Use of Wearable Sensors to Measure Improvements in Real World Upper Extremity Activity After Brachial Plexus Reconstruction

Abstract

Purpose

Current functional assessments after brachial plexus (BP) reconstruction, including surgeon-graded active range of motion and strength scores, and patient-reported outcomes (PROs), have limitations. This study evaluated the feasibility of wearable sensors to quantify recovery and functional outcomes following BP reconstruction. We hypothesized that sensor-derived measures of daily upper-extremity use would correlate with PROs and that real-world upper-extremity use would improve.

Methods

Traumatic BP injury patients wore bilateral wrist accelerometers for 7-day periods at 3- to 6-month intervals. Data were grouped relative to timing of reconstruction. Duration and magnitude of upper-extremity activity were expressed as ratios of injured to uninjured limb (use and magnitude ratios). Disabilities of the Arm, Shoulder, and Hand; Impact of BP Injury; 36-Item Short Form Health Survey; and Patient-Reported Outcomes Measurement Information System-Upper Extremity scores were collected. Correlations assessed relationships between accelerometry data and PROs. Linear mixed models characterized changes in these relationships over time and the effect of time from surgery on accelerometry data.

Results

Twenty-seven participants were enrolled (82% male, 78% White). Mean age at injury and surgery were 41.7 ± 14.0 years and 42.5 ± 14.3 years, respectively. The use ratio correlated significantly with Disabilities of the Arm, Shoulder, and Hand (r = −0.45, P < .05), Impact Disability (r = −0.56, P < .05), 36-Item Short Form Health Survey Physical Function (r = 0.26, P < .05), and Patient-Reported Outcomes Measurement Information System-Upper Extremity (r = 0.38, P < .05) scores. The magnitude ratio significantly correlated with Disabilities of the Arm, Shoulder, and Hand (r = −0.44, P < .05), Impact Disability (r = −0.50, P < .05), and Patient-Reported Outcomes Measurement Information System-Upper Extremity (r = 0.47, P < .05) scores. Linear mixed models supported these associations but showed no effect of time from surgery on use or magnitude ratios.

Conclusions

Wearable sensor-derived use and magnitude ratios demonstrate convergent validity with PROs following BP reconstruction.

Clinical relevance

Wearable sensors may offer a more direct, objective assessment of functional recovery to support postoperative assessment, prognostication, treatment planning, and counseling beyond traditional clinical measures.

TRAUMATIC BRACHIAL PLEXUS injuries (BPIs) are devastating injuries that severely affect quality of life, both physically and psychologically.^1–3 Brachial plexus (BP) reconstructive surgery aims to reduce pain and improve function. Expectations for outcomes and perceptions of recovery after surgery vary tremendously among patients.^4,5 It is challenging for the surgeon to predict the speed and extent of improvement a patient will achieve, often leading to patient frustration during their lengthy recovery.⁶

The traditional method of monitoring recovery and reporting outcomes after BP reconstruction relies on active range of motion (AROM) measurements and surgeon-graded muscle strength via the Medical Research Council (MRC) scale.^7,8 Although these tools help quantify improvements postreconstruction, they have limitations. They test joints in isolation, contrary to the normal functional actions of the upper extremity (UE), which require concurrent motion at multiple joints.⁹ These measurements are also prone to inconsistency, inaccuracy, and interrater and intrarater variability and may not reflect real-world functionality because testing is performed in controlled clinical settings.⁹

In light of these shortcomings, there has been a shift to incorporate patient-reported outcome (PRO) measures into longitudinal functional assessment. Patient-reported outcomes, such as the Disabilities of the Arm, Shoulder, and Hand (DASH) survey and the Impact of Brachial Plexus Injury Questionnaire, offer insight into patients’ perceptions of their recovery.¹⁰ Combined with MRC and AROM measurements, surgeons try to craft a more holistic assessment of patient progress and use this information to adjust postoperative management. Nonetheless, these PROs are limited by subjectivity, prone to bias and less than ideal psychometric properties, and are, at best, an indirect quantification of UE activity in daily life.^11,12 For instance, PROs may be associated with variability over time and inconsistencies in the relationship between self-reported and direct measured performance within the stroke population.¹³ To address this measurement gap, clinicians need better tools. Accelerometry data from wearable wrist sensors have been validated in quantifying bilateral UE function in the real-world and are becoming increasingly used in the treatment of patients with UE neurological deficits.^7,14–19 Initially used in stroke patients, these sensors have also been used successfully in analyzing function after nerve reconstruction in the neonatal BP palsy population as well as in a small cohort of adult patients with traumatic BPIs.^17–19

