Does frailty predict postoperative outcomes in geriatric patients receiving surgery for colorectal cancer? A systematic review and meta-analysis

Background Surgery remains the mainstay of colorectal cancer (CRC) and substantially reduces cancer-related morbidity and mortality. Preoperative assessment for frailty in geriatric patients is critically important in risk stratication and clinical decision-making. In this systematic review and meta-analysis, we aimed to quantitatively summarise the effect of frailty on postoperative outcomes in geriatric patients receiving surgery for CRC. Methods A systematic literature search was conducted in MEDLINE, Cochrane and EMBASE from inception to 30 April 2020. Fully published articles reporting risk estimate(s) of frailty on postoperative complication(s), readmission and/ or mortality in patients aged ≥ 65 years who received surgery for CRC were eligible for qualitative and quantitative analyses. Results Across 10 articles of 9 unique studies (n = 69332) that were eventually included in the systematic review and meta-analysis, overall prevalence of frailty was 23.0% (95% CI: 11–43%, I 2 = 100%). Odds ratios (ORs) on overall and severe postoperative complications were respectively increased by 2.36- (95% CI: 1.66–3.35, P <0.01; I 2 = 12%) and 2.35-fold (95% CI: 1.30–4.27, P <0.01; I 2 = 72%) in frail patients compared to non-frail counterparts. On pooled analysis, frailty was signicantly associated with an increased risk of postoperative readmission (OR:1.91; 95% CI: 1.35–2.70, P <0.01; I 2 = 6%). Whilst a signicantly higher risk of frailty on mortality during 12 months after CRC surgery was observed (OR: 5.52; 95% CI:4.40–6.92, P <0.01; I 2 = 89%), the summary OR on 30-day/ inpatient mortality crossed the null line (OR: 1.65; 95% CI: 0.56–4.93, P = 0.37; I 2 = 55%). Funnel plot and Duval-Tweedie’s trim and ll test did not reveal signicant publication bias. Conclusions In the Systematic Reviews and Meta-analyses; MOOSE: the Meta-Analysis of observational studies in Epidemiology; HR: hazard ratio; MeSH: Medical Subject Headings; ASA: the American Society of Anesthesiologists; NOS: Newcastle-Ottawa Scale; NS: non-signicant; LOS: length of stay; identication of Seniors at Risk for Hospitalised Patients; IL: TNF: tumour necrosis factor; Colorectal Cancer Aging Frailty GFI: SIOG: the Society of GA: geriatric assessment; survival.


Background
In 2012, approximately 700,000 deaths from colorectal cancer (CRC) were recorded, making CRC the fourth leading cause of cancer death worldwide [1]. The incidence of CRC dramatically rises up to 1.8 million new cases annually due to a rapidly ageing population, shift in dietary patterns and reduction in physical activity [2]. The risk of CRC increases sharply with age. Individuals aged >60 years have an over 50-time higher risk of CRC compared to those younger than 40 years, contributing to 80% of cases diagnosed [3]. Comprehensive geriatric assessment is currently regarded as the gold standard to globally evaluate older adults' health status; however this assessment may not be feasible due to time constraints and lack of expertise. Screening for frailty is more practicable in routine clinical practice [4]. Given that surgery remains the mainstay of CRC and effectively decreases distant metastasis, local recurrence and cancer-related mortality, accurate evaluation on general health status and identi cation of frailty preoperatively is of great importance in clinical decision-making [5]. Previously a meta-analysis has demonstrated comorbidity is associated with higher risks of overall and cancer-speci c mortality [6].
However, the impact on postoperative outcomes in elderly patients with CRC has never been quantitatively examined by using formal frailty assessment instrument(s) [7]. Also, the underlying mechanisms by which frailty confer a poorer prognosis need to be determined yet [8][9][10]. In this systematic review and meta-analysis, we aimed to summarize current evidence and quantitatively evaluate the effect of frailty on postoperative outcomes in geriatric patients receiving surgery for CRC.

