Abstract
-
Objective
To evaluate whether combining clinical frailty with magnetic resonance imaging (MRI)-derived posterior paraspinal muscle degeneration identifies perioperative risk phenotypes in adults aged ≥75 years undergoing lumbar fusion.
-
Methods
We retrospectively studied patients aged ≥75 years undergoing lumbar fusion with preoperative lumbar MRI. Frailty was assessed using the Fried phenotype (frail: score ≥3). Posterior paraspinal muscle degeneration across L1–S1 was quantified using automated segmentation and a composite posterior frailty index (PFI); severe degeneration was defined as the upper quartile of PFI. Patients were classified into 4 frailty×muscle phenotypes. Primary outcomes were any in-hospital complication and prolonged length of stay (LOS ≥16 days).
-
Results
Among 248 patients, phenotypes A–D (A, nonfrail/nonsevere; B, frail/nonsevere; C, nonfrail/severe; D, frail/severe) comprised 132, 54, 20, and 42 patients, respectively. Any in-hospital complication occurred in 18.2% of phenotype A compared with 50.0%–57.1% in phenotypes B–D (p<0.001). Prolonged LOS (≥16 days; cohort 75th percentile) occurred in 0.8% of phenotype A versus 38.9% (B), 35.0% (C), and 78.6% (D) (p<0.001), corresponding to absolute risk increases of +34.2 to +77.8 percentage points. After adjustment, higher-risk phenotypes remained independently associated with increased odds of any complication and prolonged LOS; however, the prolonged-LOS odds estimates were imprecise due to sparse events in the reference group. Phenotype was not independently associated with 90-day readmission. Pain improvement (ΔVAS [visual analogue scale]) was attenuated in phenotypes B and D, while differences in ΔODI (Oswestry Disability Index) were not statistically significant.
-
Conclusion
Integrating frailty and MRI-based posterior paraspinal degeneration provides actionable stratification of complication and prolonged LOS risk after lumbar fusion in older adults.
-
Keywords: Frailty, Paraspinal muscles, Magnetic resonance imaging, Spinal fusion, Sarcopenia, Postoperative complications
INTRODUCTION
Lumbar fusion is increasingly performed in older adults, a population in whom perioperative trajectories vary widely and resource utilization is substantial [
1,
2]. Readmissions and prolonged length of stay (LOS) after lumbar fusion are common and clinically meaningful, reflecting a mixture of surgical stress, comorbidity burden, mobility limitations, and discharge barriers [
3]. Traditional risk assessment tools often emphasize chronological age and comorbidities, yet these constructs do not fully capture physiologic reserve.
Frailty—conceptualized as vulnerability to stressors from cumulative physiologic decline—has emerged as a robust predictor of adverse surgical outcomes [
4-
6]. Multiple spine-surgery syntheses have demonstrated that frailty is associated with higher complication rates, longer hospitalization, and increased costs [
7,
8]. However, frailty phenotyping alone may still be insufficient for precision risk stratification, particularly when operative risk and recovery are shaped by musculoskeletal reserve and functional capacity.
Sarcopenia and muscle quality deterioration represent complementary vulnerability domains. Contemporary consensus emphasizes that muscle quality (e.g., fatty infiltration) and strength may be as consequential as muscle quantity [
9]. In spine surgery, imaging-derived body composition markers have been linked to complications and diminished recovery [
10-
12]. In parallel, paraspinal muscle degeneration—especially fatty infiltration of the posterior paraspinal compartment—has gained attention as a local marker of spinal support capacity and a systemic surrogate of musculoskeletal health [
13-
15]. Automated or semi-automated magnetic resonance imaging (MRI) segmentation further enables scalable, reproducible quantification of muscle volume and composition across lumbar levels [
14,
16].
We hypothesized that integrating clinical frailty with posterior paraspinal muscle degeneration would better identify high-risk subgroups among adults aged ≥75 years undergoing lumbar fusion. Accordingly, we developed a combined phenotype framework using the Fried frailty score and an MRI-derived posterior frailty index (PFI) and evaluated its associations with complications, LOS, readmission, and early improvements in pain and disability.
MATERIALS AND METHODS
1. Study Design and Participants
We conducted a retrospective cohort study of adults aged ≥75 years who underwent lumbar fusion for degenerative lumbar disease and had preoperative lumbar MRI available for analysis. To reduce heterogeneity in surgical approach, the cohort was restricted to patients treated with open posterior interbody fusion (transforaminal lumbar interbody fusion). Patients with non-degenerative indications (e.g., infection, tumor, trauma) or missing key exposure/outcome data were excluded, as applicable. Symptom timing was captured as symptom duration (months) extracted from the chief complaint in the medical record. In routine care, patients were typically indicated for surgery after failure of conservative management, which commonly included analgesic/anti-inflammatory medications and physical/kinetic therapy; because specific regimens and adherence were not standardized and not consistently captured in structured form, preoperative therapy exposure was not quantitatively modeled. The workflow is summarized in
Fig. 1.
The study protocol was reviewed and approved by the institutional review board (IRB) of Capital Medical University Xuanwu Hospital (IRB No. 2018-086). Written informed consent was waived due to the retrospective design. This study is reported in accordance with the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) guidelines for observational cohort studies.
2. Frailty Assessment
Frailty was assessed using the Fried frailty phenotype (score 0–5), classifying patients as frail when the score was ≥3, scores 0–2 were categorized as nonfrail (robust or prefrail) [
4]. Frailty scoring was derived from the preoperative clinical evaluation and/or standardized frailty assessments recorded in the medical record.
