Background
Craniotomy is used to treat a number of intracranial conditions, including brain tumors, arteriovenous malformations, arterial aneurysms, acute and chronic hemorrhage, and a number of congenital conditions [
1]. Planning for surgery requires an informed risk discussion, including the benefits of surgery and the likelihood of complications. In patients undergoing craniotomy, there are limited risk assessment tools to assist clinicians and patients with decision-making. There is a clinical need for a standardized, validated, preoperative risk assessment tool to provide informed consent and aid in patients’ decision-making. As such, pre-operative evaluation inclusive of frailty should be considered part of the process of informed consent for medical intervention. This active discussion and quantifiable risk provision would pertain to outcomes should frail patients decide to proceed with surgical management. Frailty is now an established risk assessment tool in a multitude of surgical specialties [
2‐
5]. Frailty was defined as a clinical syndrome in which three or more of the following criteria were present: unintentional weight loss (10 pounds in the past year), self-reported exhaustion, weakness (grip strength), slow walking speed, and low physical activity [
6]. Measurement of frailty indices and correlation with the likelihood of peri-operative complications may assist clinicians and patients with decision-making during the surgical process [
5].
Two commonly recognized conceptual frameworks for frailty are the phenotypic framework and the deficit accumulation model [
7]. The phenotype framework is based on a group of physical signs and symptoms, including physical characteristics (weight loss, weakness, exhaustion, slowness, and low activity), and is associated with reduced levels of energy [
8]. Examples of instruments using the phenotype framework include the Physical Frailty Phenotype, Frailty/Vigor Assessment, and Clinical Frailty Scale [
9]. The deficit accumulation model is a multiple aggregate domain model that relies on the number rather than the nature of health problems [
10,
11]. A correlation exists between the two constructs of frailty, with studies demonstrating overlap of the two classifications of frailty measurement instruments. Studies have demonstrated both construct and content validity among frailty instruments [
7,
10‐
12]. Of these, the modified Frailty Index-5 derived from the modified Frailty Index-11 and John Hopkins Frailty Instrument, among others, are frequently used. Other systematic reviews and meta-analyses have included measurements of different instruments of frailty to allow for pooled effect estimates [
3,
13,
14].
In the published individual studies on the association of frailty with cranial surgery, patients who are considered frail experience higher rates of complications, operative mortality, and hospital length of stay [
15‐
17]. While frailty is a spectrum, in this study we sought to differentiate frail from non-frail patients. There is a clinical need to summarize the pooled data on the impact of frailty in cranial surgery in a quantitative manner. This would improve the incorporation of frailty into risk assessment modeling in cranial surgery. We sought to address this unmet clinical need by conducting a systematic review and meta-analysis of the association of frailty with perioperative outcomes, including the overall rate of complications within 30 days, perioperative mortality and discharge disposition within 30 days, and length of stay in adult patients undergoing open cranial surgery (craniotomy).
Methods including statistical analysis
Protocol and registration
The protocol for this review was registered prospectively within the International Prospective Register of Systematic Reviews CRD42023405240. This systematic review was conducted in accordance with the methodology for meta-analysis of observational study design [
18]. The present review is being reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) Statement [
19].
Eligibility criteria
We included prospective and retrospective observational studies that reported frailty in patients undergoing open cranial surgery. Patients who were deemed to be not frail were considered the observational control group. We included all patients older than 18 who had undergone a craniotomy. We included patients undergoing craniotomy procedures for any of the following pathologies or locations: brain tumor, benign or malignant, aneurysm surgery, intracranial hemorrhage, or anterior or posterior fossa surgery. We excluded patients undergoing minimally invasive surgery and burr hole surgery.
Previous literature has noted that systematic reviews and meta-analyses should include all available evidence to avoid selection bias and to increase the power of analyses of primary effects by differences in patients and interventions. We considered any tool to measure frailty as eligible. We defined frailty as per the study’s definition as being frail and non-frail. The absence of frailty was identified by studies as the absence of frailty qualifiers [
20]. We pooled the non-frail or lowest frailty score group as the reference group. We included studies published in the English language. We did not introduce a time limit in our eligibility criteria. Studies were excluded if they were not an original research contribution. Studies were also excluded if they were single-arm studies only, systematic reviews, conference presentations, or letters to the editor.