Methods
Research question and eligibility criteria for literature search The systematic review and meta-analysis was conducted following the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guidelines and the Meta-Analysis of observational studies in Epidemiology (MOOSE) guidelines [11,12] (PRISMA checklist available in Additional le 1: Table S1). The study focused on the following research question: Does frailty predict clinical outcomes in patients who received surgery for CRC? Studies were included in the systematic review and meta-analysis when they met the following criteria: 1) Patients received surgery for CRC either for curative or palliative purpose; 2) Patients were aged 65 years or above; 3) Studies were conducted as case-control, cohort or nested design; 4) Formal frailty assessment instruments were utilized and assessed preoperatively; 5) Risk estimates on clinical outcomes including postoperative complication(s), readmission and/or mortality compared between frail and non-frail patients were expressed as either hazard ratio (HR) or odds ratio (OR); 6) Studies were published as full articles; 7) Studies were conducted in human population; 7) Articles were published in English. Studies were excluded if 1) Patients with CRC did not receive surgical therapy; 2) No formal frailty assessment was available in the articles; 3) Risk estimates compared between frail and non-frail patients were unavailable in the articles; 4) Studies were published as conference abstracts; 5) Studies were conducted in animals; 6) Articles were written in language(s) other than English.

Search strategy
Literature search through MEDLINE, Cochrane and EMBASE databases was systematically performed from inception to 30 Apr 2020 to identify studies that compared the risk estimates on clinical outcomes between frail and non-frail patients who received surgery for CRC. The following key Medical Subject Headings (MeSH) terms were used for literature search: 'frailty' OR 'frail' AND 'colorectal carcinoma' OR 'colorectal cancer' OR 'colorectal malignancy' OR 'bowel surgery' OR 'colectomy' OR 'colorectomy' OR 'colon resection' OR 'colorectal resection'. The identi ed records were independently scanned and assessed by two research investigators (XY & CX) to identify the eligible studies. Discrepancies between the two researchers were discussed and resolved by two senior researchers (JW & LY).

Data extraction
Information on study population, study design, frailty assessment instruments, prevalence rate of frailty, comorbidities, the American Society of Anesthesiologists Physical classi cation System (ASA classi cation), prevalence rates of postoperative complications, length of in-hospital stay, risk estimates on clinical outcomes including, postoperative complication(s) and mortality was extracted from the eligible studies. The severity of complications was graded according to prede ned criteria [13,14]. When necessary, corresponding authors of the publications were contacted for either clari cation of data redundancy or data recalculation for associated risk estimates.

Assessment of publication bias
Quality of study assessment was conducted using Newcastle-Ottawa Scale (NOS) for cohort studies for assessing risk of bias by two researchers (XY & CX) independently [15]. High-quality studies were graded when 7 or above out of 9 points were obtained using NOS. Discrepancies between the two authors were resolved by two senior researchers (JW & LY).

Statistical analysis
Meta-analyses were conducted using the inverse of variance method and random-effects models by assuming that the true value of the effect size of each study is sampled from a probability distribution rather than identical. Data from the same study and population were calculated only once based on the most recent available publication and/ or the largest sample size to avoid data redundancy [16,17]. Forest plots were used to present the pooled estimates in graphical form. P values lower than 0.05 were regarded as statistically signi cant. Low heterogeneity was de ned by using the Cochrane Q test with <25% of I 2 , while moderate and high heterogeneity were de ned as <50% and ≥50%, respectively [18]. Egger's linear regression, Begg's rank correlation test and leave-one-out sensitivity analysis were further conducted to identify potential reason(s) for high heterogeneity (i.e. I 2 ≥50%) [19,20].
Risk estimates from the models that adjusted for the maximum numbers of covariates were selected for data-analysis, where alternative adjusted estimates were available. When multivariate risk estimate was unavailable, univariate risk estimate was used instead. Indirect methods of extracting estimates were used, where risk effect was not directly reported in the publication [16,21,22]. The OR and 95% con dence interval (CI) were calculated. Funnel plot and Duval-Tweedie's trim and ll test were used to visually ad quantitatively assess publication bias. Statistical analyses were performed by Review Manager Version 5.3 (Cochrane IMS, Copenhagen, Denmark) and R Version 4.0.0 (R Foundation for statistical Computing).