MRI acquisition. Preoperative lumbar MRI was performed as part of routine surgical workup using 1.5T scanners. The analysis focused on axial T2-weighted imaging used for automated muscle segmentation, with typical acquisition parameters of slice thickness 3–5 mm, interslice gap ≤1 mm, and in-plane resolution approximately 0.5–0.8 mm. For each patient, the most recent preoperative MRI available prior to the index surgery was used for analysis. Preoperative lumbar MRI was processed with an automated segmentation workflow to quantify posterior paraspinal muscle volume and fat infiltration (FI) across L1–S1, using bilateral posterior compartment measures. To ensure segmentation accuracy, all automated masks were visually reviewed by an orthopaedic fellow and manually corrected where necessary under the supervision of a fellowship-trained spine surgeon. Reliability of this workflow was validated in a random subset of 20 patients, yielding high interobserver agreement with mean Dice similarity coefficients of 0.88 for multifidus, 0.91 for erector spinae, and 0.93 for psoas and quadratus lumborum. We derived a composite PFI=z(FI)−z(volume) to jointly encode poorer muscle quality (higher FI) and lower muscle quantity (smaller volume). “Severe” posterior paraspinal muscle degeneration was defined as the upper quartile of PFI in the cohort. This workflow leveraged MuscleMap-consistent quantitative muscle MRI principles [
16] (A segmentation and reconstruction schematic is provided in
Fig. 2).
Patients were stratified into 4 prespecified phenotypes based on frailty (nonfrail vs. frail) and posterior paraspinal degeneration (nonsevere vs. severe): A, nonfrail/nonsevere; B, frail/nonsevere; C, nonfrail/severe; D, frail/severe.
5. Outcomes and Statistical Analysis
Primary outcomes were (1) any in-hospital complication and (2) prolonged LOS, defined as LOS≥16 days (cohort 75th percentile). Secondary outcomes included major and minor in-hospital complications, 90-day readmission, and changes in patient-reported outcomes (ΔVAS, ΔODI). Pain intensity was assessed using the visual analogue scale (VAS) for low back pain (0–10), as this metric is most mechanistically relevant to paraspinal muscle status. Disability was assessed using the Oswestry Disability Index (ODI). Postoperative patient-reported outcomes were obtained at the first routine outpatient follow-up within 1–3 months after surgery. Changes (ΔVAS and ΔODI) were calculated as baseline minus postoperative values. In-hospital complications were abstracted from medical records and discharge summaries. Postoperative adverse events were categorized according to the Clavien-Dindo classification system. Events graded as Clavien-Dindo grades I–II were considered minor complications, whereas grades III–IV were considered major complications [
17].
Baseline characteristics and perioperative variables were compared across phenotypes using analysis of variance or Kruskal-Wallis tests for continuous variables, and chi-square or Fisher exact tests for categorical variables, as appropriate. Multivariable logistic regression was used for binary outcomes (complications, prolonged LOS, readmission) and linear regression for ΔVAS and ΔODI, with phenotype A as the reference category. Models adjusted for key clinical and operative covariates (e.g., age, sex, body mass index [BMI], Charlson Comorbidity Index, American Society of Anesthesiologists physical status classification, operative time, and fused levels). Two-sided p<0.05 defined statistical significance. To clarify which component of posterior paraspinal degeneration primarily drives the association with prolonged LOS, we conducted additional multivariable logistic regression models evaluating posterior paraspinal FI and muscle volume separately, as well as jointly (
Supplementary Table 1). FI and volume were standardized (z-scores). Prolonged LOS was defined as LOS≥16 days. Results are reported as odds ratios (ORs) with 95% confidence intervals (CIs). As sensitivity analyses for the primary endpoint of prolonged LOS (≥16 days), we additionally adjusted the prolonged-LOS logistic regression models for baseline clinical severity, including baseline VAS and baseline ODI (separately and jointly), to assess whether the phenotype associations were attributable to preoperative symptom burden.
RESULTS
1. Cohort and Phenotype Distribution
A total of 248 patients aged ≥75 years were included. Phenotype distribution was: A (nonfrail/nonsevere) n=132, B (frail/nonsevere) n=54, C (nonfrail/severe) n=20, and D (frail/severe) n=42 (
Fig. 3).
Baseline characteristics differed across phenotypes (
Table 1). Symptom duration was chronic overall (median, 48.5; interquartile range [IQR], 21.0–92.0 months) and did not significantly differ across phenotypes (p=0.124). Compared with phenotype A, higher-risk phenotypes were older (p<0.001) and had lower BMI (p<0.001). Baseline symptom burden also differed: baseline VAS and ODI were higher in phenotypes B and D than in phenotype A (both p<0.001).
Posterior paraspinal muscle size was reduced in severe-degeneration phenotypes (C and D), with lower posterior volume and cross-sectional area (both p<0.001). Muscle quality showed a stepwise deterioration across phenotypes, with higher posterior FI and higher PFI from phenotype A through phenotypes C/D (all p<0.001). Anterolateral FI also differed across phenotypes (p<0.001) (
Table 1).
