"Bring Your Own Device"-A New Approach to Wearable Outcome Assessment in Trauma.

Background and Objectives: Outcome data from wearable devices are increasingly used in both research and clinics. Traditionally, a dedicated device is chosen for a given study or clinical application to collect outcome data as soon as the patient is included in a study or undergoes a procedure. The current study introduces a new measurement strategy, whereby patients' own devices are utilized, allowing for both a pre-injury baseline measure and ability to show achievable results. Materials and Methods: Patients with a pre-existing musculoskeletal injury of the upper and lower extremity were included in this exploratory, proof-of-concept study. They were followed up for a minimum of 6 weeks after injury, and their wearable outcome data (from a smartphone and/or a body-worn sensor) were continuously acquired during this period. A descriptive analysis of the screening characteristics and the observed and achievable outcome patterns was performed. Results: A total of 432 patients was continuously screened for the study, and their screening was analyzed. The highest success rate for successful inclusion was in younger patients. Forty-eight patients were included in the analysis. The most prevalent outcome was step count. Three distinctive activity data patterns were observed: patients recovering, patients with slow or no recovery, and patients needing additional measures to determine treatment outcomes. Conclusions: Measuring outcomes in trauma patients with the Bring Your Own Device (BYOD) strategy is feasible. With this approach, patients were able to provide continuous activity data without any dedicated equipment given to them. The measurement technique is especially suited to particular patient groups. Our study's screening log and inclusion characteristics can help inform future studies wishing to employ the BYOD design.

Only limited correlations between patient-reported outcomes and objectively monitored physical activity 10-years after THA.

In orthopaedics, patient reported outcomes (PROMs) are designed to quantify changes in pain and to assess physical function, most often after joint arthroplasty. However, PROMs have some disadvantages, most important is their subjective nature. The aim was to investigate how patient-self-reported-outcomes of general health, disease-specific outcome and physical function, joint-awareness and self-perceived activity- levels are correlated with objectively-measured physical-activity (PA) parameters derived from wearable activity-monitors (AM) in subjects with a hip-arthroplasty. A prospective cohort study was conducted in a group of 32 patients, with a mean follow-up of 10 years after total hip arthroplasty. To assess different domains, the SF-36 (general health), HOOS-PS (pain/functional outcome), FJS-12 (joint awareness) and SQUASH (physical activity) were chosen. Activity-monitoring was performed using a 3-axis accelerometer, gyroscope and magnetometer. No significant correlations between PA-parameters and the FJS-12 and SQUASH were found. The HOOS- PS was significant correlated with BMI, the daily time walking and total-time active and the amount of daily steps. The physical functioning-subscale of the SF-36 was significant negative correlated with BMI and time sitting, but significant positive correlated with time walking, total-time active and the amount of daily steps. Considering the value of PA for maintaining general health, the value of using sensor-based AMs to assess efficacy of treatments in this health related dimension or use it as a tool for patient education, awareness and communication, seems very high.

Patients with hip resurfacing arthroplasty are not physically more active than those with a stemmed total hip.

Background and purpose - Hip resurfacing arthroplasty (HRA) was designed for the highly active patient because of the various theoretical advantages compared with stemmed total hip arthroplasty (THA), but has shown high failure rates. Physical activity (PA) after arthroplasty is frequently determined with the use of questionnaires, which are known for their subjective nature, recall bias, and ceiling effect. These disadvantages are not applicable to physical activity monitoring (AM) using sensors. We compared objectively measured PA at long-term follow-up in a matched cohort of HRA and stemmed THA subjects.Patients and methods - We compared 2 groups of 16 patients (12 males) in each group, one having received unilateral HRA (median age 56 years at surgery) and a matched group having received unilateral stemmed THA with a small diameter femoral head (28 mm) on conventional polyethylene (median age 60 years at surgery) with osteoarthritis as indication for surgery, 10 years after surgery. Groups were matched by sex, age at surgery, and BMI. The daily habitual PA was measured over 4 consecutive days in daily living using a 3-axis accelerometer, gyroscope, and magnetometer. Both quantitative parameters (time standing, sitting, walking, number of steps, and sit-stand transfers) and qualitative parameters (walking cadence) were determined.Results - The AM was worn for a median 13 (11-16) hours per day. The median daily step count was 5,546 (2,274-9,966) for the HRA group and 4,583 (1,567-11,749) for the stemmed THA-group with 39 (21-74) versus 37 (24-62) daily sit-stand transfers respectively. The other PA parameters were also similar in both groups.Interpretation - We found similar median PA levels and also identical ranges. While short-term effects may exist, ageing and related behavioral adaptations or other effects seem to render the theoretical activity benefits from HRA irrelevant at longer follow-up.

