Phenotyping Degenerative Lumbar Spine Disease Through Big Data: First-order Derivative Analysis of Smartphone-based Accelerometry Data

Introduction: Spinal surgery has historically relied on patient-reported outcome measures to assess post-operative outcomes. To supplement these subjective questionnaires, patient mobility data harvested from built-in accelerometry within smartphones can provide granular information about patient’s functional status before and after surgery. In this study, we utilized first-order derivatives of activity data to phenotype the pre-operative and post-operative courses of patients who underwent either lumbar decompression (LD) or fusion (LF) surgery.

Methods: LD and LF patients were retrospectively consented and enrolled. Activity data (steps-per-day) recorded in Apple Health (Apple Inc., Cupertino, CA) over 2 years peri-operatively was classified into temporal epochs representing distinct functional states including pre-operative baseline, pre-operative decline, and post-operative recovery. The first-order derivatives of patient activity magnitude across time were then calculated for all epochs.

Results: A total of 21 LD and 31 LF patients were included, encompassing over 70,000 datapoints. 66.7% (14/21) of LD and 67.7% (21/31) of LF patients experienced at least one period of activity decline pre-operatively, defined as diminished physical activity compared to baseline. During these declines, mean first-order derivative for LD patient activity, representing the rate of activity decline during disease progression, was significantly more positive than for LF (0.043 vs. -0.123, p = 0.045), indicating a greater rate of decline for LF compared to LD. During post-operative recovery, LD first-order derivative was significantly higher than LF (0.003 vs. -0.060, p = 0.041), suggesting a more gradual functional recovery in LF compared to LD.

Conclusion : First-order derivative analysis of patient activity data is a promising technique in building towards phenotyping patient activity profiles to differentiate between different pathologies and surgical treatments. Comparing patient activity data between different surgical procedures can help establish common presenting patterns of pre-operative patient activity as well as the unique contributions of specific surgical interventions to patient outcomes.