Operative time, estimated blood loss, and fused levels were similar across phenotypes (operative time p=0.092; blood loss p=0.718; fused levels p=0.164) (
Table 2). LOS increased substantially with increasing phenotype risk (median [IQR]: A, 8.000 [5.000–11.000] days; B, 15.000 [12.250–17.000] days; C, 14.000 [11.000–16.250] days; D, 19.000 [16.000–21.750] days; p<0.001). Prolonged LOS (LOS≥16 days) occurred in 0.8%, 38.9%, 35.0%, and 78.6% of phenotypes A–D, respectively (p<0.001). Ninety-day readmission did not significantly differ across phenotypes (A, 16.7%, B, 24.1%, C, 20.0%, D, 28.6%; p=0.349) (
Fig. 4;
Table 2).
In-hospital complications were more frequent in higher-risk phenotypes (any complication: A, 18.2%; B, 50.0%; C, 55.0%; D, 57.1%; p<0.001). Major complication rates differed across phenotypes (p<0.05), while minor complications showed a marked gradient (A, 3.8%; B, 18.5%; C, 30.0%; D, 42.9%; p<0.001).
Patient-reported outcomes varied across phenotypes. Postoperative VAS and ODI were highest in phenotype D (median [IQR]: VAS, 5.000 [2.000–6.000]; ODI, 33.976±21.642; both p<0.001). Pain improvement (ΔVAS) decreased across phenotypes (A, 3.053±1.280; B, 2.130±1.518; C, 2.700±1.302; D, 1.357±1.144; p<0.001). Disability improvement (ΔODI) showed a similar trend but did not reach statistical significance (p=0.064).
4. Adjusted Associations by Phenotype
In multivariable models (reference: phenotype A), phenotypes B, C, and D were independently associated with higher odds of any complication (OR [95% CI]: B vs. A, 5.034 [2.386–10.617]; p<0.001; C vs. A: 5.329 [1.710–16.611], p<0.01; D vs. A: 6.537 [2.789–15.322], p<0.001) and prolonged LOS (B vs. A: 64.702 [12.821–326.517], p<0.001; C vs. A: 61.863 [8.793–435.254], p<0.001; D vs. A: 499.818 [80.496–3,103.491], p<0.001) (
Table 3;
Fig. 5).
For major complications, only phenotype B remained significantly associated versus A (2.980 [1.337–6.646], p<0.01), whereas C and D were not significant. For minor complications, phenotypes B, C, and D all showed increased odds versus A (all p≤0.01). Phenotype was not independently associated with 90-day readmission.
For pain improvement, phenotypes B and D demonstrated smaller adjusted ΔVAS compared with phenotype A (B vs. A: -1.024 [-1.490 to -0.558], p<0.001; D vs. A: -2.009 [-2.462 to -1.556], p<0.001), while the C vs. A comparison was not significant. Adjusted differences in ΔODI were not significant across phenotypes (
Table 3).
In adjusted analyses (reference: phenotype A), a strong phenotype gradient was observed for prolonged LOS (≥16 days; cohort 75th percentile). Prolonged LOS occurred in 1 of 132 patients (0.8%) in phenotype A versus 21 of 54 (38.9%) in phenotype B, 7 of 20 (35.0%) in phenotype C, and 33 of 42 (78.6%) in phenotype D (p<0.001) (
Table 2). The corresponding absolute risk differences (ARDs) versus phenotype A were +38.1, +34.2, and +77.8 percentage points, respectively (unadjusted risk ratios: 51.3, 46.2, and 103.7). Consistent with these absolute risk gradients, phenotypes B–D remained independently associated with prolonged LOS in multivariable logistic regression (
Table 3;
Fig. 5). However, due to the extremely low event rate in the reference group (1 of 132, 0.8%), the ORs are mathematically inflated (sparse data bias). More clinically relevant are the ARDs: compared to phenotype A (0.8%), the absolute risk of prolonged LOS increased by +38.1% in phenotype B (38.9%), +34.2% in phenotype C (35.0%), and +77.8% in phenotype D (78.6%). These robust absolute gradients confirm the high risk despite the instability of the relative point estimates. Sensitivity analyses further adjusting for baseline symptom severity yielded consistent findings for prolonged LOS (≥16 days). In the model additionally adjusted for both baseline VAS and baseline ODI, phenotypes B, C, and D remained strongly associated with prolonged LOS compared with phenotype A (B vs. A: 109.174 [95% CI, 13.239–900.292]; C vs. A: 78.643 [6.578–940.278]; D vs. A: 840.983 [89.449–7,906.749]; all p<0.001). Results were similar when adjusting for baseline VAS alone or baseline ODI alone (
Supplementary Table 2).
To disentangle the relative contributions of muscle quality and quantity, posterior paraspinal FI and muscle volume were examined in additionally adjusted models for prolonged LOS (≥16 days). Higher FI showed a strong association with prolonged LOS (per 1-standard deviation [SD] higher FI: 12.066 [5.974–24.368]; p<0.001), and lower muscle volume was also associated (per 1-SD lower volume: 2.648 [1.346–5.208]; p=0.005). When FI and volume were entered jointly, both remained independently associated (FI: 13.926 [5.969–32.486]; p<0.001; volume: 4.091 [1.099–15.222]; p=0.036), suggesting complementary contributions of muscle quality and quantity (
Supplementary Table 1).
