This was a prospective cohort study of 85 adults undergoing elective cardiac surgery at a university teaching hospital between April 2020 and May 2021. The study was reported according to the STROBE guidelines (Elm et al. 2008) and registered on the Chinese Clinical Trials Registry (ChiCTR2000031098). Approval for the study was obtained from the Joint Chinese University of Hong Kong – New Territories East Cluster Clinical Research Ethics Committee (CRE no.: 2019.711). All adult patients scheduled for elective cardiac surgery gave written informed consent for the study.

Patients were admitted to the cardiothoracic surgical ward a day before surgery and were admitted to the intensive care unit (ICU) for early postoperative care with later care in a high dependency cardiac ward. Patients who were undergoing elective coronary artery bypass grafting, valve surgery or aortic intervention were included. Patients undergoing emergency cardiac surgery; patients with known musculoskeletal or neurological disorders that were associated with lower limb muscle atrophy (e.g. poliomyelitis, stroke, peripheral neuropathy), or previous major surgery of a lower extremity (e.g. hip replacement, metal fixation, amputation), localised infection, skin disorders, and cognitive impairment (inability to provide consent) were excluded.

Standardised ultrasound examination was performed on all recruited patients one to ten days before surgery. Ultrasound measurements of the RFM (MTRFM, CSARFM and EchoRFM) were performed by a physiotherapist who had certain previous experience of soft tissues ultrasound assessment. Under the guidance of a certified specialist radiologist, the study ultrasound operator was instructed in hands-on ultrasound of the RFM for three sessions (each lasting 60 min) with three patients scanned for this learning exercise. Subsequently, five patient scans were performed under supervision before independent scanning proceeded.

Each participating patient underwent RFM measurements on the enrolment day. The ultrasound technique utilised the B-mode of the HD11 XE ultrasound system (Philips Healthcare, Best) and a linear multi-frequent transducer (5–12 MHz, Philips Healthcare, Best). Participants were positioned lying supine in a relaxed position with both knees supported with a rolled towel in extension (in the natural resting position of 15 degrees) and the toes pointing upwards. Measurements were taken at the halfway point between the greater trochanter of femur and the proximal border of the patella (Perkisas et al. 2021). The transducer was placed perpendicular to the long axis of thigh with ample use of transmission gel to maintain acoustic contact with the skin surface and applying minimal pressure on the thigh soft tissues. The mid-portion of the RFM myofascia was used as the boundary for muscle thickness (Fig. 1A-1 and B-1) and cross-sectional area measurements (Fig. 1A-2 and B-2). Three consecutive measurements were obtained on each leg and the mean MTRFM and CSARFM of both legs combined were used. To address inherent phenotypic variation in muscle mass across different body physiques, measurements of MTRFM and CSARFM were normalised by adjusting for body mass index (BMI) and body surface area (BSA) with normalised values reported for analysis and comparison.

Typical transverse ultrasound images of (A) frail and (B) non-frail participants. (A-1) muscle thickness, (A-2) cross-sectional area and (A-3) echogenicity of rectus femoris muscle of a 62-year-old frail male participant. (B-1) muscle thickness, (B-2) cross-sectional area and (B-3) echogenicity of rectus femoris muscle of a 61-year-old non-frail male participant. F, femur; RFM, rectus femoris muscle; SF, subcutaneous fat; VIM, vastus intermedius muscle

Depth, overall gain, and time-gain compensation settings were kept constant when capturing images for echogenicity measurement. Images were processed with image normalisation, which is an image processing technique that distributes image intensities evenly by setting the maximum and minimum intensity in the image as 0–255 arbitrary units [au] (with background, black = 0 au and text, pure white = 255 au respectively) (Li et al. 2015; Li et al. 2012), before intensity measurement using ImageJ software version 1.52 (National Institute of Health, Bethesda). Echogenicity, i.e. the mean pixel intensity within a given region of interest, was calculated using histogram analysis and expressed in grayscale values from 0 to 255 au. In each image, a ‘Polygon selection’ tool was used to outline a region of interest within the confines of the RFM myofascia (Fig. 1A-3 and B-3). The average value of three echogenicity measurements was used.

Preoperative frailty status was assessed before surgery using CFS and GST5m, both previously used frailty assessment tools in clinical setting (Aucoin et al. 2020; Rockwood et al. 2005; Afilalo et al. 2010; Turner and Clegg 2014; Wilson et al. 2013). CFS was categorised as ‘Non-frail’ (CFS ≤ 4) and ‘Frail’ (CFS > 4) (Rockwood et al. 2005). For GST5m, patients were asked to walk 5 m at a comfortable pace and the walking time recorded (Afilalo et al. 2010). This test was repeated 3 times and the mean time calculated. High-risk status for frailty and poor outcome was defined as taking 6 seconds (s) or more to complete the 5-m distance (Afilalo et al. 2010).