Study selection
The initial search identi ed 2035 records, of which 1035, 165 and 835 were collected from MEDLINE, Cochrane and EMBASE, respectively ( Figure 1). After 2025 records were excluded due to duplicates, reviews, conference abstracts and irrelevant contents, 10 articles of 9 unique studies were eventually included in the systematic review and meta-analysis.

Study characteristics
A total of 69332 patients (males: 50.9% (35240/69249) reported in 9 unique studies) who received surgery for CRC were eventually included in the systematic review and meta-analysis (Table 1). Seven articles of 6 studies [16,17,[22][23][24][25][26] presented a prospective observational design reporting longitudinal data, while data from the remaining 3 studies were retrospectively analyzed [8,21,27]. Study characteristics including population, study design, sample size frailty assessment instrument and risk estimates were summarized in Table 1.

Study quality and Publication bias
Assessment of study quality yielded the 10 articles were all scored ≥7 out of 9 points using NOS (Additional le 1: Table S2). On estimating prevalence of frailty, 66.7% (6/9) of the included studies fell outside the 95% CI of funnel plot control limits. We therefore proceeded to performing trim and ll test and found publication bias was trivial (adjusted prevalence of frailty: 34.3%; 95% CI: 20.58-51.27, P <0.01; I 2 = 99.8%) (Additional le 1: Figure S1). Egger's linear regression or Begg's rank correlation test did not reveal strong evidence of publication bias in prevalence of frailty (Egger's test: P = 0.431; Begg's test: P =0.532).
Funnel plot, Egger's or Begg's test was not performed in the forest plot for pooled risk estimate of frailty on severe complications, 30-day/ inpatient or 1-year mortality due to small number of studies. When leave-one-out sensitivity analysis was performed, the overall prevalence of frailty did not signi cantly differ, indicating an acceptable robustness of pooled estimate (Additional le 1: Table S3). Nevertheless, the heterogeneity identi ed in OR of frailty on severe postoperative complications, 30-day/ inpatient mortality and mortality within one year was signi cantly diminished by the leave-one-out approach (Additional le 1: Tables S4-S6).

