Lasse Schlör
lasse-schloer@unibo.it

University of Bologna
Bologna, 2025-08-21

PhD: Individually tailored digital-motor outcomes in real life

Funded by the European Union’s Horizon Europe Marie Skłodowska-Curie Actions scheme, GA no. 101120256.

My Background

My background in Tübingen

Until 2018	B.Sc. in Computer Science
Until 2021	M.Sc. in Computer Science
Until 2024	Research Assistant in Cognitive Science

CCVEP decoding

Audio engineering

Tools I like to work with

My background in Bologna

Since June 2024	PhD in Health & Technologies

Medicine Made to Measure (MMM)

Doctoral network

Funded by the Marie Curie Actions Scheme
10 PhD candidates
22 participating organisations

Aim: to develop

Antisense oligonucleotide treatments (ASOs)
Tailored to single patients with nano-rare disease mutations

MMM group photos

First General Assembly – Barcelona, June 2024

Mid-Term Meeting – Tübingen, December 2024

Rare diseases

A rare disease is any disease with a prevalence of <0.05%.

There are >6000 rare diseases.

Taken together, their prevalence is ~4%.

>70% have genetic causes.

Nano-rare mutations: Worldwide <30 patients.

(approximate numbers, for Europe)

“rare individually, common collectively”

MMM pipeline

Supervisors

Dr. Sabato Mellone
University of Bologna

Prof. Dr. Matthis Synofzik
University of Tübingen

Prof. Dr. Mats Karlsson
University of Uppsala

Co-supervisors

Carlo Tacconi
mHealth Technologies s.r.l., Bologna

Prof. Dr. Lorenzo Chiari
University of Bologna

Secondments

Uppsala, Sweden

Aug 2025 - Sep 2025

Tübingen, Germany

Sep 2024 - Nov 2024
Three more months in the coming years

Introduction

Individualized ASOs for nano-rare mutations:
Trial design challenges

ASO treatment is taylor-made for the mutation
Tiny number of patients worldwide
ASO treatment has slow, disease-modifying effect

Traditional clinical trial designs infeasible
Traditional N-of-1 trial designs infeasible

Spinocerebellar Ataxia (SCA)

Inherited neurodegenerative disease
Affects the cerebellum and spinal cord
Rare disease: ≤5 people per 100,000

Increased gait variability

Widened base of support

Impaired foot placement

Loss of trunk control

Deliverables

D3.1	Outcomes of the large cohort study Completed Jan 2025
D3.2	Outcomes of focused study Due May 2027 (postponed)
D3.3	Robust capture of n-of-1 movement changes Due Sep 2027

Gait feature extraction

Inertial measurement units

Can be attached to various body parts
Measure movement

Accelerometer
Gyroscope
(Magnetometer, Barometer, …)

IMU hardware suppliers:

	mHealth Technologies
	APDM
	Any smartphone can be used as IMU

Sensor configurations

Lower back and feet

Lower back only

3D IMU trajectory plot

Gait cycle analysis

Gait analysis pipelines

Software	Mobility Lab	gaitmap	mobgap
Organization
Configuration	Feet & trunk	Feet only	Trunk only
Open source	✗	✓	✓
Validated*	✓	✗	✓
Stride DMOs	58	12	4

*Note:
Clinical validation is outcome- and population-specific.
Only Mobility Lab has been validated in ataxias.

Gait cycle analysis

Double support

Spatial analysis

Elevation at mid-swing

Foot contact angles

Circumduction

Lateral step variability

Walking bouts

Various possible definitions.

Here: Sequence of strides not interrupted for longer than two median stride durations.

Walking segments

Subdivision of walking bouts into more homogeneous segments:

Drop first two and last two strides.
Split into as many segments of ≥25 strides as possible.