We searched MEDLINE via Ovid SP; EMBASE via Ovid SP; and the Cochrane Library (Cochrane Database of Systematic Reviews and CENTRAL). We searched the grey literature [
21‐
23]. We completed our searches in March 2023. For the search strategy, we combined keyword and subject heading combinations in the predetermined databases (
Supplementary file-Search Strategy) [
18].
References of included studies were searched for any other potentially eligible studies for inclusion.
Data management
Study information was stored and managed using Endnote X9 [
24]. Two reviewers independently screened titles and abstracts for inclusion. When there was disagreement, this was reviewed and determined by the third author. Data were extracted by a single reviewer. Relevant outcome data were collected as presented by the studies. Extracted data were confirmed by a second reviewer. We included articles for full-text review unless both reviewers deemed them irrelevant. Data were extracted using a prespecified extraction form. The results of the data search have been presented in a PRISMA flow diagram. Authors of primary publications were contacted for data clarifications or missing outcome data.
Outcomes and prioritization
We used the Clavien–Dindo model to grade and define perioperative complications [
25,
26]. This classification system has been identified to be reliable and reproducible in the surgical literature (Table
1).
Table 1
Classification of surgical complications
Grade I | Any deviation from the normal postoperative course without the need for pharmacological treatment or surgical, endoscopic, and radiological interventions Allowed therapeutic regimens are: drugs as antiemetics, antipyretics, analgetics, diuretics, electrolytes, and physiotherapy. This grade also includes wound infections opened at the bedside. |
Grade II | Requiring pharmacological treatment with drugs other than such allowed for grade I complications. |
Blood transfusions and total parenteral nutrition are also included. |
Grade III | Requiring surgical, endoscopic, or radiological intervention. |
Grade IIIa | Intervention is not under general anesthesia. |
Grade IIIb | Intervention under general anesthesia. |
Grade IV | Life-threatening complication (including CNS complications)a requiring IC/ICU management. |
Grade IVa | Single organ dysfunction (including dialysis). |
Grade IVb | Multiorgan dysfunction. |
Grade V | Death of a patient. |
Our primary outcome was defined as the overall number of complications experienced by patients within 30 days. We defined the complications according to the Clavien–Dindo classification system. We considered any systemic, neurological, or regional complication as included in the overall complication number (Clavien–Dindo 1–4). This has been defined as any composite score of cardiovascular, pulmonary, neurological, thromboembolic, or infectious complications. We used the weighted composite model to calculate this outcome [
27]. Our secondary outcomes consisted of length of hospital stay, frailty and discharge disposition, hospital readmission within 30 days, and mortality within 30 days.
Measures of association
For dichotomous outcomes, we obtained an odds ratio (OR) from the group with the exposure (frail group) and control (non-frail) group event rates. Dichotomous outcomes included complications in the frail versus non-frail group as defined by the primary outcome, frailty and discharge disposition, hospital re-admission within 30 days, and mortality within 30 days. For continuous data, we obtained the mean difference (MD) with the associated standardized mean difference (SMD). Continuous outcomes included length of hospital stay.
Risk of bias in individual studies
The risk of bias in the included studies was assessed using the Cochrane “Risk of Bias” tool in non-randomized studies [
28]. We included a “risk of bias” table. We generated a “risk of bias summary” [
29].
Data synthesis—planned summary measures and methods of handling and combining data
As different measurement tools were used to assess common outcomes, the results were pooled using random effects meta-analyses to calculate summary estimates using Revman software. We used Review Manager 5.3 Software for statistical analysis. We used the inverse variance weighting summary of continuous outcomes and Mantel–Haenszel methods for dichotomous outcomes [
30]. We generated odds ratios for binary outcomes and standardized mean differences for continuous outcomes. The outcomes were presented with 95% confidence intervals. We used the data presented by the studies, as clinical frailty assessment is typically employed as a risk stratification tool, as opposed to as a part of a multivariable risk model.
We performed a sensitivity analysis, where we excluded studies at the highest risk of bias. As part of sensitivity analysis, we planned to analyze the effects of methodology—primary outcome only in the patients undergoing cranial surgery in studies with prospective methodology only. Where we identified significant heterogeneity, we planned to conduct separate subgroup analyses to explore potential causes of heterogeneity and account for inherent bias due to selection, classification, and confounders among the different studies. For all tests, significance was defined as p < 0.05.
