Accelerometry Data Delineates Phases of Recovery and Supplements Patient-Reported Outcome Measures Following Lumbar Laminectomy

Abstract

Objective

Patient-reported outcome measures (PROMs) are traditionally used to track patients’ recoveries after spine surgery. Wearable accelerometers have adjunctive value because of the continuous, granular, and objective data they provide. We conducted a prospective study of lumbar laminectomy patients to 1) determine if time-series data from wearable accelerometers could delineate phases of recovery and 2) compare accelerometry data to PROMs during recovery tracking.

Methods

We prospectively recruited lumbar stenosis patients indicated for lumbar laminectomy. Subjects wore accelerometers that recorded their daily step counts from at least 1 week preoperatively to 6 months postoperatively. Subjects completed the Oswestry Disability Index (ODI) and the 12-Item Short Form Survey (SF-12) preoperatively and at 2 weeks, 1, 3, and 6 months postoperatively. Daily aggregate median steps (AMS) and individual visit-specific median steps (VSMS) were calculated. The Pruned Linear Exact Time method was used to segment AMS into distinct phases. Associations between VSMS and PROMs were identified using Spearman rank correlation testing.

Results

Segmentation analysis revealed three distinct postoperative phases: step counts rapidly increased for the first 40 days postoperatively (acute healing), then gained more slowly for the next 90 days (recovery), and finally plateaued at preoperative levels (stabilization). VSMS was significantly correlated with PROMs throughout the postoperative period. However, PROMs significantly exceeded baseline at 6 months postoperatively while step counts did not (all p < 0.05).

Conclusion

Continuous data from accelerometers allowed for identification of 3 distinct stages of the postoperative period after lumbar laminectomy. PROMs remain necessary to capture subjective elements of recovery.

INTRODUCTION

Recovery after lumbar spine surgery involves reduced mobility, which impacts patients’ independence, activities of daily life, and employment. To counsel patients about the trajectories of their recoveries, surgeons rely upon existing outcomes literature -mainly patient-reported outcome measure (PROM) data- and their own clinical experiences. However, PROMs fail to track outcomes, and specifically physical function, in a continuous and objective manner outside of the clinic. Moreover, the interpretation of PROM values is dependent on when during recovery they are measured but, there is substantial heterogeneity in the current literature on the timing of postoperative PROM measurements.^1–3

Data from wearable fitness trackers (ex. Fitbit, Garmin) may fill these evidence gaps using continuously measured physical activity data that could inform 1) optimal timing of postoperative PROM measurement by identifying distinct stages in the recovery process, and 2) preoperative counselling strategies. Previous studies in orthopaedic populations, including lumbar spine patients, have established that wearable accelerometers are capable of tracking postoperative activity levels.^4–6 However, it is unclear to what extent wearable accelerometer data represent a novel PROM as the relationship between accelerometry-measured activity levels and traditional PROMs -such as the Oswestry Disability Index (ODI) and 12-Item Short Form Survey (SF-12)- has varied between studies.^4,7,8

This prospective cohort study tracked the physical function of lumbar stenosis patients indicated for lumbar laminectomy using wearable accelerometers. The aims of our study were to 1) identify specific phases of postoperative recovery using a change point detection analysis of accelerometry data and 2) compare accelerometry-defined and PROM-defined recovery patterns. We hypothesized that our analysis of accelerometry data would define distinct stages of recovery after surgery and accelerometry-measured physical activity levels would not correlate with traditional PROM data, thus signifying a novel measure quantifying patients’ recoveries.

MATERIALS AND METHODS

Study Design and Patient Recruitment

Institutional review board (IRB) approval was obtained prior to commencing recruitment. The study cohort was recruited from patients that presented to a single high-volume urban, academic center (Mount Sinai Hospital, New York, NY) from July 2018 to March 2020 for evaluation of lumbar stenosis with or without spondylolisthesis who were subsequently indicated for primary lumbar laminectomy with or without fusion up to two levels. Recruitment was suspended in March 2020 due to the COVID-19 pandemic. Patients underwent surgery with one of four surgeons, including the principal investigator.