The purpose of this study was to explore the feasibility of using wearable sensors to directly quantify recovery trends and outcomes postsurgery in adult patients with traumatic BPIs. Our goals were twofold. First, we examined relationships between measurable UE use from accelerometry data and PRO instruments to assess convergent validity. Although the accelerometer variables cannot determine in which specific actions people are engaged, they can directly quantify UE activity in daily life.²⁰ We hypothesized that increases in real-world upper-extremity use, as quantified by wearable sensor data, would correlate with self-reported improvement. Second, in quantifying use both before and after surgery during real-world activities, we evaluated how much improvement in the injured extremity occurred postreconstruction at selected timepoints, assessing longitudinal validity. We hypothesized that there would be a direct improvement in real-world use with time from reconstruction.

MATERIALS AND METHODS

After study approval by the institutional review board of the authors’ home institution, a convenience cohort of individuals with traumatic BPIs sustained as early as September 2018 was recruited from a BP and peripheral nerve clinic staffed by two orthopedic surgeons at a level 1 trauma and tertiary center. Brachial plexus injury was defined as injury to at least two major peripheral nerves proximal to the pectoralis major insertion. A priori, it was determined that a sample size of 25 would be sufficient to detect a correlation of r = 0.53 between self-reported and direct measure of change, with 95% confidence and 80% power.

Participants aged 18—99 years with a confirmed diagnosis of traumatic unilateral BPI were included in this study if they had undergone reconstructive surgery after September 2018 or were scheduled for reconstruction in the near future. Exclusion criteria included pursuing nonsurgical management, history of prior BP reconstruction at another institution, and birth-related or nontraumatic etiologies of BPI (eg, secondary to radiation or neuritis). Written informed consent was obtained from each participant, and Health Insurance Portability and Accountability Act authorization was obtained as required. Demographic data including BPI pattern were collected.

Participants were followed longitudinally every 3—6 months for up to 2 years from initial recruitment. Those recruited before surgery or within 3 months after surgery were included in the preoperative group because of the limited recovery typically observed in early postoperative periods, amplified by restrictions implemented by the care team. Additionally, given the nature of neuroregeneration after surgery, the authors did not expect any meaningful recovery within the first 3 postoperative months, therefore reasoning that these baseline data will be equivalent to preoperative collection.^21,22

At each timepoint, participants completed four surveys: DASH, measuring perceived disability across pain, impact and function subscales, with higher scores indicating worse disability; Impact of BPI Questionnaire, measuring psychological and physical consequences of BPI, also with greater scores dictating more dysfunction; 36-Item Short Form Health (SF-36) survey, measuring health-related quality of life across eight categories (with focus on physical function scores); and Patient-Reported Outcomes Measurement Information System (PROMIS) survey.^10,23,24 Although data from pain interference, UE function, depression, and anxiety subgroups were collected, analysis was performed using PROMIS-UE scores alone, in which lower scores indicated higher levels of self-reported dysfunction.

