Comparison of Cox Regression to Machine Learning in Predicting Cancer-Specific Survival of Fibroblastic Osteosarcoma

Comparison of Cox Regression to Machine Learning in Predicting Cancer-Specific Survival of Fibroblastic Osteosarcoma | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Comparison of Cox Regression to Machine Learning in Predicting Cancer-Specific Survival of Fibroblastic Osteosarcoma Longteng Chao, Xinmiao Ye, Junyuan Chen, Guorong She, Zhengang Zha This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-3839137/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background Bone cancer called osteosarcoma (OS), especially its fibroblastic type, makes things very hard in the world of bone diseases. This happens because of its fierce character and the complexity involved in deciding outcomes. Current prognostic models, like the American Joint Committee on Cancer (AJCC) system and Tumor Node Metastasis (TNM) Staging System, don't always fully include important individual patient factors such as age, sex and race. These things are very important for making a correct prediction. Methods A total of 394 patients with fibroblastic osteosarcoma were included in the study, adhering to specified inclusion and exclusion criteria. The cohort was subsequently segregated into training and validation sets at a 7:3 ratio. X-tile software facilitated the determination of optimal age and tumor size cutoffs. Missing data were managed using multiple imputation and K-Nearest Neighbor (KNN) methods. The primary endpoint was cancer-specific survival (CSS), categorized into binary data (survival status at 3 and 5 years) and time-to-event data. Independent prognostic factors were ascertained using the Boruta algorithm, which informed the construction of predictive models employing Cox regression and diverse machine learning algorithms such as Survival Tree, Extra Survival Trees, Random Survival Forest, Gradient Boosting Survival Analysis, Fast Kernel Survival SVM, and Minlip Survival Analysis. Model performance metrics included the concordance index (C-index), accuracy, recall, F1 score, and time-dependent Area Under the Curve (AUC). A calibration plot was generated to validate the accuracy of the most proficient machine learning model. Decision curve analysis (DCA) was implemented to ascertain the model's clinical utility. Additionally, we used the SHapley Additive exPlanations (SHAP) method to show how important our model found key things that can predict outcomes. Results For age, the determined optimal cutoff points were established at 40 and 57 years. Regarding tumor size, these points were set at 60mm and 103mm. Our study identified nine significant independent prognostic factors impacting the cancer-specific survival in patients with fibroblastic osteosarcoma. These included age group, tumor stage, tumor size group, radiation, surgery type, primary site, sex, chemotherapy, and grade group. Comparative analysis of different algorithms, utilizing metrics such as accuracy, recall, F1 score, C-index, and time-dependent AUC, highlighted the Extra Survival Trees model as the superior predictive tool for machine learning. This model demonstrated high efficiency (3-year CSS accuracy: 0.91, 5-year CSS accuracy: 0.89), notable recall rates (3-year: 0.81, 5-year: 0.74), and robust F1 scores (3-year: 0.83, 5-year: 0.80), along with an average AUC of 0.89 and a C-index of 0.92 for training and 0.80 for validation. The calibration curve for this model indicated high predictive accuracy, and its clinical usefulness was further corroborated by decision curve analysis (DCA). SHAP analysis identified 'age group', 'tumor stage', and 'tumor size group' as the three most influential variables impacting cancer-specific survival predictions in fibroblastic osteosarcoma. Our study suggested otherwise than previous ones. It showed that radiation and chemotherapy may not work for treating this type of bone cancer called fibroblastic osteosarcoma. Conclusion Research indicates that predictive analysis using machine learning outperforms traditional methods in forecasting outcomes for patients with fibroblastic osteosarcoma. This development offers considerable promise for enhancing tailored therapeutic approaches and prognostic outcomes in fibroblastic osteosarcoma. Fibroblastic Osteosarcoma Cancer-Specific Survival Machine Learning Cox Regression SHAP Analysis Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Introduction Osteosarcoma (OS) represents the most prevalent malignant sarcoma of the skeletal system, primarily affecting children and adolescents, thus constituting the foremost bone cancer in these demographic groups [ 1 ]. The manifestation of osteosarcoma is marked by a variety of characteristics, influenced by aspects such as tumor size, gene expression patterns, treatment responses, and histological subtypes. Fibroblastic osteosarcoma, a particular histological subtype of conventional osteosarcoma, accounts for approximately 10% of cases. This subtype is typified by spindle-shaped malignant cells, which occasionally exhibit epithelioid traits. While these cells frequently present pronounced cytological atypia, this manifestation is not universally observed. The unique cellular attributes of fibroblastic osteosarcoma delineate it as an aggressive tumor variant, often correlating with a worse prognosis for patients [ 2 , 3 ]. Identifying risk factors for early mortality in patients with fibroblastic osteosarcoma is, therefore, of paramount importance. Fibroblastic osteosarcoma, as a distinct subtype, presents significant challenges in the development of precise prognostic predictive models. These challenges primarily stem from difficulties in obtaining adequate case samples and the lack of reliable biomarkers. Furthermore, various studies have highlighted the limitations of the American Joint Committee on Cancer (AJCC) staging system in accurately predicting outcomes for osteosarcoma patients [ 4 ]. The Tumor, Node, Metastasis (TNM) Staging System, despite its focus on tumor size, regional lymph node involvement, and distant metastasis, does not adequately consider critical individual factors such as patient age, which are essential for a comprehensive prognosis assessment [ 5 ]. Wei Zhong's study explored prognosis and prognostic factors in patients with telangiectatic osteosarcoma, with a specific emphasis on extremity-affected cases [ 6 ]. Additionally, Bo Chen investigated the risks and prognostic factors of distant metastasis in patients newly diagnosed with osteosarcoma [ 7 ]. Current survival prediction methodologies predominantly utilize established clinical and sociodemographic predictors, frequently applying Cox proportional hazards regression analysis to develop nomograms [ 7 – 11 ]. Although some models demonstrate promising concordance index (C-index) estimates, there is an inherent risk of overfitting associated with these assessments. In the context of precision medicine's rapid evolution, machine learning (ML) techniques are becoming progressively prevalent in medical imaging analysis and treatment planning [ 12 ]. ML's approach, not requiring predefined relationships between input and output variables [ 13 ], offers a more adaptable and comprehensive analysis framework. This method effectively considers all interactions and potential modifications of effects among these variables. Significantly, ML emerges as a highly efficient and accurate tool compared to traditional semi-parametric and parametric models, enhancing the capabilities of data analysis [ 14 ]. The application of machine learning (ML) in the healthcare sector encompasses diagnosis, prognostic evaluation, and risk factor identification [ 15 ]. Yet, data specific to the use of ML for prognostic analysis in fibroblastic osteosarcoma patients remains limited. In this study, we extracted data on fibroblastic osteosarcoma cases from the period of 1998 to 2019, utilizing the Surveillance, Epidemiology, and End Results (SEER) database. We primarily focused on cancer-specific survival (CSS) as our endpoint to highlight the direct impact of osteosarcoma on patient outcomes. The goal was to find out what medical things go along with CSS, then make a good predicting model for when it might happen. We compared Cox regression and machine learning methods to determine which one was more effective in predicting patient survival duration. In this research, the main aim is to find out which predictive model works best. It helps clinicians quickly spot dangers and choose right treatment plans faster. The work aims to improve how we make choices about treating and managing Fibroblastic Osteosarcoma. Materials and Methods 2.1 Study population In this study, we used the Surveillance, Epidemiology and Results (SEER) database as our source of information. This public database doesn't need approval from an ethics group or agreement to be part of research. Following the rules set out by SEER, our research gathered details from patients with fibroblastic osteosarcoma between 1998 and 2019. The criteria for inclusion were: (1) a pathologically confirmed diagnosis of fibroblastic osteosarcoma; (2) primary tumor localization to bones and joints; and (3) availability of detailed survival status and duration data. On the other hand, exclusion criteria were applied as follows: (1) a survival time of less than one month; (2) cases identified solely through autopsy or death certificates; (3) lack of surgery information; and (4) cases with unknown primary site, survival status, survival time, or cause of death. 2.2 Predictor variables and outcomes Our study involved the collection of comprehensive data, including various parameters such as age, gender, ethnicity, tumor laterality, primary tumor location, grade group, tumor dimensions, spread extent, stage, surgical procedures, and the use of radiation and chemotherapy. We utilized X-tile software for analyzing and optimally categorizing continuous variables like age and tumor size, establishing these categories based on the most impactful cut-off values. For tumor primary site classification, we divided sites into limbs and joints (codes C40.0; C40.1; C40.2; C40.3; C40.8), axial bones (C41.2; C41.3; C41.4), skull and mandible (C41.0; C41.1), among others. The tumor's extent was delineated as either confined within the periosteum, extending beyond it, or exhibiting further spread. Tumor stages were segmented into regional, localized, and distant categories. Regarding treatment, we considered various surgical approaches including amputation (AMP), limb-salvage surgery (LSS), local treatment, or no surgical intervention, along with radiation and chemotherapy options. To enhance analytical clarity, smaller groups were combined: tumor grades were grouped into Grade I/II and Grade III/IV, reflecting the level of tumor differentiation. The primary outcome measured in our study was cancer-specific survival (CSS), defined as the duration from diagnosis to death attributed solely to the tumor. 2.3 Data preprocessing In our investigation, any unreported data points were categorized as missing. We closely checked the amount of data not available for each prognostic factor. We decided to add those with less than 25% missing info only. For the management of these missing values, we applied two distinct methods depending on the rate of data absence: For rates lower than 20%, the system used a method called multiple imputation. For rates more than 20%, we employed an approach known as K-Nearest Neighbor (KNN). This two-part method helped us do a more accurate and detailed handling of missing data in our study [ 14 , 16 ]. We evaluated multicollinearity among variables by computing the variance inflation factor (VIF). A VIF value below 10 was deemed indicative of negligible or non-significant multicollinearity in this analysis. Additionally, we performed Spearman correlation analysis to ascertain the interrelationships among variables. In this analysis, a correlation coefficient exceeding 0.5 signified a notable correlation between the variables [ 17 ]. 