Objective assessment of physical activity and sedentary behaviour in knee osteoarthritis patients - beyond daily steps and total sedentary time.

BACKGROUND: Knee osteoarthritis patients may become physically inactive due to pain and functional limitations. Whether physical activity exerts a protective or harmful effect depends on the frequency, intensity, time and type (F.I.T.T.). The F.I.T.T. dimensions should therefore be assessed during daily life, which so far has hardly been feasible. Furthermore, physical activity should be assessed within subgroups of patients, as they might experience different activity limitations. Therefore, this study aimed to objectively describe physical activity, by assessing the F.I.T.T. dimensions, and sedentary behaviour of knee osteoarthritis patients during daily life. An additional goal was to determine whether activity events, based on different types and durations of physical activity, were able to discriminate between subgroups of KOA patients based on risk factors. METHODS: Clinically diagnosed knee osteoarthritis patients (according to American College of Rheumatology criteria) were monitored for 1 week with a tri-axial accelerometer. Furthermore, they performed three functional tests and completed the Knee Osteoarthritis Outcome Score. Physical activity levels were described for knee osteoarthritis patients and compared between subgroups. RESULTS: Sixty-one patients performed 7303 mean level steps, 319 ascending and 312 descending steps and 601 bicycle crank revolutions per day. Most waking hours were spent sedentary (61%), with 4.6 bouts of long duration (> 30 min). Specific events, particularly ascending and descending stairs/slopes, brief walking and sedentary bouts and prolonged walking bouts, varied between subgroups. CONCLUSIONS: From this sample of KOA patients, the most common form of activity was level walking, although cycling and stair climbing activities occurred frequently, highlighting the relevance of distinguishing between these types of PA. The total active time encompassed a small portion of their waking hours, as they spent most of their time sedentary, which was exacerbated by frequently occurring prolonged bouts. In this study, event-based parameters, such as stair climbing or short bouts of walking or sedentary time, were found more capable of discriminating between subgroups of KOA patients compared to overall levels of PA and sedentary time. Thereby, subtle limitations in physical behaviour of KOA-subgroups were revealed, which might ultimately be targeted in rehabilitation programs. TRIAL REGISTRATION: German Clinical Trials Registry under ' DRKS00008735 ' at 02.12.2015.

Leg power, pelvic movement and physical activity after periacetabular osteotomy. A prospective cohort study.

To investigate changes in leg power, pelvic movement and patient-reported outcome in patients with hip dysplasia one year after periacetabular osteotomy. Forty-one patients (7 males) with a mean age of 28.8 years scheduled for periacetabular osteotomy were tested before surgery, and 4 and 12 months after. Leg power, pelvic range of motion and the Hip and Groin Outcome Score (HAGOS) was collected. One year after surgery, power in the operated leg had improved (p = 0.004) and there was no significant difference between power in the operated leg and contralateral leg (p = 0.22). In the frontal plane, pelvic range of motion decreased significant during stair-climbing and stepping down. The same pattern was seen in the sagittal plane but the changes were non-significant. All subscales on the HAGOS improved significantly over time (p < 0.001). Leg power and pelvic range of motion in patients with symptomatic hip dysplasia improved 12 months after periacetabular.

Femoral Head Size is Correlated With Head Position in Very Elderly Patients.

BACKGROUND: In cases of femoral neck fracture, it is often not possible to accurately determine the original position of the head center to assess appropriate restoration of leg length. The aim of this study was to determine the accuracy of predicting the position of the femoral head center based on new and established correlations between the femoral head diameter (FHD) and the distance between the lesser trochanter and the femoral head center (LT-FHC) in the very elderly (aged ≥80 years) as the mainly affected but yet underinvestigated group. METHODS: The FHD and the LT-FHC distance were determined in 148 subjects (104 males, 44 females); 90 aged ≥80 years and 58 aged <80 years. For each age and gender subgroup one specific (LT-FHC)/FHD ratio was determined. The accuracy of the new determined ratios and the established ratios by others were compared by recalculating the LT-FHC distance of each individual subject. RESULTS: The FHD and the LT-FHC were significantly correlated, most strongly in elderly females (R = 0.554, P < .001). Using the new age- and gender-specific ratios, the LT-FHC distance could be predicted within 10 mm of the true value in 95% of the cases and in only 77% using previously reported formulas. CONCLUSION: Age- and gender-specific formulas yield higher accuracy than generic formulas. The formulas presented in this study can offer a practical, easy to use instrument for orthopedic surgeons performing hip arthroplasty in very elderly patients in addition to classic techniques to prevent significant leg-length discrepancy.