Participants received two GT3X+ wrist-worn devices at each interval, either at their office visit or via post. They were instructed to wear one device on each wrist for 7 consecutive days at each timepoint. The main sensor in the device is a 3-dimensional accelerometer with a range of ±8 g. Sampling frequency was 30 Hz. Data were downloaded, visually inspected to confirm wearing, and filtered using ActiLife software (v6.13.3; ActiGraph LLC). Data were bandpass filtered (0.5—2 Hz) to remove accelerations because of machines (eg, elevators, moving cars), converted to activity counts, and downsampled 1 second epochs using ActiLife proprietary algorithms.²⁵ Custom-written software in R was used to compute two variables: use ratio and magnitude ratio.⁷ Briefly, the 7-day, 1 second time-series data were converted to vector magnitudes [√(x² + y² +z²)], and a <2 activity count threshold was used to determine if movement occurred or did not occur.^26,27 The duration of UE activity was the sum of the seconds during movement, and the magnitude of UE activity was the median value when the limb was moving. These values were expressed as ratios of the affected side to the unaffected side and called the use ratio and the magnitude ratio, respectively.⁷ Importantly, in people without neurological or physical impairment, the use ratio and magnitude ratio are narrowly distributed and stable across age groups.²⁸ In a sample of adult control participants, the use ratio between nondominant and dominant UEs was 0.95 ± 0.06, whereas the non-transformed magnitude ratio (reported by Bailey and Lang as log-transformed) was 0.97 ± 0.11.¹⁵

Feedback from accelerometry data processing was not provided to participants to avoid influencing their future PRO reporting and avoid introducing response bias.

Statistical analysis

Data were grouped by time intervals (before surgery to 3 months after surgery, 3—6 months after surgery, 6—12 months, 12—18 months, etc). A Spearman’s correlation analysis was performed including all timepoints to assess the correlation between direct measures of UE activity in daily life (via accelerometry data) and self-reported daily activity and function (via PROs), representing cross-sectional convergent validity. A linear mixed model analysis was performed to assess the longitudinal validity over time, accounting for correlation of observations within subjects. A secondary linear mixed model analysis was also performed to evaluate the relationship between accelerometry data and general mental health (emotional well-being) scores over time.

A third exploratory linear mixed model analysis was used to evaluate the effect of time from surgery on UE use. Additionally, the relative use and magnitude ratios at each timepoint with respect to first recorded baseline was calculated and plotted to depict percent change over time. Because of heterogeneity in injury patterns and surgical treatment, subgroup analyses based on type of surgical treatment were not conducted. All statistical tests were two-sided with a significance level of 0.05.

RESULTS

Forty-four patients with traumatic BPIs treated with surgery or scheduled for reconstruction were identified in the clinic. Figure 1 shows the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) chart of participant enrollment. Data from 27 enrolled participants were further analyzed, with 10 participants undergoing baseline collection in the preoperative period. The remaining 17 were excluded for varying reasons including missed appointments and incomplete questionnaires. Notably, there were 28 instances of data collection where PROMIS-UE data were not obtained with other questionnaire data because of participants not returning/completing the PROMIS-UE. Mean age at time of injury and surgery were 41.7 ± 14.0 years and 42.5 ± 14.3 years, respectively. 81.5% were males (n = 22). 77.8% identified as White (n = 21), whereas the remainder were Black or African American (n = 6). Fourteen participants (51.9%) sustained injury to their dominant hand. Four did not report hand dominance. Participants underwent reconstruction on average 9.6 months from injury and were followed for an average of 19.4 months postreconstruction. Brachial plexus injury pattern and therefore type of surgical reconstruction varied greatly among participants (Table 1). The number of data points per individual is reported in Table 2. Further descriptions of the sample are reported in the Appendix (available online on the Journal’s website at www.jhandsurg.org).

FIGURE 1:

Participant enrollment.

TABLE 1.

Patterns of Brachial Plexus Injury

Pattern of BPI	Number of Participants Who Sustained This Injury
Panplexopathy	7
Upper trunk	9
Combined postganglionic upper and preganglionic lower trunk	1
Combined C5 preganglionic and C6—C8 postganglionic nerve	1
Posterior and medial cord	1
Lateral and medial cord	1
Posterior cord	3
Medial cord	1
High radial nerve, distal lateral cord, and distal medial cord	1
High radial nerve and lateral cord injury	1
Axillary and suprascapular nerve injury	1

TABLE 2.

