*2.4. Data Processing*

' The raw data of the sensors were corrected, filtered, synchronized, and evaluated using Matlab (R2020b, MathWorks, Natick, MA, USA) as follows. Calibration was performed to compensate for imperfections in magnitude and directional errors of each sensor such that the acceleration was measured in stationary position equal to 1.0 g (gravity, i.e., 9.81 m/s 2 acceleration) in each direction, and the orientation of the measured vector is perpendicular to the planar sides of the sensor's housing when resting on a horizontal flat surface [24]. The detailed description of these calibrations is provided in Appendix A. The quantified imperfections were used to correct the raw data.

Since the arm and chest sensors were recording independently, their data needed to be synchronized in time. This required shifting and scaling operations based on the time landmarks defined by the knocking events at attachment and detachment. Shifting was achieved based on the starting point of the activity assessment, which was dictated by five knocking events on both sensors directly after attachment. Scaling was evaluated and corrected based on the lengths of the measurement periods of the two sensors determined by the time difference of the initial and final knocking events; interpolation ensured that the data were available at the same time points for both sensors.

The measurement noise of the raw data was alleviated using a combination of a lowpass Butterworth filter with a cut-off of 5 Hz to remove high-frequency noise, which was followed by a smoothing step using a moving average filter utilizing quadratic regression with a window size of 60 ms.

Shoulder angle was calculated as the rotational difference between the coordinate systems of the arm sensor versus the chest sensor (Figure 1, right). The calculation method was validated in an experimental setup to ensure <2 ◦ accuracy in the angles between the two sensors. The 0 ◦ angle was defined at the initial knocking event performed directly after sensor attachment, at neutral position of the shoulder with the patient being in an upright position. Due to the limitation that accelerometer sensors can determine their orientation only with respect to the gravity vector in steady states, it was not possible to discriminate the different anatomical components of shoulder rotation. Thus, only a single shoulder angle integrating components of flexion–extension and adduction–abduction could be determined. Large acceleration events, i.e., >1.5 g, were excluded to ensure reliable orientation assessment for the sensors. Moreover, the part of the data related to the sleeping and resting periods of the patients was excluded, as these periods were not of primary interest and could not be reliably assessed due to the technical limitations of accelerometers and their attachment to the skin. Therefore, the final evaluation was restricted to periods when the upper body posture was between −30◦ and +30◦ , as assessed by the chest sensor.

Shoulder elevations, referred to below as "events", were determined as peaks between increasing and subsequent decreasing angle in the shoulder activity data with a minimum prominence of 10◦ .

The average magnitude of shoulder angles and number of shoulder elevation events were evaluated for each post-operative day. Additionally, the changes compared to the direct postoperative status; i.e., the average of the 2–5 postoperative days were evaluated for both the average shoulder angle and number of elevation events. These relative values allowed for a more direct comparison between patients. The two different rehabilitation protocols were compared by averaging the data of all patients per hospital and comparing the outcomes of the hospitals.

#### **3. Results**

This feasibility study included 14 patients (11 women and 3 men), with seven patients treated in each hospital H1 and H2 (Table 1). One patient from hospital H2 had to be excluded from follow-up because of discomfort wearing the activity trackers. Mean age at time of surgery for the remaining thirteen patients was 63 ± 8 years. Five patients were living alone. The injured and tracked arm was the dominant arm in eight patients. Sling removal time was 23 ± 4.5 days and 2 ± 1.5 days in hospitals H1 and H2, respectively.

**Table 1.** Demographic data, shoulder angles and number of daily shoulder elevation events of the patients involved in the study. Sex: F = female and M = male. SD refers to standard deviation.


The total recording time was on average 31 ± 10 days (mean ± standard deviation (SD)). Ten patients had measurements in both first and second 3-weeks periods. Two patients had mild adverse event in form of skin irritation or reactions at the sensor attachment site during the first recording period and could not complete the second period. All adverse events were fully resolved by three months. Another four patients experienced mildly irritated, red skin but were able to take part in the second recording period.

