1. Introduction
Degenerative cervical myelopathy (DCM) is characterized by progressive compression of the spinal cord in the cervical canal, producing debilitating neurological deficits in upper limb function, gait instability, sphincteric disturbance, and ultimately spastic quadriparesis. It is the most common cause of adult spinal cord dysfunction worldwide and its prevalence increases significantly with age [
1]. With the projected shift in demographics over the next 30 years, the burden of (potentially treatable) neurological dysfunction in the elderly has become a major public health concern of the 21st century [
1,
2,
3,
4]. The Institute of Medicine has declared that DCM includes 3 of the top 100 national priorities for comparative effectiveness in research, and efforts to address the rise of disability from DCM have been implemented worldwide, including the establishment of international consensus treatment guidelines [
5,
6,
7].
DCM is an umbrella term that encompasses a number of degenerative pathologies that include osteoarthritis (spondylosis), ligament disease (ossification of the posterior longitudinal ligament (OPLL), ligament hypertrophy), and degenerative listhesis or instability. These entities have pronounced effects on the functional abilities and quality of life of impaired individuals, which may be comparable to serious health conditions, such as cancer or heart disease [
8]. The mainstay of treatment for DCM is decompressive surgery, which arrests the progression of the disease and provides a sustained and meaningful improvement in functional and quality of life measures [
2,
3,
7,
9,
10,
11,
12,
13]. Clinical factors such as duration of symptoms prior to surgery and severity of baseline functional impairment correlate strongly with the chances of a substantial clinical benefit after intervention [
9,
11,
12,
14,
15].
Decisions regarding the best application of surgical intervention for DCM have become an important focus of clinical study [
3,
6]. In the elderly population, this issue becomes complex as the potential impact of preventing neurological disability in the elderly needs to be balanced against the healthcare costs and complication profile [
4,
6,
7,
13]. Although increasing age is associated with poorer surgical outcomes, in DCM patients many studies have shown sustained long-term functional and quality of life improvements after surgery [
4,
10,
11,
15,
16,
17]. Moreover, when elderly patients are matched for co-morbidities and baseline functional impairment, their complication profile is equivalent [
3]. This has led to the evolving opinion that age alone has become less relevant for the purpose of estimating perioperative risk profile and prognosis after surgery [
18,
19,
20,
21,
22].
Efforts to move past age alone as a predictor of outcomes has led to the development of measures of physiological reserve (or ‘frailty indices’). The most commonly cited index in spine surgery is the Modified Frailty Index (11-point or 5-point,
Table 1) [
23,
24,
25,
26] which is derived from the original 70-point Canadian Study of Health and Aging Frailty Index (CSHA-FI) [
27]. Other measures include the American Society of Anesthesiologists Grading scale (ASA) and the modified Charlson Co-morbidity Index (mCCI). ASA is a well-established subjective estimate of overall illness severity that has an uncertain role in the prediction of perioperative outcomes after spine surgery [
18]. The mCCI, like the MFI, is also matched to NSQIP variables but produces different weightings according to increasing age and certain co-morbidities. Although used widely in general surgery, the mCCI has received limited application in spine surgery [
19,
20,
28]. In 2017, Shin et al. published a study of 6965 patients who underwent either anterior cervical discectomy or posterior cervical fusion, and showed frailty (as measured by the 11-point MFI) was associated with an increased risk of adverse events [
23]. However, this study did not distinguish between radiculopathy or myelopathy patients, and it has been previously demonstrated that myelopathy patients have a higher risk of perioperative complications compared to patients with purely radicular symptoms [
24]. To date, no study exists that models the impact of frailty specifically on surgical DCM patients, or that compares the discriminatory ability of different measures of frailty.
The objectives of the current study were to (1) define the effect of age on the perioperative outcomes of mortality, unplanned readmission/reoperation, major complication, length of stay and discharge to non-home destination for patients undergoing surgery for DCM, (2) directly compare measures of frailty in the same cohort to determine which factor exhibits a greater influence on the observed outcomes, and (3) define the potential correlation between MFI-5 and MFI-11 in DCM patients. We hypothesize that after adjustment for common surgical factors, frailty is a better predictor of perioperative complications compared to age alone. Frailty as a predictor of perioperative complications would have important implications for the clinical management of elderly patients with DCM.
