James Skoric

BACKGROUND Digital tools for tracking health and health-related behaviours are increasingly available and helpful to individuals managing chronic conditions, including pain. Despite the growing use and observed benefits of mHealth, sustained engagement is an issue. Little is known about users of widely adopted apps such as ManageMyPain (MMP). User characteristics and engagement with MMP were assessed in 2017, however, improvements to the app and the broader digital health landscape warrant an updated investigation. OBJECTIVE To investigate the relationship between MMP app engagement and user characteristics including sociodemographic variables and pain-related factors to understand how reach and clinical impact can be maximized for individuals living with pain. METHODS Machine learning techniques were used to cluster MMP user profiles into 5 distinct engagement groups based on longevity, number of records, and frequency of app use. Differences among clusters were assessed using Chi-square tests, ANOVAs, and logistic regression as appropriate. Secondary analyses assessed interactions with gender for each cluster. RESULTS A total of N=86,429 app users consented to have their data used for research purposes. Gender differences for all engagement groups emerged (p<.001), with a higher proportion of male users in the most engaged group. Statistically significant differences emerged for all variables investigated including age, number of pain conditions, number of medications, average pain severity, daily reflection score, and pain volatility (all p<.001). Cluster membership significantly predicted cannabinoid use, with all other clusters having significantly lower odds of use than the high longevity, high engagement cluster (all p<.001). High longevity cluster membership predicted opioid medication use, with other clusters having significantly lower odds (all p<.001). Self-reported depression was more likely to be reported by users with higher longevity (all p≤.001), less likely to be reported by males (p=.024), and more likely reported by older app users (p<.001). Anxiety showed increased odds of being reported by higher longevity and higher engagement clusters (both p<.001), those who reported their gender as other (p<.001), those with higher pain severity (p<.001), and younger app users (p=.002), and had lower odds of being reported by males (p<.001). Comorbid depression and anxiety had reduced odds in the lower longevity clusters (all p≤.004), and for males (p<.001), and increased odds for those who reported their gender as other (p<.001) and those who reported higher mean pain severity (p<.001). CONCLUSIONS Results demonstrate that different dimensions of engagement capture distinct aspects of the pain experience and may represent varied unmet needs. Our findings indicate potential opportunities to develop tailored digital interventions to support differing pain self-management needs and bolster engagement among those with lower involvement. Effective digital interventions could address some of the gaps in care that exist for individuals living with pain, improving access, communication, reducing inequities, and providing self-care support.

The effective management of chronic pain remains a significant challenge due to its complex nature. This study explores the utility of digital health tools, specifically the Manage My Pain app, to not only monitor symptoms but also collect valuable information that may be used to predict significant improvements in user outcomes through the application of machine learning techniques. In this study, a comprehensive set of features, including demographic details, pain descriptions, and app usage were extracted from one-month of self-reported data collected from 6,413 users of the Manage My Pain app. These features along with temporal sequences of pain and function scores were used to train and validate multiple models aiming to predict significant functional improvements the following month. We found that combining extracted and temporal features led to superior models, regardless of model architecture. On a held-out test set, a random forest achieved a balanced accuracy of 0.75 and an area under the receiver operating characteristic curve (AUC) of 0.85. A convolutional neural network with multilayer perceptron demonstrated a balanced accuracy of 0.79 and an AUC of 0.88. Finally, a time-series transformer combined with TabNet achieved a balanced accuracy of 0.77 and an AUC of 0.84. By integrating machine learning with digital health data from the Manage My Pain app, significant functional improvements in individuals with chronic pain can be predicted. This study highlights the potential of forecasting outcomes using regularly self-reported outcome information captured by patient-facing digital health tools. These forecasts could significantly alter treatment strategies and improve chronic pain management, underscoring the transformative impact of digital health technology in chronic pain care.

Multimodal cardiac monitoring systems require multiple sensors. This complexity renders them impractical in scenarios requiring continuous monitoring in everyday life. To overcome this limitation, we demonstrate a generative modeling framework that leverages the vibrational cardiography (VCG) signal from a single, wearable sensor to estimate multiple cardiac waveforms. We recorded VCG signals at the xiphoid process of 20 subjects, along with electrocardiography (ECG), impedance cardiography (ICG), non-invasive blood pressure (NIBP), and photoplethysmography (PPG). To broaden the range of cardiac, respiratory, and hemodynamic responses, we included breath holding, deep breathing, and the cold pressor test. A conditional Generative Adversarial Network was trained to reconstruct normalized ECG, ICG, NIBP, and PPG signals from VCG inputs. The model was evaluated using a leave-one-subject-out validation scheme to assess calibration-free generalization across individuals. We show that the reconstructed waveforms exhibit strong alignment with their respective targets, capturing both morphological structure and temporal dynamics, with median Pearson's correlation coefficients of 0.808, 0.907, 0.833, and 0.929 for ECG, NIBP, ICG, and PPG, respectively. We demonstrate that accuracy is consistent across interventions. The fiducial point analysis demonstrated the ability to capture key cardiac features within each waveform. Hence, we demonstrate the viability of combining single-sensor VCG with generative modelling to estimate multiple cardiac waveforms, offering a streamlined alternative to conventional multimodal sensor systems. This approach has the potential to improve the practicality of continuous cardiac monitoring in daily life.

