Introduction

Among metabolic diseases, type 2 diabetes mellitus (T2DM) ranks as one of the most widespread worldwide, currently influencing roughly one in ten adults and an estimated 828 million people [1–3]. Its complications extend beyond glucose metabolism and include detrimental effects on bone quality, angiogenesis, and delaying wound healing [4–7]. Sustained hyperglycemia facilitates the accumulation of advanced glycation end-products, inhibits osteoblast function, and induces microvascular impairment, thereby delaying bone regeneration and reducing fusion stability following spinal procedures [8,9]. Consequently, the diabetic cohort showed an increased incidence of surgical complications, including delayed union and pseudarthrosis, accompanied by higher overall healthcare costs driven by prolonged inpatient care after spinal fusion [10–12]. As the prevalence of both diabetes and spinal degenerative diseases continues to rise [13,14], understanding the interaction between these conditions has become increasingly relevant to surgical decision-making and postoperative management.

Most prior research has focused on the association between diabetes and fusion outcomes in open and minimally invasive lumbar fusion surgery [15,16]. These studies consistently show that T2DM is associated with inferior fusion rates and higher reoperation risk [17,18]. However, its influence on endoscopic lumbar interbody fusion (Endo-LIF) performed through a posterior approach has not been clearly defined. Throughout this report Endo-LIF refers to single-level procedures recorded as percutaneous endoscopic interbody fusion approached from behind, and is used as a descriptive label for the operation as recorded in administrative coding rather than as a designation of a specific endoscopic technique. In contrast to open or minimally invasive approaches, Endo-LIF provides direct endoscopic visualization with minimal soft-tissue trauma and smaller incisions. These features support a faster recovery and reflect the current trend toward less invasive, anatomy-preserving spine surgery [19–21]. Whether T2DM is associated with poorer long-term fusion outcomes among patients undergoing this operation has not been examined.

This retrospective cohort analysis explored the association between T2DM and outcomes after single-level Endo-LIF, with attention to perioperative complications, medical resource use, and fusion success. Clarifying this association may help identify at-risk populations and guide perioperative management strategies.

Materials and methods

Patient selection

Adult patients who underwent single-level Endo-LIF between January 1, 2015, and January 1, 2025, were identified within the TriNetX Global Collaborative Network (n = 15,994). Patient identification was performed using ICD-10-PCS procedure codes for single-level percutaneous endoscopic lumbar interbody fusion performed through a posterior approach (0SG047J, 0SG04KJ, 0SG04JJ, 0SG04AJ, and 0SG04ZJ). These codes identify the cohort by coded operative approach only: they do not distinguish uniportal full-endoscopic from biportal endoscopic technique, nor a posterior interbody fusion from a transforaminal endoscopic variant, and Endo-LIF is therefore used throughout as a code-defined descriptor rather than a technical designation.

Patients with T2DM were defined as those who met both of the following criteria:

1. (a) presence of an ICD-10-CM diagnostic code for T2DM (E11) to exclude other types of diabetes (E08, E09, E10, E12, and E13), and
2. (b) a documented HbA1c ≥ 6.5% recorded at any time prior to surgery.

This combined definition was applied to ensure diagnostic accuracy and consistency with the American Diabetes Association (ADA) Standards of Medical Care in Diabetes 2025 [23].

Patients were excluded if they were younger than 18 years or had diagnostic codes indicating lumbar fracture (S32), malignant bone tumor (C41), scoliosis (M41), congenital spinal deformity (Q76), spinal infection and inflammatory spondylopathy (M46), or type 1 diabetes (E10). After the inclusion and exclusion criteria were applied, the database query returned 5,401 patients (894 with T2DM and 4,507 without diabetes). Propensity-score matching on the TriNetX platform is performed on complete cases; 562 patients (112 with T2DM and 450 without diabetes) lacked a recorded value for at least one matching covariate, were therefore not eligible for matching, and were excluded at this step. The analytic cohort comprised 4,839 patients, of whom 782 had T2DM and 4,057 had no diabetes. Propensity-score matching (PSM) at a 1:1 ratio yielded two cohorts of 755 patients each, which were used for the final comparison (Fig 1).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Flowchart patient selection.

https://doi.org/10.1371/journal.pone.0358422.g001

Results

Baseline characteristics

In total, 4,839 individuals satisfied the study’s eligibility criteria and had complete data on the matching covariates, of whom 782 had T2DM and 4,057 did not. Following 1:1 propensity-score matching, 755 patients were retained in each group for comparison.