The tracker of four patients ran out of battery before the end of the measurement period, and thus, the collected data were not complete. A single sensor broke and did not

allow the data to be accessed. Six patients reported incidents of tape detachment; these were reattached by the patients themselves. Wherever possible, the date and time of deand reattachments were assessed by the patient and the study personnel. Analysis of the sensor data allowed to identify and correct or exclude these parts of the data.

The overall mean shoulder angle ranged between 11 ◦ and 23 ◦ in all patients (Table 1). The evolution of daily average of shoulder angle over time showed no longitudinal change for most patients (Figure 2 top and middle). This trend was confirmed by the evolution of the relative change of shoulder angle compared to the direct postoperative days, remaining smaller than 5 ◦ (Figure 2, bottom).

**Figure 2.** Daily average shoulder angle results. (**Top**): the mean (blue line) and standard deviation (gray zone) of shoulder angle for each postoperative day of a patient (H2\_P04). The overall mean ± standard deviation of shoulder angle for the entire tracking period of this patient was 16.3 ± 9.3 ◦ . (**Middle**): absolute daily average of the shoulder angles for each patient. (**Bottom**): normalized (compared to the direct postoperative state) daily average of the shoulder angles for each patient. Note that the data were not available throughout the entire six-week period for some patients.

The overall average number of daily shoulder elevation events ranged between 547 and 5756 in all patients (Table 1). The number of daily events increased for most but not all patients (Figures 3 and 4, top). The relative change of daily event numbers compared to the direct postoperative days showed an increasing trend over time, reaching up to 300% increase (Figure 4, bottom).

**Figure 3.** Number of shoulder elevation events of one patient (H2\_P04). (**Top**): evolution of the number of shoulder elevation events throughout the tracking period. (**Bottom**): evolution of the number of shoulder elevation events throughout the tracking period categorized into three ranges according to the maximum shoulder angle reached. The daily mean ± standard deviation of shoulder elevation events of this patient was 4109 ± 1684.

There were no characteristic differences between the two clinical sites, i.e., rehabilitation protocols, in terms of the longitudinal evolution of the change in the average shoulder angle (Figure 5). However, the evolution of the percentile changes in the number of events relative to the postoperative period was increasing for H2 but not for H1, and the differences between sites became more pronounced for higher elevation thresholds (Figure 6).

**Figure 4.** Absolute (**top**) and relative ((**bottom**), compared to the direct postoperative state) number of daily shoulder elevations larger than 10 ◦ , shown for each patient. Note that the data were not available throughout the entire six-week period for some patients.

**Figure 5.** Comparison of the average results of the two clinical sites, i.e., rehabilitation protocols, in terms of the change in the shoulder angle compared to the direct postoperative period.

**Figure 6.** Comparison of the average results of the two clinical sites, i.e., rehabilitation protocols, in terms of the change in the total number of shoulder elevation events over time. The sub-figures show the data for shoulder elevation events beyond 10 ◦ (**top**), 20 ◦ (**middle**) and 30 ◦ (**bottom**).

#### **4. Discussion**

The primary findings of this study were that in patients after proximal humerus fracture undergoing ORIF, the mean shoulder angle varied up to a factor 2 between individuals, but it hardly increases in the first six weeks. The number of events exhibited a 10-fold difference between subjects, and the time evolution of event numbers showed an increasing trend. The comparison of the two hospitals indicated that the rehabilitation protocol might affect the number of daily shoulder elevation events with patients following an immediate functional non-weight bearing rehabilitation program having a higher number of events, especially for large shoulder angles.

'

Postoperative rehabilitation protocols after proximal humerus ORIF vary substantially between different hospitals and surgical centers ranging from strict immobilization using a shoulder sling to a more progressive, functional non-weight-bearing approach without sling immobilization [25]. If immobilized, the duration of postoperative sling usage ranged from none to eight weeks [25]. A recent review summarizing five comparative studies did not find any benefit of longer sling immobilization compared to early functional therapy. While exercise and early functional mobilization is clearly advised, the amount and influence of postoperative mobilization and its effect on clinical and subjective outcomes as well as revision and failure rates are still unknown [26,27].