We reported statistical heterogeneity using the Chi
2 statistic and the
I2 statistic. Both were calculated for each of the outcomes listed above. Statistical heterogeneity was declared if the Chi
2 statistic had
P < 0.1. We evaluated the importance of
I2 depending on the magnitude and direction of effects as well as the strength of evidence for heterogeneity [
18]. We determined heterogeneity as not important for
I2 of 0 to 40%, as moderate for
I2 of 30 to 60%, as substantial for
I2 of 50 to 90%, and as considerable for
I2 of 75 to 100%. Publication bias was tested by funnel plots (Metafunnel in STATA) using Egger’s test [
31].
Confidence in cumulative evidence—summary of findings’ tables and Grading of Recommendations, Assessment, Development and Evaluation (GRADE)
The quality of the evidence needs to be appraised to the extent to which one can be confident that the estimates of effect reflect the items assessed. We used the GRADE classification system to rate the quality of the body of evidence across individual outcomes in observational studies [
32,
33]. We generated the “Summary of findings” table using GRADEpro software for observational studies. We constructed a “Summary of findings” table for the primary outcome. The “Summary of findings” table was supported by the Evidence Profile Table [
34]. There are five areas evaluated within the body of evidence: within-study risk of bias (methodological quality), indirectness, heterogeneity of data (inconsistency), imprecision of effect estimates, and risk of publication bias.
Subgroup analysis
As preplanned in the protocol, three subgroup analyses were conducted based on clinical and methodological assumptions. We planned the following subgroup analysis:
1.
Patients undergoing cranial surgery for tumor surgery only.
2.
Patients undergoing cranial surgery for non-tumor surgery only.
3.
Patients older than 65 years undergoing cranial surgery.
Discussion
In this meta-analysis of observational studies in patients undergoing craniotomy, we identified that patients with frailty have increased odds of Clavien–Dindo 1–4 complications. Evidence for this outcome was of moderate quality with very low statistical heterogeneity. This outcome maintained statistical robustness across sensitivity analysis as well as post hoc analysis according to the frailty model (phenotype or deficit accumulation model) used. Frailty was associated with adverse secondary clinical outcomes, albeit with associated statistical heterogeneity. Frail patients were twice as likely to be discharged to a location other than home; however, evidence for this was low due to marked heterogeneity. Length of stay and mortality were higher in the frail group; however, these outcomes were associated with marked heterogeneity and very low quality of evidence.
This is the first meta-analysis in patients undergoing open cranial surgery, indicating increased odds of complications in frail patients. We identified higher odds of any complication, as classified according to the Clavien–Dindo 1–4 groups, in frail patients undergoing open cranial surgery. This was a robust outcome with very persistent low statistical heterogeneity across sensitivity analysis and subgroup analysis. The results of this meta-analysis allow for quantification of the greater likelihood of complications in frail patients, thereby facilitating surgical decision-making and patient perioperative pathways. Although there are no other comparative meta-analysis data, the likelihood of complications was increased in a recently published meta-analysis of patients undergoing cardiac surgery [
3].
Frail patients had more adverse outcomes than the non-frail cohort; however, these outcomes were associated with significant heterogeneity. Frail patients were more likely to be discharged to a location other than home. Quality of life outcomes, such as discharge to a non-home location (NHD), are important to patients and should be considered part of the process of informed consent for medical intervention. NHD has been shown to be associated with decreased overall survival and significant healthcare and social costs [
43].
However, in our study for the non-home discharge outcome, the quality of evidence was downgraded due to heterogeneity impacting the inconsistency criterion. Further studies focusing on clinical homogeneity are needed. The rates of readmission failed to reach statistical significance. Our research findings show parallels with large database studies in related fields [
44,
45]. Our findings are in line with the quantitative analysis of other surgical and intensive care groups [
13]. Greater statistical heterogeneity may be due to the clinical in-group differences between frail patients in patients undergoing open cranial surgery. A more sophisticated approach to studying the outcomes in frail patients undergoing craniotomy would involve stratification according to the level of frailty in the original studies.
The pooled prevalence of frailty in this study was 48%, ranging from 21 to 85%. A diverse range of frailty may be contributed to by a diverse range of cranial pathology; therefore, the patient population needs to undergo surgery. Our study has a higher pooled prevalence of frailty compared to recent meta-analyses in other fields [
13]. The reasons for this are unclear. The development of cranial pathology may have a role in increasing preoperative frailty at the time of measurement.
