More help than harm: surgery for metastatic spinal cord compression is associated with more favorable overall survival within a propensity score analysis

This study was approved by the local Institutional Ethical Review Board (IRB) (Reference number: EK-Nr. 30-252 ex 17/18) and performed in accordance with relevant guidelines and regulations; informed consent for later retrospective analysis of their data was obtained from all participants.

Patients with symptomatic MSCC with neurologic impairment (n = 398) and timepoint of onset of symptom were identified from our data system. A decision (A) decompressive surgery with dorsal stabilization and subsequent recommendation of radiation (B) no surgery, recommendation of radiation was documented in all cases. According to our clinic internal protocol, this decision was based on the calculated mBs, time since onset of neurologic impairment, general health state, and patients’ consent to surgery. Extent of surgical intervention was performed according to mBs recommendation in most cases [6]. Radiation was recommended in all MSCC cases, if residual radiation dose was available for the region involved.

All statistical analyses were performed by FP with Stata (Windows version 17.0, Stata Corp., Houston, TX, USA). Continuous variables were reported as medians [25th–75th percentile] and count data as absolute frequencies (column %). The distribution of variables between two or more other variables was assessed with rank-sum tests, χ²-tests, and Fisher’s exact tests, as appropriate. Risks of death-from-any-cause (overall survival) and progression-free survival were estimated with 1-Kaplan–Meier estimators, compared between two or more groups with log-rank tests, and modeled with uni- and multivariable Cox regression models. Risks of radiographic disease progression were estimated with competing risk cumulative incidence estimators accounting for death-from-any-cause as the competing event of interest, compared between two or more groups with Gray’s test, and modeled with Fine and Gray competing risk regression models. Predictors of spine surgery were examined with logistic regression. The magnitude of difference in the distribution of variables between patients with and without spine surgery was quantified with standardized mean differences (SMDs). Propensity score analyses were performed with inverse-probability-of-treatment-weights (IPTW), and balance diagnostics were performed as previously described. The propensity score model was developed on multiply imputed data (chained equations algorithm, 10 imputation datasets, separate imputation models for continuous, binary, and ordinal variables) including all ten available variables of the dataset, whereas other analyses were done according to the complete-case principle. The full analysis code is available on reasonable request from FP (florian.posch@medunigraz.at).

Of 1778 referrals to our department for orthopedic evaluation of radiologically confirmed spinal metastases, 398 patients had symptomatic MSCC with neurologic impairment (Table 1). Median age was 66 years [25th–75th percentile: 56–74], and 144 patients (36%) were female. The most frequent tumor entities were prostate, lung, and breast cancer. During a median follow-up of 5.8 years according to the reverse Kaplan–Meier estimator and 0.5 years according to the median of observation times, 355 patients (89%) died. This corresponded to 3-month, 6-month, 12-month, and 3-year all-cause mortality of 34% (95%CI 30–39), 51% (95%CI 46–56), 64% (95%CI 59–69), and 85% (95%CI 81–88), respectively (Supplementary Fig. 2).

One-hundred-ninety-four patients (49%) were treated with spine surgery, whereas the remaining 204 patients were treated non-surgically. The most important predictor of spine surgery was the mBs; twenty-eight percent, 34%, 52%, 64%, and 81% of patients with mBs of 0, 1, 2, 3, and 4 points were treated with surgery, respectively (p < 0.0001). Other predictors of spine surgery were younger age, a pathologic fracture, and selected tumor entities (e.g., breast cancer, renal cancer, endocrine cancer), whereas brain metastases or a new metastatic site in the 3 prior months were associated with lower odds of spine surgery (Table 2).

A low mBs emerged as the strongest predictor for an adverse survival experience. In detail, 6-month mortality estimates were 82%, 75%, 46%, 32%, and 0% in patients with mBs of 0, 1, 2, 3, and 4 points, respectively (log-rank p < 0.0001, Fig. 1). In further univariable analysis, factors associated with more favorable survival were younger age, selected tumor entities (e.g., breast cancer, sarcoma, and renal cell cancer), no brain metastases, no new metastatic sites within the preceding 3 months, and treatment with spine surgery (Table 3).