DISCUSSION
1. Principal Findings
In this retrospective observational cohort of adults aged ≥75 years undergoing lumbar fusion, we identified 4 frailty–posterior paraspinal degeneration phenotypes and observed graded differences in postoperative LOS and minor complications across phenotypes. Because exposures were not randomized and important determinants of recovery (e.g., baseline symptom burden, postoperative rehabilitation intensity, and discharge environment) may differ between groups, these findings should be interpreted as associations that support risk stratification and hypothesis generation rather than as evidence of causal effects of muscle degeneration or frailty on outcomes. We conceptualize severe posterior paraspinal degeneration on MRI as an imaging marker that may capture aspects of local musculoskeletal reserve (myosteatosis and/or low contractile tissue) that are not fully represented by performance-based frailty screening alone. This interpretation is supported by prior comparative cohort evidence showing that paraspinal muscle morphology correlates with back pain and spinopelvic parameters, highlighting the clinical relevance of paraspinal muscle health beyond systemic risk constructs [
18]. Accordingly, phenotype differences in LOS and minor complications may reflect a combination of factors—reduced trunk extensor reserve affecting early mobilization and endurance, higher susceptibility to low-grade inpatient events that delay discharge, and heterogeneity in surgical invasiveness and postoperative care pathways. Importantly, alternative explanations such as baseline pain/disability severity, social support, and institutional discharge practices may also contribute; therefore, we avoid deterministic interpretations and emphasize the need for prospective validation.
A particularly striking finding was the elevated risk of adverse outcomes in phenotype C—patients who were not systemically frail yet possessed severe paraspinal degeneration. Despite having preserved physiological reserve (low Fried scores), these patients experienced high rates of prolonged LOS (35.0%) and minor complications (30.0%). We propose that this divergence is driven by “mechanical vulnerability” leading to a vicious cycle of immobility.
Successful discharge after lumbar fusion relies heavily on early mobilization (e.g., log-rolling, transfers, and ambulation). The posterior paraspinal muscles form the “dynamic corset” of the spine; when severe fatty infiltration compromises this local contractile capacity (as in phenotype C), patients may experience greater difficulty and pain during these essential movements. This local mechanical failure impedes adherence to physical therapy protocols, directly delaying discharge eligibility. Furthermore, this delayed mobilization likely explains the increase in “minor” complications. The specific complications observed (e.g., urinary retention, postoperative ileus, and atelectasis) are classically associated with postoperative stasis and bed rest. In phenotype C patients, the inability to mobilize early creates a susceptibility to these stasis-related events. Although non–life-threatening, these complications require medical management that further disrupts the recovery trajectory. Thus, for nonfrail patients with severe muscle degeneration, the primary driver of risk appears to be a pathophysiological cascade where local muscle deficiency leads to delayed mobilization, which in turn precipitates secondary minor complications, ultimately resulting in prolonged hospitalization.
3. Why Combining Frailty and Posterior Paraspinal Degeneration Matters
Frailty is a well-established surgical risk construct that operationalizes reduced physiologic reserve and vulnerability to stressors [
4,
5]. However, frailty tools were not designed to quantify musculoskeletal capacity directly, and this may be particularly relevant in spine surgery where postoperative recovery is tightly linked to early mobilization, trunk control, and tolerance of rehabilitation. Sarcopenia consensus frameworks underscore that muscle quality (e.g., FI) and functional performance can be as important as muscle quantity [
9]. In this context, posterior paraspinal degeneration can be viewed as both (1) a local marker of spinal extensor system integrity and (2) a systemic marker of poor musculoskeletal reserve. The phenotype framework used here makes that duality clinically actionable: it separates patients who are frail but not severely degenerated (B), severely degenerated but not frail (C), and those with convergence of both risk dimensions (D), thereby enabling more granular prediction than frailty alone.
The adjusted ORs for prolonged LOS were strikingly large. This should be interpreted with appropriate caution and clinical context. First, prolonged LOS was defined as the ≥75th percentile threshold, and the reference group (A) had very low event frequency, which inflates ORs even when absolute differences are clinically plausible. Second, when event rates approach zero in one stratum, conventional logistic regression can become unstable (quasi-complete separation), yielding large point estimates and wide CIs—an issue well recognized in surgical outcomes modeling. The key clinical message, therefore, is not the exact magnitude of the OR but the robust directionality and separation of risk: both frailty and severe posterior degeneration were independently associated with prolonged hospitalization, and their combination identified a subgroup with an exceptionally high probability of prolonged LOS.
5. Complications: Additive Vulnerability Across Domains
Complication rates increased across phenotypes, with the most consistent adjusted associations seen for any complication and for minor complications. This pattern is clinically coherent in elderly fusion populations: small physiologic perturbations (delirium, urinary complications, wound issues, transient neurologic symptoms, electrolyte disturbances) may not individually qualify as major events yet can meaningfully prolong hospitalization and impede rehabilitation. Frailty plausibly elevates risk through reduced cardiopulmonary reserve, impaired stress response, and heightened susceptibility to immobilization-related morbidity. Severe posterior muscle degeneration may compound this by reducing the capacity for early ambulation and trunk stabilization, increasing pain-related immobility and dependence, and limiting participation in postoperative physiotherapy. Together, these mechanisms provide a biologically plausible explanation for why phenotype D consistently performed worst.
6. Recovery Outcomes: Why Pain Improvement Differed More Than Disability Improvement
We observed clearer separation in pain improvement than disability improvement across phenotypes. This is clinically plausible because pain and disability capture overlapping but distinct domains: early postoperative pain is influenced by systemic vulnerability (e.g., stress response, analgesic tolerance, and broader recovery capacity), whereas disability improvement is more contingent on rehabilitation intensity, discharge environment, comorbidity burden (including non-spine limitations), and ceiling/floor effects in very elderly patients. In lumbar fusion populations, pain (particularly leg pain) may improve more rapidly than axial back pain and disability, and functional gains may lag behind pain relief over short observation windows [
19]. In addition, evidence from broader surgical cohorts suggests frailty status is associated with altered perioperative recovery needs, including analgesic-related outcomes, supporting the plausibility that systemic vulnerability may shape early pain trajectories [
20].