2.4 Model development and evaluation In our research, models were developed using Cox regression and contrasted with six distinct machine learning algorithms: Survival Tree, Extra Survival Trees, Random Survival Forest, Gradient Boosting Survival Analysis, Fast Kernel Survival SVM, and Minlip Survival Analysis. Patient data was divided, dedicating 70% for training and 30% for validation, with the random number table method ensuring randomized allocation. The comparative approach between the training and validation sets was tailored according to the type of outcome variable; continuous variables were analyzed using the t-test, while categorical variables were assessed with either the chi-square test or Fisher’s exact test. Furthermore, a blend of grid search and multiple cross-validation techniques was employed to identify the optimal parameter values that yielded the highest C-index values, thus defining the model parameters. In our study, the variable selection process was enhanced using the Boruta algorithm, an advanced method for identifying key variables in a dataset. This algorithm contrasts the importance of actual features with synthetically generated shadow features through random variables. Through iterative analysis and selection, Boruta rigorously identifies and retains only those features with statistical significance. The integration of the Boruta algorithm into our study ensured the inclusion of critical variables, thereby greatly improving the accuracy and dependability of our predictive model, leading to more reliable and precise results [ 18 ]. To safeguard against overfitting, our model was rigorously evaluated, giving precedence to the validation set while also meticulously examining the training set, thus upholding a comprehensive evaluation framework. The model's discriminative efficacy is gauged using the concordance index (C-index). This index reveals the generalization capacity of the model, where a C-index ranging from 0.5 to 0.7 indicates weak predictive ability, 0.7 to 0.85 suggests moderate ability, and 0.85 to 1.0 represents strong predictive capacity. To assess our model's ability to distinguish between outcomes accurately, we applied various metrics: accuracy, which indicates the proportion of correct predictions; time-dependent Area Under the Curve (AUC) for assessing classification efficacy; precision, denoting the percentage of true positives among positive predictions; recall or sensitivity, pivotal for identifying true positives; and the F1 score, harmonizing precision with recall. The calibration of the model was ascertained using a calibration curve, where alignment with the 45° diagonal suggests improved calibration. Additionally, Decision Curve Analysis (DCA) was utilized to measure the net clinical benefit, reinforcing the model's statistical soundness and clinical relevance. 2.5 Model interpretation In our study, the SHapley Additive exPlanation (SHAP) framework was instrumental in interpreting the outcomes of our machine learning model [ 19 ]. We applied SHAP to elucidate the insights gained from the final model, specifically focusing on a detailed analysis of the risk factors affecting mortality in patients with fibroblastic osteosarcoma. This approach involved assessing the impact of various features in the predictive process and their significance in influencing patient outcomes. Our study meticulously adhered to the guidelines outlined in the Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD) statement [ 20 ], ensuring a clear and transparent depiction of our predictive modeling approach. All statistical analyses were executed using R software (version 4.3.0) and Python software (version 3.11.6), with a P value less than 0.05 deemed to indicate statistical significance. Result 3.1 Patient characteristic Between 1998 and 2019, the SEER database provided records for 459 individuals diagnosed with fibroblastic osteosarcoma. Adhering to the specified exclusion criteria, 394 patients were ultimately selected for inclusion in the study. A flowchart depicting the process of data selection and exclusion is illustrated in Fig. 1. In our study, instances of unknown data within the final sample were handled as missing values. It was crucial to note that the missing data percentage for each variable was kept below 25%, a key aspect in preserving the integrity and robustness of our analysis. This approach was effective in managing the challenges posed by incomplete data. Considering that the missing data rate for each variable did not exceed 25%, we opted to retain all variables without excluding any due to low rates of missing data. Among these, Tumor Size had the highest missing rate at 23%, followed by Grade Group at 16%, and Extension at 11%. To effectively manage these missing values, we applied methods such as multiple imputation and the K-Nearest Neighbor (KNN) technique for the imputation of missing data. Utilizing the X-tile program, we determined the optimal cutoff points for age as 40 and 57 years, leading to the categorization of age into three groups: under 40 years, between 41 and 57 years, and over 57 years. In a similar vein, cutoff points for tumor size were identified, classifying it into three categories: up to 60mm, between 61mm and 103mm, and over 103mm. Establishing these categories was instrumental in enhancing the precision and depth of our data analysis. Table 1 presents the demographic and clinical characteristics of the fibroblastic osteosarcoma patients included in our study. The majority, or 66.2%, were aged 0–40 years, with the age groups of 41–57 years and over 57 years comprising 22.8% and 10.9% of the cohort, respectively. The distribution of gender was nearly equal, with females representing 48.0% and males 52.0%. Most patients were White (82.7%), with Black (9.4%) and other racial groups (7.9%) also represented. Tumor laterality was almost evenly split between left-sided (44.9%) and right-sided (39.3%), and 15.7% of cases were not applicable to a specific paired site. The majority of tumors were located in limbs and joints (75.6%), while a smaller proportion occurred in axial bones (11.4%) and skull and mandible regions (12.9%). Most patients presented with higher-grade tumors (Grade III-IV, 83.0%), while a smaller percentage had Grade I-II tumors (15.0%). The sizes of the tumors were distributed as follows: ≤60mm (30.5%), 61-103mm (43.7%), and > 103mm (25.9%). Over half of the tumors were confined within the periosteum (53.8%), while some extended beyond it (39.6%) or demonstrated further spread (6.6%). The staging of the tumors varied, with regional stage being most common (45.7%), followed by localized (40.6%) and distant stages (13.7%). Surgical procedures included limb salvage surgery (LSS, 52.8%), amputation (AMP, 17.3%), and local treatments (20.3%), with a minority undergoing no surgery (9.6%). The majority of patients did not receive radiation therapy (90.6%), while a small number did (9.4%). Chemotherapy was administered to a significant portion of the patients (74.5%), with the remainder having no chemotherapy or unknown status (24.6%). Statistical analysis using P-value indicated no significant differences between the training and validation groups, suggesting a balanced distribution of these characteristics. Table 1 Demographic characteristics of patients with fibroblastic osteosarcoma Overall (N = 394) Training Group (N = 275) Validation Group (N = 119) P-value Age Group age0-40 261 (66.2%) 180 (65.5%) 81 (68.1%) 0.770 age41-57 90 (22.8%) 63 (22.9%) 27 (22.7%) age > 57 43 (10.9%) 32 (11.6%) 11 (9.2%) Sex Female 189 (48.0%) 137 (49.8%) 52 (43.7%) 0.314 Male 205 (52.0%) 138 (50.2%) 67 (56.3%) Race Group White 326 (82.7%) 233 (84.7%) 93 (78.2%) 0.091 Black 37 (9.4%) 20 (7.3%) 17 (14.3%) Other 31 (7.9%) 22 (8.0%) 9 (7.6%) Laterality Left 177 (44.9%) 125 (45.5%) 52 (43.7%) 0.428 Right 155 (39.3%) 111 (40.4%) 44 (37.0%) Not a paired site 62 (15.7%) 39 (14.2%) 23 (19.3%) Primary Site Axial bones 45 (11.4%) 33 (12.0%) 12 (10.1%) 0.854 Limbs and joints 298 (75.6%) 207 (75.3%) 91 (76.5%) Skull and mandible, etc 51 (12.9%) 35 (12.7%) 16 (13.4%) Grade Group Grade I-II 59 (15.0%) 42 (15.3%) 17 (14.3%) 0.922 Grade III-IV 335 (85.0%) 233 (84.7%) 102 (85.7%) Tumorsize Group ≦ 60 120 (30.5%) 84 (30.5%) 36 (30.3%) 0.535 61–103 172 (43.7%) 124 (45.1%) 48 (40.3%) > 103 102 (25.9%) 67 (24.4%) 35 (29.4%) Extension Inside periosteum 157 (39.8%) 113 (41.1%) 44 (37.0%) 0.701 Beyond periosteum 211 (53.6%) 145 (52.7%) 66 (55.5%) Further extension 26 (6.6%) 17 (6.2%) 9 (7.6%) Tumor Stage Localized 160 (40.6%) 112 (40.7%) 48 (40.3%) 0.473 Regional 180 (45.7%) 129 (46.9%) 51 (42.9%) Distant 54 (13.7%) 34 (12.4%) 20 (16.8%) Surgery Type AMP 68 (17.3%) 46 (16.7%) 22 (18.5%) 0.725 Local treatment 80 (20.3%) 60 (21.8%) 20 (16.8%) LSS 208 (52.8%) 143 (52.0%) 65 (54.6%) No surgery 38 (9.6%) 26 (9.5%) 12 (10.1%) Radiation Group Overall (N = 394) Training Group (N = 275) Validation Group (N = 119) P-value None/Unknown 357 (90.6%) 246 (89.5%) 111 (93.3%) 0.314 Yes 37 (9.4%) 29 (10.5%) 8 (6.7%) Chemotherapy No/Unknown 97 (24.6%) 70 (25.5%) 27 (22.7%) 0.648 Yes 297 (75.4%) 205 (74.5%) 92 (77.3%) 3.2 Feature selection Through the application of Z-values, our study pinpointed ten critical variables significantly associated with fibroblastic osteosarcoma, which included Tumor Stage, Tumor Size Group, Radiation, Age Group, Primary Site, Extension, Surgery Type, Sex, Grade Group, and Chemotherapy, as depicted in Fig. 2 . Multicollinearity assessments, using Variance Inflation Factors (VIF) with all values under 10, indicated no substantial inter-variable conflicts. Spearman correlation analysis identified a significant correlation between Tumor Stage and Extension (0.50), while correlations among other variables remained relatively lower (all below 0.5), as shown in Fig. 3 . Consequently, to eliminate redundancy, Extension was excluded from further analysis. The remaining nine variables were thus recognized as independent prognostic factors for fibroblastic osteosarcoma. Detailed in Table 1 , these factors were used in different models to predict cancer-specific survival. 3.3 Model performance comparisons In the comparative analysis against the Cox regression model, the machine learning models demonstrated significantly enhanced performance, as detailed in Table 2 , and illustrated in Figs. 4 and 5 . The machine learning models excelled across various critical performance metrics, including Accuracy, Recall, F1 score, and the Concordance index (C-index), consistently surpassing the Cox regression model in these aspects. Table 2 The performance metrics of the of Cox regression and cross-validated machine learning algorithms Accuracy Recall F1 score C-index Mean AUC CSS 3 year CSS 5 year CSS 3 year CSS 5 year CSS 3 year CSS 5 year training validation Cox regression 0.77 0.77 0.55 0.53 0.56 0.57 0.74 0.68 0.77 SurvivalTree 0.88 0.87 0.75 0.72 0.76 0.76 0.90 0.81 0.86 ExtraSurvivalTrees 0.91 0.89 0.81 0.74 0.83 0.80 0.92 0.80 0.89 RandomSurvivalForest 0.89 0.87 0.78 0.72 0.79 0.76 0.89 0.79 0.88 GradientBoostingSurvivalAnalysis 0.88 0.88 0.77 0.73 0.78 0.78 0.92 0.79 0.86 FastKernelSurvivalSVM 0.93 0.90 0.89 0.80 0.87 0.82 0.94 0.79 0.87 MinlipSurvivalAnalysis 0.91 0.92 0.81 0.79 0.83 0.85 0.92 0.81 0.87 For assessing the time-dependent Area Under the Curve (AUC) values, our analysis selected 15 key time points spanning from the 5th to 95th percentile of the survival time distribution. This methodology facilitated an extensive examination of the dataset. The findings indicated that, at most of these selected time points, the machine learning model demonstrated enhanced performance in comparison to the Cox regression model, as delineated in Fig. 5 . The AUC scores of the different models showed variation over time, with some models initially having high scores that later declined, while others either increased or maintained a consistent level. The Extra Survival Trees Model notably achieved the highest average AUC score. In an in-depth evaluation of the Extra Survival Trees Model, calibration curves and Decision Curve Analyses (DCA) were conducted, as illustrated in Fig. 6 . These calibration curves demonstrated a strong congruence between the model’s predictions and actual 5-year cancer-specific survival rates, affirming the model's precision and dependability (as shown in Figs. 6 E and F). The DCA for predicting 3-year and 5-year cancer-specific survival rates by this model, applied to both training and validation cohorts (depicted in Figs. 6 A to D), highlighted the model's substantial clinical applicability. 