Prior to matching, participants with T2DM were generally older and exhibited greater prevalence of hypertension, hyperlipidemia, ischemic heart disease, chronic kidney disease, and osteoporosis (all p < 0.05). Post-matching, baseline characteristics demonstrated satisfactory comparability, with standardized mean differences (SMDs) below 0.10 across all variables (Table 1).

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Patient demographics and comorbidities for Type 2 diabetes mellitus and No diabetes cohorts before and after propensity score matching.

https://doi.org/10.1371/journal.pone.0358422.t001

Early postoperative complications

After matching, the two cohorts showed comparable rates of pneumonia, urinary tract infection, anemia, blood transfusion, and pressure ulcer within 90 days of surgery, and none of these differences reached statistical significance (all p > 0.05; Table 2).

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Early postoperative complications (within 90 days after surgery).

https://doi.org/10.1371/journal.pone.0358422.t002

Surgical site infection and postoperative mortality were not analyzed because each event occurred in fewer than ten individuals per group, consistent with TriNetX privacy rules that restrict reporting of very small cell counts.

Overall, short-term complication rates after Endo-LIF were similar in patients with and without type 2 diabetes mellitus, with no statistically significant differences observed.

Healthcare utilization

During the first postoperative year, the rates of emergency department visits and unplanned hospital readmissions were similar between the T2DM and non-diabetes groups, with no significant differences observed in either outcome (both p > 0.05; Table 3). Short-term healthcare-utilization rates after Endo-LIF were therefore similar between the two cohorts, with no statistically significant differences observed.

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Healthcare utilization outcomes (3-, 6- and 12-month follow-up).

https://doi.org/10.1371/journal.pone.0358422.t003

Code-defined pseudarthrosis

Across all follow-up intervals, code-defined pseudarthrosis was consistently more frequent among patients with T2DM than among those without diabetes. By 6 months the event had been recorded in 73 individuals (9.7%) with T2DM compared with 40 (5.3%) in the non-diabetic group (p = 0.001, RR = 1.83, 95% CI 1.26–2.65). At 12 months, the corresponding rates were 10.1% versus 5.7% (p = 0.002, RR = 1.77, 95% CI 1.23–2.53), and by 36 months, 11.5% versus 6.2% (p < 0.001, RR = 1.85, 95% CI 1.32–2.60) (Table 4). These values are observed proportions and are not adjusted for censoring. The corresponding Kaplan–Meier estimates of cumulative incidence were 9.7% versus 5.3% at 6 months, 10.2% versus 5.8% at 12 months, 11.3% versus 6.1% at 24 months, and 12.0% (95% CI 9.8–14.6) versus 6.4% (95% CI 4.9–8.5) at 36 months (Table 5). The censoring-adjusted estimates were numerically close to the observed proportions, differing by no more than 0.5 percentage points.

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Code-defined pseudarthrosis: observed event proportions (6-, 12- and 36-month follow-up).

https://doi.org/10.1371/journal.pone.0358422.t004

Download:

• PNG
  larger image
• TIFF
  original image

Table 5. Kaplan–Meier estimates of the cumulative incidence of code-defined pseudarthrosis (6-, 12-, 24- and 36-month follow-up).

https://doi.org/10.1371/journal.pone.0358422.t005

Kaplan–Meier analysis revealed a significantly higher cumulative incidence of code-defined pseudarthrosis in the diabetic cohort relative to non-diabetic controls (log-rank χ² = 13.09, p = 0.0003), and a Cox proportional hazards model yielded a hazard ratio of 1.90 (95% CI 1.33–2.71) (Fig 2). In absolute terms, the censoring-adjusted difference in cumulative incidence at 36 months was 5.5% (12.0% versus 6.4%; 95% CI 2.6% to 8.5%); the difference calculated from observed proportions was 5.3% (95% CI 2.4% to 8.2%). The proportional hazards assumption was satisfied (proportionality test χ² = 0.031, df = 1, p = 0.86), and the complementary log–log curves of the two cohorts were approximately parallel across follow-up (S1 Fig). The analytics platform does not provide patient-level follow-up times and does not export the number of patients remaining at risk at each time point; median follow-up duration and time-specific risk-set counts could therefore not be reported, and a number-at-risk row could not be placed beneath Fig 2.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Kaplan–Meier curves showing the cumulative incidence of code-defined pseudarthrosis within 3 years after single-level Endo-LIF.