Patients were excluded if they were not fluent in English, had previous spine surgery within 5 years of recruitment, or had other medical conditions that could significantly inhibit mobility (including cardiovascular conditions and other orthopaedic conditions). Patients were also excluded if they did not own a smartphone capable of connecting to a consumer-grade fitness tracker device via Bluetooth.

Patients were enrolled in clinic or over the phone at least 1 week prior to surgery. Informed consent was obtained at the time of enrollment. Patients were permitted to withdraw from the study at any time and could be withdrawn by the study team due to noncompliance. Data from withdrawn patients was included in the analysis if the patient had completed the preoperative data collection period with suitable compliance (i.e. recorded accelerometry data for at least 4 of 7 days prior to surgery).⁹ Patient demographic and perioperative data were collected via chart review of electronic medical records.

Accelerometry Measurements and PROMs

The primary outcome measure was steps taken per day, which was recorded using the Fitbit™ Flex 2 wrist-worn accelerometer, a consumer-grade device, and used as a proxy for daily physical activity level. The Fitbit™ Flex 2 was chosen because of its comparable validity and inter-device reliability to research-grade accelerometers as well as the added benefit of a more consumer-friendly design which would, in theory, make the device easier for subjects to use and thereby comply with the study protocol.^10,11 Other physical activity measures such as distance travelled per day or calories burned per day were not used because the Fitbit™ Flex 2 derives these measurements using steps per day as the primary measure.

Subjects were provided with an accelerometer upon enrollment and were instructed to wear the device on the non-dominant wrist daily until 6 months after surgery. They were also instructed to follow the postoperative care guidelines provided by their surgeon including any recommendations for activity avoidance and/or physical therapy. Subjects were not required to wear the device while sleeping. Accelerometry data from each subject’s device was recorded continuously and study compliance was monitored using a wellness portal data management platform. Compliance was defined as the ratio of days that a subject recorded data to the total number of days in each study interval. Patients who failed to record data for three consecutive days were contacted by the study team. Patients with less than 57% (i.e. 4 out of 7 days) compliance between two study time points were excluded from any statistical analyses involving that study interval.

Secondary outcomes included lower back pain-related level of disability as measured by the Oswestry Disability Index (ODI), walking disability as measured by the ODI Section 4 subscore, and overall physical health status as measured by the 12-Item Short Form Survey (SF-12) Physical Component Score (PCS). Subjects recorded responses to the ODI and SF-12 upon enrollment at least 1 week prior to surgery and at the major postoperative milestones: 2 weeks (ODI only), 1 month, 3 months, and 6 months after surgery. The SF-12 was not completed at the 2-week follow-up because the survey requires a recall period of 4 weeks. All PROMs were administered via email.

Statistical Methods

All statistical analyses were performed using the Python^© Pandas data analysis library and Microsoft Excel. P-values less than 0.05 were considered significant.

To quantify activity levels for the entire cohort over the full study period, we developed a metric termed aggregate median steps per day (AMS), which we calculated by determining the median value of all subjects’ step counts on a given study day. Change points in AMS were identified using change point detection analysis with a Python implementation of the Pruned Exact Linear Time (PELT) search method.¹² Time intervals between consecutive change points were considered to represent distinct phases of the recovery period. An average value for AMS was calculated for each identified phase of recovery. Student’s two-sample t-tests were used to evaluate for differences in average AMS between the preoperative period and each postoperative recovery phase. Day-to-day variability of AMS during each phase was quantified using the coefficient of variation (CV), the ratio of standard deviation to average. Linear regression was used to quantify the change over time in AMS during each identified phase of recovery; the regression slope (β) was calculated for each phase and its statistical significance was determined using one-sample Student’s t-test. To assess the effect of spinal fusion on recovery patterns, the cohort was stratified into two groups: patients who underwent fusion of one or more levels and those who did not undergo fusion. The change in AMS between the preoperative phase and the final recovery phase was compared between the two groups using Student’s t-test.