The Femoral Head Center Shifts in a Mediocaudal Direction During Aging.

BACKGROUND: Accurate reconstruction of hip joint biomechanics is the key stone in total hip arthroplasty. Although proximal femur morphology is known to vary with both age and gender, few studies investigated this in the very elderly (ie, ≥80 years). The purpose of this study was to compare basic morphological parameters describing the position of the femoral head between very elderly and middle-aged subjects. METHODS: Computed tomographic scans of the femur of 90 very elderly (mean 84 years, range 80-105 years) and 58 middle-aged subjects (mean 52 years, range 20-79 years) were made. After 3-dimensional reconstruction, the neck-shaft angle, femoral neck anteversion angle, femoral head height, femoral neck length, and mediolateral offset (ML-offset) were determined. RESULTS: The neck-shaft angle was on average 3.6° less in elderly males (125.9° ± 5.0°) than in middle-aged males (129.5° ± 5.1°) (P < .001). The femoral neck anteversion angle was not significantly different between both age and gender groups. The femoral head height was -12.3 mm in elderly females compared to -10.5 mm (Δ 17%) in elderly males (P = .284) and even -8.0 mm (Δ 54%) in middle-aged males. The ML-offset was 10% (4.1 mm) larger in elderly compared to middle-aged males (P < .001). CONCLUSION: These findings suggest that the femoral neck and head shift in a relative varus position during aging. Femoral prostheses with increased ML-offset and a lower caput-collum-diaphyseal angle are needed to accommodate the morphology of the femur in the very elderly. Care must be taken not to lengthen the operated leg, particularly in very elderly females.

Quantifying Habitual Levels of Physical Activity According to Impact in Older People: Accelerometry Protocol for the VIBE Study.

Physical activity (PA) may need to produce high impacts to be osteogenic. The aim of this study was to identify threshold(s) for defining high impact PA for future analyses in the VIBE (Vertical Impact and Bone in the Elderly) study, based on home recordings with triaxial accelerometers. Recordings were obtained from 19 Master Athlete Cohort (MAC; mean 67.6 years) and 15 Hertfordshire Cohort Study (HCS; mean 77.7 years) participants. Data cleaning protocols were developed to exclude artifacts. Accelerations expressed in g units were categorized into three bands selected from the distribution of positive Y-axis peak accelerations. Data were available for 6.6 and 4.4 days from MAC and HCS participants respectively, with approximately 14 hr recording daily. Three-fold more 0.5-1.0g impacts were observed in MAC versus HCS, 20-fold more 1.0-1.5g impacts, and 140-fold more impacts ≥ 1.5g. Our analysis protocol successfully distinguishes PA levels in active and sedentary older individuals.

Adjustments in 2011 KSS increase the clinical suitability.

The 2011 KSS is a valid clinical TKA questionnaire, but with a low completion rate (42%). Adjustments, focusing on optimizing scale features, are required to improve its clinical use. The low completion rates, non-optimal scale features, lacking rules or a combination of these factors where addressed, leading to the development of the adjusted 2011 KSS (2011 KSS-A). Four-hundred-ninety-nine primary TKA patients were addressed pre- and postoperative by mail. Clinimetric quality was evaluated. Seventy percent responded and 90% completed the scale. Internal consistency proved excellent with Cronbach's Alpha≥0.79 for all subscales. Strong correlations were found between the Functional Activity subscales and KOOS-PS (r=-0.63 to -0.87). All subscales improved significantly after intervention (r-range 14-33%, effect size 0.50-2.85). Postoperatively, ceiling effects were found in the subscales Symptoms (16%) and Walking & Standing (26%). Adjustments led to a shortened and simplified questionnaire while maintaining its clinimetric quality.

The Morphology of the Proximal Femoral Canal Continues to Change in the Very Elderly: Implications for Total Hip Arthroplasty.