Background: Chronic pain is a complex condition that affects more than a quarter of people worldwide. The development and progression of chronic pain are unique to each individual due to the contribution of interacting biological, psychological, and social factors. The subjective nature of the experience of chronic pain can make its clinical assessment and prognosis challenging. Personalized digital health apps, such as Manage My Pain (MMP), are popular pain self-tracking tools that can also be leveraged by clinicians to support patients. Recent advances in machine learning technologies open an opportunity to use data collected in pain apps to make predictions about a patient’s prognosis. Objective: This study applies machine learning methods using real-world user data from the MMP app to predict clinically significant improvements in pain-related outcomes among patients at the Toronto General Hospital Transitional Pain Service. Methods: Information entered into the MMP app by 160 Transitional Pain Service patients over a 1-month period, including profile information, pain records, daily reflections, and clinical questionnaire responses, was used to extract 245 relevant variables, referred to as features, for use in a machine learning model. The machine learning model was developed using logistic regression with recursive feature elimination to predict clinically significant improvements in pain-related pain interference, assessed by the PROMIS Pain Interference 8a v1.0 questionnaire. The model was tuned and the important features were selected using the 10-fold cross-validation method. Leave-one-out cross-validation was used to test the model’s performance. Results: The model predicted patient improvement in pain interference with 79% accuracy and an area under the receiver operating characteristic curve of 0.82. It showed balanced class accuracies between improved and nonimproved patients, with a sensitivity of 0.76 and a specificity of 0.82. Feature importance analysis indicated that all MMP app data, not just clinical questionnaire responses, were key to classifying patient improvement. Conclusions: This study demonstrates that data from a digital health app can be integrated with clinical questionnaire responses in a machine learning model to effectively predict which chronic pain patients will show clinically significant improvement. The findings emphasize the potential of machine learning methods in real-world clinical settings to improve personalized treatment plans and patient outcomes.

Goal: Seismocardiography (SCG) offers critical insights into cardiac performance, but its analysis often faces challenges due to the limited availability of data. This study aims to generate synthetic SCG heartbeats which can augment existing datasets to enable more research avenues. Methods : We trained a Wasserstein generative adversarial network (GAN) with gradient penalty on authentic SCG heartbeats. It was conditioned with embedded subject-specific identifiers to create individualized heartbeats. We employed linear permutations in the latent and conditional spaces to control signal features, and a convolutional network to classify lung volume states from real and synthetic data separately. Results : The model effectively replicated SCG signal morphology, while maintaining a level of variance which matches the variability of cardiac activity. Comparisons with real SCG waveforms yielded Pearson's r-squared correlation of 0.62 for average heartbeats. Linear manipulations were successful in controlling simple features although they were limited in more complex characteristics. Additionally, the model demonstrated strong performance in practical applications, with the synthetic data achieving an accuracy of 88% in lung volume classification as compared to 89% achieved with real data. Augmenting real data with additional synthetic data improved performance by 3%. Conclusions : GANs for artificial SCG heartbeat generation produce realistic and diverse results that have the potential to overcome data limitations, thereby enhancing SCG-based research.

Wearable sensing has become a vital approach to cardiac health monitoring, and seismocardiography (SCG) is emerging as a promising technology in this field. However, the applicability of SCG is hindered by motion artifacts, including those encountered in practice of which the strongest source is walking. This holds back the translation of SCG to clinical settings. We therefore investigated techniques to enhance the quality of SCG signals in the presence of motion artifacts. To simulate ambulant recordings, we corrupted a clean SCG dataset with real-walking-vibrational noise. We decomposed the signal using several empirical-mode-decomposition methods and the maximum overlap discrete wavelet transform (MODWT). By combining MODWT, time-frequency masking, and nonnegative matrix factorization, we developed a novel algorithm which leveraged the vertical axis accelerometer to reduce walking vibrations in dorsoventral SCG. The accuracy and applicability of our method was verified using heart rate estimation. We used an interactive selection approach to improve estimation accuracy. The best decomposition method for reduction of motion artifact noise was the MODWT. Our algorithm improved heart rate estimation from 0.1 to 0.8 r-squared at −15 dB signal-to-noise ratio (SNR). Our method reduces motion artifacts in SCG signals up to a SNR of −19 dB without requiring any external assistance from electrocardiography (ECG). Such a standalone solution is directly applicable to the usage of SCG in daily life, as a content-rich replacement for other wearables in clinical settings, and other continuous monitoring scenarios. In applications with higher noise levels, ECG may be incorporated to further enhance SCG and extend its usable range. This work addresses the challenges posed by motion artifacts, enabling SCG to offer reliable cardiovascular insights in more difficult scenarios, and thereby facilitating wearable monitoring in daily life and the clinic.

