Toward daily-life assessment of rehabilitation treatments translation Alessandra Pedrocchi, Politecnico di Milano 31 October 2013 This Paper proposes a method to evaluate motor capacity and its possible changes using wearable sensors. Eventually, the utmost goal is to register daily-life activities with miniaturized sensors in a non-standardized environment and to recognize on the recorded data valuable information to monitor the level of impairment and the translation of positive outcomes of rehabilitation treatments into functional improvements in everyday tasks. This ultimate goal will be achieved if we will be able to define intermediate aims. These middle steps will indeed pave the way for the final destination by introducing partial benefits and validating the proper methodologies and technologies. Material point of the of the whole path is to create teams that gather specialists with different backgrounds: clinicians, technicians, statisticians, and mathematicians. Also, maximal consideration should be devoted to users' feedbacks, taking into account usability, feasibility, and users' acceptance, especially if aiming at monitoring motor capacity changes during children real daily life. The present contribution by Strohrmann and colleagues represents a valuable step forward, especially because it targets pediatric population, which has been under-investigated so far. TECHNOLOGY - From the technological point of view, the wearable sensors based on accelerometers, gyroscopes, and inertial measures are currently validated and provide reliable measures. The authors used a set of ten sensors, rather cumbersome and obtrusive, even if improved in shape as much as possible (Fig. 1). Nevertheless, they eventually suggest the effectiveness of a three sensors setup (two wrists and hip), based on evidence from recorded data. This conclusion is very positive in the framework of daily-life activity assessment and could represent a good starting point for further investigations. In particular, the proposed position of the remaining sensors allows to further develop their shape and design, through a more acceptable final project (i.e., wrists sensors could be designed as modern watches). SAMPLE SIZE- Besides the discussion of any specific result of new methods, the most important elements founding the robustness of any validation are the sample size and selection of gold standards. Statistical methods to define the sample size power are currently used in clinical trials and should be applied to any new outcome measure definition. In the definition of the tested sample, two different and somehow competing perspectives needs to be addressed. On one hand, there is the strong need to seek for solutions capable to be applied to a wide population. On the other hand, the wider is the target population, the greater should be the number of tested subjects to achieve a generalized validation. In addition, while considering a finalized research context, the time needed to achieve conclusive results and to publish them to be used in the scientific and clinical community could be strongly delayed when imposing very strict inclusion criteria to the target population. In my opinion, the best solution is to enlarge the collaborating team involving multiple clinical centers. This scenario allows us to both focus the research on a well defined population and to reach a significant sample size in realistic time span. Therefore, even if an accurate training to all involved clinicians would be needed to assure comparability of results, the robustness of results would be largely improved. GOLD STANDARD DEFINITION - A very crucial point when considering validation of new methodologies is the selection of the gold standard. The selection of widespread clinical scales would be preferable (such as in their references 15; 18 25;26). The authors of the considered paper used as ground truth reference a custom 4-scores scale for clinical evaluation of the tasks that was performed by only two experts. This solution is a preliminary validation step, but further studies are required to address better this issue PARAMETERS SELECTION - One of the main issue currently open in bioengineering is to define significant parameters from the tan of measures that can be even easily recorded. Extracting few significant outcomes parameters from all available data requires accurate validation, strong medical support, and criticism. Tackling this issue, the authors started their analysis with a pool of many parameters. They designed a regression model to fit the parameters into the ground truth reference, so as to be able to identify the most significant features while evaluating motor capability changes. They eventually selected, for the hand tasks, a subset of parameters (three) which can be a good reference, avoiding task dependency (fig.11). The results described in the paper represent a good step forward to be extensively validated in further studies with proper sample size, a more recognized gold standard and then a robust statistical validation. FROM PREDEFINED PROTOCOLS TO DAILY RECORDINGS - Lastly, another element still to be considered is the transition from the acquisition of supervised predefined tasks (capability, as defined by authors) to daily-life activities (performances in authors' terminology). Indeed, data in daily unsupervised recordings are expected to include some "activities" and some "non-activities" periods. The authors are completely ignoring this problem and they rely their analysis on the a priori defined relevant fragments of recordings. To tackle this open point, I would like to scratch three possible levels of solution (which could be desirably combined): 1) gradually move to a completely unsupervised experimental conditions. Start to get rid from a definition of start-and-end of tasks, still considering a reduced well defined set of possible "activities" that could be somehow identified 2) design and validate dedicated algorithms to distinguish between the fragments of "activities" and "non-activities" on recorded data; 3) validate the assessment parameters over the mixture data of "activities" and "non-activities". In the near future, this reviewer would appreciate that studies aiming at the development of rehabilitation assessment platforms in daily-life activities would include all the discussed elements. In particular, it would be of great interest the study and the discussion of the use of a reduced set of wearable sensors (a priori defined), the acquisition of data over a predefined experimental setup but not limited to well defined tasks (avoiding start-and-end definition) and the validation of methods against generally recognized outcome measures over a significant number of subjects, statistically verified. In this field of research, a useful and effective approach is to develop open source datasets with a clear description of the performed experiments. Even if barely used so far, this technique may allow many experts to test different approaches and methodology of data processing and to enlarge datasets to allow more robust validations. I thank the authors for providing some interesting elements which are paving the way to more complete future studies in this direction, that could really lead to a proper evaluation of daily-life activities in patients real environment. Competing interests I declare I have no competing interests.