A Smartphone Application to Measure Walking Cadence before Major Abdominal Surgery in Older Adults Open Access

Older adults undergoing major surgery face increased risks of complications, functional decline, and new or worsening disability [1, 2]. Preoperative physical functional assessments, defined as assessments that measure the capability to perform physical activity, are integral to estimate perioperative risk for older adults. Enhanced assessment aids risk stratification, shared decision-making, and identification of older adults who may benefit from preoperative physical activity or exercise interventions to enhance post-surgical functional outcomes [3]. Presently, the current standard for pre-surgery risk assessment involves a blend of questionnaires, such as self-reports of activity, and objective functional assessments, including the 6-min walk test, 5 sit-to-stands, and gait speed measurements, typically conducted in clinical settings. However, these assessments are inconsistently performed, with studies suggesting that fewer than 10% of clinics routinely administer them to older adults [4].

Walking is the most performed mode of exercise among older adults and underpins key objective functional assessments that involve mobility such as the 6-min walk test and the 400-m walk test. Walking cadence, or the number of steps walked in a given time period (i.e., steps/min), is positively correlated with activity intensity such that a cadence ≥100 steps/min identifies moderate-intensity activity in older adults [5‒8]. As such walking cadence may also serve as a preoperative functional assessment to identify older adults at risk of perioperative complications and functional decline [8, 9].

However, measurement of walking cadence is challenging. Commonly used wearable devices and pedometers do not currently measure walking cadence but rather the total volume of steps (i.e., total steps/day). Further, these devices passively measure activity throughout the day and would only measure cadence during routine physical activity which would not provide an accurate representation of physical capability. As the vast majority of older adults spend less than 5 min a day walking at a cadence greater than 100 steps/min passive data collection measurement of cadence during routine activities would not identify physical capability. A physical functional assessment that measures usual- and fast-paced walking cadence would be an ideal measure to assess physical capability.

Smartphones are a potential tool to perform an objective physical functional assessment of walking cadence. Smartphones are owned by 70% of the older adult population and contain built-in accelerometers and computing power which allow them to measure and analyze walking activity [10]. Advances in open-source accelerometer algorithms allow researchers to use accelerometry data collected from smartphones to measure walking cadence [11]. Further, smartphones can leverage their large screens, user-friendly interfaces, and voice recordings that can guide older adults through self-administered functional assessments. Additionally, their data transfer capabilities enable seamless integration with study databases, facilitating efficient data transfer. Self-guided functional assessments, like a walk test, would overcome current barriers to performance of in-clinic assessments. Our primary aim was to detail the feasibility of “Walk Test,” a mobile application created to guide older adult surgical patients through usual- and fast-paced walks to measure walking cadence. We defined feasibility as ≥80% of participants being able to: (1) successfully install the application and enter the correct study identification number, (2) perform the usual- and fast-paced walk at-home and complete all 3 of the questionnaires, and (3) transfer the data back to the server. Additionally, we evaluated for the acceptability of Walk Test as measured by the Post-Study System Usability Questionnaire (PSSUQ) and the accuracy of Walk Test to measure walking cadence as compared to an in-clinic activPAL accelerometer.