Estimates account for censoring; a number-at-risk row could not be added because the analytics platform does not export the number of patients at risk at each time point.

https://doi.org/10.1371/journal.pone.0358422.g002

These findings indicate that type 2 diabetes mellitus was associated with an approximately two-fold higher long-term incidence of code-defined pseudarthrosis following single-level Endo-LIF, despite comparable perioperative complication rates between cohorts.

Discussion

This multicenter cohort study found that patients with T2DM had a higher incidence of code-defined pseudarthrosis after single-level Endo-LIF. Early postoperative complications and short-term healthcare use were comparable between diabetic and non-diabetic individuals, yet the cumulative incidence of code-defined pseudarthrosis remained higher in the T2DM group throughout three years of follow-up. Similar early outcomes were observed in the two cohorts, whereas the long-term outcome differed. Because the study included no comparator surgical approach, these similar early outcomes cannot be attributed to features of the endoscopic technique itself. In absolute terms the difference was moderate: the Kaplan–Meier cumulative incidence at 36 months was 12.0% versus 6.4%, an absolute risk difference of 5.5% (95% CI 2.6% to 8.5%). The near two-fold relative estimate should therefore be interpreted alongside this absolute difference when counseling patients before surgery. Both measures describe a comparison between exposure groups and should not be read as the expected consequence of an intervention, since T2DM is an observed patient characteristic rather than something assigned.

Our results are consistent with earlier reports in open and minimally invasive lumbar fusion, which have shown impaired bone healing and greater nonunion rates among patients with diabetes [10,28]. Prior studies frequently reported higher perioperative medical complications in diabetic individuals [15,29], a pattern not seen in our matched analysis. Lower tissue disruption has been described for endoscopic approaches [30], but because this study included no comparator surgical approach, the similar early profiles observed in the two cohorts cannot be attributed to features of the endoscopic technique, and they need not extend to the biological processes that determine fusion integrity.

Several metabolic factors associated with chronic hyperglycemia could plausibly contribute to the observed association with code-defined pseudarthrosis. Reduced osteoblast activity, impaired microvascular perfusion, and changes in collagen quality have been implicated in delayed bone remodeling [31,32]. Although these mechanisms were not directly evaluated in our dataset, they align with the persistent association found after adjusting for demographic and clinical variables.

Few studies have examined diabetes-specific outcomes after endoscopic lumbar fusion. By focusing on a population undergoing single-level Endo-LIF as defined by procedure coding, our findings suggest that systemic metabolic health may be associated with long-term fusion outcomes in this population. Because no comparator approach was studied, we cannot say whether this pattern differs from that seen after open or minimally invasive fusion.

Other patient-level factors not captured in claims data may also contribute; for example, lower preoperative muscle strength has been associated with poorer functional outcomes after lumbar interbody fusion [33]. Factors of this kind may be related both to diabetes status and to recovery after surgery, were unavailable for matching, and may therefore contribute to residual confounding.

The timing of the coded events also deserves comment. Most of them appeared within the first six months, whereas radiographic fusion usually takes twelve to twenty-four months to mature. A diagnosis of nonunion at six months is more consistent with delayed union, and because the outcome rests on a diagnosis code, some early entries may reflect transient axial pain or a slow but progressing fusion rather than established nonunion. This supports interpreting the endpoint as code-defined pseudarthrosis rather than radiographically confirmed nonunion. The pattern can be described more precisely. Approximately 84% and 85% of events were recorded within the first six months in the T2DM and non-diabetes cohorts respectively (73 of 87 and 40 of 47). A further 14 and 7 coded events were recorded after six months in the T2DM and non-diabetes cohorts respectively; without the corresponding risk sets, however, these counts cannot support a formal comparison of late incidence. The same pattern governs the absolute difference between the cohorts, which is already 4.4% at six months and rises only to 5.5% by thirty-six months (Table 5). Approximately four-fifths of the three-year absolute difference had therefore already emerged by six months, that is, within the interval in which a coded nonunion is more consistent with delayed union than with established nonunion. The association reported here rests largely on this early coding and should not be read as evidence that T2DM impairs fusion progressively across three years of follow-up. A landmark or sensitivity analysis confined to events first coded after 6 or 12 months would be the appropriate way to separate delayed union from established nonunion, and we regard this as a priority for prospective work in which fusion can be confirmed radiographically.