3.4 Model interpretation of machine learning In our study, the Extra Survival Trees Model was analyzed using SHAP (SHapley Additive exPlanation) to understand the model's rationale. The SHAP summary plots, illustrated in the figure, highlighted 'Age Group', 'Tumor Stage', and 'Tumor Size Group' as the most influential factors in predicting survival outcomes for patients with fibroblastic osteosarcoma. The plots (Fig. 6 ) revealed several critical insights: Age Group: Elevated ages were found to correlate with increased risk scores. Tumor Stage: More advanced tumor stages were associated with higher risk scores. Tumor Size Group: Larger tumor sizes linked to escalated risk scores. Radiation Group: Previous radiation therapy indicated higher risk scores, while its absence was associated with lower scores. Surgery Type: Surgical interventions typically resulted in lower risk scores, contrasting with higher scores in the absence of surgery. Primary Site: Tumors located in limbs and joints generally showed lower risk scores compared to those in axial bones or the skull/mandible. Sex: Male patients exhibited higher risk scores than female patients. Chemotherapy: Receiving chemotherapy was typically associated with higher risk scores. Grade Group: Elevated tumor grades corresponded to increased risk scores. These insights from the SHAP analysis provide a detailed understanding of the factors influencing the predictive model’s outcomes. Discussion OS is acknowledged as the most common malignant bone tumor, primarily impacting adolescents and children. It is characterized by a poor prognosis and limited treatment options [ 2 ]. Given the rarity of OS instances, the SEER database [ 21 ] was employed as the principal data source to secure adequate data for comprehensive analysis. In this study, Cox regression and various machine learning algorithms were utilized and compared to predict cancer-specific survival in patients with fibroblastic osteosarcoma. The applied machine learning algorithms encompassed Survival Tree, Extra Survival Trees, Random Survival Forest, Gradient Boosting Survival Analysis, Fast Kernel Survival SVM, and Minlip Survival Analysis. This integrated approach facilitated a detailed assessment of the predictive efficacy of these diverse methods in the context of fibroblastic osteosarcoma. Upon comparing various survival analysis models, the "Extra Survival Trees" model emerged as the most superior in terms of overall performance. This conclusion is primarily derived from the model's mean Area Under the Curve (AUC), an essential metric for evaluating predictive accuracy. Within our dataset, the Extra Survival Trees model exhibited a mean AUC of 0.89, markedly surpassing that of other models. This average AUC demonstrates the model's consistent predictive capacity over time. The notable score attained by the Extra Survival Trees model in this measure underscores its significant superiority in predictive precision and reliability, particularly in handling intricate survival data. Besides the mean AUC, other critical metrics, including accuracy, recall, F1 score, and Concordance Index (C-index) in both training and validation phases, were also considered. In these respects, the Extra Survival Trees model demonstrated remarkable performance, solidifying its position as the premier choice in our study. Furthermore, the calibration curve indicated the model's high predictive accuracy, and the Decision Curve Analysis (DCA) validated its clinical utility. To identify independent prognostic factors with utmost precision, a comprehensive analysis of clinical variables was conducted. Optimal cutoff values were established at 40 and 57 years for age, and 60 and 103 millimeters for tumor size, respectively. Following the conversion of continuous variables into categorical ones, the Boruta algorithm was employed in conjunction with the variance inflation factor (VIF) and Spearman correlation analysis to enhance our variable selection process. This meticulous methodology enabled the selection of pivotal clinical attributes, including Age Group, Primary Site, Sex, Grade Group, Tumor Size Group, Tumor Stage, Surgery Type, Radiation Group, and Chemotherapy. Subsequently, these attributes were systematically integrated into both Cox regression and machine learning models to advance prognostic evaluations. In our thorough study of fibroblastic osteosarcoma, we found several separate prognostic factors that greatly influence patient results. The age factor became very important, and people over 57 years old were more likely to have a bad outlook. This increase might be because of late detection and a bigger chance for bone tumors in the axial osteosarcomas, which spread more easily. This usually happens to older people. Also, older patients may find it harder to handle tough treatments because they might have other health problems or long-lasting conditions. This makes their treatment and getting better more difficult [ 22 – 25 ]. Similarly, tumor staging, particularly the presence of metastasis, became a very important independent part that affects how long the patients with this disease live. Metastasis means cancer being more dangerous. It lets the malignant cells move out of their first place and go far away, making it harder to treat them later on. Spreading can happen in different places like lungs, bones, even brain or some organs. This shows the complex steps involved when cancer spreads [ 26 , 27 ]. Tumor size also played a pivotal role in determining prognosis. Larger tumors were consistently associated with inferior outcomes [ 22 ]. This association could be attributed to the heightened likelihood of metastasis and the increased complexity in achieving complete surgical resection. Furthermore, larger tumor dimensions may signify a more aggressive tumor biology, further complicating the development of effective treatment strategies [ 24 ]. In the context of treatment strategies, radiation therapy has been identified as a high-risk factor for fibroblastic osteosarcoma. This is largely attributed to the relative radio-resistance of these tumors. Radiation therapy is often administered for palliative purposes in cases where tumors are inoperable. Consequently, patients undergoing radiation therapy typically exhibit a poorer prognosis. This outcome is partly because radiation therapy is not the preferred modality for treating osteosarcoma and is generally reserved for scenarios where surgical intervention is not feasible [ 28 , 29 ]. Surgical treatment type significantly impacts the prognosis of osteosarcoma patients [ 30 , 31 ]. Although osteosarcoma is a rare tumor, surgery remains its primary effective treatment modality [ 32 ]. Our study found that limb-salvage surgery (LSS) is more effective in reducing risk scores compared to amputation. LSS not only preserves limb function and reduces psychological burden, aiding in social integration, but also due to its less invasive nature and lesser blood loss, it reduces the risk of postoperative complications, thus aiding in quicker patient recovery, decreasing the likelihood of postoperative complications, and enhancing treatment success rates. These advantages make LSS beneficial in terms of physical, psychological, and social aspects, improving patient rehabilitation and quality of life [ 33 , 34 ]. Treating axial or the skull/mandible osteosarcomas is very hard, mostly because they have a high chance of spreading to other parts of the body and their locations are complex. These lumps usually live near big blood channels, making it more possible for cancer cells to spread through the blood [ 35 ]. Also, symptoms linked to spine tumors are often misunderstood. This can cause late diagnoses and increased chances of spreading the cancer throughout the body [ 36 ]. The confusing structure of the axial part makes major surgery harder. This can cause problems during the operation like too much blood loss, implant failure or paralysis which all make things bad for patients and their future health outlook. This is particularly the case when the tumor infiltrates crucial regions like the upper cervical area, vertebral artery foramen, or odontoid process. Such invasion significantly escalates the complexity of margin resection and complicates the treatment process [ 37 , 38 ]. Gender differences matter a lot in how bad osteosarcoma is, with males usually getting worse results [ 39 – 41 ]. This disparity might arise from several factors: firstly, male patients might exhibit lower adherence to recommended treatment protocols [ 40 ], and secondly, they may demonstrate less favorable responses to neoadjuvant chemotherapy compared to female patients [ 42 ]. Next, natural things like sex hormones inside the body could help make osteosarcoma cells grow or shrink in different ways between male and female patients [ 43 , 44 ]. Also, some research shows that poor habits like smoking and drinking in men can raise cancer risk and make the condition worse. This then affects how well they are expected to get better [ 45 ]. Chemotherapy is commonly considered a fundamental component of osteosarcoma treatment, with numerous studies highlighting its potential advantages in enhancing prognosis [ 6 , 42 ]. Nonetheless, our investigation proposes that chemotherapy may not consistently confer benefits. While chemotherapy aids in controlling tumor growth and dissemination, it can also compromise the immune system, increasing vulnerability to infections and other illnesses, thereby exerting adverse effects on patients [ 46 , 47 ]. Moreover, chemotherapy medications frequently entail a spectrum of side effects, including cardiotoxicity, nephrotoxicity, and the risks of infertility and secondary tumors, which can significantly impact patients' quality of life and adherence to treatment [ 48 ]. Our study also presents an alternative interpretation: the challenge of ascertaining the specific chemotherapy regimens and patients' adherence levels. Additionally, the introduction of a 'No/Unknown' category in the chemotherapy variable introduced ambiguity, impeding our ability to confirm whether patients indeed underwent chemotherapy. This limitation potentially impacted the statistical robustness of our findings. Ultimately, the grade of the tumor stands as an independent prognostic determinant [ 7 , 49 ]. Elevated tumor grades are intrinsically linked to an augmented propensity for lung metastasis [ 50 ], underscoring the pivotal role of tumor grade in shaping both prognosis and strategic treatment considerations. Our research addresses a critical gap in predictive modeling for fibroblastic osteosarcoma. Through the comprehensive collection of disease-related predictors, encompassing baseline patient data, clinical diagnoses, and therapeutic interventions, including both medical and surgical treatments, we harnessed the formidable capabilities of Machine Learning (ML). In recent years, ML has demonstrated remarkable proficiency in handling multifaceted variables and has found extensive application in cancer detection and prognostication [ 51 , 52 ]. Within this study, we meticulously crafted Cox regression and ML models, and an evaluation of their performance, employing five metrics (C-index, accuracy, recall, F1 score, and time-dependent AUC), was conducted. Our findings unequivocally indicate the impressive predictive capabilities of ML algorithms in the context of cancer-specific survival for fibroblastic osteosarcoma. To conduct a comprehensive assessment of ML model performance, we specifically chose to focus on a detailed analysis of the Extreme Survival Tree model, which demonstrated the best overall performance across various metrics. Within our analysis, we focused our attention on two fundamental metrics: the Calibration Curve and Decision Curve Analysis (DCA). The Calibration Curve was employed to gauge the degree of alignment between the model's predictions and actual events, shedding light on the model's predictive consistency. Simultaneously, DCA was utilized to quantify the clinical utility of the model across varying decision thresholds, aiding in the evaluation of its practical significance in clinical settings. The examination of these metrics brought to the forefront the strengths of the Extreme Survival Tree model. Notably, it showcased a high level of accuracy, closely mirroring real-world outcomes, and also demonstrated substantial clinical effectiveness, thereby offering robust decision support for clinicians. Moreover, through the amalgamation of SHAP values, the utilization of machine learning (ML) techniques effectively identified pivotal predictive factors, ultimately culminating in the development of an exceedingly precise survival prediction model. This groundbreaking approach furnishes a visual and intuitive means for discerning both protective and risk factors, offering invaluable guidance for the realm of clinical judgment and decision-making. Our ML models consistently exhibited exceptional performance metrics, underscoring their formidable generalization capabilities and profound clinical relevance. These models are characterized by their provision of lucid explanations, thereby facilitating the prediction of survival rates and consequently enhancing the comprehension of the decision-making process pertaining to the assessment of disease severity. As is widely acknowledged, the Cox Proportional Hazards regression model stands as one of the most extensively employed clinical prediction models, playing a pivotal role in assisting clinicians in making informed decisions [ 7 , 8 , 53 ]. Nonetheless, a fundamental assumption underpinning the Cox model is the proportionality of hazards, signifying that the relative hazard remains consistent over time across various predictor or covariate levels. It is imperative for this assumption to hold true for the effective utilization of the Cox Proportional Hazards regression model. Nevertheless, in real-world practice, hazard ratios frequently exhibit temporal variability [ 54 ], which has the potential to diminish the predictive accuracy of the Cox regression model under specific circumstances. In the age of big data, machine learning (ML) algorithms enhance the quantification of individual patient risks and play a pivotal role in assessing disease prognosis. When compared to traditional statistical methods, ML models offer a key advantage – their accurate applicability at both the individual and population levels, coupled with a high potential for clinical application. ML algorithms do not take into account factors such as non-proportionality, multicollinearity, or nonlinearity, thereby reducing prediction bias resulting from modeling uncertainty. However, their adoption in clinical practice faces a hurdle due to a lack