Discussion
Impact of frailty on postoperative outcomes in patients receiving surgery for CRC To our knowledge, this study was rst to systematically synthesize quantitative evidence and evaluate the effects of frailty on postoperative outcomes in patients who received surgery for CRC. The results showed that frailty negatively impacted postoperative complications by ~2.35-fold. Slightly higher ORs were observed in patients receiving elective surgery. Moreover, frailty was associated with 1.91-and 6.16fold increased risks of postoperative readmission and mortality during 12 months, whereas the effect of frailty on 30-day/ inpatient mortality reached statistical signi cance in patients who underwent elective surgery only.
Studies previously reported frailty was an independent factor in predicting postoperative complications, readmission and mortality in abdominal and cardiothoracic surgeries in older adults [28,29]. Because aggressive treatment (e.g. curative surgery) contributes to dysregulation of systemic immune system, causing progressive organ damage and exacerbating decline in physiological reserve [21], identifying frail patients in elderly population is essentially crucial in decision-making and therapeutic options. Moreover, postoperative outcome(s) leading to prolonged inpatient stay incurs substantial additional costs of medical care, and at the same time leading to poorer quality of life and increased risk of emergency readmission [30].
Several mechanisms potentially explain the association between frailty and postoperative adverse outcomes in patients undergoing surgery for CRC. First, frailty characterised by vulnerability and impaired physiological preserve is accompanied by up-regulation of in ammatory and proin ammatory cytokines (e.g. interleukin (IL)-6, C-reaction protein (CRP), tumour necrosis factor (TNF)-α, etc.), leading to the development of systemic in ammatory response [31]. This positive association between in ammatory markers and postoperative complications are particularly evident in occurrence of sepsis, anastomotic leakage and pulmonary insu ciency [32].
Second, frailty is frequently accompanied by the coexistence of multi-morbidity, being more frequent with increasing degrees of frailty [33]. However, while preoperative comorbidities and ASA classi cation are widely accepted as perioperative risk strati cation tools and associated with higher risks of complications and mortality [34], the predictive value of frailty assessment instruments in conjunction with co-morbidity or perioperative complications on mortality is still limited in literature [35].
Third, increased risks of anemia, hypoalbuminemia and poor nutritional status associated with frailty have been well documented [36][37][38]. Condition of being anemic or malnourished directly prolonged postoperative recovery due to surgical complications (e.g. dehiscence, anastomotic leakage, surgical site infection, etc.) in colorectal and other major abdominal surgeries [39][40][41]. Nevertheless, the risk estimate on major surgical complications(s) such as anastomotic leakage and/or abscess formation crossed the line of null effects summarized from 4 studies in this study. There are many factors thought to affect anastomotic leakage following colorectal surgery. Evidence has showed general factors (e.g. advanced age, obesity, malnutrition, corticosteroid use, etc.), local factors (e.g. vascular ow insu ciency, infection, etc.) as well as technical and experience factors (e.g. tumour location, operative modalities (i.e. laparoscopic vs. open surgery), anastomotic tension, etc.) all in uence the integrity of conduit and anastomotic healing [39,42,43]. Subanalyses on tumour site, operation modality, etc. were not performed in this study due to data insu ciency. The association between frailty and postoperative anastomotic leakage is yet to be investigated in future studies.
Fourth, reduction in muscle mass and mitochondrial enzyme activity exacerbated by curative operation in turn activates muscle in ammation and predisposes to frailty process [44]. Although pre-habilitation training effectively reduced morbidity and mortality after intra-abdominal cancer surgery [45], its role in improving postoperative muscle strength and functional capacity is yet to be determined. Furthermore, the degree of frailty severity uctuating over time is rarely reported in CRC patients.
Fifth, altered composition of commensal microbiota, if present, potentiates the development of frailty and bowel oncogenicity [46,47]. Although causal evidence showing interlinks between changes in gut microbial communities, frailty process and CRC formation is scarce, long-term alteration in gut microbiota after colorectal surgery may disturb skeletal muscle protein synthesis, resulting in muscle atrophy and attenuation [48].
Frailty assessment instruments and prevalence of frailty in patients receiving surgery for CRC Frailty generally affects 12.8% of older individuals aged ≥60 years and is more prevalent in cancer patients [49][50][51]. In this systematic review and meta-analysis, we reported that preoperative prevalence rate of frailty in CRC patients aged ≥65 years was 26.0% (95% CI: 16-38% ranging from 4.4-52.2%) in 9 unique studies, with a very high level of heterogeneity identi ed. While dozens of frailty assessment instruments are used worldwide [52], no standard frailty index is speci cally validated for CRC patients in predicting postoperative prognosis. Pandit et colleagues [21] recently proposed a modi ed frailty index (Colon Cancer Frailty Index (CCFI) in conjunction of previously validated Canadian Study of Health and Aging Frailty Index (CSHA-FI) and two additional variables (i.e. anemia and weight loss), guiding the strategy on treatment options and complications [21,53].
Another indicator of frailty is the cumulative de cit index, using clinical and/or laboratory variables and expressing the index as the number of abnormal variables over the total number of variables tested [54]. The predictive value of this index in evaluating complications and mortality reached ~0.70 after gastrointestinal tract surgery [55]. However, this index is more readily available for large-scale epidemiological data than in routine clinical practice. In contrast, magnitude of impairment of physiological reserve by objective/ semi-objective measures (i.e. muscle strength, gait speed, energy expenditure, etc.) is assessed in the Fried frailty phenotype. In this meta-analysis, only one study used Fried frailty phenotype [56]. More frequently, other frailty assessment instruments (e.g. Groningen Frailty Index (GFI), the John Hopkins' Adjusted Clinical Groups (ACG) de nition of frailty, etc.) with multidimensional aspects (i.e. physical, functional and psychosocial domains) are currently used in CRC researches, in line with recommendations of the International Society of Geriatric Oncology (SIOG) suggesting that geriatric assessment (GA) incorporated with some initial screening tools (e.g. G8, VES-13, etc.) be assessed in CRC patients with ≥65 years of age who are candidates for surgical procedures [57], given that increasing age coupled with coexisting medical conditions and psychosocial issues makes therapeutic decision(s) more challenging.

