This retrospective cohort analysis examines data collected at the Campus Charité Mitte of the Charité – Universitätsmedizin Berlin, Germany, between June 2016 and March 2017. As part of routine pre-surgical assessment, patients undergoing elective surgery were seen at the anaesthesia preoperative clinic of the Department of Anaesthesiology and Intensive Care Medicine. The analysis was approved by the ethical committee (EA1/227/16) of the Charité Universitätsmedizin – Berlin, Berlin, Germany (Chairperson Prof. R. Uebelhack), on August 8th, 2016. Due to the retrospective nature, the requirement for written informed consent was waived by the ethics committee. The trial has been registered retrospectively at ClinicalTrials.gov (NCT03382054).
During the implementation period of this routine assessment, patients ≥65 years of age were offered a frailty assessment either at the preoperative anaesthesia clinic or on the peripheral wards. Surgical disciplines involved included general/gastrointestinal, orthopaedic, oral and maxillofacial surgery, as well as urology, gynaecology, otorhinolaryngology, and dermatology. This analysis does not include patients with emergency procedures or procedures without anaesthesia contribution or operation. Patients unable, unwilling, or unavailable to undergo the frailty assessment (patient refusal, language barrier, insufficient data, patient not found in room or unavailable due to other tests/examinations) were not recorded. Patients with multiple assessments, cancelled operation or cardiac surgery were excluded. Ultimately, one additional medical assistant position was required to establish a routine frailty assessment. This assistant, as well as two nurses from the preoperative anaesthesia clinic (as substitutes during vacation or illness), were trained in the frailty assessment (see Table 1) by a senior physician-scientist responsible for quality management (OB). The first several assessments were performed under supervision by the trainer, so as to corroborate understanding and quality. Training for the 5-point frailty criteria was deemed simple and required little training. The screening was done electronically via our hospital program, where all patients requiring anaesthesia must be registered. The assistant screened registered patients for inclusion criteria, and assessed eligible patients visiting the preoperative anaesthesia clinic prior to the visit with the physician. Patients were taken by the assistant to a designated room, which included the necessary equipment and dimensions for the frailty assessment (i.e. paper-based questionnaire, hand dynamometer, stopwatch, and > 5 m available for walking, with appropriate markings on the floor). The results were placed in the patient file and the patients returned to the waiting room. This assistant was also responsible for visiting the peripheral wards to assess the patients not visiting the clinic. After noting the name, station and room number of a registered patient, the assistant would take the necessary equipment in a “frailty bag”, which included the aforementioned equipment as well as measuring tape and small cones to mark distances. The assessment took place at the bedside and the walking test at the nearest hallway. After the assessment, the results were placed in the patient file and the assistant returned to the station. If no eligible patients were present, this assistant supported the remaining staff with the normal preoperative clinic program. Overall, the equipment required was durable and inexpensive. The workload was deemed low, with an average frailty assessment time of under 10 min and an average of 7–8 eligible patients per day.
General patient information was gathered, including age, sex, height, weight, smoking status, polypharmacy (routine intake of > 5 medication), American Society of Anesthesiologists Physical Status (ASA PS) classification, as well as comorbidities assessed by the Charlson Comorbidity Index (CCI) [15], surgical discipline, and preoperative creatinine levels. The surgical risk was classified according to European Society of Cardiology (ESC)/European Society of Anaesthesiology (ESA) guidelines on non-cardiac surgery into low, medium, or high risk [16]. Diagnoses for the entire hospitalization period and comorbidities were derived from our hospital database according to the International Statistical Classification of Diseases and Related Health Problems (ICD-10).