A shape mismatch between cementless stems and the femoral canal of very elderly patients could partly explain the inferior performance of cementless compared to cemented stems in this age group. Influences of age and gender on canal morphology were investigated by measuring coronal/sagittal dimensions on CT-scans of subjects ≥80 years old (n = 117) and subjects < 80 years old (n = 51). Absolute canal dimensions like width were larger in octogenarians than in younger subjects. These differences were larger in the sagittal plane than in the coronal plane (P < 0.001). Canal flaring changed asymmetrically with increased medial and less lateral flaring in octogenarians. Age-related changes were mainly observed in females and should be taken into consideration for implant design, selection and implantation in octogenarians.

Strong correlation between the morphology of the proximal femur and the geometry of the distal femoral trochlea.

PURPOSE: Previous investigations suggested that the geometry of the proximal femur may be related to osteoarthritis of the tibiofemoral joint and various patellofemoral joint conditions. This study aims to investigate the correlation between proximal and distal femoral geometry. Such a correlation could aid our understanding of patient complications after total knee arthroplasty (TKA) and be of benefit for further development of kinematic approaches in TKA. METHODS: CT scans of 60 subjects (30 males, 30 females) were used to identify anatomical landmarks to calculate anatomical parameters of the femur, including the femoral neck anteversion angle, neck-shaft angle (NSA), mediolateral offset (ML-offset), condylar twist angle (CTA), trochlear sulcus angle (TSA) and medial/lateral trochlear inclination angles (MTIA/LTIA). Correlation analyses were carried out to assess the relationship between these parameters, and the effect of gender was investigated. RESULTS: The CTA, TSA and LTIA showed no correlation with any proximal parameter. The MTIA was correlated with all three proximal parameters, mostly with the NSA and ML-offset. Per 5° increase in NSA, the MTIA was 2.1° lower (p < 0.01), and for every 5 mm increase in ML-offset, there was a 2.6° increase in MTIA (p < 0.01). These results were strongest and statistically significant in females and not in males and were independent of length and weight. CONCLUSIONS: Proximal femoral geometry is distinctively linked with trochlear morphology. In order to improve knowledge on the physiological kinematics of the knee joint and to improve the concept of kinematic knee replacement, the proximal femur seems to be a factor of clinical importance. LEVEL OF EVIDENCE: III.

Continuation rates of the subdermal contraceptive Implanon(®) and associated influencing factors.

OBJECTIVES To investigate the continuation rates of the etonogestrel subdermal contraceptive implant among well-informed women, and the reasons for early discontinuation. STUDY DESIGN Retrospective consecutive cohort design. METHODS Women who had the implant inserted between 1 January 2006 and 1 January 2010 at the Atrium Medical Centre, the Netherlands were followed up for at least three years. The dates of insertion and removal were recorded, as were the reasons for removal. Statistical analysis was performed using the independent samples t-test and the Pearson alpha correlation test. RESULTS The implant was inserted in 230 women. Follow-up was possible in 214 women, with an average age of 26.7 years. Most of them were nulliparous and the majority had used a combined oral contraceptive, Implanon®, or Mirena® previously. The mean overall continuation period was 23.5 months (95% confidence interval: 21.7-25.3), with a median of 25 months. The continuation rate after 12 months was 72%; after 24 months, 53%; and after 36 months, 25%, with all women concerned having a new implant placed. The previously used contraceptive method Implanon® was associated with the highest continuation rates. An erratic bleeding pattern was the main reason for early removal. CONCLUSIONS Despite adequate counselling before insertion, the continuation rate of the etonogestrel implant was rather low compared to those reported by other investigators. The main reason for discontinuation was an irregular bleeding pattern.

Wearable activity data can predict functional recovery after musculoskeletal injury: Feasibility of a machine learning approach.