To compare accelerometry-defined and PROM-defined recovery patterns at each clinic visit, we quantified the activity levels of each patient prior to each visit (preoperative period, 2-week follow-up, etc.) using a metric termed visit-specific median steps (VSMS). For each patient, preoperative VSMS was calculated as the median of their daily step counts for the entire preoperative period. Postoperative VSMS values were calculated at each follow-up (2 weeks, 1 month, 3 months, and 6 months) as the median of each patient’s daily step counts during the week prior to the clinic visit.

Paired Wilcoxon signed-rank tests were used to evaluate changes in VSMS and PROMs between the preoperative period and each follow-up visit. Correlations between VSMS and PROMs at each time point were measured using the Spearman’s rank correlation coefficient (rho). Rho values between 0 – 0.39, 0.40 – 0.59, and 0.60 – 1 indicated weak, moderate, and strong correlations respectively.

RESULTS

Study Cohort

25 patients were enrolled in the study between August 2018 and January 2020; 5 patients were withdrawn prior to completing the preoperative period (Table 1). Of the withdrawn patients, 1 patient had poor compliance during the preoperative period and was lost to follow-up after surgery, 3 patients were indicated for surgery but decided to switch surgeons, not to have the surgery, or had other conditions that prevented them from having surgery. In addition, the study team was unable to contact 1 patient for device setup after enrollment.

Table 1.

Demographic characteristics of the study cohort (n = 20). Averages are reported with standard deviations.

Age at time of surgery (years)	Average	64.5 ± 8.8
	Range	50 – 79
Sex	Male	11 (55%)
	Female	9 (45%)
BMI	Average	29.2 ± 4.7
	Range	19.7 – 39.2
Smoking status	Never	16 (80%)
	Former	3 (15%)
	Current	1 (5%)
ASA physical status classification	I	0 (0%)
	II	13 (65%)
	III	7 (35%)
	IV	0 (0%)
Presence of spondylolisthesis	Yes	15 (75%)
	No	5 (25%)
Number of spine levels indicated for fusion	0	6 (30%)
	1	11 (55%)
	2	3 (15%)
Calendar quarter of enrollment	Q3 2018	4 (20%)
	Q4 2018	3 (15%)
	Q1 2019	1 (5%)
	Q2 2019	6 (30%)
	Q3 2019	4 (20%)
	Q4 2019	1 (5%)
	Q1 2020	1 (5%)

Abbreviations: BMI - body mass index; ASA - American Society of Anesthesiologists

Of the 20 patients active in the study after having surgery, 1 patient voluntarily withdrew after being discharged to a rehabilitation facility and 1 patient died months after surgery due to an unrelated condition. 3 patients were withdrawn by the study team prior to completing the study due to the onset of the COVID-19 pandemic. Any data collected up to the date of withdrawal was included in the analysis.

Median compliance with wearing the accelerometer was 90% (IQR = 20%). Non-compliance was most commonly due to subjects forgetting to charge the device, having issues with Bluetooth connectivity between their device and their smartphone, or temporarily misplacing the device.

Accelerometry-Defined Recovery Phases

Cohort AMS for each day of the study was calculated (Figure 1). Day of surgery was designated “day 0” and preoperative and postoperative days were designated with negative and positive numbers, respectively. Change point detection analysis with the PELT search method identified postoperative change points at days 40 and 130. Three phases of recovery were defined using these change points: phase I (days 0 to 40), phase II (days 40 to 130), and phase III (days 130 to 184).

Figure 1.

Aggregate median steps (AMS) of the cohort for each day of the study period. Day 0 refers to the day of surgery. Preoperative days are assigned negative numbers and postoperative days are assigned positive numbers. Time points at which patient-reported outcome measures (PROMs) were administered are indicated at the bottom of the graph. Distinct phases of the study period as indicated by change point detection analysis with Pruned Exact Linear Time (PELT) search method are shaded in alternating grey and white. Names for each phase are indicated at the top of the graph.

Average AMS and AMS regression slope values were obtained for each phase (Table 2). Average AMS decreased significantly between the preoperative period and postoperative phases I and II (p <= 0.002) but did not exceed the preoperative baseline in phase III of recovery (p = 0.23). Day-to-day variability was highest during phase I (CV = 0.42) and lowest during phase III (CV = 0.14). AMS most rapidly increased during phase I of the postoperative period (β = 66; p < 0.001), with a slower rate of increase in phase II (β = 11; p = 0.003), followed by a plateau in phase III (β = −3; p = 0.74).