Additional features

Signal

Position

Left foot
Right foot
Trunk

×

Frame

Sensor
Body
World

×

Quantity

Acceleration

Velocity
Position

Angular velocity

Angle

(Magnetometer)
(Barometer)

×

Dimension

x
y
z

↓

Window

Reference

Left foot
Right foot

×

Cycle segment

Stride
Swing
Stance

↓

Feature classes

Stride-to-stride comparisons
Discrete transform coefficients
…

↓

Window

Walking bout
Walking segment

↓

Feature classes

Complexity measures
…

Stride-to-stride comparisons

TODO

Discrete transform coefficients

TODO

Complexity measures

TODO

Context stratification

TODO

TODO: Weather?

Feature transformation

TODO

Feature aggregation

TODO

Context strata

Keywords	Covariate	Threshold(s)
slow / fast	Gait speed	1.2m/s
sporadic / continuous	Number of steps in a 1-minute window	45
straight / curvy	Number of turns in a 1-minute window	1
short / long	Walking bout duration	30s

Gait feature extraction summary figure

Data

Dataset

Gait recordings from Tübingen University Hospital

27 genetic spinocerebellar (pre-)ataxia patients
36 healthy controls
Lab-based & real-life
Yearly follow-ups over several years

Patients further categorized into:

6 pre-symptomatic (SARA 2 ± 1)
21 symptomatic (SARA 10 ± 3)

In total:

~200 visits
~20000 walking segments
~570,000 gait cycles

Other potential datasets

PROSPAX

Lab-based movement recordings only
≥74 ARSACS (spastic ataxi) patients
≥113 SPG7 (spastic paraplegia) patients
≥50 healthy controls

Planned prospective study

≥10 ataxia (SCA or AT) patients

Outcomes

Clinician- / patient-rated scores (per visit)

SARA
ABC
Others, but incomplete (SPRS, INAS, PGI, ADL)

Digital motor outcomes

Core temporal and spatial gait measures
Additional gait features

(Discrete transform coefficients, …)

Transformations, aggregations, and (per-visit) context stratifications of the above

Large space of combinations, small fraction suitable

Further covariates

Demographics (age, gender)
Diagnosis / gene affected
Anthropometrics at first visit (weight, height, handedness, …)
Estimated age of disease onset (only 19 of 27 patients)

SARA score

“Scale for the assessment and rating of ataxia”

Item	Possible responses
1) Gait	0 - 8
2) Stance	0 - 6
3) Sitting	0 - 4
4) Speech disturbance	0 - 6
5) Finger chase	0 - 4
6) Nose-finger test	0 - 4
7) Fast alternating hand movements	0 - 4
8) Heel-shin slide	0 - 4
Total	0 - 40

ABC score

TODO

Disease onset estimates

Tezenas du Montcel et al. 2014

Data: Per-visit level

Raw data, visit level, lateral step deviation, preferred speed task

Data: Per-visit level

Data: Per-segment level

Raw data, segment level, lateral step deviation, preferred speed task

Qualitative characteristics

Generally constant over time in the absence of disease

Effect of age, if present, is small

Roughly linear rise at disease onset
Theoretical saturation not/barely represented in the dataset

Clinical scales bounds not reached
DMOs said to saturate around SARA 12-15

Data are potentially “ill-behaved” in several ways:

Selection/attrition bias
Disease onset estimation error (and missingness)
Depending on model, residuals might be:

Heteroskedastic
Non-normal
Serially correlated

Data: Per-walking-segment level

TODO

Simulating outcome data

TODO

Simulating outcome data

Traditional trial design

Cross-sectional testing

Cross-sectional box plot: SARA (with simulated controls data)

Cross-sectional box plot: lateral step deviation

Cross-sectional box plot: stride length CV

Cross-sectional box plot: stride length d2

(lab based)

(real life)