Patients who were treated with spine surgery had more favorable survival outcomes than patients being treated non-surgically (6-month mortality: 46% vs. 81%, log-rank p < 0.0001, Hazard ratio (HR) = 0.43, 95%CI 0.34–0.54, p < 0.0001, Fig. 2A). However, spine surgery was associated with many favorable prognostic factors for survival such as a higher mBs (Table 1). The association between spine surgery and lower mortality prevailed after accounting for the mBs in stratified analysis (stratified log-rank p < 0.0001) as well as multivariable analysis (HR for all-cause death for spine surgery adjusted for the mBs = 0.57, 0.45–0.72, p < 0.0001). Nonetheless, we still anticipated bias in comparing outcomes between patients with and without spine surgery due to the non-random indication of this procedure in our retrospective study. To control this anticipated selection bias, we performed a propensity score analysis. The propensity score model (Supplementary Table 1) included all ten variables available to us, and the resulting propensity score covered the whole probability range from 0 to 1 (Supplementary Fig. 3A). The resulting IPTW (Supplementary Fig. 3B) led to a considerable reduction in standardized mean differences and statistical significances between patients who were treated with and without spine surgery including in the most important prognostic variables such as the mBs and brain metastases (Supplementary Fig. 4, Table 1). In IPTW-analysis, spine surgery was still associated with more favorable survival outcomes (IPTW-adjusted log-rank p = 0.021, Fig. 2B, IPTW-Adjusted HR for death with spine surgery = 0.68, 0.49–0.94, p = 0.021).

In subanalysis, spine surgery appeared to be associated with favorable overall survival across all mBs groups except in those patients with a mBs of 0 points (p for interaction = 0.058, Fig. 3). Also, in patients with a mBs of 1 point (n = 91), patients who were treated with spine surgery (n = 31) experienced a more favorable OS prognosis than patients not treated with spine surgery (1-year mortality: 63% vs. 92%, log-rank p = 0.0003, Fig. 4). This suggested that there may be a subgroup within this mBs group that may benefit from spine surgery. To further delineate those patients, we explored predictors of OS in patients with an mBs of 1 point (Table 4). Here, lower age, female sex, non-genitourinary malignancies, and no new metastatic sites within the preceding 3 months emerged as variables for delineating patients with an mBs of 1 point and a more favorable OS prognosis.

One-hundred-and-five patients experienced radiographic disease progression after spine surgery. The risk of radiographic disease progression was significantly higher in patients undergoing spine surgery than in patients treated without spine surgery (6-month competing risk cumulative incidences: 20.9% vs. 11.8%, Gray’s test p = 0.016, Supplementary Fig. 5A), and this association weakened after adjusting for the mBs (Subdistribution hazard ratio (SHR) for radiographic disease progression with spine surgery adjusted for mBs = 1.51, 0.86–2.64, p = 0.151). Given the more favorable OS with spine surgery reported above, this association likely indicated reverse causality. We therefore performed a joint analysis of radiographic progression and mortality [i.e., progression-free survival (PFS)], in which spine surgery was associated with more favorable PFS (6-month risk of radiographic progression or death: 46.4% vs. 74.9%, log-rank p < 0.0001, HR for radiographic progression or death with spine surgery = 0.45, 0.34–0.58, p < 0.0001, Supplementary Fig. 5B).

A subset of 71% of our cohort (n = 281) had data on short-term neurologic outcome available (described in the Methods section). Here, any neurologic improvement was seen in 126 (75%) of the 168 patients undergoing spine surgery, and 21 (19%) of the 113 patients treated without spine surgery (χ² p < 0.0001). Spine surgery and the mBs were the strongest determinants of neurologic improvement (Fig. 5, Supplementary Table 2).