For the analysis, patients were classified into three groups according to the number of preoperative pathological frailty criteria described by Fried (0–5 criteria, see Table 1), consisting of non-frail (0 criteria, reference group), pre-frail (1–2 positive criteria), and frail (3–5 positive criteria) groups. Slight modifications were made to Fried’s frailty assessment in an attempt to adapt and improve data collection according to European standards, as summarized in a previous publication [17]. This included estimating weight loss in kilograms instead of pounds, and using a cut-off of ≥5 kg instead of ≥10 pounds (ca. 4.5 kg). In addition, metabolic equivalent tasks (METs) [18] were used instead of kilocalories/week (kCal/w). According to Fried, it is important to classify physical activity into low, moderate, and high levels, whereas a low level of activity in kCal/w is cited as a pathological criterion [7]. METs offer a different unit to evaluate physical activity, can also be classified into low, moderate, and high levels, and have the advantage of being faster and easier to use in clinical practice. Fried has defined physical activity in terms of METs [19], whereas a MET under 3 was considered low (and therefore pathological). As suggested by Fried, patients with > 2 missing criteria were removed from the analysis [7].
The primary outcome was the incidence and type of postoperative complications, which was selected in accordance with the Veteran Affairs’ National Surgical Quality Improvement Program (NSQIP) [20, 21] for purposes of comparability. Their standardized list of complications included pneumonia, pulmonary embolism, acute kidney injury, cerebrovascular accident, coma, superficial and deep wound surgical site infections, urinary tract infection, sepsis, deep vein thrombosis, reoperation, and reintubation due to respiratory/cardiac failure, myocardial infarction, cardiac arrest, and death. Although frailty assessments were performed by a trained staff assistant, outcome parameters were documented by healthcare documentation specialists into the hospital databank, who were not affiliated with his study. The hospital diagnoses were examined retrospectively by the authors for the presence or absence of ICD-10 codes corresponding to the NSQIP complications.
Although the frailty status of the patients were documented in the physical patient file, it was not noted in the electronic file nor the premedication records due to a missing interface, and no specific recommendations were made for the treatment of frail patients (minimizing performance bias). Outcome parameters were obtained from our hospital database, which were neither assessed nor documented by the frailty screening staff (minimizing measurement bias).
The evaluation of the data was carried out in an explorative approach. All data collected during the implementation period of the routine assessment (between June 2016 and March 2017) were available and were analysed considering the exclusion criteria. Due to the retrospective nature of this analysis, a sample size calculation was performed post-hoc, showing that 788 patients would be required to evaluate a difference between two groups (healthy vs pre-frail/frail) with a confidence of 80 and 5% alpha. Descriptive analyses and statistical testing were performed using the R Project of Statistical Computing, version 3.3.1. When normal distributions were ruled out using the Kolmogorov-Smirnov test, results were given as medians and interquartile ranges (IQR), otherwise as mean ± standard deviation (SD). Binary and ordinal variables were expressed by numbers with percentages. Differences in binary and ordinal variables between two independent groups were analysed by the exact chi-square test. In metric, non-normally distributed variables, differences between two independent groups were assessed with the Mann-Whitney-U-test and in ≥3 independent groups using the Kruskal-Wallis test. In metric, normally distributed variables, differences between groups were assessed using Student’s t-tests.
We removed the effect of baseline confounder variables by pairwise next neighbour matching (1:1:1). This includes a propensity score creation and next neighbour matching for the first and second group, followed by an additional propensity score creation and matching for the second and third groups, with group order 0 (non-frail), 1 (pre-frail) and 2 (frail). Propensity score matching was performed using the R package “MatchIt” version 3.0.2, based on Ho et al. [22]. The following baseline characteristics were included, as there were considered to be major confounders: age, sex, body mass index, ASA PS, surgical risk, type of anaesthesia, CCI, surgical discipline, smoking status, polypharmacy, as well as preoperative creatinine levels and glomerular filtration rates (GFR) as surrogates for chronic kidney injury. Additionally, the following comorbidities were also included: coronary artery disease, peripheral artery disease, diabetes mellitus, liver disease, tumour, cardiac failure, cerebrovascular accident, asthma bronchiale, and chronic obstructive pulmonary disease. Baseline characteristics that remained statistically significant after propensity score matching were included in a subsequent logistic regression model with frailty status as further explanatory variable. Since propensity score matching presents a method of regression analysis itself, subjecting variables that have already been successfully controlled in propensity score matching (i.e. p < 0.05) to a subsequent logistic regression would not improve the analysis. The regression’s target variable was compound complications. A two-tailed p-value < 0.05 was considered statistically significant. All tests should be understood as constituting explorative analysis; no adjustment for multiple testing has been made.
