Leiomyosarcoma (LMS) is a common subtype of soft tissue sarcoma (STS) derived from smooth muscle. Generally, incidence of LMS is greatest in patients 70 and older [1]. LMS often presents as an asymptomatic, painless mass, and comprises 20–25% of all STS and appears anywhere smooth muscle is present [2]. The retroperitoneum and intra-abdominal regions encompass 35% of cases, the uterus approximately 30%, and the extremities about 10–20% [3]. Because the mainstay of treatment of uterine LMS is en bloc total hysterectomy, the focus of orthopedic oncology lies primarily in management of extremity LMS [4]. 5-year overall survival (OS) of LMS of the extremities is highly varied ranging from 30-75% depending on multiple factors including age, tumor grade, size, and excision margins [3,[5], [6], [7], [8], [9]]. Treatment involves surgical resection with wide negative margins often with radiotherapy [10].

Determining patient prognosis is key to choosing patient treatment. Previous work has identified prognostic factors in attempts to determine patient outcomes including age, primary site, stage, histologic grade, surgical margins, and tumor depth [2,10,11]. Nomograms are predictive models designed to leverage these prognostic factors to calculate event probability and have been utilized in LMS to determine patient outcomes [12]. Xue et al. created a nomogram predicting 5- and 10- year overall survival and cancer-specific survival in extremity LMS, utilizing age, tumor grade, distant metastasis, tumor size, and lack of surgery as predictor variables, and Zhuang et al. developed a nomogram to predict the 1-, 2-, and 5-year overall survival of retroperitoneal LMS patients [13,14]. Despite the historic predictive utility of nomograms, machine learning has since been shown to provide similar if not greater predictive power for outcomes [15].

In recent years, machine learning (ML), a subtype of artificial intelligence, has enhanced predictive capabilities across the medical field. Whereas nomograms utilize traditional statistical methods, ML is an iterative process that recognizes complex interactions and patterns among variables. This new form of analysis provides greater accuracy than traditional methods [16]. Applications of ML are broad within the medical field and an burgeoning area of development [[17], [18], [19], [20]].

To our knowledge, no ML approach has been utilized for survival prognostication in extremity LMS specifically. Our study seeks to expand upon the work by Xue et al. in predicting extremity LMS mortality rates. The aim of this work uses ML to develop algorithms predicting 1-, 3-, and 5-year mortality rates for patients diagnosed with extremity LMS using the SEER database. The best performing algorithm was then externally validated using our institution's LMS cohort.