Wearable Sensors for Delirium Detection at an Early Stage (WeSen_delirium)

Hypothesis and primary objective The overarching aim of the planned study is to identify sensor-based activity and sleep indicators and patterns (using the sensor 'Fitbit Charge 5'), which precede a delirium and thereby may qualify as potential person-level predictors to inform a predictive algorithm to detect an evolving delirium at an early stage. In addition, clarifying the feasibility of using such data for the research purpose of this study is the second aim.

Primary objectives pertain to:

1. Sensor-based profiling of delirium: Determination of delirium onset and termination; determination of sensor profiles of delirious states in terms of key characteristics (e.g., frequency of transitions between rest and inactivity, movement), which are stable across individuals whereby taking into consideration discussions on clinical motor subtypes of delirium.
2. Within-person level - forerunners and repercussions: Characterizing and comparing activity and sleep patterns (e.g., sleep deprivation, increased movement while asleep) derived from a wearable sensor preceding and following a delirium, delirium to delirium-unrelated time periods within persons who experienced a delirium (i.e., 'Can any such indicators / patterns distinguish between periods in time proximity to a delirium compared to delirium-unrelated time periods?').
3. Between-person level - forerunners and repercussions: Comparing real-time activity and sleep patterns derived from a wearable sensor across persons who experienced a delirium to those who did not (i.e., 'Can any such indicators / patterns distinguish between persons who develop a delirium and those who do not?').
4. Predictive algorithm development: Subsequently identification of activity and sleep indicators / patterns that qualify as potential forerunners of a delirium (i.e., distinct parameters or their patterns/interplay) to inform and develop a prediction algorithm for delirium detection at an early stage (to be validated in a follow-up study).
5. Visual data analysis: The investigators will employ clustering techniques (e.g., Self-Organizing Maps, K-Means Clustering) for longitudinal data to identify similar sensor measurement trajectories across study participants (e.g., for hourly heart rate measurements, hourly step counts, etc.). Because these analyses are exploratory, they will be refined on the basis of available data.

The predictions are based on a recent review summarizing state-of-the-art research. Evidence suggests the existence of clinical delirium subtypes (i.e., motor subtypes: hypoactive, hyperactive, mixed).

However, research on subtype classification is still rare. Initial research suggests that the hypoactive and mixed type seems more distinguishable from delirium than the hyperactive type. The predictions might thus be more accurate for deliria of the hypoactive and mixed type. Specifically, the investigators anticipate that:

1. Sensor-based profiling of delirium (daytime, nighttime): The investigators expect hypoactive delirious states to be characterized by decreased activity levels in comparison to non-delirium. Instead, increased activity levels during hyperactive delirious states in comparison to non-delirium are expected. Activity levels of the mixed delirium type are anticipated to range between level of the hyper- and the hypoactive type.
2. Within-person level - forerunners and repercussions (daytime):

• Hypoactive type: The time span preceding and following a delirium during daytime will be characterized by increased activity and restlessness levels compared to delirium-unrelated time periods of an individuals during daytime.
• Hyperactive type: The time span preceding a delirium during daytime will be characterized by increased activity and restlessness levels compared to delirium-unrelated time periods of an individuals during daytime.

1. Within-person level - forerunners and repercussions (nighttime): The time span preceding and following the delirium will be characterized by increased activity levels during nighttime (indicated by, e.g., increased mean heart rate and movement, frequency, and quantity of transitions between resting/activity), reduced sleep time / efficiency, and less minutes spent resting compared to delirium-unrelated time periods of an individual during nighttime.

However, research that carefully distinguishes time-dependent within- and between-person level predictors and their interplay through leveraging a wearable sensor is currently lacking. Also, to date, no algorithm that allows detection of deliria at an early stage has been developed. Given the novelty of this line of research, this study is mainly exploratory in nature.

Primary and secondary endpoints Endpoints Endpoints concern the sensor-based assessments (Fitbit) as well as standardized measures of delirium, such the DOS/CAM assessment (for details see Delirium-related endpoints), measured between the day of the baseline assessment and the day of discharge of the study participants.

Sensor-based endpoints:

Primary endpoints concern sensor-based in real-time indicators and their patterns related to activity and sleep assessed with the Fitbit sensor, which measures a broad range of indicators relevant to the present study, including but not limited to the example indicators listed below.

Activity- and sleep parameters, e.g.:

• heart (resting) rate, heart rate variability
• oxygen saturation, breathing rate
• skin temperature
• total sleep time, time spend in different sleep stages
• heart (resting) rate, heart rate variability
• sleep latency
• time spent in light
• 'sleep efficiency'
• step count, minutes spent in activity intensity levels (high, moderate, low)

Circadian rhythm-related indices, e.g.:

• Quantity of rest-activity transitions
• Least active 5-hours in a 24hr cycle
• Most active 10-hour period in a 24hr cycle
• Relative amplitude

Delirium-related endpoints:

Confusion Assessment Method Further endpoints concern standardized instruments for the assessment of deliria. A widely-established instrument to diagnose a delirium during an inpatient stay is the Confusion Assessment Method (CAM) which enable health care professionals to identify delirium quickly and accurately in both, clinical and research settings. The CAM is a four-point scale with the following criteria (short form): (1) acute onset and/or fluctuating course, (2) attentional disturbance, (3) formal thought disorder, (4) altered level of consciousness. All items are dichotomously scored as absent or present.

Delirium Observation Screening Scale Standard screening for delirium at the University Hospital Zurich and the Klinik Hirslanden Zurich comprises the application of a Delirium Observation Screening Scale (DOS) designed to capture early delirium symptoms, three times a day by trained nurses. If an upcoming delirium will be suspected (based on the specific DOS-value of 3 or more), it will be further assessed by CAM.

Demographic variables Non-pharmacologic factors that may influence the occurrence of delirium are increasing age, male gender, ICU or surgery interventions, transfer and changes in the setting of the patient and clinical management seems to be influenced by the subtypes of delirium.

Medical risk factors for delirium In addition, well known medical risk factors for a delirium will be collected in order to conduct a descriptive analysis. These data will be obtained from patient chart reviews of the clinical information system KISIM, namely selected variables of medical history and existing diseases of the participating patients and the medication at the time of admission, laboratory values (blood collection - clinical chemistry), and further risk-factors for delirium, e.g.: serum urea, as well as the diseases and medication at the time of discharge. The selected variables are listed in the CRF.

Sample size estimation Given the exploratory nature of the planned study, a precise sample size estimation is not possible. The aim of this study is to include 10 patients that become delirious during their USZ / Hirslanden stay as well as at least 10 patients developing no delirium during their hospital stay. Based on the data collected to date, we estimate that there is a 25% chance that new participants will develop delirium during the course of the study. Therefore, we are seeking approval to increase the maximum number of participants in the study to 72. In addition, enrollment will be stopped once we have documented 10 confirmed and analyzable cases of delirium, even if this occurs before the proposed cap of 72 participants is reached. "Analyzable" means that the participant's data are adequate for evaluation. For example, data would be considered inadequate if a participant experienced delirium but rarely or never wore the Fitbit sensor during the relevant time period.