Although the 90-day all-cause readmission rate appeared high in this very elderly fusion cohort, large database studies of lumbar fusion report 90-day readmission rates in a similar range (24.8%), supporting that our estimate is plausible for an all-cause definition in higher-risk populations [
21]. Ninety-day readmission did not differ significantly across phenotypes after adjustment. This result is not necessarily discordant with the strong LOS and complication signals. Readmissions are influenced by multiple non-physiologic factors—local care pathways, outpatient access, discharge destinations, and thresholds for admission—that can attenuate or obscure biologic gradients. Additionally, some risks captured by frailty and muscle degeneration may manifest primarily as in-hospital vulnerability (prolonged recovery, minor complications, delayed mobilization) rather than postdischarge events. Finally, competing risks may operate: patients who remain hospitalized longer may have complications managed during index admission rather than after discharge. This is still a valuable negative result; it helps refine the clinical scope of the phenotype approach and supports an interpretation focused on inpatient course and early recovery rather than readmission prediction. Readmission causes were not captured in a standardized fashion in this retrospective dataset; future prospective work should classify surgical vs. medical readmission causes.
A notable methodological strength is the use of a composite PFI that jointly captures muscle quality and quantity: PFI=z(FI)−z(volume). This is conceptually aligned with modern sarcopenia thinking that emphasizes multidimensional muscle health rather than relying on a single morphometric parameter [
9]. Standardization via z-scores improves comparability and prevents scale dominance. The quartile-based severe threshold is pragmatic and reproducible, but it is also cohort-derived; thus, external validation is necessary before clinical translation. Importantly, the phenotype approach does not require complex modeling to be interpretable—clinicians can readily understand the 4 quadrants, which may facilitate adoption in perioperative decision-making.
The practical appeal of this work is that lumbar MRI is already routinely obtained in degenerative lumbar disease, and frailty screening is increasingly embedded in preoperative evaluation. A combined phenotype could therefore be implemented with minimal additional patient burden, provided segmentation and quality metrics can be generated reliably.
Potential applications include:
Risk communication and shared decision-making: Phenotypes B/C/D represent materially different postoperative trajectories that can inform patient counseling, expectation setting, and surgical planning.
Discharge planning and resource allocation: The pronounced phenotype differences in prolonged LOS support early coordination for rehabilitation placement and discharge barriers—particularly for phenotype D.
Targeted perioperative optimization: For high-risk phenotypes, perioperative pathways consistent with enhanced recovery after surgery (ERAS) principles may be prioritized (early mobilization, multimodal analgesia, delirium prevention, nutrition/hydration optimization) [
22,
23]. While ERAS is not phenotype-specific, the phenotype framework can identify those most likely to benefit from intensified pathway adherence.
10. Limitations
Several limitations merit emphasis to align with top-journal expectations. First, this study is retrospective and observational; therefore, the reported associations do not establish causality, and residual confounding from unmeasured factors (e.g., rehabilitation resources, discharge barriers, socioeconomic context) may remain despite adjustment. Second, the sample size for phenotype C (nonfrail/severe) was relatively small (n=20). While the effect size was large enough to reach statistical significance, this small sample size limits the precision of the estimates and results in wider CIs; thus, findings for this specific subgroup should be validated in larger cohorts. Third, the thresholds used for PFI (upper quartile) and prolonged LOS (75th percentile) are cohort-specific distributions. While this approach is necessary for deriving a novel index, external validation is required to establish universal cutoff values for broad clinical application. Finally, longer-term outcomes, standardized patient-reported outcomes collection timepoints, and external validation cohorts are necessary before clinical deployment.
11. Future Directions
Future work should pursue (1) multicenter external validation with harmonized MRI protocols and discharge practices; (2) evaluation of continuous PFI models (rather than quartiles) to improve calibration and minimize threshold dependence; (3) integration of additional domains such as nutritional indices, sarcopenia strength measures, and psychosocial factors to refine prediction; and (4) prospective testing of phenotype-guided enhanced recovery or prehabilitation programs, with endpoints that include LOS, complications, discharge disposition, and longer-term functional recovery.
CONCLUSION
In summary, our findings support the concept that frailty and posterior paraspinal muscle degeneration represent complementary vulnerability domains in elderly lumbar fusion patients. A simple four-quadrant phenotype framework provides clinically interpretable stratification of complications, prolonged hospitalization, and early pain recovery. With external validation and pathway integration, this approach has potential to enhance perioperative risk stratification and guide targeted optimization in geriatric spine surgery.
Supplementary Materials
NOTES
-
Conflict of Interest
The authors have nothing to disclose.
-
Funding/Support
This study received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.
-
Author Contribution
Conceptualization: SL; Data curation: MCG, SW, XC, YL; Formal analysis: MCG; Methodology: MCG, XL, CK; Project administration: SL; Visualization: SL; Writing – original draft: MCG; Writing – review & editing: MCG.
Fig. 1.Workflow of this study. MRI, magnetic resonance imaging; PFI, posterior frailty index; LOS, length of stay; PRO, patient-reported outcomes; VAS, visual analogue scale; ODI, Oswestry Disability Index.