2. Experimental Section
2.1. Data Source
The data source for this study was the American College of Surgeons (ACS) National Surgical Quality Improvement Program (NSQIP) database, for years 2010 through 2018 inclusive. The NSQIP datasets encode surgical procedures by Current Procedural Terminology (CPT) codes and diagnoses by International Classification of Diseases, Ninth/Tenth Revision, Clinical Modification (ICD-9/10-CM) codes. NSQIP collects pre-operative through 30-day post-operative data on randomly assigned patients at participating hospitals. Quality and reliability of the data are ensured through rigorous training of data abstractors and inter-rater reliability audits of participating sites [
25].
2.2. Patient Population
Eligible patients who had a primary diagnosis of DCM (ICD-9-CM 721.1 or 722.71; ICD-10-CM M47.12 or M50.00, M50.01, M50.02, M50.03) and underwent a cervical decompression and fusion operation, including anterior (CPT 22551, 22554, 63081) and/or posterior (CPT 22600, 63051, 63020) approach. ICD-9/10-CM and CPT codes used for this study are summarized in
Table 2.
2.3. Baseline Characteristics
Data relating to baseline demographic characteristics and comorbidities were extracted. The number of operated levels was determined by searching the “other procedure” fields for CPT add-on codes specifying each additional level fused (CPT 22552, 22614). Surgical approaches were separated into anterior, posterior, or combined as a categorical variable.
2.4. Calculation of Frailty Indices
The MFI-11 was calculated according to established mapping of existing variables included in the NSQIP database (see
Table 1). A total score between zero and one was calculated by dividing the number of variables present (for functional status, partial or complete dependency = 1) by 11. Afterwards, 0.09 was categorized as “Pre-Frail”, 0.18 as “Frail”, and 0.27 and above as “Severely Frail”, in line with previously established standards [
26,
29]. A modified frailty index (mFI-5) was derived according to the standard methodology described by Searle et al. [
30], calculated using NSQIP variables [
31]. Specifically, five factors within the NSQIP (functional dependence, diabetes, history of chronic obstructive pulmonary disease (COPD), history of congestive heart failure, and hypertension) map to the original Canadian Study of Health and Aging (CSHA) Frailty Index [
27]. For every patient, each of these five factors (deficits) were coded as absent (0) or present (1). The mean score across all deficits was calculated, resulting in an index ranging from 0 (least frail) to 5 (most frail), with a score of 1 as “Pre-Frail”, 2 as “Frail”, and 3 or more as “Severely Frail” as categorical variables. ASA score was taken directly from the NSQIP database for each patient. The mCCI score was calculated according to previous methods mapped directly from the corresponding NSQIP variables [
18], creating a score from 0 to 23 depending on the age and presence of defined co-morbidities.
2.5. Outcomes
Outcomes evaluated were 30-day mortality, unplanned readmission, unplanned reoperation, and major complication, as well as total hospital length of stay (LOS) and routine discharge (home). Major complication was a composite outcome of pneumonia, deep vein thrombosis (DVT), pulmonary embolism, myocardial infarction, cardiac arrest, wound infection or dehiscence, stroke, and sepsis.
2.6. Statistical Analysis
All statistical analyses were performed using Stata 16 (Stata Corp, College Station, TX, USA) with an a priori specified significance level of p = 0.05 (two-tailed). Descriptive statistics were by mean and standard deviation (SD) for continuous variables as well as count and percentage for categorical variables.
The effect of age, MFI-5, MFI-11, CCI, and ASA were each analyzed by univariate analysis using simple logistic regression for dichotomous outcomes or linear regression for continuous outcomes. Effect sizes were summarized by odds ratio (OR) (dichotomous outcomes) or beta coefficients (continuous outcomes) and associated 95% confidence intervals (95% CI). The independent effect of age and frailty on outcomes was further evaluated by multivariable regression. Again, for each outcome, a logistic or linear regression model was constructed that included both variables and additionally adjusted for sex, type of fusion, and number of levels as covariates. To weigh the relative importance of age versus frailty in predicting each outcome, standardized regression coefficients were calculated and their magnitudes directly compared. The margins of interaction of the final model of age (by decade) and frailty (continuous variable) were calculated for all adverse events to assess how the burden of frailty was affected by increasing age.