Table 2.

Averages and linear regression slopes (β) of the aggregate median steps (AMS) of the cohort for each accelerometry-defined recovery phase.

a) Average AMS during each phase with standard deviation. Two-sample Student’s t-tests comparing average AMS for each postoperative phase to the preoperative baseline were performed and p-values were calculated. P-values <0.05 are in bold.
	Preoperative	Phase I	Phase II	Phase III
Average	5536 ± 1314	2102 ± 878	4692 ± 907	5856 ± 898
p-value		< 0.001	0.002	0.23
b) Regression slopes of AMS during each phase with standard error. One-sample t-tests of the correlation coefficient were performed and p-values were calculated. P-values <0.05 are in bold.
	Preoperative	Phase I	Phase II	Phase III
Regression slope (β)	−78 ± 24	66 ± 6	11 ± 4	−3 ± 8
p-value	0.003	< 0.001	0.003	0.74

Patients who underwent fusion saw a greater increase in average AMS between the preoperative period and phase III than those who did not (p < 0.001). However, neither group of patients exceeded their respective baselines by the end of phase III (p > 0.05).

Associations between PROMs and Step Counts at Study Milestones

VSMS and PROM scores were obtained for each postoperative milestone (Figure 2). While VSMS only returned to baseline at 6 months after surgery (p = 0.75), ODI, ODI Section 4, and SF-12 PCS scores significantly improved relative to baseline by 6 months after surgery (all p< 0.01).

Figure 2.

(a) Average visit-specific median steps (VSMS), (b) average Oswestry Disability Index (ODI), (c) average ODI Section 4 (Walking) subscore, and (d) average 12-Item Short Form Health Survey Physical Component Score (SF-12 PCS) of the cohort at each study milestone. Error bars represent standard error. Changes in metrics between the preoperative baseline and each postoperative milestone were evaluated using paired Wilcoxon signed rank tests and p-values were calculated. Postoperative milestones with significant differences in metrics from baseline (i.e. p-value <0.05) are marked with an asterisk (*).

Correlations between VSMS and PROMs were evaluated at each study milestone (Table 3). Preoperative VSMS was not significantly correlated with any preoperative PROMs. Correlations between VSMS and ODI were significant and of moderate strength at 2 weeks (p = 0.04), 1 month (p = 0.04), and 3 months (p = 0.02) after surgery. Similarly, VSMS and ODI section 4 subscore were moderately correlated at 3 months after surgery (p = 0.02). VSMS had a strong and consistent positive correlation with the SF-12 PCS throughout the postoperative period, specifically at 1 month (p = 0.02), 3 months (p < 0.001), and 6 months (p = 0.008) after surgery.

Table 3.

Spearman’s rank correlations (rho) between visit-specific median steps (VSMS) and patient-reported outcome measures (PROMs) for the cohort at each clinic visit. PROMs include the Oswestry Disability Index (ODI), ODI section 4 subscore, and Short Form-12 Health Survey (SF-12). The statistical significance of each correlation coefficient was determined using one-sample Student’s t-test and p-values were calculated. Statistically significant rho values and p-values <0.05 are in bold.

	Preoperative (n = 20) rho (p-value)	Postoperative
		2 weeks (n = 17) rho (p-value)	1 month (n = 18) rho (p-value)	3 months (n = 16) rho (p-value)	6 months (n = 14) rho (p-value)
VSMS vs ODI	−0.22 (0.35)	−0.49 (0.04)	−0.49 (0.04)	−0.57 (0.02)	−0.10 (0.73)
VSMS vs ODI Section 4 Subscore	−0.18 (0.45)	−0.26 (0.32)	−0.44 (0.07)	−0.58 (0.02)	−0.11 (0.71)
VSMS vs SF-12 PCS	0.34 (0.14)	- -	−0.56 (0.02)	0.77 (< 0.001)	0.67 (0.008)