of interpretability. Consequently, the integration of SHAP (SHapley Additive exPlanations) aims to elucidate how machine models process outputs in an easily comprehensible manner, mitigating the aforementioned limitations. To date, there has been no targeted application of machine learning algorithms for predicting the survival of patients with fibroblastic osteosarcoma. In the era characterized by the ubiquity of big data, machine learning (ML) algorithms play a pivotal role in significantly enhancing the precision of quantifying individual patient risks and facilitating the assessment of disease prognosis. In contrast to conventional statistical methods, ML models offer a substantial advantage due to their precise applicability at both the individual and population levels, coupled with their immense potential for clinical utility [ 55 ]. These ML algorithms adeptly tackle issues such as non-proportionality, multicollinearity, and nonlinearity within the data, thereby minimizing prediction bias arising from modeling uncertainties. Nonetheless, their widespread adoption in clinical practice often encounters hindrances stemming from challenges in interpretability. To address this limitation, the adoption of SHAP (SHapley Additive exPlanation) seeks to elucidate the decision-making process of machine models in a comprehensible format, effectively mitigating the aforementioned obstacles. As of the current state of knowledge, the targeted utilization of ML algorithms for predicting survival outcomes in patients afflicted with fibroblastic osteosarcoma remains an uncharted territory. While our study offers valuable insights into the prediction of survival in fibroblastic osteosarcoma using both Cox regression and machine learning algorithms, it is essential to acknowledge several limitations. One notable constraint is the relatively small sample size, which not only limits the generalizability of our findings but also impacts the robustness of our statistical analysis. Furthermore, the absence of external validation raises concerns regarding the applicability of our results to diverse populations. Although the incorporation of SHAP (SHapley Additive exPlanation) values enhances the interpretability of our models, the inherent complexity of these machine learning approaches may still present challenges when translating them into practical clinical settings. Additionally, the presence of ambiguity introduced by the inclusion of a 'No/Unknown' option in chemotherapy data complicates the assessment of its influence on patient outcomes. It is important to note that limitations related to data inherent in the SEER database constrain the depth of our analysis. In anticipation of the future, it becomes imperative to conduct extensive studies involving larger and more diverse patient cohorts to validate and refine our predictive models. Subsequent research endeavors should be geared towards augmenting the interpretability and applicability of machine learning models within clinical contexts, ensuring their effective integration into healthcare practice. Moreover, addressing the intricacies associated with treatment data, particularly pertaining to chemotherapy regimens and patient adherence, will be pivotal for a more precise evaluation of treatment outcomes. By surmounting these challenges, we can advance our comprehension and management of fibroblastic osteosarcoma, thereby making a substantial contribution to the enhancement of patient care and ultimate clinical outcomes. Conclusion In short, we did a study that compared methods of cox regression and machine learning (ML) to predict the survival outcomes of patients with fibroblastic osteosarcoma. Importantly, this research looks at a part that has never been studied before in predicting results for fibroblastic osteosarcoma. Our findings indicate that ML algorithms, when coupled with SHapley Additive exPlanation (SHAP) values, demonstrate a promising predictive capability for survival in fibroblastic osteosarcoma. We need to be careful when looking at these results because a small sample size limits them and there's no outside checking. By adding to what's not well studied, our study sets up for more work and use in this area. Declarations Funding Medical Scientific Research Foundation of Guangdong Province (A2021074). Competing Interests The authors have no relevant financial or non-financial interests to disclose. Author Contributions All authors significantly contributed to the conception and design of the study. Longteng Chao was responsible for data analysis and drafting the manuscript. Xinmiao Ye and Junyuan Chen played a key role in data collection. Guorong She and Zhengang Zha made substantial contributions to the study's conception and design. All authors were actively involved in revising earlier versions of the manuscript and have read and approved the final version for publication. Data Availability The datasets generated and analyzed during this study are accessible in the SEER database, which can be found at seer.cancer.gov. References Philip T, Blay JY, Brunat-Mentigny M, Carrie C, Chauvot P, Farsi F, Fervers B, Gentet JC, Giammarile F, Kohler R, Mathoulin S, Patricot LM, Thiesse P (2001) French National Federation of Cancer (FNCLCC). Osteosarcoma. Br J Cancer. 2001;84 Suppl 2(Suppl 2):78–80. doi: 10.1054/bjoc.2000.1770 . WHO Classification of Tumours Editorial Board (2020). WHO Classification of Tumours. Soft Tissue and Bone Tumors. 5th ed. Volume 3. International Agency for Research on Cancer; Lyon, France: 2020. Bacci G, Bertoni F, Longhi A, Ferrari S, Forni C, Biagini R, Bacchini P, Donati D, Manfrini M, Bernini G, Lari S (2003) Neoadjuvant chemotherapy for high-grade central osteosarcoma of the extremity. Histologic response to preoperative chemotherapy correlates with histologic subtype of the tumor. Cancer. 2003;97(12):3068-75. doi: 10.1002/cncr.11456 . Kim MS, Lee SY, Lee TR, Cho WH, Song WS, Koh JS, Lee JA, Yoo JY, Jeon DG (2009) Prognostic nomogram for predicting the 5-year probability of developing metastasis after neo-adjuvant chemotherapy and definitive surgery for AJCC stage II extremity osteosarcoma. Ann Oncol. 2009;20(5):955 – 60. doi: 10.1093/annonc/mdn723 . Epub 2009 Jan 19. Balachandran VP, Gonen M, Smith JJ, DeMatteo RP (2015) Nomograms in oncology: more than meets the eye. Lancet Oncol. 2015;16(4):e173-80. doi: 10.1016/S1470-2045(14)71116-7 . Zhong W, Luo W, Lin Z, Wu Z, Yuan Y, He Y (2023) Prognostic analysis of telangiectatic osteosarcoma of the extremities. Front Oncol. 2023;12:1105054. doi: 10.3389/fonc.2022.1105054 . Chen B, Zeng Y, Liu B, Lu G, Xiang Z, Chen J, Yu Y, Zuo Z, Lin Y, Ma J (2021) Risk Factors, Prognostic Factors, and Nomograms for Distant Metastasis in Patients With Newly Diagnosed Osteosarcoma: A Population-Based Study. Front Endocrinol (Lausanne). 2021;12:672024. doi: 10.3389/fendo.2021.672024 . Lee RJ, Arshi A, Schwartz HC, Christensen RE (2015) Characteristics and prognostic factors of osteosarcoma of the jaws: a retrospective cohort study. JAMA Otolaryngol Head Neck Surg. 2015;141(5):470-7. doi: 10.1001/jamaoto.2015.0340 . Liu Z, Yin J, Zhou Q, Yang J, Zeng B, Yeung SJ, Shen J, Cheng C (2020) Survival after pulmonary metastasectomy for relapsed osteosarcoma. J Thorac Cardiovasc Surg. 2022;163(2):469–479.e8. doi: 10.1016/j.jtcvs.2020.10.137 . Epub 2020 Nov 24. Wang J, Zhanghuang C, Tan X, Mi T, Liu J, Jin L, Li M, Zhang Z, He D (2022) A Nomogram for Predicting Cancer-Specific Survival of Osteosarcoma and Ewing's Sarcoma in Children: A SEER Database Analysis. Front Public Health. 2022;10:837506. doi: 10.3389/fpubh.2022.837506 . Xue W, Zhang Z, Yu H, Li C, Sun Y, An J, Qi L, Zhang J, Liu Q (2023) Development of nomogram and discussion of radiotherapy effect for osteosarcoma survival. Sci Rep. 2023;13(1):223. doi: 10.1038/s41598-023-27476-9 . Jiang F, Jiang Y, Zhi H, Dong Y, Li H, Ma S, Wang Y, Dong Q, Shen H, Wang Y (2017) Artificial intelligence in healthcare: past, present and future. Stroke Vasc Neurol. 2017;2(4):230–243. doi: 10.1136/svn-2017-000101 . Khan K, Ahmad W, Amin MN, Ahmad A, Nazar S, Alabdullah AA (2022) Compressive Strength Estimation of Steel-Fiber-Reinforced Concrete and Raw Material Interactions Using Advanced Algorithms. Polymers (Basel). 2022;14(15):3065. doi: 10.3390/polym14153065 . Xu L, Cai L, Zhu Z, Chen G (2023) Comparison of the cox regression to machine learning in predicting the survival of anaplastic thyroid carcinoma. BMC Endocr Disord. 2023;23(1):129. doi: 10.1186/s12902-023-01368-5 . Erratum in: BMC Endocr Disord. 2023;23(1):174. Jiang F, Jiang Y, Zhi H, Dong Y, Li H, Ma S, Wang Y, Dong Q, Shen H, Wang Y (2017) Artificial intelligence in healthcare: past, present and future. Stroke Vasc Neurol. 2017;2(4):230–243. doi: 10.1136/svn-2017-000101 . Janssen KJ, Vergouwe Y, Donders AR, Harrell FE Jr, Chen Q, Grobbee DE, Moons KG (2009) Dealing with missing predictor values when applying clinical prediction models. Clin Chem. 2009;55(5):994–1001. doi: 10.1373/clinchem.2008.115345 . Epub 2009 Mar 12. Tu YK, Kellett M, Clerehugh V, Gilthorpe MS (2005) Problems of correlations between explanatory variables in multiple regression analyses in the dental literature. Br Dent J. 2005;199(7):457 – 61. doi: 10.1038/sj.bdj.4812743 . Yue S, Li S, Huang X, Liu J, Hou X, Zhao Y, Niu D, Wang Y, Tan W, Wu J (2022) Machine learning for the prediction of acute kidney injury in patients with sepsis. J Transl Med. 2022;20(1):215. doi: 10.1186/s12967-022-03364-0 . Lundberg SM, Lee S-I (2017) A unified approach to interpreting model predictions. Adv Neural Inf Process Syst 2017, 30. Heus P, Reitsma JB, Collins GS, Damen JAAG, Scholten RJPM, Altman DG, Moons KGM, Hooft L (2020) Transparent Reporting of Multivariable Prediction Models in Journal and Conference Abstracts: TRIPOD for Abstracts. Ann Intern Med. 2020 Jun 2. doi: 10.7326/M20-0193 . Epub ahead of print. Che WQ, Li YJ, Tsang CK, Wang YJ, Chen Z, Wang XY, Xu AD, Lyu J (2023) How to use the Surveillance, Epidemiology, and End Results (SEER) data: research design and methodology. Mil Med Res. 2023;10(1):50. doi: 10.1186/s40779-023-00488-2 . Xin S, Wei G (2020) Prognostic factors in osteosarcoma: A study level meta-analysis and systematic review of current practice. J Bone Oncol. 2020;21:100281. doi: 10.1016/j.jbo.2020.100281 . Iwata S, Ishii T, Kawai A, Hiruma T, Yonemoto T, Kamoda H, Asano N, Takeyama M (2014) Prognostic factors in elderly osteosarcoma patients: a multi-institutional retrospective study of 86 cases. Ann Surg Oncol. 2014;21(1):263-8. doi: 10.1245/s10434-013-3210-4 . Epub 2013 Aug 23. Tsuda Y, Ogura K, Shinoda Y, Kobayashi H, Tanaka S, Kawai A (2018) The outcomes and prognostic factors in patients with osteosarcoma according to age: a Japanese nationwide study with focusing on the age differences. BMC Cancer. 2018;18(1):614. doi: 10.1186/s12885-018-4487-2 . Tsuchie H, Emori M, Nagasawa H, Miyakoshi N, Murahashi Y, Shimizu J, Mizushima E, Yamashita T, Shimada Y (2019) Prognosis of Primary Osteosarcoma in Elderly Patients: A Comparison between Young and Elderly Patients. Med Princ Pract. 2019;28(5):425–431. doi: 10.1159/000500404 . Epub 2019 Apr 17. Wang J, Zhanghuang C, Tan X, Mi T, Liu J, Jin L, Li M, Zhang Z, He D (2022) A Nomogram for Predicting Cancer-Specific Survival of Osteosarcoma and Ewing's Sarcoma in Children: A SEER Database Analysis. Front Public Health. 2022;10:837506. doi: 10.3389/fpubh.2022.837506 . Beird HC, Bielack SS, Flanagan AM, Gill J, Heymann D, Janeway KA, Livingston JA, Roberts RD, Strauss SJ, Gorlick R (2022) Osteosarcoma. Nat Rev Dis Primers. 2022;8(1):77. doi: 10.1038/s41572-022-00409-y . Erratum in: Nat Rev Dis Primers. 2022;8(1):82. Hristov B, Shokek O, Frassica DA (2007) The role of radiation treatment in the contemporary management of bone tumors. J Natl Compr Canc Netw. 2007;5(4):456 – 66. doi: 10.6004/jnccn.2007.0040 . Schwarz R, Bruland O, Cassoni A, Schomberg P, Bielack S (2009) The role of radiotherapy in oseosarcoma. Cancer Treat Res. 2009;152:147 – 64.doi: 10.1007/978-1-4419-0284-9_7 . Mavrogenis AF, Abati CN, Romagnoli C, Ruggieri P (2012) Similar survival but better function for patients after limb salvage versus amputation for distal tibia osteosarcoma. Clin Orthop Relat Res. 2012;470(6):1735-48. doi: 10.1007/s11999-011-2238-7 . Epub 2012 Jan 24. Mason GE, Aung L, Gall S, Meyers PA, Butler R, Krüg S, Kim M, Healey JH, Gorlick R (2013) Quality of life following amputation or limb preservation in patients with lower extremity bone sarcoma. Front Oncol. 2013;3:210. doi: 10.3389/fonc.2013.00210 . Ritter J, Bielack SS (2010) Osteosarcoma. Ann Oncol. 2010;21 Suppl 7:vii320-5. doi: 10.1093/annonc/mdq276 . Evans DR, Lazarides AL, Visgauss JD, Somarelli JA, Blazer DG 3rd, Brigman BE, Eward WC (2020) Limb salvage versus amputation in patients with osteosarcoma of the extremities: an update in the modern era using the National Cancer Database. BMC Cancer. 