Delayed functional recovery after injury is associated with significant personal and socioeconomic burden. Identification of patients at risk for a prolonged recovery after a musculoskeletal injury is thus of high relevance. The aim of the current study was to show the feasibility of using a machine learning assisted model to predict functional recovery based on the pre- and immediate post injury patient activity as measured with wearable systems in trauma patients. Patients with a pre-existing wearable (smartphone and/or body-worn sensor), data availability of at least 7 days prior to their injury, and any musculoskeletal injury of the upper or lower extremity were included in this study. Patient age, sex, injured extremity, time off work and step count as activity data were recorded continuously both pre- and post-injury. Descriptive statistics were performed and a logistic regression machine learning model was used to predict the patient's functional recovery status after 6 weeks based on their pre- and post-injury activity characteristics. Overall 38 patients (7 upper extremity, 24 lower extremity, 5 pelvis, 2 combined) were included in this proof-of-concept study. The average follow-up with available wearable data was 85.4 days. Based on the activity data, a predictive model was constructed to determine the likelihood of having a recovery of at least 50 % of the pre-injury activity state by post injury week 6. Based on the individual activity by week 3 a predictive accuracy of over 80 % was achieved on an independent test set (F1=0,82; AUC=0,86; ACC=8,83). The employed model is feasible to assess the principal risk for a slower recovery based on readily available personal wearable activity data. The model has the potential to identify patients requiring additional aftercare attention early during the treatment course, thus optimizing return to the pre-injury status through focused interventions. Additional patient data is needed to adapt the model to more specifically focus on different fracture entities and patient groups.

Wearable Sensors in Other Medical Domains with Application Potential for Orthopedic Trauma Surgery-A Narrative Review.

The use of wearable technology is steadily increasing. In orthopedic trauma surgery, where the musculoskeletal system is directly affected, focus has been directed towards assessing aspects of physical functioning, activity behavior, and mobility/disability. This includes sensors and algorithms to monitor real-world walking speed, daily step counts, ground reaction forces, or range of motion. Several specific reviews have focused on this domain. In other medical fields, wearable sensors and algorithms to monitor digital biometrics have been used with a focus on domain-specific health aspects such as heart rate, sleep, blood oxygen saturation, or fall risk. This review explores the most common clinical and research use cases of wearable sensors in other medical domains and, from it, derives suggestions for the meaningful transfer and application in an orthopedic trauma context.

Best practice in digital orthopaedics.

Digitization in orthopaedics and traumatology is an enormously fast-evolving field with numerous players and stakeholders. It will be of utmost importance that the different groups of technologists, users, patients, and actors in the healthcare systems learn to communicate in a language with a common basis. Understanding the requirements of technologies, the potentials of digital application, their interplay, and the combined aim to improve health of patients, would lead to an extraordinary chance to improve health care. Patients' expectations and surgeons' capacities to use digital technologies must be transparent and accepted by both sides. The management of big data needs tremendous care as well as concepts for the ethics in handling data and technologies have to be established while also considering the impact of withholding or delaying benefits thereof. This review focuses on the available technologies such as Apps, wearables, robotics, artificial intelligence, virtual and augmented reality, smart implants, and telemedicine. It will be necessary to closely follow the future developments and carefully pay attention to ethical aspects and transparency.

Continuous Shoulder Activity Tracking after Open Reduction and Internal Fixation of Proximal Humerus Fractures.

Postoperative shoulder activity after proximal humerus fracture treatment could influence the outcomes of osteosynthesis and may depend on the rehabilitation protocol. This multi-centric prospective study aimed at evaluating the feasibility of continuous shoulder activity monitoring over the first six postoperative weeks, investigating potential differences between two different rehabilitation protocols. Shoulder activity was assessed with pairs of accelerometer-based trackers during the first six postoperative weeks in thirteen elderly patients having a complex proximal humerus fracture treated with a locking plate. Shoulder angles and elevation events were evaluated over time and compared between the two centers utilizing different standard rehabilitation protocols. The overall mean shoulder angle ranged from 11° to 23°, and the number of daily elevation events was between 547 and 5756. Average angles showed longitudinal change <5° over 31 ± 10 days. The number of events increased by 300% on average. Results of the two clinics exhibited no characteristic differences for shoulder angle, but the number of events increased only for the site utilizing immediate mobilization. In addition to considerable inter-patient variation, not the mean shoulder angle but the number of elevations events increased markedly over time. Differences between the two sites in number of daily events may be associated with the different rehabilitation protocols.

1st EFORT European Consensus "Medical & Scientific Research Requirements for the Clinical Introduction of Artificial Joint Arthroplasty Devices": Background, Delphi Methodology & Consensus process.