Data Points Log

Number of Data Points	Number of Participants
	Enrolled Before Surgery	Enrolled After Surgery	Total
1	5	4	9
2	2	5	7
3	2	0	2
4	1	4	5
5	0	4	4

N = 27

Correlations between PRO data and the use ratio and magnitude ratio were statistically significant, suggesting cross-sectional convergent validity (Table 3). Longitudinal validity was suggested by the linear mixed model analysis, which demonstrated statistically significant decreases in DASH and Impact Disability scores when the use ratio and magnitude ratio increased over time (Table 4). Similar relationships were found with the SF-36 physical function and PROMIS-UE scores and the use ratio. In contrast, there was no significant effect of time from surgery on the relationship between accelerometry data and emotional well-being scores (Table 4).

TABLE 3.

Spearman Correlation Analysis Modeling the Relationship Between Upper-Extremity Function and Patient-Reported Outcome Data

Score	Spearman’s Correlation
	Number of Data Points	Use Ratio	P Value	Magnitude Ratio	P Value
DASH	69	−0.45	<.05	−0.44	<.05
SF-36 physical function	69	0.26	<.05	0.16	.19
Impact disability	69	−0.56	<.05	−0.50	<.05
PROMIS-Upper Extremity	41	0.38	<.05	0.47	<.05

TABLE 4.

Linear Mixed Model Analysis Modeling the Association Between Upper-Extremity Function and Patient-Reported Outcome Measures Over Time

Effects	Solution for Fixed Effects
	Use Ratio			Magnitude Ratio
	Estimate	Standard Error	Pr > |t|	Estimate	Standard Error	Pr > |t|
DASH	−0.0025	0.00076	<0.05	−0.0021	0.00071	<0.05
SF-36 physical function	0.0028	0.00075	<0.05	0.00081	0.00076	0.29
SF-36 Emotional well-being	−0.000030	0.00094	0.98	0.00033	0.00085	0.70
Impact disability	−0.0033	0.0009	<0.05	−0.0026	0.00082	<0.05
PROMIS-UE	0.0083	0.0027	<0.05	0.0021	0.0015	0.17

Figure 2 shows the individual participant trajectories over time. By the 2-year mark, participants on average used their injured extremity 68% of the time (Fig. 2A) at 59% of the intensity of their uninjured extremity (Fig. 2B). Notably, there was considerable variability in the functional trajectories postreconstruction. An exploratory linear mixed model analysis aimed at characterizing the effect of time on changes in the ratios indicated that time from surgery did not have a significant effect on use ratio (P = .17) or magnitude ratio (P = .44), potentially because of the small sample size.

FIGURE 2:

A Individual trajectories of the use ratio. B Individual trajectories of the magnitude ratio. Normative values (±1 SD around the mean) from neurologically intact adults are indicated by the gray shaded rectangles.

DISCUSSION

Traumatic BPIs are known to dramatically alter the course of a person’s life. After reconstructive surgery, current clinical practice relies on PRO surveys in concert with AROM and strength assessments to monitor recovery.^29–31 Although jointly these metrics yield a more nuanced assessment, results from these forms of assessment should be interpreted cautiously. Accelerometry data from wrist-worn sensors potentially provide additional insight into recovery. Although limb use is not equivalent to isolated motor capacity or movement quality or does not imply task success, in the context of traumatic BPI, greater spontaneous use of the injured extremity during daily activities, as measured via accelerometry data, reflects an integrated and relevant dimension of functional recovery, indicating that the limb is sufficiently capable to be incorporated into everyday tasks. This study builds on the existing literature by tracking BPI patients longitudinally in real-world settings. The results demonstrate a complementary association between self-reported outcomes and direct outcome measures at a single point in time as well as over time. These data illustrate the potential utility of these devices for direct quantification of postoperative recovery after BP reconstruction.

