Sleep is regularly assessed in inpatient care as a core symptom of psychiatric illness [1]. Standard overnight checks rely on staff observation and are inherently subjective, sometimes disrupting patient sleep [2]. Wearable devices offer a means of obtaining continuous, objective sleep and activity data without disturbing patients, presenting an opportunity to enhance the accuracy and efficiency of inpatient monitoring [3].

This implementation aimed to address a health system challenge identified by the World Health Organization—specifically, the lack of reliable, high-quality data to support clinical decision-making (challenge 1.3) [4]. Although many have advocated for integrating digital data into psychiatric practice [5-7], most prior efforts have centered on feasibility testing or predictive modeling rather than sustained clinical integration. Consequently, little is known about how wearable-derived data can be incorporated into existing workflows or meaningfully used by clinicians within inpatient psychiatric care.

Previous research illustrates both the promise and challenges of using digital measures in hospital settings. Kooij et al [8] examined the use of wireless wearable sensors on teaching hospital wards and reported that nurses perceived benefits in the early detection of physiological deterioration among hospitalized patients. Similarly, Pickham et al [9] conducted a pragmatic randomized controlled trial evaluating wearable patient sensors for enhancing nursing care delivery. In psychiatry, Hong et al [10] used wearable-derived modeling to forecast symptom trajectories, whereas Odachi et al [11] and Durrani et al [12] evaluated the feasibility of feeding back sleep and activity data to clinicians and patients. While these studies have examined the feasibility of wearable-derived data use among clinicians within a research framework, none have attempted to integrate systems for wearable-derived data collection and delivery in clinical care through an implementation framework. By “implementation framework,” we refer to a structured, iterative process of adapting device management, data processing, and data delivery procedures based on clinician feedback with the goal of embedding the system into existing inpatient workflows. This approach extends beyond feasibility testing to understanding the organizational, usability, and workflow determinants of sustained adoption. This approach answers the call in the study by Odachi et al [11] for the development of guidelines for implementation of digital health tools through collaboration among researchers, engineers, and clinical staff—especially exploring how to incorporate data into routine care.

Our implementation was a collaboration between a research laboratory and clinical staff at a single adult inpatient unit at a major psychiatric hospital (Table 1). We sought to explore and evaluate methods for integrating wearable-derived sleep and activity feedback into psychiatric inpatient workflows through a clinician-partnered implementation. Using an iterative design framework within routine inpatient psychiatric care, we adapted project roles, data feedback formats, and delivery mechanisms throughout the implementation in response to feedback from participating clinical staff. The primary aim of this implementation was to evaluate how objective sleep and activity data could be effectively collected, reported, and incorporated into clinical decision-making within a psychiatric inpatient setting. Specifically, we sought to develop, deploy, and iteratively refine a prototypical sleep and activity report for clinician use, assessing its feasibility, usability, and integration into clinical workflow.

Implementation unfolded in 2 phases (Table 2; Multimedia Appendix 1). Phase 1 (August 2023-December 2023) tested several methods of device management, report request, and report delivery. During this phase, we began writing natural language summaries of the graphical report (Figure 1) in response to challenges of interpretability (Figure 2). Uptake metrics were tracked throughout phase 1, including number of admissions, devices offered to patients, devices accepted by patients, and reports generated. Phase 2 (January 2024-June 2024) established an email-based request model for reports and streamlined data turnaround times, fully automating the generation of graphical reports and natural language summaries. In this phase, project resources were shifted away from driving uptake on the unit and rededicated to streamlining systems for report generation, report delivery, and gathering clinician feedback. This shift was in response to low uptake in phase 1 and served as an opportunity to clarify and address the challenges identified in phase 1. Consequently, uptake metrics were not tracked, and reports were only generated at clinician request.

At the conclusion of phase 2, semistructured interviews were conducted with a convenience sample of 8 psychiatrists from across the hospital to explore their impressions of the sleep and activity report. Of these psychiatrists, only 1 had used the sleep and activity report in practice. This was the unit medical director (AGY), who is listed as an author of this paper and who participated in the implementation as a highly engaged early adopter and project champion on the unit. Additionally, the unit medical (AGY) and nursing (KH) directors were each interviewed separately to discuss the course of the implementation on the unit.

Interviews were analyzed using the rigorous and accelerated data reduction technique developed by Watkins [13]. Two members of the research team (BWC and RDP) collaboratively developed a “coding scheme” organizing the interview data into categories and then independently coded the interview transcripts and met regularly to engage in consensus-reaching discussions to resolve any discrepancies in coding. Codes were then grouped into overarching themes comprising feedback on the report, workflow integration, and barriers to and facilitators of implementation.

Sleep and activity were monitored using waterproof, wrist-worn GENEActiv devices (Activinsights Ltd). We chose this device for its capacity to collect continuous, raw triaxial accelerometer data at a fixed rate without frequent recharging. Devices sampled at 20 Hz for up to 30 days.