2020;20(1):995. doi: 10.1186/s12885-020-07502-z . Qi L, Ren X, Liu Z, Li S, Zhang W, Chen R, Chen C, Tu C, Li Z (2020) Predictors and Survival of Patients with Osteosarcoma After Limb Salvage versus Amputation: A Population-Based Analysis with Propensity Score Matching. World J Surg. 2020;44(7):2201–2210. doi: 10.1007/s00268-020-05471-9 . Fahey M, Spanier SS, Vander Griend RA (1992) Osteosarcoma of the pelvis. A clinical and histopathological study of twenty-five patients. J Bone Joint Surg Am. 1992;74(3):321 – 30. Sundaresan N, Rosen G, Boriani S (2008) Primary malignant tumors of the spine. Orthop Clin North Am. 2009;40(1):21–36, v. doi: 10.1016/j.ocl.2008.10.004 . Grimer RJ, Taminiau AM, Cannon SR (2002) Surgical Subcommitte of the European Osteosarcoma Intergroup. Surgical outcomes in osteosarcoma. J Bone Joint Surg Br. 2002;84(3):395–400. doi: 10.1302/0301-620x.84b3.12019 . Weeden S, Grimer RJ, Cannon SR, Taminiau AH, Uscinska BM (2001) European Osteosarcoma Intergroup. The effect of local recurrence on survival in resected osteosarcoma. Eur J Cancer. 2001;37(1):39–46. doi: 10.1016/s0959-8049(00)00362-2 . Williams LA, Richardson M, Marcotte EL, Poynter JN, Spector LG (2019) Sex ratio among childhood cancers by single year of age. Pediatr Blood Cancer. 2019;66(6):e27620. doi: 10.1002/pbc.27620 . Epub 2019 Feb 28. Duchman KR, Gao Y, Miller BJ (2015) Prognostic factors for survival in patients with high-grade osteosarcoma using the Surveillance, Epidemiology, and End Results (SEER) Program database. Cancer Epidemiol. 2015;39(4):593-9. doi: 10.1016/j.canep.2015.05.001 . Epub 2015 May 20. Petrilli AS, Gentil FC, Epelman S, Lopes LF, Bianchi A, Lopes A, Figueiredo MT, Marques E, De Bellis N, Consentino E, et al (1991) Increased survival, limb preservation, and prognostic factors for osteosarcoma. Cancer. 1991;68(4):733-7. doi: 10.1002/1097-0142(19910815)68:43.0.co;2-0 . Bielack SS, Kempf-Bielack B, Delling G, Exner GU, Flege S, Helmke K, Kotz R, Salzer-Kuntschik M, Werner M, Winkelmann W, Zoubek A, Jürgens H, Winkler K (2023) Prognostic factors in high-grade osteosarcoma of the extremities or trunk: an analysis of 1,702 patients treated on neoadjuvant cooperative osteosarcoma study group protocols. J Clin Oncol. 2002;20(3):776 – 90. doi: 10.1200/JCO.2002.20.3.776. Corrected and republished in: J Clin Oncol. 2023;41(27):4323–4337. Pan Y, Chen D, Hu T, Lv G, Dai Z (2019 ) Characteristics and Prognostic Factors of Patients With Osteosarcoma Older Than 60 Years From the SEER Database. Cancer Control. 2019 Jan-Dec;26(1):1073274819888893. doi: 10.1177/1073274819888893 . Wagle S, Park SH, Kim KM, Moon YJ, Bae JS, Kwon KS, Park HS, Lee H, Moon WS, Kim JR, Jang KY (2015) DBC1/CCAR2 is involved in the stabilization of androgen receptor and the progression of osteosarcoma. Sci Rep. 2015;5:13144. doi: 10.1038/srep13144 . Katzke VA, Kaaks R, Kühn T (2015) Lifestyle and cancer risk. Cancer J. 2015 Mar-Apr;21(2):104–10. doi: 10.1097/PPO.0000000000000101 . Sanford NN, Martin AM, Brunner AM, Cote GM, Choy E, DeLaney TF, Aizer AA, Chen YL (2018) Secondary Acute Leukemia in Sarcoma Patients: A Population-Based Study. Int J Radiat Oncol Biol Phys. 2018;100(3):687–694. doi: 10.1016/j.ijrobp.2017.11.011. Epub 2017 Nov 14. Anderson PM (2017) Immune Therapy for Sarcomas. Adv Exp Med Biol. 2017;995:127–140. doi: 10.1007/978-3-319-53156-4_6 . Ferrari S, Palmerini E (2007) Adjuvant and neoadjuvant combination chemotherapy for osteogenic sarcoma. Curr Opin Oncol. 2007;19(4):341-6. doi: 10.1097/CCO.0b013e328122d73f . Deng G, Chen P (2023) Characteristics and prognostic factors of adult patients with osteosarcoma from the SEER database. Medicine (Baltimore). 2023;102(37):e33653. doi: 10.1097/MD.0000000000033653 . Xie L, Huang W, Wang H, Zheng C, Jiang J (2020) Risk factors for lung metastasis at presentation with malignant primary osseous neoplasms: a population-based study. J Orthop Surg Res. 2020;15(1):32. doi: 10.1186/s13018-020-1571-5 . Lee C, Light A, Alaa A, Thurtle D, van der Schaar M, Gnanapragasam VJ (2021) Application of a novel machine learning framework for predicting non-metastatic prostate cancer-specific mortality in men using the Surveillance, Epidemiology, and End Results (SEER) database. Lancet Digit Health. 2021;3(3):e158-e165. doi: 10.1016/S2589-7500(20)30314-9 . Epub 2021 Feb 3. Li C, Liu M, Zhang Y, Wang Y, Li J, Sun S, Liu X, Wu H, Feng C, Yao P, Jia Y, Zhang Y, Wei X, Wu F, Du C, Zhao X, Zhang S, Qu J (2023) Novel models by machine learning to predict prognosis of breast cancer brain metastases. J Transl Med. 2023;21(1):404. doi: 10.1186/s12967-023-04277-2 . Xu G, Wu H, Xu Y, Zhang Y, Lin F, Baklaushev VP, Chekhonin VP, Peltzer K, Wang X, Mao M, Wang G, Cui P, Zhang C (2021) Homogenous and Heterogenous Prognostic Factors for Patients with Bone Sarcoma. Orthop Surg. 2021;13(1):134–144. doi: 10.1111/os.12851 . Epub 2020 Dec 10. Grant SW, Hickey GL, Head SJ (2019) Statistical primer: multivariable regression considerations and pitfalls. Eur J Cardiothorac Surg. 2019;55(2):179–185. doi: 10.1093/ejcts/ezy403 . Deo RC (2015) Machine Learning in Medicine. Circulation. 2015;132(20):1920-30. doi: 10.1161/CIRCULATIONAHA.115.001593 . Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-3839137","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":265922693,"identity":"ae1b15f0-44cc-42a9-986e-e02ad29db6aa","order_by":0,"name":"Longteng Chao","email":"","orcid":"","institution":"The First Affiliated Hospital of Jinan University","correspondingAuthor":false,"prefix":"","firstName":"Longteng","middleName":"","lastName":"Chao","suffix":""},{"id":265922694,"identity":"544fa425-887d-4caf-9822-205e61271126","order_by":1,"name":"Xinmiao Ye","email":"","orcid":"","institution":"The First Affiliated Hospital of Jinan University","correspondingAuthor":false,"prefix":"","firstName":"Xinmiao","middleName":"","lastName":"Ye","suffix":""},{"id":265922695,"identity":"edbee0b8-718b-44a2-ac86-d45c2388d401","order_by":2,"name":"Junyuan Chen","email":"","orcid":"","institution":"The First Affiliated Hospital of Jinan University","correspondingAuthor":false,"prefix":"","firstName":"Junyuan","middleName":"","lastName":"Chen","suffix":""},{"id":265922696,"identity":"4e6e2181-38e1-4fa7-8dad-42f4b391ca87","order_by":3,"name":"Guorong She","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA1klEQVRIiWNgGAWjYDACCTiL+cCBDxWkaWFLPDjjDGlaeIwP87YQoUN+dvPh1zwV9+TM+dd8OMDbwCDPL3YAvxaDO8fSrHnOFBtbzni74YDkDgbDmbMTCGiRyDEz5m1LSNxw4+yGA4ZnGBIMbhPQIj8j/xtUy5kHBxLbiNDCcCOH+TFYy/kehgMHidFicCPNjHHOmQRjgxtsBgcbzkgQ9ov8jOTHH95UJMgZnD/8+POfCht5fmlCDgNGISRqJMAqJfCphAPmD2CK/wBRqkfBKBgFo2AEAgCv1ExAuUy/SAAAAABJRU5ErkJggg==","orcid":"","institution":"The First Affiliated Hospital of Jinan University","correspondingAuthor":true,"prefix":"","firstName":"Guorong","middleName":"","lastName":"She","suffix":""},{"id":265922697,"identity":"6470b6bd-eeea-452f-8740-d1ff6890ab49","order_by":4,"name":"Zhengang Zha","email":"","orcid":"","institution":"The First Affiliated Hospital of Jinan University","correspondingAuthor":false,"prefix":"","firstName":"Zhengang","middleName":"","lastName":"Zha","suffix":""}],"badges":[],"createdAt":"2024-01-06 07:29:33","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-3839137/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-3839137/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":49382656,"identity":"2517c13a-09c7-411a-beae-e0d6cb1c382e","added_by":"auto","created_at":"2024-01-09 19:29:47","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":28860,"visible":true,"origin":"","legend":"\u003cp\u003eFlow chart of the screening of SEER database\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-3839137/v1/e2c5ab2ab099771be0999e7b.png"},{"id":49382655,"identity":"11a451c8-bd53-44a8-bb39-0e236eb4890f","added_by":"auto","created_at":"2024-01-09 19:29:47","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":43475,"visible":true,"origin":"","legend":"\u003cp\u003eFeature selection based on the Boruta algorithm.\u003c/p\u003e\n\u003cp\u003eNote: The horizontal axis displays the names of the variables, whereas the vertical axis represents their corresponding Z-values. The box plot visualizes these Z-values as computed in the model. Green boxes indicate the top seven variables of importance, yellow boxes represent those considered tentative, and red boxes signify variables deemed to be of low importance in the analysis.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-3839137/v1/49db6c7bd08a42645beb00a7.png"},{"id":49382658,"identity":"f5f1385c-703d-49a8-8c29-4f9bfcac8b65","added_by":"auto","created_at":"2024-01-09 19:29:47","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":125476,"visible":true,"origin":"","legend":"\u003cp\u003eSpearman correlation analysis\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-3839137/v1/de74d60660e4d112ea584006.png"},{"id":49382932,"identity":"d44fdbb1-bc35-4b7e-ac4e-32db963c3f0f","added_by":"auto","created_at":"2024-01-09 19:37:47","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":122033,"visible":true,"origin":"","legend":"\u003cp\u003ePerformance comparison of survival analysis models\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-3839137/v1/4be34e6bfae02c61ca9c6ac2.png"},{"id":49382933,"identity":"3cce57af-50ec-4c4b-aeff-62dc25d30aa5","added_by":"auto","created_at":"2024-01-09 19:37:47","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":43086,"visible":true,"origin":"","legend":"\u003cp\u003eTime-dependent AUC\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-3839137/v1/884705a8ccb192ca3b1dcc15.png"},{"id":49898167,"identity":"64e8b985-0589-48f7-b458-06d40b697469","added_by":"auto","created_at":"2024-01-20 01:07:19","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":764830,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-3839137/v1/41bd8556-65c6-4ead-b260-dab53e00d283.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Comparison of Cox Regression to Machine Learning in Predicting Cancer-Specific Survival of Fibroblastic Osteosarcoma","fulltext":[{"header":"Introduction","content":"\u003cp\u003eOsteosarcoma (OS) represents the most prevalent malignant sarcoma of the skeletal system, primarily affecting children and adolescents, thus constituting the foremost bone cancer in these demographic groups [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. The manifestation of osteosarcoma is marked by a variety of characteristics, influenced by aspects such as tumor size, gene expression patterns, treatment responses, and histological subtypes. Fibroblastic osteosarcoma, a particular histological subtype of conventional osteosarcoma, accounts for approximately 10% of cases. This subtype is typified by spindle-shaped malignant cells, which occasionally exhibit epithelioid traits. While these cells frequently present pronounced cytological atypia, this manifestation is not universally observed. The unique cellular attributes of fibroblastic osteosarcoma delineate it as an aggressive tumor variant, often correlating with a worse prognosis for patients [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Identifying risk factors for early mortality in patients with fibroblastic osteosarcoma is, therefore, of paramount importance.\u003c/p\u003e \u003cp\u003eFibroblastic osteosarcoma, as a distinct subtype, presents significant challenges in the development of precise prognostic predictive models. These challenges primarily stem from difficulties in obtaining adequate case samples and the lack of reliable biomarkers. Furthermore, various studies have highlighted the limitations of the American Joint Committee on Cancer (AJCC) staging system in accurately predicting outcomes for osteosarcoma patients [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. The Tumor, Node, Metastasis (TNM) Staging System, despite its focus on tumor size, regional lymph node involvement, and distant metastasis, does not adequately consider critical individual factors such as patient age, which are essential for a comprehensive prognosis assessment [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eWei Zhong's study explored prognosis and prognostic factors in patients with telangiectatic osteosarcoma, with a specific emphasis on extremity-affected cases [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Additionally, Bo Chen investigated the risks and prognostic factors of distant metastasis in patients newly diagnosed with osteosarcoma [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. Current survival prediction methodologies predominantly utilize established clinical and sociodemographic predictors, frequently applying Cox proportional hazards regression analysis to develop nomograms [\u003cspan additionalcitationids=\"CR8 CR9 CR10\" citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eAlthough some models demonstrate promising concordance index (C-index) estimates, there is an inherent risk of overfitting associated with these assessments. In the context of precision medicine's rapid evolution, machine learning (ML) techniques are becoming progressively prevalent in medical imaging analysis and treatment planning [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. ML's approach, not requiring predefined relationships between input and output variables [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e], offers a more adaptable and comprehensive analysis framework. This method effectively considers all interactions and potential modifications of effects among these variables. Significantly, ML emerges as a highly efficient and accurate tool compared to traditional semi-parametric and parametric models, enhancing the capabilities of data analysis [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. The application of machine learning (ML) in the healthcare sector encompasses diagnosis, prognostic evaluation, and risk factor identification [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. Yet, data specific to the use of ML for prognostic analysis in fibroblastic osteosarcoma patients remains limited.