DISCUSSION

Our analysis of accelerometry data from a prospective cohort of lumbar laminectomy patients demonstrated that the postoperative period can be divided into three distinct phases based on rate of increase and day-to-day variation in daily steps. Furthermore, we found that physical function data and PROMs were significantly correlated; specifically, accelerometry-derived metrics correlated moderately with the ODI from 2 weeks to 3 months after surgery and more strongly with the SF-12 PCS from 1 month to 6 months after surgery. Although PROMs far exceeded preoperative levels by 6 months after surgery, patient physical activity measured at both the cohort level (AMS) and individual level (VSMS) only returned to baseline.

Using accelerometry data to measure physical activity levels in patients recovering from surgery is not a novel concept, however, no studies to date have used time-series analyses to formally characterize distinct phases of recovery. We identified three postoperative phases using a change point detection method: acute healing, recovery, and stabilization. The first postoperative phase (Phase I), or the “Acute Healing” phase, lasted from the day of surgery to 40 days after surgery and was defined by a rapid increase in daily step count with considerable day-to-day variation. We posit that the observed increase in physical activity during the first 40 postoperative days reflects a rapid recovery due to acute healing from surgery. The quick rise in postoperative activity may also be driven by early mobilization protocols implemented at our institution; such protocols have been noted to significantly increase ambulation in the days immediately following surgery.¹³ Basil et al. similarly observed a rapid increase in physical activity during the first 6 weeks following lumbar spine surgery.¹⁴ The second and longest postoperative phase (Phase II), or the “Recovery” phase, lasted from postoperative day 40 to 130 and was marked by a slower rate of increase and decreased variability in daily step count compared to Phase I. Interestingly, these changes coincide with the initiation of physical therapy (PT) among our cohort at the 4–6 week mark. While one of the goals of postoperative PT is to increase levels of physical activity following surgery, a 2020 case series by Coronado et al. also found that standard PT regimens had little-to-no effect on accelerometer-measured physical activity following lumbar spine surgery.⁵ Continuously queried physical activity data could be used to provide step-based targets to patients during PT and better evaluate their progress. The third phase of recovery (Phase III), or the “Stabilization” phase, consisted of the remainder of the 6-month accelerometer recording period after day 130 and was characterized by a plateau in daily step count coupled with low day-to-day variation; physical activity during this phase returned to the preoperative baseline but did not exceed it. Interestingly, this result is inconsistent with a 2020 prospective study by Inoue et al. which found a significant improvement in step count at 6 months following lumbar spine surgery; the observed discrepancy may be due to differences in the surgical procedures included in each study¹⁵. The decrease in variation of physical activity during Phase III may be attributable to a decrease in frequency of pain flares and several lifestyle changes during this period including a return to normal activities of daily living.

While accelerometer-measured activity levels just returned to baseline, we found that the SF-12 PCS and ODI far exceeded baseline at 6 months postoperatively. These findings are in line with previous studies that have examined physical activity levels in this population.^4,6,16 The lack of improvement in daily step counts after surgery may reflect typical activity patterns in the low back pain patient population. These patients are often elderly and thus their chosen activities of daily life may not necessitate a significant gain in daily step counts for successful completion. Given the improvement in patients’ conditions as detailed by PROMs, surgery may allow these patients to live their preferred lifestyles in a more comfortable manner without a requisite increase in their mobility. A study that tests patients’ maximal exertion may be better suited to detect increases in objective measures of physical function after surgery. For instance, patients’ gains may be better quantified by investigating their time spent walking before requiring rest while completing their daily tasks. In fact, previous studies have found that lumbar stenosis patients are typically very limited in the length of their bouts of activity and that the length of these bouts often increase after surgery.^17,18

This study showcases the value of accelerometry as an adjunctive outcome measure. The technology allows for a continuous and objective source of recovery tracking data that gives surgeons insight into how patients function outside of the clinic. In identifying three stages of the postoperative period, this study provides a framework for scheduling follow-up visits after surgery. For instance, clinicians could consider prescribing only one clinic visit between 40 and 130 days after surgery because patients’ physical statuses, as described by step counts, appear to be relatively stable during the “Recovery” phase (Phase II). Additional visits for assessing pain and other elements of functional recovery could be conducted via telemedicine. The data also allows for recognition of and timely intervention for patients who are less active than expected during the postoperative period. Lastly, an understanding of typical patient behavior after surgery allows for better preoperative counseling and expectation setting for patients, which has been shown to improve postoperative outcomes.^19,20