The objectives of the 1st EFORT European Consensus on 'Medical and Scientific Research Requirements for the Clinical Introduction of Artificial Joint Arthroplasty Devices' were foremost to focus on patient safety by establishing performance requirements for medical devices. The 1st EFORT European Consensus applied an a priori-defined, modified Delphi methodology to produce unbiased, high-quality recommendation statements, confirmed by consensus voting of a European expert panel. Intended key outcomes are practical guidelines justified by the current stage of knowledge and based on a broad European Expert Consensus, to maintain innovation and optimisation of orthopaedic devices within the boundaries of MDR 2017/745. Twenty-one main research areas of relevance were defined relying on input from the EFORT IPSI WG1 'Introduction of Innovation' recommendations and a related survey. A modified Delphi approach with a preparatory literature review and work in small groups were used to prepare answers to the research questions in the form of 32 draft Consensus statements. A Consensus Conference in a hybrid format, on-site in the Carl Gustav Carus University of Dresden was organised to further refine the draft statements and define consensus within the complete group of participants by final voting, intended to further quantify expert opinion knowledge. The modified Delphi approach provides practical guidelines for hands-on orientation for orthopaedic surgeons, research institutes and laboratories, orthopaedic device manufacturers, patient representatives, Notified Bodies, National Institutes and authorities. For the first time, initiated by the EFORT IPSI (WG1 'Introduction of Innovation'), knowledge of all related stakeholders was combined in the 1st EFORT European Consensus to develop guidelines and result in a comprehensive set of recommendations.

Assessing function in patients undergoing joint replacement: a study protocol for a cohort study.

BACKGROUND: Joint replacement is an effective intervention for people with advanced arthritis, although there is an important minority of patients who do not improve post-operatively. There is a need for robust evidence on outcomes after surgery, but there are a number of measures that assess function after joint replacement, many of which lack any clear theoretical basis. The World Health Organisation has introduced the International Classification of Functioning, Disability and Health (ICF), which divides function into three separate domains: Impairment, activity limitations and participation restrictions. The aim of this study is to compare the properties and responsiveness of a selection of commonly used outcome tools that assess function, examine how well they relate to the ICF concepts, and to explore the changes in the measures over time. METHODS/DESIGN: Two hundred and sixty three patients listed for lower limb joint replacement at an elective orthopaedic centre have been recruited into this study. Participants attend the hospital for a research appointment prior to surgery and then at 3-months and 1-year after surgery. At each assessment time, function is assessed using a range of measures. Self-report function is assessed using the WOMAC, Aberdeen Impairment, Activity Limitation and Participation Restriction Measure, SF-12 and Measure Yourself Medical Outcome Profile 2. Clinician-administered measures of function include the American Knee Society Score for knee patients and the Harris Hip Score for hip patients. Performance tests include the timed 20-metre walk, timed get up and go, sit-to-stand-to-sit, step tests and single stance balance test. During the performance tests, participants wear an inertial sensor and data from motion analysis are collected. Statistical analysis will include exploring the relationship between measures describing the same ICF concepts, assessing responsiveness, and studying changes in measures over time. DISCUSSION: There are a range of tools that can be used to assess function before and after joint replacement, with little information about how these various measures compare in their properties and responsiveness. This study aims to provide this data on a selection of commonly used assessments of function, and explore how they relate to the ICF domains.

Generalizability of deep learning models for predicting outdoor irregular walking surfaces.

Observations from laboratory-based gait analysis are difficult to extrapolate to real-world environments where gait behavior is modulated in response to complex environmental conditions and surface profiles. Inertial measurement units (IMUs) permit real-world gait analysis; however, automatic detection of surfaces encountered remains largely unexplored. The aims of this study are to quantify for machine learning models the effect of (1) random and subject-wise data splitting and (2) sensor location and count on surface classification performance. Thirty participants walked on nine surface conditions (flat-even, slope-up, slope-down, stairs-up, stairs-down, cobblestone, grass, banked-left, banked-right) wearing IMUs (wrist, trunk, bilateral thighs, bilateral shanks). Data were separated into gait cycles, normalized to 101 samples, and spilt into train and test sets (85 and 15%, respectively). For random splitting, trials were randomly assigned to the train or test set. In subject-wise splitting, all trials from 4 random participants were selected for testing. Linear discriminant analysis extracted features from the IMUs. Features were delivered to a neural network. F1-score evaluated model performance. Models achieved F1 scores of 0.96 and 0.78 using random and subject-wise splitting, respectively. Random splitting performance was mainly invariant to sensor location/count; however, subject-wise splitting showed best performance using lower-limb sensors. In general, stairs and sloped surfaces were easily predicted (F1 > 0.85) while banked surfaces were challenging, especially for subject-wise models (F1 ≈ 0.6). Neural networks can detect surfaces based on subtle changes in walking behavior captured by IMUs. Data splitting approaches and sensor location/count (subject-wise) have a non-negligible effect on model performance.