\u003c/p\u003e \u003cp\u003eIn this study, we extracted data on fibroblastic osteosarcoma cases from the period of 1998 to 2019, utilizing the Surveillance, Epidemiology, and End Results (SEER) database. We primarily focused on cancer-specific survival (CSS) as our endpoint to highlight the direct impact of osteosarcoma on patient outcomes. The goal was to find out what medical things go along with CSS, then make a good predicting model for when it might happen. We compared Cox regression and machine learning methods to determine which one was more effective in predicting patient survival duration. In this research, the main aim is to find out which predictive model works best. It helps clinicians quickly spot dangers and choose right treatment plans faster. The work aims to improve how we make choices about treating and managing Fibroblastic Osteosarcoma.\u003c/p\u003e"},{"header":"Materials and Methods","content":"\u003cp\u003e2.1 Study population\u003c/p\u003e \u003cp\u003eIn this study, we used the Surveillance, Epidemiology and Results (SEER) database as our source of information. This public database doesn't need approval from an ethics group or agreement to be part of research. Following the rules set out by SEER, our research gathered details from patients with fibroblastic osteosarcoma between 1998 and 2019.\u003c/p\u003e \u003cp\u003eThe criteria for inclusion were: (1) a pathologically confirmed diagnosis of fibroblastic osteosarcoma; (2) primary tumor localization to bones and joints; and (3) availability of detailed survival status and duration data.\u003c/p\u003e \u003cp\u003eOn the other hand, exclusion criteria were applied as follows: (1) a survival time of less than one month; (2) cases identified solely through autopsy or death certificates; (3) lack of surgery information; and (4) cases with unknown primary site, survival status, survival time, or cause of death.\u003c/p\u003e \u003cp\u003e2.2 Predictor variables and outcomes\u003c/p\u003e \u003cp\u003eOur study involved the collection of comprehensive data, including various parameters such as age, gender, ethnicity, tumor laterality, primary tumor location, grade group, tumor dimensions, spread extent, stage, surgical procedures, and the use of radiation and chemotherapy. We utilized X-tile software for analyzing and optimally categorizing continuous variables like age and tumor size, establishing these categories based on the most impactful cut-off values.\u003c/p\u003e \u003cp\u003eFor tumor primary site classification, we divided sites into limbs and joints (codes C40.0; C40.1; C40.2; C40.3; C40.8), axial bones (C41.2; C41.3; C41.4), skull and mandible (C41.0; C41.1), among others. The tumor's extent was delineated as either confined within the periosteum, extending beyond it, or exhibiting further spread. Tumor stages were segmented into regional, localized, and distant categories.\u003c/p\u003e \u003cp\u003eRegarding treatment, we considered various surgical approaches including amputation (AMP), limb-salvage surgery (LSS), local treatment, or no surgical intervention, along with radiation and chemotherapy options. To enhance analytical clarity, smaller groups were combined: tumor grades were grouped into Grade I/II and Grade III/IV, reflecting the level of tumor differentiation.\u003c/p\u003e \u003cp\u003eThe primary outcome measured in our study was cancer-specific survival (CSS), defined as the duration from diagnosis to death attributed solely to the tumor.\u003c/p\u003e \u003cp\u003e2.3 Data preprocessing\u003c/p\u003e \u003cp\u003eIn our investigation, any unreported data points were categorized as missing. We closely checked the amount of data not available for each prognostic factor. We decided to add those with less than 25% missing info only. For the management of these missing values, we applied two distinct methods depending on the rate of data absence: For rates lower than 20%, the system used a method called multiple imputation. For rates more than 20%, we employed an approach known as K-Nearest Neighbor (KNN). This two-part method helped us do a more accurate and detailed handling of missing data in our study [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eWe evaluated multicollinearity among variables by computing the variance inflation factor (VIF). A VIF value below 10 was deemed indicative of negligible or non-significant multicollinearity in this analysis. Additionally, we performed Spearman correlation analysis to ascertain the interrelationships among variables. In this analysis, a correlation coefficient exceeding 0.5 signified a notable correlation between the variables [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e].\u003c/p\u003e \u003cp\u003e2.4 Model development and evaluation\u003c/p\u003e \u003cp\u003eIn our research, models were developed using Cox regression and contrasted with six distinct machine learning algorithms: Survival Tree, Extra Survival Trees, Random Survival Forest, Gradient Boosting Survival Analysis, Fast Kernel Survival SVM, and Minlip Survival Analysis. Patient data was divided, dedicating 70% for training and 30% for validation, with the random number table method ensuring randomized allocation. The comparative approach between the training and validation sets was tailored according to the type of outcome variable; continuous variables were analyzed using the t-test, while categorical variables were assessed with either the chi-square test or Fisher\u0026rsquo;s exact test. Furthermore, a blend of grid search and multiple cross-validation techniques was employed to identify the optimal parameter values that yielded the highest C-index values, thus defining the model parameters.\u003c/p\u003e \u003cp\u003eIn our study, the variable selection process was enhanced using the Boruta algorithm, an advanced method for identifying key variables in a dataset. This algorithm contrasts the importance of actual features with synthetically generated shadow features through random variables. Through iterative analysis and selection, Boruta rigorously identifies and retains only those features with statistical significance. The integration of the Boruta algorithm into our study ensured the inclusion of critical variables, thereby greatly improving the accuracy and dependability of our predictive model, leading to more reliable and precise results [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eTo safeguard against overfitting, our model was rigorously evaluated, giving precedence to the validation set while also meticulously examining the training set, thus upholding a comprehensive evaluation framework. The model's discriminative efficacy is gauged using the concordance index (C-index). This index reveals the generalization capacity of the model, where a C-index ranging from 0.5 to 0.7 indicates weak predictive ability, 0.7 to 0.85 suggests moderate ability, and 0.85 to 1.0 represents strong predictive capacity.\u003c/p\u003e \u003cp\u003eTo assess our model's ability to distinguish between outcomes accurately, we applied various metrics: accuracy, which indicates the proportion of correct predictions; time-dependent Area Under the Curve (AUC) for assessing classification efficacy; precision, denoting the percentage of true positives among positive predictions; recall or sensitivity, pivotal for identifying true positives; and the F1 score, harmonizing precision with recall. The calibration of the model was ascertained using a calibration curve, where alignment with the 45\u0026deg; diagonal suggests improved calibration. Additionally, Decision Curve Analysis (DCA) was utilized to measure the net clinical benefit, reinforcing the model's statistical soundness and clinical relevance.\u003c/p\u003e \u003cp\u003e2.5 Model interpretation\u003c/p\u003e \u003cp\u003eIn our study, the SHapley Additive exPlanation (SHAP) framework was instrumental in interpreting the outcomes of our machine learning model [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. We applied SHAP to elucidate the insights gained from the final model, specifically focusing on a detailed analysis of the risk factors affecting mortality in patients with fibroblastic osteosarcoma. This approach involved assessing the impact of various features in the predictive process and their significance in influencing patient outcomes.\u003c/p\u003e \u003cp\u003eOur study meticulously adhered to the guidelines outlined in the Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD) statement [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e], ensuring a clear and transparent depiction of our predictive modeling approach. All statistical analyses were executed using R software (version 4.3.0) and Python software (version 3.11.6), with a P value less than 0.05 deemed to indicate statistical significance.\u003c/p\u003e"},{"header":"Result","content":"\u003cp\u003e3.1 Patient characteristic\u003c/p\u003e \u003cp\u003eBetween 1998 and 2019, the SEER database provided records for 459 individuals diagnosed with fibroblastic osteosarcoma. Adhering to the specified exclusion criteria, 394 patients were ultimately selected for inclusion in the study. A flowchart depicting the process of data selection and exclusion is illustrated in Fig.\u0026nbsp;1.\u003c/p\u003e \u003cp\u003eIn our study, instances of unknown data within the final sample were handled as missing values. It was crucial to note that the missing data percentage for each variable was kept below 25%, a key aspect in preserving the integrity and robustness of our analysis. This approach was effective in managing the challenges posed by incomplete data.\u003c/p\u003e \u003cp\u003eConsidering that the missing data rate for each variable did not exceed 25%, we opted to retain all variables without excluding any due to low rates of missing data. Among these, Tumor Size had the highest missing rate at 23%, followed by Grade Group at 16%, and Extension at 11%. To effectively manage these missing values, we applied methods such as multiple imputation and the K-Nearest Neighbor (KNN) technique for the imputation of missing data.\u003c/p\u003e \u003cp\u003eUtilizing the X-tile program, we determined the optimal cutoff points for age as 40 and 57 years, leading to the categorization of age into three groups: under 40 years, between 41 and 57 years, and over 57 years. In a similar vein, cutoff points for tumor size were identified, classifying it into three categories: up to 60mm, between 61mm and 103mm, and over 103mm. Establishing these categories was instrumental in enhancing the precision and depth of our data analysis.\u003c/p\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e presents the demographic and clinical characteristics of the fibroblastic osteosarcoma patients included in our study. The majority, or 66.2%, were aged 0\u0026ndash;40 years, with the age groups of 41\u0026ndash;57 years and over 57 years comprising 22.8% and 10.9% of the cohort, respectively. The distribution of gender was nearly equal, with females representing 48.0% and males 52.0%. Most patients were White (82.7%), with Black (9.4%) and other racial groups (7.9%) also represented. Tumor laterality was almost evenly split between left-sided (44.9%) and right-sided (39.3%), and 15.7% of cases were not applicable to a specific paired site. The majority of tumors were located in limbs and joints (75.6%), while a smaller proportion occurred in axial bones (11.4%) and skull and mandible regions (12.9%). Most patients presented with higher-grade tumors (Grade III-IV, 83.0%), while a smaller percentage had Grade I-II tumors (15.0%). The sizes of the tumors were distributed as follows: \u0026le;60mm (30.5%), 61-103mm (43.7%), and \u0026gt;\u0026thinsp;103mm (25.9%). Over half of the tumors were confined within the periosteum (53.8%), while some extended beyond it (39.6%) or demonstrated further spread (6.6%). The staging of the tumors varied, with regional stage being most common (45.7%), followed by localized (40.6%) and distant stages (13.7%). Surgical procedures included limb salvage surgery (LSS, 52.8%), amputation (AMP, 17.3%), and local treatments (20.3%), with a minority undergoing no surgery (9.6%). The majority of patients did not receive radiation therapy (90.6%), while a small number did (9.4%). Chemotherapy was administered to a significant portion of the patients (74.5%), with the remainder having no chemotherapy or unknown status (24.6%). Statistical analysis using P-value indicated no significant differences between the training and validation groups, suggesting a balanced distribution of these characteristics.