To compare the discriminative ability of age and the various indices of frailty, receiver operating characteristics (ROC) curve analysis was performed for each dichotomous outcome reported. This was done first using a base model that included the type of approach, number of operated levels and gender, and then with the addition of age and each index of Frailty, before comparing to the final multivariable model incorporating age, MFI-5, and the base model. The Kappa correlation coefficient was calculated to assess the discriminative ability of the MFI-5 to predict the categorical assessment of frailty compared to the MFI-11 assessment.
2.7. Ethics Approval
Institutional Review Board approval was not required for this study, which relied on de-identified data derived from a national administrative healthcare dataset.
4. Discussion
As the burden of age-related degenerative spine conditions becomes an ever-greater public health priority, DCM has the potential to be a major cause of preventable neurological disability and poor quality of life across the world [
4,
8,
16]. Decompressive surgery remains the only treatment proven to arrest or reverse the dysfunction caused by myelopathy and has a proven benefit in the elderly population [
3,
7,
10,
32]. The diagnosis of DCM and identification of suitable surgical candidates is therefore of paramount importance, and for this reason, identifying tools to aid risk stratification and predict potential outcomes after surgery is a major focus of clinical research [
6,
9,
13,
14].
The evidence for decompressive surgery to improve functional and quality of life impairment in DCM is strong [
7,
10,
32]. Reports on the effect of age on the outcomes after DCM surgery have been variable, with many supporting the notion that increasing age is associated with negative clinical outcomes, whether increased complication rates or functional outcomes [
12,
13,
15,
33]. A recent ambispective study has demonstrated a clear and sustained benefit for surgery in DCM for both functional and quality of life outcomes in patients over the age of 70, albeit with an order of magnitude less than their younger surgery- and co-morbidity-matched cohort [
3]. The suggested mechanism for this discrepancy has traditionally been the burden of age-related co-morbidities influencing clinical outcomes and reduction of physiological reserve, which has been manifested as measures of frailty or frailty indices in recent years. This hypothesis has been further substantiated with the correlation of worsening frailty to increased complication rate after spine surgery, and poorer recovery after spinal cord injury [
28,
34,
35]. However, despite DCM being the most common indication for cervical spine surgery in North America, the effect of frailty on the outcomes after DCM surgery has not been investigated [
16].
This study is one of the most comprehensive investigations of perioperative adverse events after surgery for DCM, and is the first to present a direct comparison of age and frailty indices. Concepts of frailty, although distinct from risk stratification, are increasingly incorporated in the pre-operative assessment as a superior and comprehensive alternative to age alone. Although it is no surprise that increasing age leads to an increased risk of adverse events, it appears the burden of frailty can have an effect up to 28 times the magnitude compared to age alone. On univariate analysis, increasing frailty (MFI-5/MFI-11) had the largest effect size for mortality, major complication and discharge to non-home destination. This effect remained in the multivariable model, but the effect size was largest for mortality and unplanned reoperation. These findings suggest that physiological reserve is much more of a driver of perioperative complications when compared to age alone. This is consistent with previous studies on spine surgical patients, but also other surgical domains [
36,
37]. The degree to which frailty appears to have an influence on unplanned reoperation rates has a sound pathophysiological basis. Given the usual indications for early reoperation are compressive hematoma, infection, and hardware failure, patients who have a higher mFI score are more likely to develop post-operative infections and a higher risk of osteopenia/osteoporosis.
Frailty significantly affected the risk of perioperative adverse events for patients of all ages, but the burden of frailty becomes greater with increased age. This effect is not linear, as demonstrated by the margins of interaction between age (by decade) and categorical level of frailty (see
Figure 1). This effect (for readmission, reoperation, length of stay, and discharge to non-home destination) begins after the age of 60, which is in line with previously reported studies [
32,
34,
35].