Fig. 2.Example of paraspinal muscle segmentation. Representative axial T2-weighted lumbar magnetic resonance imaging at the L4–5 level illustrating the automated segmentation workflow. The original image (left) is processed with MuscleMap to generate color-coded masks (middle) for the psoas major (P_R, P_L; yellow), quadratus lumborum (Q_R, Q_L; blue), erector spinae (E_R, E_L; red), and multifidus (M_R, M_L; green). These masks are further used to reconstruct three-dimensional muscle models (right), from which volumetric, cross-sectional area, fat-infiltration, and radiomics features are extracted. R, right; L, left; M, multifidus; P, psoas major; Q, quadratus lumborum; E, erector spinae.
Fig. 3.Frailty×posterior paraspinal muscle degeneration phenotypes. Scatter plot of Fried frailty score (y-axis) versus posterior frailty index (PFI; x-axis), where PFI=z(FI)−z(volume). Horizontal dashed line indicates the frailty threshold (Fried score≥3), and vertical dashed line indicates the severe degeneration threshold (upper quartile of PFI). Points are colored by phenotype: A (nonfrail/nonsevere), B (frail/nonsevere), C (nonfrail/severe), and D (frail/severe), with phenotype sample sizes annotated.
Fig. 4.Prolonged length of stay and Ninety-day readmission by phenotype. (A) Proportion of patients with prolonged LOS (LOS≥16 days; cohort 75th percentile threshold) across phenotypes A–D. Numbers above bars indicate events/total in each phenotype. Error bars denote 95% confidence intervals. Between-phenotype comparison: p<0.001. (B) Proportion of patients with 90-day readmission across phenotypes A–D. Numbers above bars indicate events/total in each phenotype. Error bars denote 95% confidence intervals. Between-phenotype comparison: p=0.349. Phenotypes: A (nonfrail/nonsevere), B (frail/nonsevere), C (nonfrail/severe), and D (frail/severe).
Fig. 5.Adjusted associations by phenotype (reference: phenotype A). Forest plot of multivariable adjusted associations comparing phenotypes B, C, and D with phenotype A. Top panel shows adjusted odds ratios (log scale) with 95% confidence intervals for binary outcomes (prolonged LOS, minor complication, major complication, any complication, and 90-day readmission). Bottom panel shows adjusted mean differences (vs. phenotype A) with 95% confidence intervals for continuous outcomes (ΔVAS and ΔODI). ΔVAS and ΔODI are defined as baseline minus postoperative values. Phenotypes: A (nonfrail/nonsevere), B (frail/nonsevere), C (nonfrail/severe), and D (frail/severe). LOS, length of stay; VAS, visual analogue scale; ODI, Oswestry Disability Index.
Table 1.Baseline characteristics by phenotype
Table 1.
|
Variable |
Phenotype A (n = 132) |
Phenotype B (n = 54) |
Phenotype C (n = 20) |
Phenotype D (n = 42) |
p-value |
|
Age (yr) |
79.03 ± 2.80 |
81.19 ± 3.71 |
79.30 ± 4.11 |
80.69 ± 3.63 |
0.001 |
|
BMI (kg/m²) |
26.03 ± 3.31 |
25.27 ± 3.47 |
21.24 ± 2.41 |
22.65 ± 2.63 |
0.001 |
|
Male sex |
59 (44.7) |
16 (29.6) |
6 (30.0) |
13 (31.0) |
0.136 |
|
Symptom duration (mo) |
52.5 (27.8–82.3) |
29.5 (17.3–90.3) |
79.5 (30.3–98.3) |
39.0 (12.8–95.8) |
0.124 |
|
VAS, baseline |
3.96 ± 1.90 |
4.65 ± 2.17 |
5.55 ± 1.79 |
5.64 ± 2.32 |
0.001 |
|
ODI, baseline |
32.46 ± 19.05 |
47.19 ± 14.15 |
38.90 ± 13.65 |
52.71 ± 16.60 |
0.001 |
|
Smoking |
15 (11.4) |
3 (5.6) |
2 (10.0) |
3 (7.1) |
0.612 |
|
Drinking |
8 (6.1) |
1 (1.9) |
1 (5.0) |
2 (4.8) |
0.688 |
|
Multiple drugs |
29 (22.0) |
9 (16.7) |
4 (20.0) |
6 (14.3) |
0.677 |
|
Hypoproteinemia |
33 (25.0) |
14 (25.9) |
4 (20.0) |
6 (14.3) |
0.482 |
|
Anemia |
36 (27.3) |
18 (33.3) |
3 (15.0) |
9 (21.4) |
0.356 |
|
Anxiety |
17 (12.9) |
7 (13.0) |
3 (15.0) |
2 (4.8) |
0.488 |
|
Depression |
23 (17.4) |
4 (7.4) |
3 (15.0) |
2 (4.8) |
0.093 |
|
Cognitive disorder |
95 (72.0) |
39 (72.2) |
14 (70.0) |
27 (64.3) |
0.802 |
|
Posterior paraspinal volume (mL) (L/R sums) |