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eDemographic characteristics of patients with fibroblastic osteosarcoma\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eOverall (N\u0026thinsp;=\u0026thinsp;394)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eTraining Group (N\u0026thinsp;=\u0026thinsp;275)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eValidation Group (N\u0026thinsp;=\u0026thinsp;119)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eP-value\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge Group\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eage0-40\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e261 (66.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e180 (65.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e81 (68.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.770\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eage41-57\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e90 (22.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e63 (22.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e27 (22.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eage\u0026thinsp;\u0026gt;\u0026thinsp;57\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e43 (10.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e32 (11.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e11 (9.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSex\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFemale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e189 (48.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e137 (49.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e52 (43.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.314\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMale\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e205 (52.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e138 (50.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e67 (56.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eRace Group\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWhite\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e326 (82.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e233 (84.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e93 (78.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.091\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBlack\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e37 (9.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e20 (7.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e17 (14.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOther\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e31 (7.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e22 (8.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e9 (7.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eLaterality\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLeft\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e177 (44.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e125 (45.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e52 (43.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.428\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRight\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e155 (39.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e111 (40.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e44 (37.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNot a paired site\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e62 (15.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e39 (14.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e23 (19.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003ePrimary Site\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAxial bones\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e45 (11.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e33 (12.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e12 (10.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.854\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLimbs and joints\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e298 (75.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e207 (75.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e91 (76.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSkull and mandible, etc\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e51 (12.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e35 (12.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e16 (13.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eGrade Group\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGrade I-II\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e59 (15.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e42 (15.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e17 (14.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.922\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGrade III-IV\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e335 (85.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e233 (84.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e102 (85.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTumorsize Group\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e≦\u0026thinsp;60\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e120 (30.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e84 (30.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e36 (30.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.535\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e61\u0026ndash;103\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e172 (43.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e124 (45.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e48 (40.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026gt;\u0026thinsp;103\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e102 (25.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e67 (24.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e35 (29.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eExtension\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInside periosteum\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e157 (39.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e113 (41.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e44 (37.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.701\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBeyond periosteum\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e211 (53.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e145 (52.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e66 (55.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFurther extension\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e26 (6.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e17 (6.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e9 (7.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTumor Stage\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLocalized\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e160 (40.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e112 (40.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e48 (40.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.473\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRegional\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e180 (45.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e129 (46.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e51 (42.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDistant\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e54 (13.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e34 (12.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e20 (16.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSurgery Type\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAMP\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e68 (17.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e46 (16.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e22 (18.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.725\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLocal treatment\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e80 (20.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e60 (21.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e20 (16.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLSS\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e208 (52.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e143 (52.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e65 (54.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo surgery\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e38 (9.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e26 (9.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e12 (10.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eRadiation Group\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003eOverall (N\u0026thinsp;=\u0026thinsp;394)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003eTraining Group (N\u0026thinsp;=\u0026thinsp;275)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003eValidation Group (N\u0026thinsp;=\u0026thinsp;119)\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cb\u003eP-value\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNone/Unknown\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e357 (90.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e246 (89.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e111 (93.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.314\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e37 (9.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e29 (10.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e8 (6.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eChemotherapy\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNo/Unknown\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e97 (24.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e70 (25.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e27 (22.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.648\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e297 (75.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e205 (74.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e92 (77.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e3.2 Feature selection\u003c/p\u003e \u003cp\u003eThrough the application of Z-values, our study pinpointed ten critical variables significantly associated with fibroblastic osteosarcoma, which included Tumor Stage, Tumor Size Group, Radiation, Age Group, Primary Site, Extension, Surgery Type, Sex, Grade Group, and Chemotherapy, as depicted in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e2\u003c/span\u003e. Multicollinearity assessments, using Variance Inflation Factors (VIF) with all values under 10, indicated no substantial inter-variable conflicts. Spearman correlation analysis identified a significant correlation between Tumor Stage and Extension (0.50), while correlations among other variables remained relatively lower (all below 0.5), as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e3\u003c/span\u003e. Consequently, to eliminate redundancy, Extension was excluded from further analysis. The remaining nine variables were thus recognized as independent prognostic factors for fibroblastic osteosarcoma. Detailed in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, these factors were used in different models to predict cancer-specific survival.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e3.3 Model performance comparisons\u003c/p\u003e \u003cp\u003eIn the comparative analysis against the Cox regression model, the machine learning models demonstrated significantly enhanced performance, as detailed in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, and illustrated in Figs.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e4\u003c/span\u003e and \u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e5\u003c/span\u003e. The machine learning models excelled across various critical performance metrics, including Accuracy, Recall, F1 score, and the Concordance index (C-index), consistently surpassing the Cox regression model in these aspects.