The mCCI proved a significant predictor of all adverse events on univariate and multivariable analysis, and the effect size remained similar for both. This may suggest that it could be a useful tool in the assessment of DCM patients. However, the role of the mCCI in multivariable modeling is unclear due to the inclusion of age-related modifiers. The mCCI incorporates increasing age as a contributor to the overall score, and therefore does not provide an age-independent measure of frailty. It cannot therefore be a definitive frailty assessment when incorporated into a multivariable model that includes age as a continuous variable. The ASA score also proved to be significantly predictive of adverse events on both univariable and multivariable analysis. However, the ASA score is notoriously subjective and its practical use for pre-operative adverse event prediction modeling has not been substantiated.
There was strong correlation between the MFI-11 and MFI-5 assessment of frailty tiers, which has been echoed in previous articles in other spine pathologies [
28]. The effect size of the odds ratio or regression coefficients were larger when the MFI-5 was used compared to the MFI-11 for all dichotomous outcomes and length of hospital stay. This would suggest that to achieve an assessment of “Frail” or “Severely Frail” with the MFI-5 this would indicate a greater degree of frailty compared to the MFI-11 equivalent. This is strong evidence that the use of the MFI-5 is an effective determinant of frailty and further substantiates the MFI-5 as the standard of choice for frailty assessment for DCM patients. This has important implications for the use of MFI-5 in clinical practice and for future studies into the effect of frailty in adult spine surgery.
There are limitations to the current study. The NSQIP database carries metrics regarding peri-operative events, but no long term follow up or outcome measures are included. This restricts the analysis to short term follow up only, and therefore long-term outcomes cannot be extrapolated. However, the validity of short-term complication rates and their use for pre-operative decision-making in spine surgery is well published. In a similar vein, there is no measure of baseline functional impairment and therefore there is no way to eliminate the effect of myelopathy severity on the surgical approach, or other covariates applied to the regression model. Pre-operative neurological function is a key predictor of the functional outcomes of DCM surgery, however its relationship to perioperative adverse events and mortality is less clear. It is known that myelopathy patients have increased perioperative complications compared to radiculopathy [
26], but there is a paucity of evidence of the effect of functional impairment on perioperative outcomes in DCM.
There was an observed difference in the number of patients that were able to have the MFI-11 calculated (n = 11,758) compared to the MFI-5 (n = 41,140). This difference arose arbitrarily due to a reduction in the demographic information collected by NSQIP from 2015 onwards. This difference in numbers has the potential to skew the effects and effect size of the results presented. The authors argue that given the categorical correlation of frailty was excellent between MFI-5 and MFI-11 (97.58% agreement), the large size of the cohorts, the strength of the a priori statistical frameworks and the levels of significance observed, the risk of statistical inaccuracy is low.
The study is a retrospective analysis of prospectively collected nationwide hospital inpatient registry data. As such, it lacks the protection from confounders and bias from a true prospectively-collected data set. However, uniformity across the data collection and homogeneity amongst the selected cohort does reduce the impact of any latent confounding. The cohort selection relied on ICD-9 & 10 diagnostic codes, and therefore the potential risk of not capturing patients if they were inappropriately coded is present. Patients were removed from the final analysis if no reliable CPT code data could identify the surgical approach or number of levels, which led to a reduction in the overall number of patients in the multivariable model. It is unlikely that these occurrences have affected the findings of the study given the levels of significance seen on univariable and multivariable analysis.
The authors note that DCM carries a significant heterogeneity in presenting symptoms ranging from mild sensory disturbance in the fingers through to gait instability and eventual quadriparesis. Traditional measures of frailty, in particular those based on 5- or 11-point scales as used in this study, may not be as accurate when compared to other pathologies not affecting gait or upper limb function. Therefore, developing more appropriate indices of frailty specific to DCM should be a future research priority. Also, the authors wish to state that the purpose of this study is not to prove patients who have an increased frailty score should not be offered surgical management. In contrast, this study should be used to provide further clarification of the factors that go into informed decision-making, which should be a shared process between the patient and clinician. The concept that frailty could potentially be viewed as a modifiable risk factor is also emerging, however evidence that improving frailty index scores prior to planned spine surgery to reduce perioperative complications is not conclusive at this stage [
38].