Walk Test is a smartphone application that performs a remote walking cadence assessment. The application provides (1) visual and audio instructions to guide users to perform usual- and fast-paced timed walks, (2) timing and data recording of accelerometry during the 2-min usual- and fast-paced walks (3) post-walk surveys, and (4) data transmission. iOS and Android versions of Walk Test were developed. The Walk Test application employs a combination of text, images, and voice commands to guide participants through both the usual- and fast-paced walks (Fig. 1). In the app, users are instructed to select a long room or hallway at least 30 feet in length for the test, walk back and forth along the selected route, and are provided with guidance on how to cancel and restart the tests if needed. Safety instructions included recommendations to wear their normal walking shoes, to avoid wearing slippers, to use their walking aids (i.e., cane), and to stop in the event of shortness of breath or chest pain. During the walks voice commands are used to signal the passage of time (e.g., 30 s elapsed, 1 min remaining, 30 s remaining). Each walk was set for 2 min, as the 2-min walk test closely resembles the 6-min walk test while being more time-efficient and feasible for remote assessment, reducing the risk of participant disengagement [12]. This feature of playing a continuous audio stream also prevents the application from being suspended in the background by the smartphone’s power-saving mechanisms. At the end of each walk, the application vibrates and alarms to signal users to stop walking. If needed, users can take a short break before they start the fast-paced walk. Throughout both walks, the smartphone’s triaxial accelerometer records raw acceleration data at a frequency of 60 Hz. Data from each walk are stored in a .csv file, typically around 500 KB in size. Immediately following the walks, users are presented three surveys: (1) Duke Activity Status Index (DASI), (2) Post-Study System Usability Questionnaire (PSSUQ), and (3) post-walk questions. The DASI was included as it is a recognized standard for functional assessment in the perioperative literature [9]. The PSSUQ was administered after the walk to collect immediate feedback on the application's usability [13]. There were four post-walk questions to provide detail about how the walk was performed including (1) whether the walk took place in the home or outside of the home, (2) whether the user walked back and forth or in a continuous path, (3) whether the user was worried about falling, and (4) the user’s perceived exertion during the walks (i.e., very light, somewhat hard, hard, and very hard).

This study protocol was reviewed and approved by The University of Chicago Institutional Review Board (IRB), approval number IRB 22-0556. Written informed consent was obtained from participants to participate in the study. Participants were enrolled from January 31, 2023, through March 1, 2024.

To evaluate the feasibility of Walk Test, we performed a prospective cohort study with patients over the age of 60 years scheduled for abdominal surgery. Patients who were ambulatory with access to a safe walking space were called on the phone prior to their appointment at the Anesthesia and Perioperative Medicine Clinic. Patients were excluded if they did not have a compatible smartphone or were not able to walk independently. The study was explained, and consent was obtained electronically after participants discussed the risks and benefits of the study with the research coordinator (EY). Participants were sent hyperlinks via a text to download and install the iOS or Android application. Participants were instructed to enter their assigned study identification number upon installation and that number was sent with every data transmission, ensuring accurate identification of each participant data privacy.

A few participants preferred to discuss the risks and benefits of the study in-person. For those participants, study personnel discussed the risks and benefits of the study during the in-clinic visit. After consent was obtained, the same application deployment process described above was performed in-clinic, with no technical assistance from study personnel.

Participants were asked to complete the usual- and fast-paced walks using instructions provided by Walk Test: (1) at-home before surgery, and (2) in-clinic. At-home participants performed the walks guided by Walk Test. In-clinic participants performed the walks using Walk Test in a standardized 25-meter hallway while wearing an activPAL accelerometer attached to their right anterior thigh. Participants walked back and forth between two cones which were separated by 25 meters in a hallway. The activPAL served as the gold-standard reference to evaluate the accuracy of Walk Test.

Following the completion of the walks and questionnaires, participants were guided to submit the data to a University of Chicago server. Data transmission was unidirectional (participant to server) and was encrypted for security. The data files did not contain protected health information (Fig. 2). Files were compressed into a zip file and directly transmitted to a session database hosted by the University of Chicago. Our research coordinator then downloaded those files directly from the server for analysis.

A medical history, demographics, height, weight were obtained from the Anesthesia and Perioperative Medicine Clinic visit. Frailty was assessed using the Fried Frailty Phenotype with a 4-meter usual-paced gait speed, with a usual-paced gait speed of <1.0 m/s identifying older adults at risk of mobility disability [14, 15]. The DASI is a 12-item questionnaire (yes/no) to estimate functional capacity, with a cutoff of <34 to identify older adults at risk of adverse perioperative outcomes [16, 17]. To assess baseline disability, we used the WHO Disability Assessment (WHODAS) 2.0 which is a 12-item Likert scale (0–5) questionnaire which evaluates 6 different domains of disability and has been well validated in the perioperative literature [18, 19]. The raw score from the WHODAS 2.0 was converted into a percentile score consistent with prior analyses with a score of 35 indicating significant clinical disability [20].