Limitations of the study
There are several limitations in this systematic review and meta-analysis.
First, distribution of curative and palliative purpose that was only available in three of the included studies made it unable to in particular analyze the effect of frailty on mortality and other outcomes across the subgroups. Among patients with metastatic CRC, majority of the lesions are unresectable, despite that a few of them can still be amenable to curative resection of primary tumour and metastases. Palliative surgery is mainly reserved for these patients, which confers an operative mortality rate at 6-10% [58] and postoperative morbidity between 18-47% [59]. However, some of the studies reported that no signi cant association was observed between tumour stage and postoperative complication(s) [17].
Moreover, in multivariable analysis, frailty status remained to be an independent prognostic indicator after adjustment for TNM stage [16]. Second, only all-cause mortality was considered as a hard endpoint in this study. Non-cancer-speci c mortality has been the main form of competing risk that occurs and leads to biased results when applying survival analysis [60]. Competing risk nomograms has been established for gastrointestinal cancer diseases in the recent years in order that overall survival (OS) and cancer-speci c survival (CSS) in patients with surgically resected tumours can be predicted [61,62]. As far as we know, only two studies previously explored discrimination between physiologically t and nont patients in cancer-related and non-cancer speci c mortality after curative resection for CRC [63,64].
However, neither of them predicted outcomes by utilizing formal frailty assessment instrument(s). Competing risk models including functional and nutritional factors will be critical in classifying geriatric patients according to their frailty status [63].Third, as an important confounder affecting prognosis, emergency surgery was not absolutely eliminated in this study, although we attempted to subanalyze the patients who underwent elective surgery. The positive correlation between emergent surgical presentation and postoperative mortality in elderly patients undergoing CRC surgery was con rmed in previous studies [65,66]. Given that an emergent surgical presentation in a senior patient frequently points at a diagnosis of bowel obstruction or perforation, these two conditions often indicate an advanced stage and carry a poor prognosis [65]. Also, it is unlikely that preoperative assessment and prehabilitation training can be implemented in these patients. Fourth, frailty assessment instruments were not uniformly assessed.
Some studies mostly relied on physical components, while others used multidimensional approach, culminating in large variance in prevalence of frailty. Moreover, risk estimates were extracted directly or calculated indirectly from relatively small numbers of studies. Furthermore, univariate risk estimates were used in some of the studies when multivariate appraisals were unavailable in original publications.

Conclusions
Frailty is highly prevalent in geriatric patients receiving surgery for CRC. In the studies reviewed, frailty appeared to be associated with higher risks of postoperative complications, readmission and mortality during 12 months. Nevertheless, no signi cant association between frailty and 30-day/ inpatient postoperative mortality was observed.

Declarations
Ethics approval and consent to participate Not declared.

Consent for publication
Not applicable.

Availability of data and materials
The datasets analyzed in the study are available from the corresponding author upon reasonable request.

Competing interests
The authors declare they have no competing interests.

Funding
The study was supported by grants from Shanghai Municipal    Figure 1 Flow diagram on study identi cation and selection.   Forest plot for risk estimates of frailty on postoperative readmission Figure 5