We did not identify marked heterogeneity in the measurement instruments used. MFI-5 is a tested derivative of MFI-11, and the majority of studies used one of these instruments (7 studies in total incorporating 49,557 patients). With regard to large retrospective database studies, the mFI-5 is user-friendly. The two studies that used the phenotype frailty model utilized the John Hopkins Frailty Instrument or the related John Hopkins Adjusted Clinical Group’s frailty-defining diagnosis indicator. In the post hoc analysis, we identified the odds ratios of a primary outcome occurring as similar between the two groups, with no appreciable significant differences. Although most studies have utilized deficit accumulation model instruments, it is unclear whether deficit accumulation models or phenotype models are better instruments for frailty.
Consensus on a unified measurement instrument of frailty would eliminate instrument-related heterogeneity and streamline the research processes. However, there may be some benefits lost from multitool validation. Studies have demonstrated that the 5-factor modified frailty index (mFI-5) and the 11-factor modified frailty index (mFI-11) are equally effective in predicting adverse outcomes in the American College of Surgeons National Surgical Quality Improvement Program database [
46]. They have both been identified to be equally predictive of postoperative complications [
47]. The MFI-5 index has been deemed credible for future use to study frailty within NSQIP, within other databases, and for clinical assessment and use [
46,
48,
49]. Khallafah et al. specifically demonstrated the validity and responsiveness across the mFI-5, mFI-11, and Charleston comorbidity index in the neurosurgical field [
50].
Based on the findings of this meta-analysis, we postulate that frailty measurements could be used as an integral component of strategies to improve the quality and value of neurosurgical care for patients undergoing craniotomy surgery. Although there is a component of delivering timely care to patients undergoing cranial surgery, perioperative pathways can be streamlined to facilitate patient care. Multidisciplinary decision-making can be instituted to assess patient deficits and risks and formulate goals of care. Prehabilitation in patients facing urgent open cranial needs to be considered against the urgency of surgical management. There is evidence that prehabilitation as a component of a multidisciplinary approach improves patient-reported outcomes [
51]. When it is feasible to do so in this patient population, prehabilitation may be utilized. Importantly, an increased odds ratio of complications in patients with frailty can be used as a part of informed surgical consent balanced with surgical benefits.
Limitations of this meta-analysis include the grouping of all categories of frail patients together in most of the index studies. There is a spectrum of frailty ranging from least frail to most frail. Our study was able to estimate a more comprehensive outcome assessment for all of the frail patients grouped together. More sophisticated observational research data are required prior to stepwise risk estimation according to patients in mild, moderate, or severe frailty grouping. “Further high-quality prospective studies which stratify the frailty groups independently of one another, are needed in order to provide a more accurate odds of complications in patients suffering from increasing frailty.”
It is likely that once these groups are stratified according to the level of frailty, a clearer picture with regard to association with complications would emerge. Our study encompassed all patients requiring cranial surgery. We performed planned subgroup and sensitivity analyses. However, we were unable to differentiate on the basis of tumor type. This study therefore included patients with glioblastoma, where the extent of resection is linked to prognosis [
37]. This meta-analysis was only able to evaluate short-term outcomes for up to 30 days, with limited ability to review long-term outcomes as well as patient-reported outcomes (PROMs). Retrospective databases used by the studies (NSQIP) lack perioperative outcomes. Additional limitations of this meta-analysis include the retrospective design of seven studies. The use of retrospective chart reviewers and/or the NSQIP database may have introduced observational bias, as it relies on trained staff performing retrospective data collection. Data with regard to tumor histology, location, size, grade, and stage may be controlled for in a multivariable analysis; however, this level of detail is not available through a retrospective database.
A single study by Cloney et al. reported on the prevalence of frailty in patients with cranial pathology undergoing non-operative cranial management. Frailer patients (P 0.0002; odds ratio [OR], 0.15; 95% confidence interval [CI], 0.05–0.40) were significantly less likely to undergo surgical resection on multiple regression analysis. We were unable to source more comprehensive data on the prevalence of frailty in patients undergoing non-operative (conservative/palliative) cranial surgery. This may be valuable information to study in order to compare levels of frailty in patients undergoing craniotomy versus those who are not undergoing operative management.”
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.