To measure cadence using the smartphone accelerometer we used an open-source “one-size-fits-most” walking recognition algorithm to analyze the raw accelerometer output of the smartphone [11]. Briefly, this approach leverages the observation that regardless of the sensor location, orientation, or subject, during walking activity the device’s accelerometer signal oscillates around a local mean with a frequency equal to the performed steps [21]. To translate this information into cadence, the algorithm decomposes the vector magnitude of the accelerometer signal into its time-frequency representation using continuous wavelet transform, splits this projection into non-overlapping one-second windows, and estimates the temporal cadence as the frequency with the maximum average wavelet coefficient. The estimated frequency reflects the cadence a person walks within this time window. Finally, the average cadence was calculated as a mean of all one-second cadences over the duration of each observed walk.

For in-clinic walks, participants wore an activPAL (PAL Technologies Ltd., Glasgow Scotland) accelerometer adhered to the right thigh to serve as a gold-standard measure of cadence. The activPAL is a small, lightweight triaxial accelerometer that samples data at 20 Hz and measured cadence during the in-clinic usual- and fast-paced walks. Data from the activPAL were downloaded to a study computer using the PAL Software Suite version 8 and periods of walking were manually identified by our research coordinator (EY). Once the segments were identified, we calculated the number of total steps walked and divided by the duration of each walk to get the mean walking cadence for each 2-min walk.

Participants were asked about whether they were concerned about falling during the walks at-home and whether they experienced a fall during the walks at-home.

Descriptive summaries of categorical variables were reported as frequency and percentages; continuous variables are reported as a mean and standard deviation or median and interquartile range (IQR) if the variables values were not normally distributed. The Shapiro-Wilk test and histogram plots were used to assess for normality of the distributions.

To assess the feasibility of the app, we calculated the proportion of participants who successfully installed the application, entered the correct study identification number and completed data transfer, performed the usual- and fast-paced walks and completed all 3 questionnaires. Feasibility was defined a priori as ≥80% of participants being able to complete each step without assistance. Acceptability was assessed using the PSSUQ with a score of <4 indicating sufficient acceptability.

To establish the accuracy of Walk Test, we compared the cadence measured by Walk Test to the cadence as measured by the activPAL for the in-clinic walks. We used two methods to assess the agreement and accuracy between Walk Test and activPAL. First, we used Lin’s concordance correlation coefficient (CCC) as a measure of agreement and precision between the two measurements. The Lin’s CCC ranges between −1 and 1, with zero indicating no correlation and an absolute value of 1 indicating complete or perfect correlation. Second, we calculated the mean absolute percent error (MAPE) to measure the accuracy of our approach. The MAPE is frequently used to calculate the accuracy between accelerometer devices with <5% being an industry standard [22].

The Lin’s CCC was used to determine the agreement between at-home and in-clinic walks before surgery. We also used a paired Wilcoxon signed rank test to determine a difference in the mean between the at-home and in-clinic attempts for each of the two walks (usual- and fast-paced).

Finally, we conducted two sensitivity analyses to ensure the robustness of our findings. We stratified the usual- and fast-paced walks by BMI (<30 kg/m² vs. ≥30 kg/m²) and sex (male vs. female), as both factors are known to influence walking cadence [23]. Lin’s CCC was used to assess agreement within each subgroup, and a correlation test was performed to compare CCC values between groups [24]. R software version 4.3.3 was used for accelerometer output analysis and STATA 16 was used for all other parts of data analysis.