Longitudinal paired-samples testing

Seemann et al. 2025

Longitudinal paired-samples testing

Seemann et al. 2025

Longitudinal paired-samples testing

Longitudinal delta plot, lateral step deviation, preferred speed task

Longitudinal paired-samples testing

Pooled sample size estimates

Year pair selection criteria
Time between baseline and follow-up year

Pool 1-, 2-, and 3-year effect sizes

With inverse variance weighting
Pre-normalize to 1-year by linear interpolation

Assumes outcomes change linearly on average

Outcome ranking

outcome	ess	rb_lgt			rb_crs			ρ_ΔΔ_2y		spread
outcome	ess	1y	2y	3y	sym	sara<8	sara≥8	sara	-abc	spread
test_dw_instance_compound_5	39	.86	.75	.96	.68	.28	.67	.32	.12
test_dw_instance_compound_3	45	.82	.66	.93	.59	.18	.62	.26	.26
test_dw_instance_compound_2	55	.77	.78	.78	.49	.36	.48	.58	.21
test_dw_instance_compound_4	66	.79	.65	.36	.75	.48	.52	.33	.2
agg_adjacent_swings_resampled_sensor_lumbar_acc_x_r_d1_abs_μ /σ /curvy_long/μ	74	.71	.45	.76	.52	.32	.46	-.5	-.06	.47
agg_adjacent_swings_resampled_sensor_lumbar_acc_x_r_μ /σ /curvy_long/μ	83	.62	.62	.67	.54	.24	.45	-.32	.0055	.45
agg_adjacent_swings_resampled_sensor_lumbar_gyr_z_r_abs_μ /σ /long /μ	86	.68	.46	.54	.5	.066	.42	-.036	.2	.52
agg_adjacent_swings_resampled_sensor_lumbar_gyr_z_r_abs_d1_abs_μ/σ /long /μ	88	.72	.4	.41	.49	.09	.4	.062	.32	.57
agg_adjacent_swings_resampled_sensor_lumbar_acc_x_r_abs_μ /σ /curvy_long/μ	88	.58	.66	.65	.54	.24	.45	-.3	.21	.42
agg_adjacent_swings_resampled_sensor_lumbar_gyr_z_r_abs_d3_abs_μ/σ /long /μ	89	.71	.42	.38	.48	.067	.41	.11	.093	.54
stance_duration_μ /cv/curvy_long/μ	90	.65	.76	.38	.63	.29	.52	.25	-.43	.45
agg_adjacent_swings_resampled_sensor_lumbar_gyr_z_r_abs_d2_abs_μ/σ /long /μ	93	.71	.44	.38	.48	.067	.4	.11	.21	.54
agg_adjacent_swings_resampled_sensor_lumbar_gyr_z_r_abs_d1_abs_μ/μ /long /μ	94	.62	.52	.43	.54	.12	.42	.16	.31	.44
double_support_apdm_μ /σ /curvy_long/μ	95	.69	.46	.49	.72	.42	.46	.13	-.37	.36
coeff_swings_sensor_acc_dft_x_1_abs_μ /cv/long /μ	95	.76	.51	.43	.71	.43	.44	.1	.044	.41
agg_adjacent_swings_resampled_sensor_lumbar_acc_x_r_abs_d2_abs_μ/μ /curvy_long/μ	96	.66	.5	.56	.63	.32	.49	-.24	.088	.42
agg_adjacent_swings_resampled_sensor_lumbar_gyr_z_r_abs_d3_abs_μ/μ /long /μ	97	.65	.49	.43	.54	.12	.42	.24	.35	.46
swing_apdm_μ /cv/curvy_long/μ	98	.72	.58	.49	.7	.36	.51	.15	-.32	.36
agg_adjacent_swings_resampled_sensor_lumbar_gyr_z_r_abs_d2_abs_μ/μ /long /μ	98	.64	.51	.41	.53	.11	.42	.22	.31	.43
agg_adjacent_swings_resampled_sensor_lumbar_acc_z_r_abs_d3_abs_μ/γ /short /μ	98	-.69	-.53	-.41	-.48	-.28	-.48	-.57	-.15	.69
agg_adjacent_swings_resampled_sensor_lumbar_acc_x_r_abs_d1_abs_μ/σ /curvy_long/μ	100	.66	.45	.69	.51	.31	.44	-.49	.071	.42
initial_plus_mid_swing_apdm_μ /cv/curvy_long/μ	100	.63	.69	.6	.67	.42	.42	-.097	.082	.38
agg_adjacent_swings_resampled_sensor_lumbar_gyr_z_r_abs_μ /cv/long /μ	100	.57	.56	.47	.5	.072	.41	.1	.15	.57
gait_speed_apdm_μ /σ /curvy_long/μ	100	.57	.73	.34	.44	.21	.45	.39	0	.42
(…)
sara_sim_1	103	.53	.56	.76	1	.85	1	1	.077
(…)

