Enrico Martin

29 Treatment and survival differences across tumor sites Covariates Covariates extracted for analysis were: sex, age (≤18, 19-59, and ≥60 years), race (White, Black, Asian & other), tumor site, SEER tumor grade (I-IV), tumor size, extent of resection, administration of radiotherapy, timing of radiotherapy to surgery (prior to, after, during, prior to and after surgery), and survival. Tumor sites were recoded using ICD-O-3 site codes into: intracranial, spinal, head and neck, limbs, core (including chest, abdomen, and pelvis), and NOS (not otherwise specified or unknown, Supplemental Table 1 ). In the SEER database, tumor size is determined from pathological reports, or from radiologic reports in case of preoperative treatment, unclear pathological reports, or in case no surgery was performed. Surgical procedures were coded differently in the SEER before and after 1998 and extent of resection can be interpreted from them. A single variable was constructed using codes prior to 1998 and after 1998 to evaluate extent of surgical resection from all time periods. These were recoded into the following subgroups: no surgery, biopsy, partial resection, gross total resection, surgery not otherwise specified, and unknown status of surgery ( Supplemental Table 2 ). Statistical analysis Data were stratified per tumor site and descriptive statistics were performed on demographics. Only primary tumors were used for survival analyses. Univariable and multivariable Cox proportional hazard analyses were performed for each tumor site to evaluate possible factors of influence on overall survival (OS). Subsequently, a univariable and three multivariable Cox proportional hazard models were constructed for all primary MPNSTs combined with tumor site as a separate variable. These three models were separated to appraise influences of different therapy regimens on overall survival and avoid correlation among variables included. P-values <0.05 were considered statistically significant. Bonferroni correction was applied to correct for multiple testing. Kaplan-Meier survival curves for overall survival and disease-specific survival (DSS) were constructed for MPNSTs per site. Statistical analyses were conducted using IBM ® (Armonk, NY) Statistical Package for the Social Sciences (SPSS) ® version 24 (IBM Inc., 2016) and Kaplan-Meier curves were created using R version 3.3.3 (R Core Team, 2017). Results Patient population 3267 patients with MPNSTs were identified in the SEER database: 167 intracranial (5.1%), 119 spinal (3.6%), 449 head and neck (13.7%), 1022 limb (31.3%), 1307 core (40.0%), and 203 NOS & unknown (6.2%, Table 1 ). The mean age was 47.6 years (SD: 21.0). The majority of patients were male (54.1%) and white (78.9%). Most patients were only treated surgically (46.8%), with a combination of surgery and radiation being the second most common treatment strategy (32.8%). 53.8% were of unknown tumor grade. Most often tumors were classified as grade IV (16.8%) and the median size of all tumors was 2