We screened a total of 180 patients, 114 declined to participate, 16 were screen fails as they had a smartphone that were incompatible with Walk Test (used iOS software version 8 or lower) and 50 enrolled in our study, of whom 9 participants withdrew for a final cohort of 41 patients. Patient characteristics can be seen in Table 1. The mean age was 69 ± 5 years, with 26 (63%) female and 15 (37%) owning an Android-based smartphone. The mean WHODAS 2.0 score was 9 ± 11; DASI score was 40 ± 17, usual-paced gait speed was 0.95 ± 0.21 m/s, and 12 (29%) classified as pre-frail and 5 (12%) patients classified as frail.

For feasibility, 24 (59%) successfully installed the application at home, with 17 participants requesting informed consent to take place in-person. All of the remaining 17 participants were able to successfully install the application on their own while in-clinic without the assistance of study personnel. All participants (41/41) entered the correct study identification number, 88% (36/41) performed the usual- and fast-paced walks at-home, 100% (41/41) performed the in-clinic walks, and 100% successfully transferred the data to the University of Chicago servers.

The median (IQR) overall score of the PSSUQ was 1 (1, 1), demonstrating strong measures of acceptability and usability, with 0 participants having an overall score greater than 4. The median (IQR) subdomain score for information quality was 1 (1, 1) and the median (IQR) subdomain score for system usefulness was 1 (1, 1) which demonstrated the strength of the quality and usefulness of the information presented to participants.

The mean walking cadence of patients for the usual-paced walk was 100 ± 11 steps/min as measured by the activPAL and 102 ± 11 as measured by Walk Test. The mean walking cadence of patients for the fast-paced walk was 108 ± 14 steps/min as measured by the activPAL and 110 ± 14 steps/min as measured by Walk Test. The range of walking cadences as measured by the activPAL was between 72 and 141 steps/min. The Lin’s CCC between activPAL and Walk Test for usual-paced walk was 0.97 (95% CI: 0.96, 0.99, p < 0.001) with an average difference of 2 ± 2 steps/min with a 95% limits of agreement between −2 and 6 steps/min (Fig. 3a). The Lin’s CCC between activPAL and Walk Test for fast-paced walk was 0.96 (95% CI: 0.93, 0.98, p < 0.001) with an average difference of 2 ± 3 steps/min with a 95% limits of agreement between −5 and 9 steps/min (Fig. 3b). The MAPE between the activPAL and Walk Test for the usual- and fast-paced walks were 2.0 ± 2.0% and 2.6 ± 2.6%, respectively, and the MPE was −3 ± 5%.

A total of 88% (36/41) of the participants performed the at-home walks. Patients walked in a back-and-forth pattern for 69% (25/36) of the walks as compared to a continuous walk, and 61% (22/36) of participants performed the walk away from home as compared to inside the home. Most participants reported light exertion (23/36, 64%) after completing both walks while 2 participant reported hard exertion (2/36, 6%) and 11 reported somewhat exertion (31% 11/36) during the walks. Only 2 participants reported they were concerned about falling during the walks. The mean cadence for usual- and fast-paced walks as measured by Walk Test at-home was 103 ± 13 steps/min and 113 ± 13 steps/min, respectively.

The Lin’s CCC between the at-home and in-clinic walks for usual-paced walks was 0.70 (95% CI: 0.53, 0.86) with an average difference of −1 ± 9 steps/min and 95% limits of agreement of −20 and 17 steps/min (Fig. 4a). The Lin’s CCC between the at-home and in-clinic walks for fast-paced walks was 0.46 (95% CI: 0.21, 0.72) with an average difference of −2 ± 16 steps/min and 95% limits of agreement of −33 and 28 steps/min (Fig. 4b).

There were no adverse events reported throughout the enrollment period. Even though 34% (14/41) participants reported a fall in the last 6 months, there were no falls during any of the walks at-home or in-clinic and only 2 participants were concerned about falling during the walks at-home.