Outcome ranking: Metrics

ess	1-year estimated sample size (pooled from 1-, 2-, 3-year)
rb_lgt	Longitudinal effect sizes (Wilcoxon signed-rank rank biserial)
rb_crs	Cross-sectional effect sizes (Mann-Whitney rank biserial)
ρ_ΔΔ_2y	2-year change correlation (Spearman)
spread	Measure of within-visit variation

Outcome ranking: Filtering

Require:

A certain ability to discriminate symptomatic patients
A certain ability to discriminate low-SARA from higher-SARA
Longitudinal patient p-values to be (somewhat) significant
Reasonable number of patients considered longitudinally
Longitudinal effects in controls to be non-significant or small
Longitudinal and cross-sectional effects' signs to agree

Outcome trajectory modeling

Linear model

\[ y_{ij} = \beta_1\mathrm{BOn}_{ij} + \beta_2\mathrm{BSat}_{ij} + \beta_3\mathrm{BOn}_{ij}\left(\mathrm{EDO}_\mathrm{o}\right)_{i} + \beta_4\mathrm{BOn}_{ij}\mathrm{Gender}_{i} + \beta_4\mathrm{BOn}_{ij}\mathrm{BMI}_{i} \]

Fixed-effects panel model (linear model with de-meaning)
Done as a course exercise
Non-parametric bootstrap & Monte Carlo methods employed

Fixed-effects panel model predictions

Fixed-effects panel model: Lateral step deviation

Fit results

Outcome (lab-based)	gait_speed_apdm_μ/μ/all/μ	stride_length_apdm_μ/cv/all/μ
\(R^2\)	0.60	0.50
\(\mathrm{BOn}\) (\(p\))	-0.011 (0.92)	0.0083 (0.32)
\(\mathrm{BSat}\) (\(p\))	-0.22 (2.2e-07)	0.012 (0.00035)
\(\mathrm{BOn} \times \mathrm{Gender}\) (\(p\))	-0.095 (0.00028)	0.0044 (0.041)
\(\mathrm{BOn} \times \mathrm{BMI}\) (\(p\))	-0.0061 (0.10)	6e-04 (0.054)
\(\mathrm{BOn} \times \mathrm{EDO_o}\) (\(p\))	0.0043 (0.00013)	-0.00053 (5.3e-08)
ESS	60	106
ESS (Seemann et al. 2025)	66	67

Assumption violations

Here:

Simulated data
No bootstrapping

Composite measures

Factor analysis

TODO: Include figure from Coni et al. 2019

Pilot test on visit level resulted in subpar composites

Optimized composite measures

Idea: Formulate the search for the ideal combination of outcomes as an optimization problem.

E.g. Find linear combination that minimizes some loss.
Loss: E.g. Composite change vs. disease time change.

\[ \min_\beta \sum_{i, j} \left( \left( t_{i j} - \bar{t}_{i \cdot} \right) \cdot I_{\mathrm{symptomatic}}(i) - \sum_{k} \left( y_{k i j} - \bar{y}_{k i \cdot} \right) \cdot \beta_k \right)^2 + \Omega(\beta) \]

This is a linear regression (with optional regularizer \(\Omega(\beta)\))

From the de-meaned (and normalized) per-visit outcomes
To the de-meaned time since baseline, if symptomatic

Result \(\theta = \beta y\) is a generated regressor – likely suboptimal compared to a jointly estimated latent variable model.

TODO: Discuss cross-sectional IRT work of Hamdan et al.

