Table 2 Types of evidence provided in the EEL-II

R. Bhidayasiri. et al.

⁸⁴

Treatment monitoring

To evaluate the efficacy of rotigotine transdermal patch in RCT, using a wearable sensor features as final endpoint

Significant differences, determined by statistical test p-values, in sensor-derived features before/after intervention

M. Iijima. et al.

⁸⁵

Treatment monitoring

Objectively assessing gait disorder before/after medication change

Significant differences, determined by statistical test p-values, in sensor-derived features before/after the treatment change

H. Khodakarami et al.

⁸⁶

Treatment monitoring

Predicting levodopa response using a machine learning algorithm trained on sensor data

AUC performance of machine learning algorithm when predicting 6 different classes of levodopa responses.

R. Bouça-Machado et al.

⁸⁷

Treatment monitoring

To evaluate the efficacy of intervention using a wearable sensor features as final endpoint

Significant differences, determined by statistical test p-values, and effected size, determined as Cohen’s d, in sensor-derived features before/after intervention

N. Caballol et al.

⁸⁸

Treatment monitoring

Analyzing the responsivity of sensor measure in detecting ON-OFF time, dyskinesia, Freezing

Significant differences, determined by statistical test p-values, in sensor-derived features between medication changed group and medication stable group. Kappa agreement analysis to evaluate the agreement between clinical interview and sensor results.

I. Thomas et al.

⁸⁹

Intervention decision

Evaluating the accuracy of a sensor-based medication dosing schedules (SBDS)

Correlation and mean relative errors between sensor prediction and clinical experts’ evaluation

A. Silva de Lima et al.

³³

Patient stratification

Evaluating Fall incidence rate in PD vs Controls using IMU wearable sensors

Significant differences, determined by statistical test p-values, in sensor-derived features across different patient populations

M. Mancini et al.

⁴⁶

Patient stratification

Analyzing difference in turning between Freezers and not Freezer using IMU wearable sensors

Significant differences, determined by statistical test p-values, in sensor-derived features across different patient populations

L. Haertner et al.

⁹⁰

Patient stratification

Evaluate difference in gait parameters in PD with and without Fear of Falling using IMUs sensors

Significant differences, determined by statistical test p-values, in sensor-derived features across different patient populations

K. Srulijes et al.

³⁴

Patient stratification

Evaluating fall incidence rate in different disease population using IMU wearable sensors

Significant differences and correlations of sensor-derived features with patient diagnosis and clinical scores

M. Marano et al.

⁴³

Patient stratification

Evaluating fall incidence rate in different patient population using IMU wearable sensors

Significant differences, determined by statistical test p-values, in sensor-derived features across different patient populations

M. Mancini et al.

⁴⁵

Patient stratification

Automated detection of Freezing of Gait using IMU sensors data and detection of freezers vs non-freezers

Significant differences, determined by statistical test p-values, in sensor-derived features across different patient populations. Accuracy of FOG events detected compared with clinical rating

K. Kyritsis et al.

⁹¹

Patient stratification

Training a machine learning algorithm to distinguish PD vs Control from IMU sensor data during eating

AUC performance of machine learning algorithm in predicting patient diagnosis from sensor derived measures

A. Atrsaei et al.

⁴⁴

Patient stratification

Evaluating the effect of fear of falling on sensor derived gait parameters

Significant differences, determined by statistical test p-values, in sensor-derived features across different patient populations

A. Mirelman et al.

⁹²

Patient stratification, Disease prediction

To assess sleep disturbances in different PD vs Healthy Controls using objective sensor measures

Significant differences, determined by statistical test p-values, in sensor-derived features across different patient populations

V. Shah et al.

³²

Disease prediction

To determine which sensor-derived mobility measures of discriminate PD from healthy control

AUC performance of machine learning algorithm

L. Adams et al.

²⁶

Disease prediction

Evaluating difference in sensor-derived mobility parameters in PD vs Huntington patients

Significant differences, determined by statistical test p-values, in sensor-derived features across different disease populations

V. Shah Al.

³¹

Disease prediction

Evaluating difference in sensor-derived gait parameters in PD, Multiple Sclerosis and Healthy Controls.

Significant differences, determined by statistical test p-values, and AUC in sensor-derived features across different patient populations.

A. Nouriani. et al.

²⁴

Disease prediction

Evaluate the predictive value of sensor-derived features for prospective fall frequency.

Linear regression analysis, between sensor-derived features and fall frequency prediction, rho, standard and mean squared error, t-stat and p-values are reported.

L. Evers et al.

³⁰

Disease prediction, Symptom monitoring

To monitor motor fluctuations in PD using sensor derived measure during gait segment and to discriminate against healthy controls.

AUC performance of machine learning algorithm in predicting patient conditions and diagnosis from sensor derived measures.

H. Gaßner et al.

⁴¹

Symptom monitoring

To evaluate the reliability of supervised, standardized sensor-based gait outcomes at home compared to the hospital.

Significant differences, determined by statistical test p-values and intra class correlation coefficients, between sensor-derived features at home and in the lab.

L. Zhu et al.

⁴⁸

Symptom monitoring

Correlating patient conditions with discrepancy between PROMs and Objective measures.

Correlation of sensor derived information with self-reported outcomes, rho and p-value. Statistical analysis with p-values and intra class correlation coefficients.

A. Rodríguez- Molinero et al.

³⁸

Symptom monitoring

To investigate the link between gait sensor derived features and UPDRS III.

Correlation of sensor derived information with clinical scores, rho and p-value.

A. Rodríguez- Molinero et al.

⁴⁷

Symptom monitoring

To investigate the link between sensor-derived gait measures ON-OFF state.

Accuracy and positive/negative predictive values of machine learning algorithm results based on sensor-derived features with patient-reported outcomes.