Data were processed using DPSleep, an open-source longitudinal pipeline for sleep analysis from accelerometer data [14]. Extracted sleep parameters included sleep onset time, wake time, sleep duration, and sleep fragmentation. Fragmentation index was calculated by the slightly modified percentage of mobile minutes during the sleep epoch as described in the study by Rahimi-Eichi et al [15] and is intended as a proxy measure for sleep quality. Minute-level activity was derived and normalized within participants expressed as relative activity percentiles.

Medication administration data were retrieved from the electronic medical record (EMR) via the institution’s data warehouse. The graphical section of the report was generated in MATLAB 2019b (MathWorks, Inc). The natural language section was generated using a Python script that summarized key data from the graphical report using simple heuristics; large language models were not used in data processing or report generation.

Our implementation intended to support inpatient psychiatrists in clinical decision-making by reporting objective inpatient sleep and activity. In the implementation, we focused on discerning how psychiatrists wished to see data presented in the report and how best to integrate reports into psychiatrists’ workflow.

All protected health information was stored on encrypted institutional servers and HIPAA (Health Insurance Portability and Accountability Act)-compliant cloud storage. Data collection and management followed institutional governance procedures.

Patient participation was voluntary, and consent to wear a device was obtained verbally as part of routine clinical workflow consistent with institutional review board determination that the project qualified as an implementation initiative rather than human subject research. This designation as not human subject research covered both patient participation and the collection of feedback from clinical staff via semistructured interviews.

The implementation architecture comprised three integrated modules: (1) data acquisition using GENEActiv devices and software, (2) processing and analysis using DPSleep, and (3) report generation (MATLAB and Python) and delivery (email and secure file transfer).

The implementation was conducted at McLean Hospital, a division of Mass General Brigham and an affiliate of Harvard Medical School. The implementation was coordinated by the McLean Institute for Technology in Psychiatry and Functional Neuroimaging and Bioinformatics Laboratory in collaboration with unit leadership. All intellectual property is owned by Mass General Brigham.

The implementation was funded by discretionary institutional resources from McLean Institute for Technology in Psychiatry. The implementation period covered August 2023 to June 2024, with total expenditures confined to existing institutional budgets; no additional financial resources were required beyond routine operational support. No external grant funding was used.

Hardware costs consisted of GENEActiv Original devices (approximately US $275 per device), which were supplied from an existing inventory in the Functional Neuroimaging and Bioinformatics Laboratory purchased for use in prior studies via grants and discretionary funds. Thirty devices were allocated for this implementation. Data analysis and reporting relied exclusively on in-house–developed software. Personnel time—including a single research assistant managing device turnover and data processing—represented most ongoing costs.

Sustainability was assessed in terms of patient participation, technical feasibility, and potential institutionalization. Patient willingness to wear devices was high, and nurses reliably retrieved devices before discharge, demonstrating operational feasibility. One patient inadvertently took a device home upon discharge and promptly mailed it back; no watches were destroyed. Automation of report generation markedly reduced personnel burden, increasing scalability.

Long-term sustainability will depend on (1) embedding automated data transfer and report visualization within the EMR to ensure rapid and reliable access, (2) institutional recognition of objective sleep reporting as a quality of care metric, and (3) sustained cross-departmental engagement fostered by a shared sense of ownership among all participating staff.

Financially, the implementation could be maintained with minimal additional cost once incorporated into standard clinical workflows. Strategically, this initiative aligns with the hospital’s and the World Health Organization’s digital health objectives promoting interoperable, data-driven clinical decision support.

The implementation was conducted at a single site—a 21-bed inpatient unit within a 238-bed psychiatric hospital in Massachusetts, United States. The initiative represents a subnational implementation focused on feasibility and workflow adaptation prior to potential expansion locally and beyond.

During phase 1 of the implementation (August 2023 to December 2023), 155 patients were admitted to the unit, of whom 88 (56.8%) were offered a device and 68 (77.3%) accepted it. During this phase, 61.8% (42/68) of patients wearing a device had at least one sleep and activity report generated and delivered to their psychiatrist. Reports were not generated in all cases of device wear because report generation in this phase began with a request model and clinicians did not request reports for all participating patients (Table 2).

During phase 2 (January 2024 to June 2024), complete automation of report generation markedly improved timeliness: average report generation time decreased from approximately 5 days to less than 24 hours, and in several instances, same-day delivery was achieved. During this phase, 34 reports were generated and delivered, of which 32 (94.1%) were requested by the medical director (AGY) on the unit. One of the psychiatrists present in phase 1 had left the unit by this point, and a team of rotating clinicians covered this psychiatrist’s role on the unit for the duration of phase 2. Two reports were requested by rotating clinicians. No reports were requested by the third psychiatrist on the unit.

