PDF 275 KB) Suppl Fig. 1 Superimposed distribution of density histograms in the GSD-III group (red) and the healthy control group (blue) for age (A), height (B), weight (C) and BMI (D) with the corresponding normal distribution curve (solid line) and mean (dotted line) superimposed. Supplementary file2 (PDF 147 KB) Suppl Fig. 2 Examples of raw lower trunk acceleration signals along the antero-posterior (AP), medio-lateral (ML) and vertical (V) axes recorded during the 6MWT in male participants. The signals from the healthy 50-year-old participant are shown in the left-hand panels and the signals from the 47-year-old GSD-III participant are shown in the right-hand panels. The first 4 s of the 10 s of walking with maximum regularity are shown here. Supplementary file3 (PDF 81 KB) M2 and M4 models equations. Acknowledgements The authors would like to warmly thank the participants in this study and their families for their time and feedback on their experience of the disease. We also would like to thank the Association Francophone des Glycogénoses (AFG) for funding part of this research. We are very grateful to the contributing centres. We would also like to thank the members of the Institute of Myology (Paris, France), in particular Dr Rabah Ben Yaou for his contribution to the design of the registry, Nawal Benammar and Erwan Gasnier for their administrative and regulatory assistance, and Thomas Marques for his contribution to the analysis of the signals. We would also like to thank the students who contributed to the development of the in-house Locolike software under the supervision of engineers from the Institute of Myology or who used it in a preliminary analysis. Funding The French GSD-III registry is supported by the Association Francophone des Glycogénoses (AFG). Data availability Anonymized data and source codes will be shared on request from any qualified investigator for the sole purpose of replicating procedures and results presented in the article and as long as data transfer is in agreement with the European Union legislation on the General Data Protection Regulation. Declarations Conflicts of interest The authors have no competing interests to declare that are relevant to the content of this article Outside of this context, JYH received honoraria for lecturing, consulting, and travel expenses for attending meetings from Biogen, Pfizer, Roche, LFB, Italfarmaco, Lupin and Novartis. AG has received consulting fees for Ipsen, Sanofi and Theravia. MM has received support for attending meetings and/or travel from CSL-Behring and Roche and has served on advisory boards for Johnson&Johnson and Roche. PL has received research grants, consulting fees, speaker’s honoraria, support for attending meetings and/or travel from Biogen, Roche, Alexion, Johnson&Johnson, Ad Scientiam, Amicus, Astellas, Sanofi, Shionogi. He has served on advisory boards for Biogen, Roche, Alexion, Johnson&Johnson, Ad Scientiam, Amicus, Astellas, Sanofi, Shionogi and declares no stocks or shares. The other authors report no relevant disclosures. Ethical approval The protocols for the French GSD-III registry, Accelerix and NatHis DM1 studies were approved by the local Ethics Committee (CPP–Ile de France VI–Groupe Hospitalier Pitié Salpêtrière, Paris, France) in accordance with the 1964 Declaration of Helsinki and its later amendments. Written informed consent was obtained from all participants. References Associated Data Supplementary Materials Supplementary file1 (PDF 275 KB) Suppl Fig. 1 Superimposed distribution of density histograms in the GSD-III group (red) and the healthy control group (blue) for age (A), height (B), weight (C) and BMI (D) with the corresponding normal distribution curve (solid line) and mean (dotted line) superimposed. Supplementary file2 (PDF 147 KB) Suppl Fig. 2 Examples of raw lower trunk acceleration signals along the antero-posterior (AP), medio-lateral (ML) and vertical (V) axes recorded during the 6MWT in male participants. The signals from the healthy 50-year-old participant are shown in the left-hand panels and the signals from the 47-year-old GSD-III participant are shown in the right-hand panels. The first 4 s of the 10 s of walking with maximum regularity are shown here. Supplementary file3 (PDF 81 KB) M2 and M4 models equations. Data Availability Statement Anonymized data and source codes will be shared on request from any qualified investigator for the sole purpose of replicating procedures and results presented in the article and as long as data transfer is in agreement with the European Union legislation on the General Data Protection Regulation.