Graded response model (IRT)

\[ P\left( Y_{ij} \ge k \mid \Psi_i \right) = \mathrm{logit}^{-1}\left( a_j \left( \Psi_i - b_{jk} \right) \right) \]

\(i\)	Number of participant
\(j\)	Number of SARA item
\(k\)	A score for an item
\(Y_{ij}\)	Observed score for participant \(i\) and item \(j\)
\(\Psi_i\)	Latent disease severity for participant \(i\), with \(\Psi_i \sim \mathcal{N}\left( 0, 1 \right)\) assumed
\(a_j\)	Discrimination parameter of item \(j\), to be estimated
\(b_{jk}\)	"Difficulty" parameter of score \(k\) of item \(j\), to be estimated

\[ P\left( Y_{ij} = k \mid \Psi_i \right) = P\left( Y_{ij} \ge k \right) - P\left( Y_{ij} \ge k + 1 \right) \]

Item characteristic curves

Hamdan et al. 2024

Visual predictive check

Hamdan et al. 2024

Fisher information

Hamdan et al. 2024

Subpopulation analysis

Hamdan et al. 2024

Disease progression modeling

Latent disease severity model

\[ \theta_{i j} = \begin{cases} h \left( \left( \beta + b_i \right) \left( t_{i j} - \tau_i \right) \right) + u_{i j} & i \in \mathrm{patients} \\ 0 & i \in \mathrm{controls} \end{cases} \] \[ y_{k i j s} = \nu_k + \lambda_k \theta_{i j} + \epsilon_{k i j s} \]

\(t_{i j}\)	Time since baseline for subject \(i\) at visit \(j\)	Known
\(\theta_{i j}\)	Disease severity	Latent
\(\tau_i\)	Disease onset relative to baseline	Random effect
\(\beta + b_i\)	Progression rate	Fixed effect (population) + random effect
\(h\)	Hinge or sigmoidal function
\(u_{i j}\)	Per-visit severity baseline	Random effect
\(y_{k i j s}\)	Outcome \(k\) in walking segment \(s\)	Known
\(\mu_k, \lambda_k\)	Per-outcome intercept and slope (factor loading)	Fixed effects
\(\epsilon_{k i j s}\)	Residual error

Genetics-based disease onset estimates as priors?
Pre-symptomatic patients? (censoring?)
Outcomes not linear in \(\theta\)?
Context covariates? (let them affect variance of \(\epsilon\)?)

TODO: Update this draft

Thank you!

Go to index (relative)
Go to index (online)

PhD: Individually tailored digital-motor outcomes in real life

My Background

My background in Tübingen

CCVEP decoding

Audio engineering

Tools I like to work with

My background in Bologna

Medicine Made to Measure (MMM)

MMM group photos

Rare diseases

MMM pipeline

Supervisors

Co-supervisors

Secondments

Introduction

Individualized ASOs for nano-rare mutations: Trial design challenges

Spinocerebellar Ataxia (SCA)

Deliverables

Gait feature extraction

Inertial measurement units

Sensor configurations

3D IMU trajectory plot

Gait cycle analysis

Gait analysis pipelines

Gait cycle analysis

Double support

Spatial analysis

Elevation at mid-swing

Foot contact angles

Circumduction

Lateral step variability

Walking bouts

Additional features

Stride-to-stride comparisons

Discrete transform coefficients

Complexity measures

Context stratification

Feature transformation

Feature aggregation

Context strata

Gait feature extraction summary figure

Data

Dataset

Other potential datasets

Outcomes

Further covariates

SARA score

ABC score

Disease onset estimates

Data: Per-visit level

Data: Per-visit level

Data: Per-visit level

Data: Per-segment level

Qualitative characteristics

Data: Per-walking-segment level

Simulating outcome data

Simulating outcome data

Traditional trial design

Cross-sectional testing

Longitudinal paired-samples testing

Longitudinal paired-samples testing

Longitudinal paired-samples testing

Longitudinal paired-samples testing

Longitudinal paired-samples testing

Longitudinal paired-samples testing

Pooled sample size estimates

Outcome ranking

Outcome ranking: Metrics

Outcome ranking: Filtering

Outcome trajectory modeling

Linear model

Fixed-effects panel model predictions

Fit results

Assumption violations

Composite measures

Factor analysis

Optimized composite measures

Graded response model (IRT)

Item characteristic curves

Visual predictive check

Individualized ASOs for nano-rare mutations:
Trial design challenges