Wearable sensing has enabled physiological monitoring in everyday life. Clinically, cardiovascular and respiratory monitoring commonly requires separate devices. However, for feasibility in wearable systems, they are integrated into a single device. Since seismocardiography (SCG) records the motion of the chest wall, it has the potential to monitor both vital signs. The primary application of SCG is cardiovascular monitoring. As a result, its ability to detect respiration has been less defined. This study characterizes how SCG is affected by respiration and consequently, how respiratory metrics can be derived from these effects. The sternal motion of 40 subjects was recorded using an inertial measurement unit with a spirometer for reference. Three scenarios were examined with the subject at rest, holding their breath at peak inhalation, and holding their breath when exhaled. Three main aspects of respiratory modulation were explored: amplitude modulation, frequency modulation, and baseline wandering. The SCG amplitude was observed to be dependent on respiratory volume whereas baseline and frequency modulation were dependent on inhalation phase. All three effects were employed separately in the calculation of respiratory rate. Estimations from baseline wandering, amplitude modulation, and heart rate modulation produced r-squared values of 0.71, 0.44, and 0.66, respectively. The accuracy of tidal volume measurement was limited by a high inter-subject variability. Estimation of the respiratory waveform produced an average r-squared of 0.76 using multivariate linear regression. This demonstrates the potential of SCG as a tool for respiratory monitoring within an integrated, non-invasive cardiorespiratory monitoring system from a single point of contact.

Cardiography is an indispensable element of health care. However, the accessibility of at-home cardiac monitoring is limited by device complexity, accuracy, and cost. We have developed a real-time algorithm for heart rate monitoring and beat detection implemented in a custom-built, affordable system. These measurements were processed from seismocardiography (SCG) and gyrocardiography (GCG) signals recorded at the sternum, with concurrent electrocardiography (ECG) used as a reference. Our system demonstrated the feasibility of non-invasive electro-mechanical cardiac monitoring on supine, stationary subjects at a cost of $100, and with the SCG–GCG and ECG algorithms decoupled as standalone measurements. Testing was performed on 25 subjects in the supine position when relaxed, and when recovering from physical exercise, to record 23,984 cardiac cycles at heart rates in the range of 36–140 bpm. The correlation between the two measurements had r² coefficients of 0.9783 and 0.9982 for normal (averaged) and instantaneous (beat identification) heart rates, respectively. At a sampling frequency of 250 Hz, the average computational time required was 0.088 s per measurement cycle, indicating the maximum refresh rate. A combined SCG and GCG measurement was found to improve accuracy due to fundamentally different noise rejection criteria in the mutually orthogonal signals. The speed, accuracy, and simplicity of our system validated its potential as a real-time, non-invasive, and affordable solution for outpatient cardiac monitoring in situations with negligible motion artifact.

Objective: We present a novel seismocardiography (SCG) -based approach for real-time cardio-respiratory activity measurement called the Autocorrelated Differential Algorithm (ADA). Methods: Measurements were performed on ten male subjects in the supine position for three 7-minute-long sets each, corresponding to 14,619 heartbeats. The ADA utilized temporal variations, windowing, and autocorrelation to produce physiological measurements corresponding to heart rate (HR), and left ventricular ejection time, and estimations of respiration rate, volume, and phase. The versatility of the ADA was investigated in two contexts: physical exertion and heart rate variability. Results: The accuracy of HR measurements at a sampling frequency of 200 Hz resulted in a correlation coefficient (r 2 ) of 0.9808 when compared with a manual annotation of all datasets. Its reproducibility was tested on externally obtained SCG and electrocardiography datasets, which produced an r 2 of 0.8224. The accuracy and computational time was also characterized for different sampling frequencies to quantify performance. The recommended sampling frequency is 200 Hz corresponding to a computation time of 0.05 s per instantaneous measurement using a standard desktop computer. Conclusion: The ADA delivered real-time SCG measurements with a refresh rate that was dependent on the computational time per measurement, which could be decreased by lowering the sampling frequency. Significance: The presented algorithm offers a novel tool towards real-time physiological monitoring in clinical and everyday scenarios.

We present an ultra-broadband and compact asymmetrical beam splitter using subwavelength grating based multi-mode interference coupler. We experimentally demonstrate splitting ratios from 50:50 to 90:10 over 100 nm bandwidth using a 24µm×3µm design.