The coded rate of pressure ulcers (7.6% and 8.6%) was higher than expected after a single-level minimally invasive procedure. It may reflect coding of lesions present before admission, or lesions related to a patient’s care setting such as admission from institutional care, rather than an acute complication of the index operation; notably, the rate was similar in both groups. This pattern illustrates the ascertainment limitation that applies to the primary endpoint: a diagnosis code recorded after the index procedure need not denote a condition arising from it. The same reasoning applies to M96.0 and is a further reason for describing the outcome as code-defined pseudarthrosis throughout.

Several limitations should be acknowledged. First, pseudarthrosis was defined from a diagnosis code rather than confirmed on imaging [34], so the endpoint is code-defined and may misclassify true radiographic nonunion. Second, ICD-10-PCS establishes that the procedure was a single-level percutaneous endoscopic interbody fusion performed from a posterior route, but records nothing further about how it was performed: it does not separate uniportal full-endoscopic from biportal endoscopic technique, nor a posterior interbody fusion from an endoscopic transforaminal variant, and it does not reliably record the graft or implant used. The cohort therefore cannot be characterized at the level of surgical technique, and the findings should be regarded as preliminary. Third, factors that strongly influence fusion, including cage and graft type, BMP use, fixation, endplate preparation, bone quality, surgeon experience, and perioperative bone-active medication such as teriparatide, romosozumab or antiresorptive agents, are not recorded in the database and could not enter the matching model, so part of the observed difference may reflect these rather than diabetes itself. Fourth, HbA1c was available mainly for diagnosis and perioperative values were too sparse to stratify, so the study addresses the presence of T2DM rather than glycemic control or disease duration. Fifth, the study period spanned 2015–2025, over which endoscopic fusion technique, implants, graft strategies and perioperative diabetes management all changed substantially. Year of surgery was not included in the propensity-score model and the distribution of index year across the two cohorts could not be extracted, so secular change in practice remains an unmeasured potential confounder. Sixth, because the outcome is ascertained only through diagnosis coding, differences in follow-up intensity may contribute to the association: patients with T2DM may have more frequent healthcare encounters and therefore more opportunities to receive an M96.0 code, independently of true fusion status. In addition, the index window extended to January 1, 2025 while the data were accessed on July 7, 2025, so patients treated late in the period could not contribute a full three years of potential follow-up; Kaplan–Meier estimates are reported for this reason, but the number of patients remaining at risk at each time point could not be exported from the platform. Finally, the platform compares matched cohorts as independent samples and does not permit patient-level export, so the paired structure could not be modeled directly, and because the network is continuously updated the exact cohort sizes reported here would not be reproduced by an identical query run at a later date. Prospective studies with imaging-confirmed fusion and recorded surgical and metabolic detail are needed to clarify how specific features of diabetes relate to fusion outcomes.

Despite these limitations, the study has several strengths. The large multicenter dataset improves generalizability, and standardized data extraction helps reduce inter-institutional variability. Propensity score matching reduced baseline imbalance, and limiting the cohort to single-level procedures reduced heterogeneity in the number of levels fused, although the surgical technique itself could not be characterized. Taken together, these findings indicate that although perioperative outcomes were similar between the two cohorts, T2DM remained associated with the long-term outcome.

Conclusion

In this database cohort, T2DM was associated with a higher rate of code-defined pseudarthrosis after single-level Endo-LIF, while short-term complications and healthcare use were similar in the two groups. Because the outcome was defined by a diagnosis code rather than imaging, the operation could not be characterized by technique, and important surgical variables were unavailable, these preliminary findings should be interpreted as an association and do not establish a causal effect of T2DM on radiographic nonunion. They are consistent with, but do not demonstrate, a contribution of systemic metabolic factors to long-term fusion, and require confirmation in prospective studies that verify fusion on imaging and record technique and graft detail.

Supporting information

Acknowledgments

The authors acknowledge the Can Tho University of Medicine and Pharmacy for institutional support for English editing assistance.