219.27 ± 26.76 |
210.56 ± 18.31 |
183.13 ± 13.92 |
187.59 ± 21.88 |
0.001 |
|
Posterior paraspinal CSA (mm²) (L/R sums) |
10,316.78 ± 1,238.44 |
9,826.25 ± 762.09 |
8,595.80 ± 602.01 |
8,809.10 ± 942.33 |
0.001 |
|
Posterior paraspinal FI (MuscleMap combined) |
0.31 ± 0.04 |
0.35 ± 0.03 |
0.38 ± 0.03 |
0.38 ± 0.03 |
0.001 |
|
PFI |
-0.94 ± 1.38 |
0.22 ± 0.94 |
1.83 ± 0.58 |
1.80 ± 0.53 |
0.001 |
|
PFI (simple average MF+ES) |
0.32 ± 0.04 |
0.36 ± 0.04 |
0.38 ± 0.02 |
0.39 ± 0.03 |
0.001 |
|
Anterolateral FI (PM+QL) |
0.16 ± 0.03 |
0.18 ± 0.04 |
0.18 ± 0.04 |
0.21 ± 0.05 |
0.001 |
|
Fried frailty score (0–5) |
1 (0–2) |
3 (3–4) |
2 (1–2) |
4 (3–4) |
0.001 |
|
CCI |
|
|
|
|
0.018 |
|
0–1 |
81 (61.4) |
28 (51.9) |
18 (90.0) |
22 (52.4) |
|
|
≥2 |
51 (38.6) |
26 (48.1) |
2 (10.0) |
20 (47.6) |
|
|
ASA PS classification grade |
|
|
|
|
0.148 |
|
I |
- |
- |
1 (5.0) |
- |
|
|
II |
42 (31.8) |
15 (27.8) |
5 (25.0) |
13 (31.0) |
|
|
III |
87 (65.9) |
38 (70.4) |
14 (70.0) |
29 (69.0) |
|
|
IV |
3 (2.3) |
1 (1.9) |
- |
- |
|
Table 2.Postoperative outcomes by phenotype
Table 2.
|
Variable |
Phenotype A (n = 132) |
Phenotype B (n = 54) |
Phenotype C (n = 20) |
Phenotype D (n = 42) |
p-value |
|
Operative time (min) |
228.65 ± 86.88 |
224.70 ± 71.29 |
188.80 ± 60.64 |
220.57 ± 75.32 |
0.092 |
|
Estimated blood loss (mL) |
300.00 (115.00–500.00) |
260.00 (150.00–500.00) |
200.00 (132.50–500.00) |
200.00 (100.00–400.00) |
0.718 |
|
Fused levels |
2.14 ± 1.12 |
2.11 ± 1.21 |
1.75 ± 0.64 |
2.07 ± 0.92 |
0.164 |
|
LOS (day) |
8.00 (5.00–11.00) |
15.00 (12.25–17.00) |
14.00 (11.00–16.25) |
19.00 (16.00–21.75) |
0.001 |
|
VAS, postoperative |
0.91 ± 1.44 |
2.52 ± 2.23 |
2.85 ± 1.87 |
4.29 ± 2.37 |
0.001 |
|
ODI, postoperative |
10.46 ± 14.85 |
23.50 ± 17.08 |
13.25 ± 16.16 |
33.98 ± 21.64 |
0.001 |
|
ΔVAS, baseline−postoperative |
3.05 ± 1.28 |
2.13 ± 1.52 |
2.70 ± 1.30 |
1.36 ± 1.14 |
0.001 |
|
ΔODI, baseline−postoperative |
22.00 ± 11.83 |
23.69 ± 9.64 |
25.65 ± 8.58 |
18.74 ± 12.23 |
0.064 |
|
Prolonged LOS, ≥ cohort 75th percentile |
1 (0.8) |
21 (38.9) |
7 (35.0) |
33 (78.6) |
0.001 |
|
90-Day readmission |
22 (16.7) |
13 (24.1) |
4 (20.0) |
12 (28.6) |
0.349 |
|
Any complication, in-hospital |
24 (18.2) |
27 (50.0) |
11 (55.0) |
24 (57.1) |
0.001 |
|
Major complication |
19 (14.4) |
17 (31.5) |
5 (25.0) |
6 (14.3) |
0.050 |
|
Minor complication |
5 (3.8) |
10 (18.5) |
6 (30.0) |
18 (42.9) |
0.001 |
Table 3.Multivariable adjusted associations (reference: phenotype A)
Table 3.
|
Variable |
aOR (95% CI) |
p-value |
|
Any complication |
|
|
|
B vs. A |
5.034 (2.386–10.617) |
0.001 |
|
C vs. A |
5.329 (1.710–16.611) |
0.01 |
|
D vs. A |
6.537 (2.789–15.322) |
0.001 |
|
Prolonged LOS, ≥ 75th percentile |
|
|
|
B vs. A |
64.702 (12.821–326.517) |
0.001 |
|
C vs. A |
61.863 (8.793–435.254) |
0.001 |
|
D vs. A |
499.818 (80.496–3,103.491) |
0.001 |
|
Major complication |
|
|
|
B vs. A |
2.980 (1.337–6.646) |
0.01 |
|
C vs. A |
2.160 (0.611–7.640) |
0.232 |
|
D vs. A |
1.067 (0.375–3.035) |
0.903 |
|
Minor complication |
|
|
|
B vs. A |
5.923 (1.994–17.597) |
0.01 |
|
C vs. A |
10.578 (2.415–46.324) |
0.01 |
|
D vs. A |
19.879 (6.287–62.854) |
0.001 |
|
90-Day readmission |
|
|
|
B vs. A |
1.478 (0.649–3.369) |
0.352 |
|
C vs. A |
0.850 (0.210–3.438) |
0.820 |
|
D vs. A |
1.870 (0.753–4.648) |
0.178 |
|
ΔVAS |
|
|
|
B vs. A |
-1.024 (-1.490–-0.558) |
0.001 |
|
C vs. A |
-0.558 (-1.259–0.144) |
0.119 |
|
D vs. A |
-2.009 (-2.462–-1.556) |
0.001 |
|
ΔODI |
|
|
|
B vs. A |
1.163 (-2.349–4.675) |
0.516 |
|
C vs. A |
2.784 (-2.307–7.875) |
0.284 |
|
D vs. A |
-4.324 (-8.974–0.327) |
0.068 |
REFERENCES
- 1. Martin BI, Mirza SK, Spina N, et al. Trends in lumbar fusion procedure rates and associated hospital costs for degenerative spinal diseases in the United States, 2004 to 2015. Spine (Phila Pa 1976) 2019;44:369-76.