A key aspect of this study was the relationships between PROs and accelerometry data. Though functional improvement postsurgery was neither uniform nor predictable, the data show an association between measurable use and PRO data.^{10,23,29,32–34} The magnitude of the relationships found here is comparable to those found in other neurologically impaired populations.¹³ Higher disability scores (ie, DASH and Impact scores) correlated with lower use and magnitude ratios, whereas more favorable scores on SF-36 and PROMIS-UE aligned with greater functional use of the injured extremity, demonstrating cross-sectional, convergent validity between the two measurement types. Convergent validity was further substantiated by the longitudinal association of improved PROs with higher measurable usage over time. Understanding relationships between self-reported outcomes and direct measurements of functional recovery has several potential implications. First, these findings support the clinical relevance of PRO tools, showing that patients’ perceptions often align with or inform their actual functional performance. As such, treatment regimens can be reliably adjusted based on individual instruments in conjunction with standard AROM and MRC scores. Second, in situations where sensors could be used, combining accelerometry data with PROs and clinician-based assessments may allow for a more holistic assessment of recovery and inform future goal setting and expectations counseling. Third, direct quantification of recovery in daily life using accelerometry could be a powerful research tool for identifying improved surgical techniques in the future. Lastly, it is important to note that if future investigations demonstrate continued strong associations between accelerometer data and PROs, there may not be substantial advantage to using both forms of assessment given the increased cost and time associated with the former. However, there may be a complementary role for the two forms of assessment.

Following reconstruction, there was immense variability in recovery among participants over time. Although some demonstrated large increases in use and magnitude ratios, others plateaued or experienced smaller increases or even decline. This unpredictability is important for surgeons to note in preoperative counseling sessions to manage expectations after surgery. The authors believe that if these devices can be accessed and afforded, they are likely to contribute helpful data that can tailor care. This has been successfully incorporated into clinical rehabilitation care and the electronic medical record in the physical and occupational therapy program of the authors’ home institution. For patients who exhibit a high recovery of function or continued improvement in use over a long period, they can be encouraged to progress further with therapy and reenter the workforce. The authors believe that visualizing progress may serve as a motivational tool for some patients. For those slower to progress or with declining function, clinicians may use data from these sensors to adjust treatment regimens and therapy practices on an individualized basis. Despite the lack of a clear time effect on the relationship between accelerometry metrics and the SF-36 general mental health score, the authors believe it prudent to center emotional well-being during treatment for the following reasons: existing studies have shown notable mental health comorbidity rates in the traumatic BPI population, whereas others have demonstrated an association between function and emotional well-being, with preoperative depression boding poorly for postoperative function and with restoration of elbow function seen in relation with improved quality of life and disability.^35–37

This study has some limitations. First, the sample size was small and composed mostly of male participants in a single urban institution limiting the generalizability of the findings. The variability in BPI pattern and the predominantly male cohort are representative of the traumatic BPI population. Second, inconsistencies in data collection in this convenience sample because of clinical follow-up time and/or participant noncompliance led to a substantial amount of missing data. Because of incomplete collection of AROM and MRC measurements, we were unable to evaluate their relationship with the use ratio and magnitude ratio. Missing data and the small sample may have limited the ability to detect statistically significant change over time in the accelerometry variables. Third, the devices do not detect sensory recovery and are limited to motor activity as a measure of acceleration. Last, wrist-worn devices cannot measure finger movement, potentially underestimating recovery in tasks requiring hand dexterity. However, this effect is likely minimal given that the majority of everyday tasks require coordinated motion across the joints of the upper limb, not just finger movements in isolation.³⁸ This is corroborated by Bailey et al¹⁴ who demonstrated the ability of these sensors to quantify most UE activities, except highly skilled typing when the wrists were still.

With more robust data collection procedures and the potential to build a large BPI database, further studies may be able to incorporate a subgroup analysis for the various patterns of BPI encountered and/or their surgical treatments. Additionally, as wrist sensors become more commercially accessible, we anticipate incorporating their use into the general recovery course of these patients. This may help surgeons and therapists to assess outcomes in a more meaningful way, enabling them to adjust therapy protocols and/or make informed decisions on adjunct surgeries. Clinicians will have access to another tool that will inform expectation setting and serve as a foundation for preoperative counseling. Integration of these data collection methods into a larger, multicenter study could ultimately enable comparative evaluation of surgical techniques and predictive models for amount and pace of recovery after BP reconstruction.