To assess the robustness of our findings, we conducted two sensitivity analyses stratifying the usual- and fast-paced in-clinic walks by BMI and sex, comparing activPAL to the Walk Test. The Lin’s CCC for usual-paced walks by BMI was high and not significantly different between groups (BMI ≤30: 0.97, 95% CI: 0.94–0.99 vs. BMI >30: 0.97, 95% CI: 0.95–0.99; p = 0.72). Similarly, for fast-paced walks, the Lin’s CCC remained consistent across BMI groups (BMI ≤30: 0.95, 95% CI: 0.90–0.99 vs. BMI >30: 0.97, 95% CI: 0.94–0.99; p = 0.42). When stratified by sex, there were no significant differences in the Lin’s CCC for usual-paced walks (male: 0.96, 95% CI: 0.92–0.99 vs. female: 0.98, 95% CI: 0.96–0.99; p = 0.41) or fast-paced walks (male: 0.98, 95% CI: 0.95–0.99 vs. female: 0.94, 95% CI: 0.90–0.98; p = 0.20).

The current study was designed to assess the feasibility, acceptability, and accuracy of a new smartphone application, Walk Test, to measure self-administered walking cadence. This initial step of our project serves as a strong foundation to develop a cadence-based assessment and intervention to guide older adults to increase moderate-intensity walking before surgery, thereby potentially improving functional outcomes. The independent installation of the app among older adults underscores the feasibility of delivering and conducting self-administered functional tests to their home. Considering the challenges many clinics face to integrate functional performance measures into their workflow, enabling functional tests at home pre-visit may become a critical part of patient assessment. Favorable scores on the PSSUQ corroborate the acceptability and ease of use of the app. Future iterations of Walk Test will need to overcome the variability seen at-home in order to become clinically useful.

Our study demonstrated a very high level of accuracy between Walk Test and activPAL to measure cadence in the clinic setting, but there was moderate agreement between the in-clinic and at-home cadence. The strong correlation and low MAPE between Walk Test and activPAL measured walks suggest that Walk Test accurately measures walking cadence; however, there is marked differences in performance of the at-home walks likely due to variations in participant effort and the walking space used for the tests. At-home functional performance tests have not been widely implemented clinically as validation of at-home performance is challenging due to variation in patient effort. In a study by Keating et al. [25], participants walked an in-clinic and at-home 6-min walk test using a piece of 10-meter string to mark off the track. Despite standardizing the course, there was marked variation between the at-home and in-clinic distance walked likely due to participant effort, with only 38% of participants having the mean difference between the two walks within the minimum clinically important difference (<30 meters) for 6-min walk tests. Walking cadence has demonstrated large variability between in-clinic and at-home assessments as measured by wrist-worn ActiGraph accelerometers. In a secondary analysis of accelerometer data from the Study to Understand Fall Reduction and Vitamin D (STURDY), participants demonstrated slower fast-paced cadence at-home than in-clinic. The at-home walking cadence was more predictive of falls than any of the in-clinic functional evaluations, including cadence during the in-clinic 6-min walk test [24]. Future studies will be necessary to identify the optimal measure (at-home vs. in-clinic), to reduce the variability between the at-home and in-clinic tests and determine appropriate cutoffs necessary to identify at-risk older adults prior to major abdominal surgery.

Our study focused on walking cadence and diverged from existing mobile applications and other wearables that primarily focused on total daily step counts. While daily steps are easily accessible as many smartphones already collect and record these data and serve as a simple marker of physical activity behavior, their analytical and clinical validation remains unknown [26, 27]. Further, the reliability of daily step counts is limited due to variability in smartphone usage patterns as smartphones are not always carried or worn, and this can alter the overall daily step count considerably. In contrast, our approach to measure cadence required active engagement from older adults, potentially reducing participation but offering more accurate insights into the capability of the participant. Additionally, our approach to use a smartphone rather than a wearable has the advantage of not requiring additional hardware provided to the patient. Commercial wearable devices measure daily step count but older adults, and particularly vulnerable older adults with poor functional capacity, are the least likely to own and use these devices [28]. Thus a smartphone based approach that requires the participant to perform a specific activity may provide a more accurate assessment of capability.