The postoperative recovery outcome measured was days (alive and) at home within 30 days of surgery (DAH30), a patient-centred composite measure that incorporates the postoperative hospital length of stay, discharge destination (rehabilitation centre or nursing home), hospital readmission, and postoperative death (Myles et al. 2017; Moonesinghe et al. 2019). Construct validity has been established in perioperative studies involving cardiac surgical patients with half a day difference in DAH30 considered clinically meaningful (Myles et al. 2017).

Demographic data included age, gender, height, weight, haemoglobin and albumin levels, physical performance (including lower limb strength using the 30-s chair rise test (Rikli and Jones 1999), and total weekly physical activity level using the International Physical Activity Questionnaire (Macfarlane et al. 2007)), predicted mortality using the logistic European System for Cardiac Operative Risk Evaluation (EuroScore) (Roques et al. 2003), details of the surgical procedure, duration of anaesthesia, cardiopulmonary bypass time, ICU admission severity of illness (Acute Physiology and Chronic Health Evaluation III (Knaus et al. 1991)), duration of mechanical ventilation postoperatively, major adverse cardiac and cerebrovascular events, ICU and hospital length of stay, and 30-day mortality.

Based on the preliminary results from the ongoing PREQUEL trial (Yau et al. 2019), 10% of study patients were expected to be frail (i.e. CFS > 4). A sample size of 85 patients provided 80% power to determine whether a correlation coefficient (0.30) between muscle ultrasound findings and frailty differs from zero with a 2-sided α error of 0.050.

Descriptive statistics with mean (SD), or median (IQR) for continuous variables, and count (percentage) for categorical variables were reported. The Shapiro–Wilk test was used to check data for normality. Comparisons of ultrasound measurements between frailty groups were examined using Student’s t-test or Mann–Whitney U test as appropriate. The Chi-squared test was used to compare categorical data between CFS groups. To test the reliability of ultrasound measurements, the intraclass correlation coefficient (ICC) was used to test interrater reliability between the study operator and the experienced radiologist. Repeated ultrasound assessments were performed on five patients on two separate occasions by the two operators during the same day. Spearman’s rho correlation (rs) and Pearson correlation (r) were estimated between CFS and GST5m respectively with ultrasound measurements to determine their relationship. The receiver-operating characteristic analysis was performed to determine and compare the discriminative ability of each ultrasound measurement variable (MTRFM, CSARFM and EchoRFM) to identify frailty using the criteria of CFS > 4 and GST5m ≥ 6 s. Exploratory cut-offs for each ultrasound measurement variable were estimated using the Youden’s index and the corresponding performance measures: sensitivity, specificity, positive and negative likelihood ratios, and area under receiver-operating characteristic curve (AUROC) with associated 95% confidence intervals (95% CI) were reported. The receiver-operating characteristic analysis was also used to determine and compare the discriminative performance of the various frailty measures for the prediction of DAH30.

Finally, the predictive performance of each of the ultrasound RFM measurements in combination with GST5m as an add-on test to another objective test (GST5m) for identifying frailty was assessed, and compared with the two index measures (CFS and GST5m). The ‘both test positive’ rule was used to evaluate if the add-on tests (i.e. combining two objective assessment tools: GST5m ≥ 6 s followed by the ultrasound-derived RFM measures at the threshold for frailty) increased the specificity (Hayen et al. 2010). For the purposes of this study, CFS was considered to be the reference test for frailty as it is extensively used to provide predictive screening for clinical outcomes, including in cardiac surgery and ICU settings (Shimura et al. 2017; Afilalo et al. 2017; Muscedere et al. 2017). McNemar’s test was used to compare the difference in diagnostic yield (proportion of true-positives in the study population), sensitivity (%), and specificity (%) between each add-on test and GST5m alone. The relative positive and negative likelihood ratios were calculated to determine if the add-on tests outperformed the GST5m alone test (Hayen et al. 2010). The performance of CFS, GST5m, and add-on test to predict DAH30 was also estimated. Using quantile regression with robust standard errors (Staffa et al. 2019), the changes in DAH30 distribution from 10 to 90th percentiles between CFS, GST5m and the best add-on test frailty measure across frailty groups were described after adjusting for age, sex and logistic EuroScore. Calibration belts (Nattino et al. 2014) were drawn to assess the calibration performance of the four Firth logistic regression models of frailty measures (CFS or best add-on test) on DAH30 (binary outcome cut-off at 10th percentile or 50th percentile), adjusting for age, sex and logistic EuroScore. Statistical analyses were performed using SPSS software version 26 (IBM, New York), Stata software version 17 (StataCorp, College Station) and MedCalc software version 20.023 (MedCalc Software, Ostend). The level of significance was set at p < 0.050.