- 2. Kuo CC, Royse KE, Prentice HA, et al. Are octogenarians at higher risk of complications after elective lumbar spinal fusion surgery? Analysis of a cohort of 7880 patients from the Kaiser Permanente Spine Registry. Spine (Phila Pa 1976) 2022;47:1719-27.
- 3. Chen L, Ding J, Chen Z, et al. Rates, causes, and predictive factors of hospital readmissions after spine surgery for lumbar spinal stenosis: a nationwide retrospective cohort study. Neurospine 2025;22:523-39.
- 4. Fried LP, Tangen CM, Walston J, et al. Frailty in older adults: evidence for a phenotype. J Gerontol A Biol Sci Med Sci 2001;56:M146-56.
- 5. Xue QL. The frailty syndrome: definition and natural history. Clin Geriatr Med 2011;27:1-15.
- 6. Jang HJ, Chin DK, Park JY, et al. Influence of frailty on life expectancy in octogenarians after lumbar spine surgery. Neurospine 2021;18:303-10.
- 7. Baek W, Park SY, Kim Y. Impact of frailty on the outcomes of patients undergoing degenerative spine surgery: a systematic review and meta-analysis. BMC Geriatr 2023;23:771.
- 8. Veronesi F, Borsari V, Martini L, et al. The impact of frailty on spine surgery: systematic review on 10 years clinical studies. Aging Dis 2021;12:625-45.
- 9. Cruz-Jentoft AJ, Bahat G, Bauer J, et al. Sarcopenia: revised European consensus on definition and diagnosis. Age Ageing 2019;48:16-31.
- 10. Chen MJ, Lo YS, Lin CY, et al. Impact of sarcopenia on outcomes following lumbar spine surgery for degenerative disease: an updated systematic review and meta-analysis. Eur Spine J 2024;33:3369-80.
- 11. Han G, Wu H, Dai J, et al. Does paraspinal muscle morphometry predict functional status and re-operation after lumbar spinal surgery? A systematic review and meta-analysis. Eur Radiol 2023;33:5269-81.
- 12. Sun K, Zhu H, Huang B, et al. MRI-based central sarcopenia negatively impacts the therapeutic effectiveness of single-segment lumbar fusion surgery in the elderly. Sci Rep 2024;14:5043.
- 13. Chiapparelli E, Caffard T, Medina SJ, et al. Relationship between posterior paraspinal muscle fat infiltration and early conversion to lumbar spinal fusion following decompression surgery. Eur Spine J 2025;34:3031-7.
- 14. Wesselink EO, Elliott JM, Coppieters MW, et al. Convolutional neural networks for the automatic segmentation of lumbar paraspinal muscles in people with low back pain. Sci Rep 2022;12:13485.
- 15. Kim HB, Kim HS, Kim SJ, et al. Spine muscle auto segmentation techniques in MRI imaging: a systematic review. BMC Musculoskelet Disord 2024;25:716.
- 16. McKay MJ, Weber KA, Wesselink EO, et al. MuscleMap: an open-source, community-supported consortium for whole-body quantitative MRI of muscle. J Imaging 2024;10:262.
- 17. Ghannam AD, Crandall ML, Woodruff G, et al. National surgical quality improvement program adverse events combined with Clavien-Dindo scores can direct quality improvement processes in surgical patients. J Patient Saf 2022;18:e900-2.
- 18. Giordan E, Drago G, Zanata R, et al. The correlation between paraspinal muscular morphology, spinopelvic parameters, and back pain: a comparative cohort study. Int J Spine Surg 2023;17:627-37.
- 19. Koivunen K, Arokoski J, Widbom-Kolhanen S, et al. Disability and pain after lumbar surgery-group-based trajectory analysis. PLoS One 2025;20:e0313528.
- 20. Latack KR, Howard R, Bicket MC, et al. Frailty status, not just age, is associated with postoperative opioid consumption: a retrospective, population-based analysis. Ann Surg Open 2024;5:e496.
- 21. Baaj AA, Lang G, Hsu WC, et al. 90-day readmission after lumbar spinal fusion surgery in new york state between 2005 and 2014: a 10-year analysis of a statewide cohort. Spine (Phila Pa 1976) 2017;42:1706-16.
- 22. Debono B, Wainwright TW, Wang MY, et al. Consensus statement for perioperative care in lumbar spinal fusion: Enhanced Recovery After Surgery (ERAS®) Society recommendations. Spine J 2021;21:729-52.
- 23. Volkert D, Beck AM, Cederholm T, et al. ESPEN practical guideline: clinical nutrition and hydration in geriatrics. Clin Nutr 2022;41:958-89.
, Xiangyu Li1,2