This implementation explored methods of collecting and reporting objective, wearable-derived sleep and activity data within inpatient psychiatric care. Most patients (68/88, 77.3% of admits who were offered a device) were willing to wear the device. Reports were particularly useful for reconciling discrepancies between patient and nursing reports of sleep, offering an additional, objective reference point to support clinical decision-making. However, uptake among psychiatrists was limited, with sustained use only by the medical director (AGY) who co-led the initiative. Therefore, the demonstrated clinical utility of the report was concentrated in a highly engaged early adopter and may not yet reflect routine uptake among average end users. Uptake of report use was primarily limited due to issues of clarity, lack of EMR integration, speed and reliability of report delivery, and limited buy-in from clinical staff.

These findings echo earlier reports that highlight both the promise and difficulty of integrating digital measures into clinical workflows. Kooij et al [8] and Pickham et al [9] similarly found that, while wearable systems can enhance early detection and clinical awareness, sustained use depends on how seamlessly data fit into existing work routines. Similarly, Odachi et al [11] and Durrani et al [12] observed clinician interest in sleep feedback but noted practical barriers related to usability and workflow integration.

Our findings complement those of prior work by moving beyond feasibility and prediction studies to an implementation of wearable-derived sleep and activity feedback within psychiatric inpatient care using an iterative design framework that evolved procedures in response to clinician feedback and operational constraints. This approach allowed us to refine multiple methods for device management, data processing, and report delivery, yielding a more comprehensive understanding of the technical and interpersonal factors that shape feasibility in real-world inpatient psychiatric contexts.

Our implementation aligns with calls from Mohr et al [5], Westheimer et al [6], and Oudin et al [7] to move beyond feasibility demonstrations toward pragmatic integration of digital data into psychiatric decision-making. Specifically, clinicians’ preference for simply displayed data that appear directly in the EMR prior to rounds reinforces that usability and timing are critical.

Our experience underscores the need to address both technical and interpersonal infrastructure in digital health implementation in accordance with previous inpatient monitoring studies [8,9]. Achieving buy-in at all levels of staff participating in the implementation is crucial. To this end, we recommend recruiting “project champions” at all levels and keeping in communication with them throughout the implementation. Once buy-in and compliance are achieved, it is crucial that technical infrastructure be able to reliably meet the demands and expectations of users. Our system’s reliance on software in-development and nonnetworked data collection limited report timeliness. The lack of near–real-time data transfer constrained the immediacy that clinicians require in fast-paced inpatient settings. These limitations are consistent with prior observations that technical latency and data opacity undermine clinical confidence in wearable-derived data [10,12].

Successes and indications of feasibility were also evident. Most patients (68/88, 77.3% of admits who were offered a device) were willing to wear it for the duration of their stay. The reports were used by 1 of the 3 psychiatrists on the unit, who also participated in this implementation as a key member. He found the reports useful—both as a tool for insight into patient sleep and activity and for discussing medication and sleep with patients.

Our implementation was conducted on a single unit and is therefore limited in its generalizability. Our capacity for interviewing was limited, so our interviews reviewing the implementation primarily included the medical (AGY) and nursing (KH) directors on the unit—both of whom are authors of this paper—introducing potential bias in the perspectives we gathered regarding the barriers and facilitators encountered by the implementation, as well as the suggestions for further implementation initiatives. Additionally, much of the feedback we received on the utility and design of our reports was derived from prototype review with psychiatrists not directly involved in the implementation rather than real-world users. These psychiatrists were selected by the implementation team and are thereby subject to potential sampling bias. Finally, conducting this implementation using technologies in development limits our ability to assess the feasibility of implementing commercial-grade wearable-derived sleep and activity reporting on inpatient units.

Future implementations should pursue both technical optimization and organizational integration. Technically, establishing real-time, automated data pipelines and embedding outputs directly into EMR interfaces would improve timeliness and accessibility. Organizationally, fostering multidisciplinary ownership—through sustained involvement of all staff levels, regular feedback, and identification of “project champions”—could enhance buy-in and sustainability.

Future research should more closely assess the clinical impact of such tools on treatment outcomes, clinician behavior change, and patient engagement. Evaluation of how objective sleep and activity data influence diagnostic decisions, medication adjustments, and patient satisfaction will be essential to justify broader adoption. Finally, educational initiatives to familiarize clinicians with the interpretation of wearable-derived metrics may help bridge the conceptual divide between digital and traditional psychiatric assessment.

This implementation suggests that wearable-derived sleep and activity data reporting is feasible in inpatient psychiatry, as demonstrated by a high and consistent rate of patient participation and technical report generation. This data reporting potentially offers clinically meaningful insights, particularly when patient and staff sleep reports conflict, as demonstrated by limited but consistent use among 1 clinician of 3 on the unit where the implementation took place, who served as an early adopter and contributor to the implementation (AGY). Additionally, reports were thought to be potentially useful by a convenience sample of 7 psychiatrists at the hospital who reviewed the reports but did not use them in clinical practice. Sustainable use and broader uptake are more likely with reliable, near-instantaneous data transfer; EMR integration; and shared implementation ownership across staff levels.