When recording physical activity or joint motion, wearable activity trackers are frequently used, and single day data acquisition was our method of choice [28]. While this might be less burdensome for the patient in comparison to long-term recording, the informative value of these data is limited and long-term recordings are recommended. To the best of the authors' knowledge, the present work represents the first study providing detailed insights into the longitudinal evolution of postoperative shoulder activity of surgically treated patients with proximal humerus fracture. The relevance of long-term measurement was underlined by the time evolution of the assessed parameters. Although the average daily shoulder angles remained fairly constant over time, the number of shoulder angles showed an increasing trend over time, exhibiting important differences between patients, especially after three weeks postoperatively and in particular for large shoulder angle events. These may be related to the different rehabilitation protocols adopted by both investigation sites. While in the first two weeks, there is no difference with regard to the total number of elevation events >10◦ , an increased number of events for patients following the unrestricted, i.e., non-weight bearing, rehabilitation protocol was recorded thereafter. For non-operatively treated proximal humerus fractures, early active rehabilitation yields equal complications and shoulder functions as prolonged sling immobilization and restricted rehabilitation [26,29]. Similarly, the present data suggest a potential benefit of early active rehabilitation in terms of faster return to motion and function compared to a more restricted rehabilitation protocol. Nevertheless, the influence of a patients' preoperative activity level on the amount of shoulder activity in the early postoperative phase is still unknown, and therefore, caution is needed when interpreting these results.

Recording detailed postoperative shoulder activity, using wearables is challenging, and little high-quality knowledge exists. Van de Kleut et al. investigated daily shoulder activity before and after reverse Total Shoulder Arthroplasty (rTSA) using Inertial Measurement Units (IMUs) [30]. Their results showed an increased frequency of arm elevations to higher angles but no difference in the amount of time spent in the elevation. Moreover, shoulder elevation accounted for less than 1% of daily shoulder motion, and even after one year postoperatively, patients spent more than 95% of the day in shoulder angles below 60◦ [30]. These results compare to the present work, where patients spent 94% of the time in shoulder angles below 40◦ . The initial increase in shoulder events seen in the present study may be due to postoperative physiotherapy, which is in line with previously reported data showing a significant increase in events only in the early postoperative period but not thereafter [30]. This can be explained by the fact that physiotherapy is adapted to the state of the patient starting with simple exercises that become more challenging over time. Therefore, it is more likely to see a general increase in activity over time which is the case in the present work. Furthermore, physiotherapy is performed only during a limited amount of time during the day and might therefore have only a limited effect on daily shoulder activity. In general, the present data show that after open reduction and internal fixation, the shoulder activity level of patients is low and that the early return to full range of motion is not seen in the first weeks.

This study has some limitations. The small number of patients included into this pilot study did not allow for meaningful statistical analysis to be performed, but the indicated trends can be used to design more specific and focused investigations. Moreover, the feasibility of long-term tracking was assessed, providing novel insights and highlighting

potential pitfalls. Technical limitations included the issue that the battery of the activity trackers did not always last for the desired time window of three weeks. Since for the calculation of the shoulder angle, the recordings of both activity trackers are necessary, the analysis could only be conducted as long as both activity trackers were functional. Detachment of the tape fixing the sensors occurred due to the loss of adhesion to the skin or ruptures of the material, causing a partial unavailability of data until reattachment. The assessed shoulder angles were not validated against optical tracking techniques. However, the method for calculating the shoulder angle used here was similar to the one applied by Chapman et al. [31], who validated their results against a laboratory motion capture system and reported errors smaller than 2◦ for abduction, forward flexion, internal and external rotation. In addition, over the long-term, i.e., days to weeks, activity monitoring application with thousands of events, the accuracy of a single event is less critical as the focus is on behavioral change and the large data sample compensates for a potentially lower accuracy compared to what would be needed during a single functional test. Shoulder activity monitoring by counting events beyond certain joint angle thresholds may be affected by the lifestyle of the subject. Thus, correct interpretation of the absolute number of events would require a pre-trauma reference. With the latter being hardly possible, in future studies, the unaffected shoulder could be monitored simultaneously for an intra-subject reference and potential transfer of activities during the rehabilitation phase.
