Study design and participants

This prospective study was conducted at the Geriatric Outpatient Clinic of the Radboudumc in Nijmegen, the Netherlands, between February 2023 and January 2024. We recruited older (≥ 70 years) participants through flyers, advertisements, and registries of previous studies at the geriatric department with consent to be contacted for future research. Based on a screening laboratory visit, participants were enrolled in two groups: participants with OH and OH-related symptoms (“OH group”) and participants without OH and without severe OH-related symptoms in daily life or symptoms (“no OH group”). OH was defined as a drop in systolic BP of ≥ 20 mmHg and/or diastolic BP of ≥ 10 mmHg between 1 and 3 min after standing up using a 5-s moving average window [25], compared to the supine BP (1 min to 30 s before standing), measured with a continuous BP device (Finapres NOVA, Finapres Medical Systems, Enschede, the Netherlands), during at least one supine-stand transition. The presence of daily life symptoms was defined as reporting one or more OH-related symptoms, according to the OH symptom assessment (OHSA), and any impairment in daily life, according to the OH daily activity scale (OHDAS), defined severe symptoms [26]. Other inclusion criteria were the ability to provide informed consent and understand oral instructions and a functional ambulation category (FAC) score of at least 4 [27]. Participants were excluded when physically unable to perform supine-stand transitions, having moderate-to-severe dementia (clinical dementia rating ≥ 2 or Montreal Cognitive Assessment (MoCA) < 12), severely frail (clinical frailty scale ≥ 7) [28], or participating in an intervention study. The local ethics committee (CMO Radboudumc) concluded the study did not fall within the scope of the Medical Research Involving Human Subjects Act (WMO), thereby exempting the need for review by a central ethical committee. All participants signed written informed consent. The study was performed in accordance with the Declaration of Helsinki.

Data collectionScreening laboratory visit

Participants were asked to complete The Older Persons and Informal Caregiver Survey-Short Form (TOPICS-SF) questionnaire about activities of daily living (ADL) and comorbidities [29] and the Technology Experience Profile (TEP) about technology use [30]. The MoCA was completed as a cognitive screening tool [31]. Moreover, information about age, height, weight, medication use (type and number of medications), alcohol use (units per week), smoking habits (yes/no), history of falls in the last year, and OH symptoms (OHSA and OHDAS) was obtained [26]. All participants performed a maximum grip strength, grip work (sustained grip strength), and five-times chair-stand test to indicate physical fitness. Participants performed one supine-stand transition (5 min supine; 3 min standing) while instructed to lie and stand still and perform the transition as fast as possible. Following the enrolment of 17 participants, this protocol was modified by involving 3 supine-stand transitions, to better account for the variability in orthostatic BP responses. BP was measured continuously using volume-clamp photoplethysmography on the digital artery of the left middle finger and intermittently (1 min before and 1 min and 3 min after standing up) using oscillometry on the contralateral brachial artery (Omron M4 Intelli IT, OMRON Healthcare, Kyoto, Japan). The hand wearing the Finapres device was placed in a sling to prevent hydrostatic pressure artifacts. Cerebral oxygenation was measured simultaneously by two NIRS sensors (PortaLite MkII, Artinis Medical Systems, Elst, The Netherlands) attached to the forehead bilaterally, approximately 2 cm above the eyebrows. The sensors consisted of three light-emitting diodes (LEDs) and two detectors, placed at inter-optode distances of 2.9, 3.5, and 4.1 cm (long channels) and 0.70, 0.80, and 0.74 cm (short channels). The NIRS sensor had an embedded inertial measurement unit (IMU) with triaxial accelerometry and gyroscope. Sensors were kept in place and covered by a black bandana to prevent ambient light interference. The control unit was placed in a belt around the waist.

At-home measurements

Measurements were performed at the participants’ homes on two different days, maximally 5 weeks apart. A measurement day started at 9:00 when the researcher visited the participant, and information was obtained about medication changes and falls since the previous visit. The researcher equipped the participant with the NIRS sensor, in the same way as during the screening visit, and a smartphone worn in a waist belt or pocket if possible. The smartphone contained an application (Krane™: the Orikami digital biomarker platform, Orikami, Nijmegen, the Netherlands) that guided participants in performing the supine-stand tests, registered the time stamps, and asked for OH-related symptoms. The NIRS sensors and phone application are depicted in Supplementary Material Fig. S1. Two ActivPALs (PAL technologies, Glasgow, Scotland) were placed on the thigh and rib to measure the participant’s posture (supine, sitting, standing, or stepping). The researcher explained the use of the NIRS sensor and phone application and turned all devices on. During a measurement day, participants performed three supine-stand tests (5 min supine; 3 min standing) spread over the day. The first of these was supervised by the researcher, who left the home afterwards; the remainder of the day was unsupervised, including the second and third repetition of the supine-stand tests. Occurrence of OH-related symptoms throughout the day was recorded in a diary and by pressing the event button on the NIRS device (see Fig. S1), after which the smartphone application asked for the type and severity of the symptoms and during what activity they occurred. At 20:00, participants could remove and turn off the NIRS sensors, ActivPALs, and smartphone. No other interference with the equipment, including charging of the batteries, was required during the day. The next day, the researcher picked up the equipment, and the system usability scale (SUS) was asked to be completed, together with questions for feedback, user comfort, and user-friendliness of the NIRS device and application [32].