The present study is a prospective, within-subject cohort study that analyzed continuous data over an extended period. It included a homogenous cohort, which ensured that participants would not be significantly affected by confounding factors (i.e. different procedures, other health conditions). The study’s sample size was limited because we were forced to suspend study enrollment due to the Covid-19 pandemic. Furthermore, we identified that the restrictions associated with lockdowns during the pandemic may have had an impact on activity levels after surgery.²¹ We therefore consider that this cohort must stand alone with its current numbers. Ultimately, the sample size was sufficient to use PELT change point detection to identify 3 distinct stages of recovery and to detect significant correlations between VSMS and PROMs. Future studies with increased sample size are necessary to validate the results of the present study and may further refine break points in the recovery stages and/or enable detection of additional recovery stages.

A larger cohort would also allow for stratification of patients into subgroups based on demographics and other clinical factors. For instance, the necessity for fusion in addition to laminectomy in approximately two thirds of our subjects may impact their recovery profiles. We found that patients who underwent fusion had a larger improvement in AMS from baseline compared to those who did not, yet neither group surpassed their respective preoperative baselines. The improved outcome of the fusion group is a consideration warranting additional studies with greater numbers of patients in each stratum to validate these findings and confirm potential differences between subgroups of patients.

Additional limitations include that patients were required to own a smartphone with access to the internet and were allowed to have access to their Fitbit accounts. Despite introducing an increased potential for selection bias and influence on patient behavior, these choices helped to ensure a high level of compliance throughout the study’s duration and provided important information throughout the study. The threshold for the level of compliance necessary to be included in the study was set in accordance with previous fitness tracking literature.⁹ Given the study’s high compliance (~90%), we are confident that the study captured true subject behavior. Use of a fitness tracker with a longer battery life and that requires less frequent syncing with the internet may further increase compliance. Lastly, seasonal differences in activity levels were not controlled for, however, by recruiting patients across all four seasons with 6 months of follow-up, we believe that this factor was largely negated.

CONCLUSIONS

Temporal data from accelerometers allowed for identification of 3 stages (acute healing, recovery, and stabilization) of the postoperative period independent of traditional follow-up visits (ex. 1-month follow-up, 3-month follow-up) after lumbar laminectomy. We also found that the SF-12 PCS correlated with patient activity levels from 1–6 months after surgery and the ODI correlated with activity levels from 1–3 months after surgery. However, both PROMs continued to improve from 3–6 months after surgery while activity levels stabilized. This divergence suggests that subjective disability can continue to be reduced even as patients lack further gains in physical function during the “Stabilization” phase (Phase III) of the postoperative period; patients reach a point in recovery in which they can comfortably complete their daily activities without necessarily increasing their objective physical output compared to before surgery. Therefore, measurement of recovery following lumbar laminectomy should ideally be multi-modal in nature, incorporating both accelerometry and PROMs. Our use of accelerometry to define 3 important stages of the postoperative period may help inform re-alignment of patients’ in-person clinic visits to timepoints when they will be most impactful in supporting recovery. Telehealth may be used during stable postoperative phases. Furthermore, postoperative activity data may be useful in designating targeted postoperative goals to better ensure timely increases in activity levels after surgery and identify patients who do not meet these goals for further management.

Abbreviation List

AMS

Aggregate median steps

ASA

American Society of Anesthesiologists

BMI

Body mass index

CV

Coefficient of variation

IRB

Institutional review board

ODI

Oswestry Disability Index

PELT

Pruned Exact Linear Time

PROM

Patient-reported outcome measure

PT

physical therapy

SF-12

12-Item Short Form Survey

SF-12 PCS

12-Item Short Form Survey Physical Component Score

VSMS

Visit-specific median steps