Walking cadence also differs from existing clinical applications that have focused on measuring 6-min walk test (6MWT) distance [13, 29, 30]. While the 6MWT is a validated test for assessing functional capacity, estimating distance using smartphones poses practical challenges, especially in home settings where participants may walk in confined spaces [31]. The two common methods of estimating distance from a 6MWT include measuring the number of steps walked and multiplying by estimated stride length or measuring the distance walked using the global position system. Our initial approach with a different application, Step Test, highlighted significant inaccuracies in estimating distance from stride length, particularly at slower walking cadences [32]. Utilizing the global positioning system may work in certain situations where the participant walks a longer distance in a continuous fashion rather than the likely back-and-forth pattern. However, half of our participants performed the walks in their home and half walked back and forth which would render this approach unusable as satellite signals may be absorbed or attenuated by walls and ceilings [26]. Thus, utilizing walking cadence as an intensity measure during a timed walk avoids the need to estimate distance, offering a more accurate depiction of capability.

A significant limitation in our study was the lack of a more standardized condition for performance of at-home walks, leading to low levels of agreement between the at-home and in-clinic cadence. Future research will focus on standardizing at-home protocols to ensure accuracy and consistency, including guidance on choosing suitable spaces for effective walk performance. While our participant cohort skewed slightly healthier and more active, the usual-paced gait speed was 0.95 m/s and half of our patient population was either pre-frail or frail. That the in-clinic accuracy of the application was high despite an overall slower walking cohort, as evidenced by their usual-paced gait speed, strongly supports our approach to measuring cadence. Finally, our study was not able to determine whether the in-clinic or at-home cadence was better at identifying at-risk older adults. Future research will be needed to determine what measurement is best suited to identify older adults at increased risk of functional decline and complications.

Our study’s successful demonstration of the feasibility, acceptability, and accuracy of Walk Tests to measure usual- and fast-paced walking cadence in older adults’ pre-surgery lays a strong foundation. Future work will refine at-home protocols, ensuring consistency and accuracy during the at-home measurement of walking cadence. This study paves the way to utilize this approach to guide older adults through a walking intervention, utilizing walking cadence as an intensity guide to enhance functional outcomes post-surgery.

We acknowledge the contributions of Fritz Anderson and Alex Weiss for building the iOS and Android version of Walk Test.

This study protocol was reviewed and approved by the University of Chicago Institutional Review Board (IRB), Approval No. IRB 22-0556. Written informed consent was obtained from participants to participate in the study.

The authors have no conflicts of interest to report.

Daniel Rubin received funding from the National Institute of Health National Institute on Aging (R03AG078957), Foundation for Anesthesia Education and Research, Carol and George Abramson Fund for Aging and Longevity, Chicago, IL, provided funds for the development and study of the mobile application.

Daniel Steven Rubin helped with the conception, acquisition of data, analysis, drafting and revising for content, final approval, and held accountable for all aspects of the work. Marcin Straczkiewicz helped with the conception, analysis, drafting and revising for final content, and held accountable. Emi Yamamoto helped with data acquisition, analysis, revising for final content, and held accountable for all aspects of the work. Maria Lucia L. Madariaga, Sang Mee Lee, and Margaret Danilovich helped with the analysis, drafting and revising for final content, and held accountable for all aspects of the work. Mark Ferguson, Jennifer S. Brach, and Nancy W. Glynn helped with the analysis, revising for final content, and held accountable for all aspects of the work. Megan Huisingh-Scheetz helped with the conception, analysis, drafting and revising for final content, and held accountable for all aspects of the work.

The data that support the findings of this study are not publicly available due to their containing information that could compromise the privacy of research participants but are available from the corresponding author (D.S.R.) upon reasonable request.