Study design and protocol

Elective post-surgical patients with a planned post-operative admission to the Oxford Churchill Hospital ICU were considered for inclusion. We aimed to recruit 25 patients. Eligible patients were recruited from the pre-operative assessment clinics servicing the following specialities: maxillofacial surgery, gastrointestinal surgery, hepatobiliary surgery, renal transplant, pancreatic transplant, urology, and gynaecology. Candidate patients were screened by members of the clinical team against inclusion/exclusion criteria, and eligible patients were approached for informed consent. The exclusion criteria in Table ​Table44 were considered.

Participant exclusion criteria.

Recording was initiated as soon as practically possible after admission to the ICU following surgery. Each patient was monitored continuously for up to 48 h, or until discharge from the ICU. For the duration of the study, concurrent video and physiological data collected by the patient monitor system used as part of routine care were recorded. Equipment was chosen such that it would not affect the experience of the patient or clinical staff. Vital sign estimates from the non-contact monitoring software were computed retrospectively and were not available to clinical staff during the study.

A privacy blind (an opaque sheet of plastic attached to a hinged fitting of the video camera) was included. This blind could be used to obscure the view of the patient from the video camera and thus prevent the recording during dignity-compromising situations (such as routine hygiene). Patients could ask for the blind to be drawn at any time. Patients, their visitors, or staff members could also ask for the equipment to be turned off for any reason throughout the study.

Select clinical information was documented by clinical research staff on paper-based case report forms: patient demographics, admission category, recording start and end times, APACHE II score⁶¹, Richmond Agitation-Sedation Scale (RASS), Fitzpatrick score, and free-text clinical notes. The Fitzpatrick Scale is a numerical skin classification system, from Type I (pale white, always burns, never tans) to Type VI (darkest brown, never burns, never tans)⁴⁹.

The monitoring equipment is shown in Fig. ​Fig.6a.6a. As per the standard of care, vital-sign monitoring was provided using a IntelliVue^® MP70 patient monitor (Philips, Eindhoven, Netherlands). Using proprietary software, reference values of HR and RR were derived at 1-s increments. HR was derived from the single-lead ECG (sampled at 250 Hz), while RR was derived from the bipolar IP signal collected (at 62.5 Hz) using the same set of electrodes. Pulse rate was obtained from the photoplethysmogram (PPG) signal acquired by a pulse oximetry Philips ^® M1941A finger probe (Philips, Eindhoven, Netherlands) and sampled at 125 Hz.

a Patient monitor at the bedside and the non-contact vital-sign monitoring equipment at the end a patient’s bed. b Set-up of the trolley containing the non-contact vital-sign monitoring equipment. The components are shown in their respective positions—(A) optical camera; (B) infrared LED source; (C) thermal camera; (D) interface equipment; (E) workstation (in cabinet); (F) privacy blind (folded up).

Patients were monitored using a 3.2 megapixel Point Grey Research Grasshopper3^® GS3-U3-32S4M-C monochrome camera (FLIR Instruments, Oregon, USA). A non-polarised lens with a 12.5 mm fixed focal length and an aperture range of f/1.4-16 was attached (Goyo Optical^® GMHR412514MCN, Saitama, Japan), as well as an infrared-pass filter (Schneider Kreuznach^® 098, Germany). The imaging device was set to acquire raw uncompressed 8-bit image data at a resolution of 1024 × 768 pixels and a nominal sampling rate of 100 Hz. Illumination was provided using a Metaphase^® FL201-IRN-24 850 nm infrared illuminator (Metaphase, Bristol, Pennsilvania, USA). Since 850 nm is outside the visible range, the illuminator did not emit any noticeable glow light. The infrared illuminator powered by a Metaphase^® ULC-2 LED controller (Metaphase, Bristol, Pennsylvania, USA) connected to a Gardasoft^® CC320 sequence controller (Gardasoft, Cambridge, UK) was strobed at 50 Hz to illuminate alternate video frames.

The equipment was designed to be non-disruptive to standard care. A medical-grade hospital trolley was used as the basis of the non-contact monitoring system. Patients in the ICU are accommodated on any available bed, and so equipment needed to be mobile rather than installed at one particular bed space. The video camera devices and the infrared illuminator were mounted on a hospital trolley positioned at the end of the patient’s bed, as shown in Fig. ​Fig.6a.6a. The trolley was easily movable so it would not impede access to the patient. As is usual practice, patients had access to a chair beside the bed, on which they would occasionally sit. Data were captured for both locations.

All recorded video and physiological data were relayed to an 8 TB Hewlett-Packard (HP) workstation. The workstation was replaced every 12 h with an identical one, due to the large storage requirements (approximately 6.7 GB per minute for the digital video data). All video data were encrypted using industry standard methods.