Data acquisition and processingScreening laboratory visit

Cerebral oxygenation signals were acquired in OxySoft (version 3.4.12) at a sampling frequency of 100 Hz. BP signals were acquired in Acqknowledge at 200 Hz (version 5.0, BioPac Systems Inc., Goleta, USA). Both acquisition programs were synchronized using analog pulses (PortaSync MkII, Artinis Medical Systems, Elst, the Netherlands). BP and NIRS data were processed in MATLAB (2022a, MathWorks Inc., Natick, USA) using custom-written semiautomatic scripts [33]. Signal quality was assessed visually, and signals of insufficient quality were discarded. For BP, insufficient quality was defined as inability to visually distinguish any peaks and troughs. For cerebral oxygenation, the signal quality index (SQI) was used, rating signal quality between very bad (1) and very good (5) based on the presence of a heartbeat wave [34]. Channels were excluded when the average SQI was below 3. This threshold was set empirically. Sensitivity analyses to evaluate the effect of SQI threshold on the obtained results were performed using thresholds of respectively 3.5, considered good in resting situations [34], and 2, previously used at the neonatal ICU, a setting even more prone to artifacts [35]. In addition, signals were discarded when there was flatlining, a baseline shift larger than 10 µM, or an irregular amplitude. Heart rate, systolic BP (SBP), and diastolic BP (DBP) were obtained over time by peak and trough detection. BP, heart rate, and cerebral oxygenation signals were resampled at 10 Hz and filtered using a 5-s moving average filter [25]. All available long NIRS channels measuring cerebral oxygenation were averaged, while the short channels with an inter-optode distance of 0.80 cm of both sides were averaged when available.

At-home measurements

Cerebral oxygenation signals were acquired offline at a sampling frequency of 100 Hz, stored locally on the NIRS device, and transferred to MATLAB. Cerebral oxygenation signals were resampled at 10 Hz, filtered using a 5-s moving average filter, and linearly detrended to remove slow drift. Measurement starting time was retrieved retrospectively from the offline data. Oxygenation data were used in the analyses when they fulfilled the same criteria as for the screening measurement.

ActivPAL data were stored on the device and offline transferred to Microsoft Excel (Office 16) using PALconnect before processing in MATLAB. ActivPAL and NIRS/IMU data were synchronized by clock times at the start of the measurement. The signal vector magnitude (from now on called “total acceleration”) of the triaxial accelerometer signal was calculated by \(\sqrt^+^+^}\), with acceleration x in the vertical (superior-inferior) direction, acceleration y in the horizontal (left–right) direction, and acceleration z in the horizontal (anterior–posterior) direction [36]. Total acceleration was band-pass filtered (0.1–1.3 Hz cutoff, zero-phase second-order Butterworth filter) to remove high-frequency noise and slow baseline drifts.

Data analysisFeasibility

Average user comfort and SUS scores were calculated for both measurement days separately. The SQI was determined over each supine-stand repetition and each of the entire measurement days.

Supine-stand tests in the laboratory and at home

Exact times of the standardized tests at home were retrieved from the data platform linked to the smartphone application and, if necessary, corrected using NIRS-device accelerometer data. Within these time frames, accelerometer (obtained by NIRS sensor) and cerebral oxygenation data were reviewed, likewise to the accelerometer, cerebral oxygenation, and BP data as measured during the test at the laboratory. Cerebral oxygenation outcome parameters (O2Hb) were determined relative to baseline values defined as the average at 60 to 30 s before standing up: (1) maximum drop amplitude and recovery respectively at (2) 30–40 s (early), (3) 50–60 s (1 min), and (4) 60–170 s (late) after standing up.

Daily activities and daily postural changes

The unsupervised and uncontrolled postural changes were identified via the ActivPAL readouts. The transition was classified as a sit-stand when the leg sensor code changed from sitting to standing or sitting to stepping. In a window of 40 s around the ActivPAL-detected sit-stand transition, the highest peak of the filtered total acceleration was identified and assumed to be the exact sit-stand transition. The cerebral oxygenation and total accelerometer curves were extracted 30 s before the detected postural change and 1 min after the detected postural change and determined relative to the sitting baseline (45 to 15 s before standing). Signals with a minimum–maximum long-channel oxygenation difference larger than 20 µmol/L were discarded as being physiologically improbable.

OH-related symptoms

For each reported event, possible cerebral oxygenation drops and postural changes (IMU data) were identified from the prior 5 to the succeeding 5 min.

Statistical analysis

Statistical analyses were performed in MATLAB (R2022a), RStudio (2022.02.1, R version 4.1.3), and IBM SPSS Statistics 29. All continuous variables are presented as mean (standard deviation) when normally distributed or as median (interquartile range) when distributed otherwise. Categorical variables are presented as number (percentage). We used two-sided testing with a significance level of 0.05 for all analyses.

Three linear mixed models were created for each outcome parameter (maximum drop amplitude, early recovery, 1-min recovery, and late recovery) of the long- and short-channel oxygenation during a supine-stand transition. Model 1 used fixed effects for group (“OH” or “no OH”) and condition (“laboratory” or “at home”) and random effects for participants. Model 2 included only OH participants (random effects) and used fixed effects for symptoms (“yes” or “no”) and condition (“laboratory” or “at home”). Model 3 included only the at-home condition, separated into “day 1” and “day 2,” and the repetitions within a measurement day (“repetition 1,” “repetition 2,” or “repetition 3”), with random effects for participants. In all models, when interaction effects were nonsignificant, the simplest model without interaction effects was used. Reliability was represented by the two-way mixed-effect intraclass correlation coefficient (ICC) determined from all models, expressing the proportion of total variance attributed to the between-subject variability.