A. Lígia Silva de Lima et al.

³⁶

Symptom monitoring

To investigate the link between motor fluctuations and sensor derived parameters.

Linear regression analysis, between sensor-derived features and clinical scores, rho and p-value, R² before and after levodopa intake.

B. Boroojerdi et al.

²⁷

Symptom monitoring

Evaluating correlation between total motor activity during sleep measured through sensors and patient reported sleep quality.

Descriptive statistic between sensor-derived features and patient reported outcomes. Difference between at home and at clinic behavior.

M. Knudson et al.

⁹³

Symptom monitoring

Predicting from objective sensor measures activity of daily life impairment.

R² performance of multiple regression algorithm, based on sensor-derived features, when predicting clinical scores. Significant differences in sensor values for patients with and without bradykinesia and dyskinesia, p-values reported.

A. Rodríguez-Molinero et al.

⁶⁷

Symptom monitoring

Estimate dyskinesia severity from sensor-derived measurements.

Correlation of sensor derived information with Unified Dyskinesia Rating Scale, rho and p-value and confidence intervals.

Ravichandran. et al.

²²

Symptom monitoring

Evaluate if sensor derived sensor derive features are sensitive to ON-OFF state.

Descriptive statistic of sensor derived value in different states.

A. Papadopoulos et al.

⁹⁴

Symptom monitoring

Detecting tremor episode using smartphone IMU during phone calls.

Average and standard deviation of: Precision, Sensitivity, Specificity, F1-score performance of machine learning algorithm.

M. Heijmans et al.

⁶¹

Symptom monitoring

To evaluate symptom severity of people with PD through wearables and digital questionnaires.

AUC performance of logistic regression algorithm on a patient used as case study.

D. Gatsios et al.

⁶⁰

Symptom monitoring

Feasibility study aiming to collect relevant clinical data in the wild.

Significant differences and correlations between sensor use and clinical scores, rho and p-value.

R. San-Segundo et al.

⁹⁵

Symptom monitoring

Detecting tremor episode from sensor-derived data using deep learning algorithms.

AUC performance and False Positive Rate at 0.90 True Positive Rate of machine learning algorithm in predicting patient symptom severity

A. Abrami et al.

²¹

Symptom monitoring

To evaluate correlation between movement symbolic representation and clinical scores.

Significant differences and correlations between of sensor derived information with clinical scores, rho and R².

R. Bouça-Machado et al.

⁶²

Symptom monitoring

Feasibility study aiming to collect relevant clinical data in the wild via wearables.

Correlation of sensor derived information with clinical scores, rho and p-value.

M. Corrá et al.

³⁵

Symptom monitoring

Evaluate if unsupervised sensor derived gait speed and walking bout duration detect ON-OFF state.

Significant differences and correlations between of sensor derived information with clinical scores, rho, p-value and R².

Y. Raykov et al.

²³

Symptom monitoring

To predict from gait sensor data if segment recorded happened before-after medication intake.

Mean and standard error in predicting patient conditions from sensor derived measures.

J. Habets et al.

⁹⁶

Symptom monitoring

To detect significant difference in wrist IMU data during bradykinesia fluctuations and predicting them with a machine learning algorithm.

Significant differences, determined by statistical test F, Wilk’s Lambda and p-values. AUC performance of machine learning algorithm in predicting patient conditions from sensor derived measures

G. Oyama. et al.

⁴²

Symptoms monitoring

To evaluate the test-retest reliability in the home environment of a digital sensor-based assessment. Correlation with in lab clinical scores.

Evaluation test-retest reliability across different study periods, determined by Intra class correlation coefficients. Spearman correlation between sensor measure at home and clinical scores at lab.

M. Hssayeni et al.

⁶⁸

Symptom monitoring

To continuously predict UPDRS III from sensors recordings.

Correlation, determined by rho and p-value, and mean absolute errors between sensor prediction and clinical scores.

F. Lipsmeier et al.

³⁷

Symptoms monitoring

Monitoring patient condition at home using a combination of unsupervised tasks and passive monitoring using objective sensor measures.

Spearman correlation, determined by rho and p-value, between sensor derived feature and clinical scores. Intra class correlation coefficients, evaluating test-retest reliability across different study periods. Significant differences, determined by statistical test p-values.

M. Burq et al.

⁴⁰

Symptoms monitoring

Monitoring patient condition at home using a combination of unsupervised tasks and passive monitoring using objective sensor measures.

Correlation, determined by rho and confidence intervals, between sensor prediction and clinical scores. Intra class correlation, evaluating test-retest reliability across different study periods.

D. Safarpour et al.

³⁹

Symptoms monitoring

Evaluating rigidity, postural instability and gait difficulties from wearables measures.

Correlation, determined via multivariable linear regression by rho and p-value, between sensor prediction and clinical scores measured in the lab.

FS. Kanellos. et al.

⁵¹

Symptom monitoring

Investigate the correlation between at lab and at home clinical assessment.

Correlation, determined by rho and p-value, between sensor prediction and clinical scores. Bland-Altman test between sensor prediction and clinical score.

JL. Adams. et al.

⁵⁰

Disease prediction, Symptoms monitoring, Progression monitoring

Evaluate differences in digital features between early PD and controls. Links of sensor features and patient symptom severity.

Significant differences, determined by statistical test p-values, in sensor-derived features across different disease populations and patient conditions.

R. Powers et al.

²⁵

Symptom monitoring, Progression monitoring, Intervention decision

Monitoring motor symptom severity using objective sensor measures.

Correlation of sensor derived information with clinical scores, rho and p-value. Significant differences, determined by statistical test p-values, in sensor-derived features across different patient conditions and populations. Accuracy comparison between sensor prediction and clinician’s expectation.