TIHM: An open dataset for remote healthcare monitoring in dementia

A secure digital platform was developed to integrate in-home IoT and remote monitoring technologies to collect routine physiological, sleep, movement, and ambient data¹¹.

Digital markers

Digital markers are measurable physiological and in-home movement data gathered and assessed by digital devices, including portable and passive monitoring sensors. Digital markers can deliver novel and useful insights into an individual’s activity patterns and physiological health, allowing for continuous and non-invasive healthcare monitoring. Remote monitoring technologies also provide a novel approach to monitoring the effect of new interventions in clinical trials and observational and interventional studies¹².

In TIHM, sensory devices were installed in participants’ homes, and activity data was continuously recorded via passive infrared (PIR) sensors (installed in the hallway and living room), movement sensors (on kitchen, bedroom and bathroom doors), door sensor (installed in the main entrance), and an under-the-mattress sleep-mat (for monitoring sleep and in-out of bed activity). Participants were supplied Bluetooth-enabled devices to measure their blood pressure, heart rate, body temperature, weight, and hydration daily. Figure 1 shows an example of the residential setting of a participant in the study equipped with the sensors. Details of the devices and digital markers are shown in Tables 1, 2.

Demonstration of a residential setting equipped with PIR sensors for in-home activity monitoring and other sensors for sleep and physiology monitoring in the TIHM project. PIR and door sensors are included in each room of the house. An under-the-mattress sensor is used for sleep and in-out-of-bed monitoring. Connected devices which are operated manually are also used in the setting to acquire physiology data.

Participants

To be eligible for this study, participants needed to meet the inclusion criteria of being a person over 50 years old, with a verified diagnosis of dementia (of any type) or mild cognitive impairment, who has the capacity to provide informed consent to participate in the study, and either received treatment from an Old Age Psychiatry department in the past or is currently on their caseload. In addition, participants required a study partner or caregiver who had known the PLWD for at least six months and was able to attend research assessments with them. If a participant was unable to provide information about their health, their partner or caregiver completed the necessary assessments on their behalf. Individuals with unstable mental states, including severe depression, severe psychosis, agitation, anxiety, active suicidal thoughts, or those receiving treatment for terminal illnesses were not included in the study. A total of 56 people were selected as participants. All the participants have granted the publication of this dataset. The demographic details of the participants in the dataset is shown in Table 3. Some participants in the study requested not to share all or part of their information outside the study. For these cases, the corresponding information is represented by “N/A” (Not Available) in Table 3 and their data was not included in the dataset.

Ethical approval

The TIHM study received ethical approval from the London-Surrey Borders Research Ethics Committee; TIHM 1.5 REC: 19/LO/0102. The study is registered with National Institute for Health and Care Research (NIHR) in the United Kingdom under Integrated Research Application System (IRAS) registration number 257561. To the best of our knowledge, this is the first publicly available dataset for remote healthcare monitoring for PLWD that includes in-home activity and sleep data, physiological measurements, and labelled health and care-related events during the monitoring period. TIHM is also currently being offered as a service by the Surrey and Borders National Health Service (NHS) Trust in the United Kingdom.

Dataset collection

We combined in-home sensory data with individuals’ healthcare information extracted from General Practitioner (GP) records and hospital visits to create a holistic view of their well-being and care needs.

The sensor deployment relied on off-the-shelf devices to monitor in-home activities and physiology. These sensors continuously collected and communicated the data to a data collection and integration platform. The data from the sensors in this release are de-identified, cleaned (removing redundant and multiple records) and merged based on their categories into four different tables which are further explained in Section Data Records. The annotations and data labels for this study were collected by a monitoring team who contacted the participants to determine if they had experienced a health-related event. The data was labelled as true if the monitoring team validated the presence of a health-related event and false if there was no event.

The initial alert generation for triaging a healthcare event was governed by a set of rules and thresholds applied to physiological measurements and the output of an analytical model designed to analyse in-home activity and physiology data¹³. This initial analytical model was only intended to guide the monitoring team in identifying episodes of agitation and creating a labelled dataset for further data analysis and machine learning developments.

By combining the data from the in-home sensors, we obtained a comprehensive understanding of an individual’s home activity and health, and used this information to determine the risk or presence of health related conditions⁹. For example, we detected changes in an individual’s activity patterns, such as a change in room usage that may indicate social isolation or agitation⁵.

Dataset de-identification

Two types of de-identification have been applied to data. During the study, the data was pseudo-anonymised for the clinical monitoring team and for developing analytical models. The data includes the demographics (age and sex) in addition to raw sensory observations and measurements. Information governance and control methods and procedures were applied to the data during the project. An NHS-approved Data Processing and Impact Assessment was conducted for the data collection, storage and access procedures. Before making the TIHM dataset¹⁰ available online, the data was then fully anonymised by removing all personally identifying information or identifiable attributes. Participants are randomly assigned with a universally unique identifier (UUID) to increase security in the de-identification.