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eThe performance metrics of the of Cox regression and cross-validated machine learning algorithms\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"10\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c10\" colnum=\"10\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c5\" namest=\"c4\"\u003e \u003cp\u003eRecall\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c7\" namest=\"c6\"\u003e \u003cp\u003eF1 score\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c9\" namest=\"c8\"\u003e \u003cp\u003eC-index\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c10\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eMean AUC\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCSS 3 year\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCSS 5 year\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eCSS 3 year\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eCSS 5 year\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eCSS 3 year\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eCSS 5 year\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003etraining\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003evalidation\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eCox regression\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.77\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.77\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.55\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.53\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.56\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.57\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.74\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.68\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.77\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eSurvivalTree\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.88\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.87\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.75\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.72\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.76\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.76\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.90\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.81\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.86\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eExtraSurvivalTrees\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.91\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.89\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.81\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.74\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.83\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.80\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.92\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.80\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.89\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eRandomSurvivalForest\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.89\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.87\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.78\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.72\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.79\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.76\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.89\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.79\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.88\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eGradientBoostingSurvivalAnalysis\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.88\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.88\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.77\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.73\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.78\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.78\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.92\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.79\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.86\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eFastKernelSurvivalSVM\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.93\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.90\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.89\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.80\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.87\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.82\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.94\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.79\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.87\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eMinlipSurvivalAnalysis\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.91\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.92\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.81\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.79\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.83\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.92\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e0.81\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e0.87\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFor assessing the time-dependent Area Under the Curve (AUC) values, our analysis selected 15 key time points spanning from the 5th to 95th percentile of the survival time distribution. This methodology facilitated an extensive examination of the dataset. The findings indicated that, at most of these selected time points, the machine learning model demonstrated enhanced performance in comparison to the Cox regression model, as delineated in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e5\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eThe AUC scores of the different models showed variation over time, with some models initially having high scores that later declined, while others either increased or maintained a consistent level. The Extra Survival Trees Model notably achieved the highest average AUC score.\u003c/p\u003e \u003cp\u003eIn an in-depth evaluation of the Extra Survival Trees Model, calibration curves and Decision Curve Analyses (DCA) were conducted, as illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e6\u003c/span\u003e. These calibration curves demonstrated a strong congruence between the model\u0026rsquo;s predictions and actual 5-year cancer-specific survival rates, affirming the model's precision and dependability (as shown in Figs.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e6\u003c/span\u003eE and F). The DCA for predicting 3-year and 5-year cancer-specific survival rates by this model, applied to both training and validation cohorts (depicted in Figs.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e6\u003c/span\u003eA to D), highlighted the model's substantial clinical applicability.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e3.4 Model interpretation of machine learning\u003c/p\u003e \u003cp\u003eIn our study, the Extra Survival Trees Model was analyzed using SHAP (SHapley Additive exPlanation) to understand the model's rationale. The SHAP summary plots, illustrated in the figure, highlighted 'Age Group', 'Tumor Stage', and 'Tumor Size Group' as the most influential factors in predicting survival outcomes for patients with fibroblastic osteosarcoma. The plots (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e6\u003c/span\u003e) revealed several critical insights:\u003c/p\u003e \u003cp\u003eAge Group: Elevated ages were found to correlate with increased risk scores.\u003c/p\u003e \u003cp\u003eTumor Stage: More advanced tumor stages were associated with higher risk scores.\u003c/p\u003e \u003cp\u003eTumor Size Group: Larger tumor sizes linked to escalated risk scores.\u003c/p\u003e \u003cp\u003eRadiation Group: Previous radiation therapy indicated higher risk scores, while its absence was associated with lower scores.\u003c/p\u003e \u003cp\u003eSurgery Type: Surgical interventions typically resulted in lower risk scores, contrasting with higher scores in the absence of surgery.\u003c/p\u003e \u003cp\u003ePrimary Site: Tumors located in limbs and joints generally showed lower risk scores compared to those in axial bones or the skull/mandible.\u003c/p\u003e \u003cp\u003eSex: Male patients exhibited higher risk scores than female patients.\u003c/p\u003e \u003cp\u003eChemotherapy: Receiving chemotherapy was typically associated with higher risk scores.\u003c/p\u003e \u003cp\u003eGrade Group: Elevated tumor grades corresponded to increased risk scores.\u003c/p\u003e \u003cp\u003eThese insights from the SHAP analysis provide a detailed understanding of the factors influencing the predictive model\u0026rsquo;s outcomes.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eOS is acknowledged as the most common malignant bone tumor, primarily impacting adolescents and children. It is characterized by a poor prognosis and limited treatment options [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. Given the rarity of OS instances, the SEER database [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e] was employed as the principal data source to secure adequate data for comprehensive analysis.\u003c/p\u003e \u003cp\u003eIn this study, Cox regression and various machine learning algorithms were utilized and compared to predict cancer-specific survival in patients with fibroblastic osteosarcoma. The applied machine learning algorithms encompassed Survival Tree, Extra Survival Trees, Random Survival Forest, Gradient Boosting Survival Analysis, Fast Kernel Survival SVM, and Minlip Survival Analysis. This integrated approach facilitated a detailed assessment of the predictive efficacy of these diverse methods in the context of fibroblastic osteosarcoma.\u003c/p\u003e \u003cp\u003eUpon comparing various survival analysis models, the \"Extra Survival Trees\" model emerged as the most superior in terms of overall performance. This conclusion is primarily derived from the model's mean Area Under the Curve (AUC), an essential metric for evaluating predictive accuracy. Within our dataset, the Extra Survival Trees model exhibited a mean AUC of 0.89, markedly surpassing that of other models. This average AUC demonstrates the model's consistent predictive capacity over time. The notable score attained by the Extra Survival Trees model in this measure underscores its significant superiority in predictive precision and reliability, particularly in handling intricate survival data. Besides the mean AUC, other critical metrics, including accuracy, recall, F1 score, and Concordance Index (C-index) in both training and validation phases, were also considered. In these respects, the Extra Survival Trees model demonstrated remarkable performance, solidifying its position as the premier choice in our study. Furthermore, the calibration curve indicated the model's high predictive accuracy, and the Decision Curve Analysis (DCA) validated its clinical utility.\u003c/p\u003e \u003cp\u003eTo identify independent prognostic factors with utmost precision, a comprehensive analysis of clinical variables was conducted. Optimal cutoff values were established at 40 and 57 years for age, and 60 and 103 millimeters for tumor size, respectively. Following the conversion of continuous variables into categorical ones, the Boruta algorithm was employed in conjunction with the variance inflation factor (VIF) and Spearman correlation analysis to enhance our variable selection process. This meticulous methodology enabled the selection of pivotal clinical attributes, including Age Group, Primary Site, Sex, Grade Group, Tumor Size Group, Tumor Stage, Surgery Type, Radiation Group, and Chemotherapy. Subsequently, these attributes were systematically integrated into both Cox regression and machine learning models to advance prognostic evaluations.\u003c/p\u003e \u003cp\u003eIn our thorough study of fibroblastic osteosarcoma, we found several separate prognostic factors that greatly influence patient results. The age factor became very important, and people over 57 years old were more likely to have a bad outlook. This increase might be because of late detection and a bigger chance for bone tumors in the axial osteosarcomas, which spread more easily. This usually happens to older people. Also, older patients may find it harder to handle tough treatments because they might have other health problems or long-lasting conditions. This makes their treatment and getting better more difficult [\u003cspan additionalcitationids=\"CR23 CR24\" citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eSimilarly, tumor staging, particularly the presence of metastasis, became a very important independent part that affects how long the patients with this disease live. Metastasis means cancer being more dangerous. It lets the malignant cells move out of their first place and go far away, making it harder to treat them later on. Spreading can happen in different places like lungs, bones, even brain or some organs. This shows the complex steps involved when cancer spreads [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eTumor size also played a pivotal role in determining prognosis. Larger tumors were consistently associated with inferior outcomes [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. This association could be attributed to the heightened likelihood of metastasis and the increased complexity in achieving complete surgical resection. Furthermore, larger tumor dimensions may signify a more aggressive tumor biology, further complicating the development of effective treatment strategies [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eIn the context of treatment strategies, radiation therapy has been identified as a high-risk factor for fibroblastic osteosarcoma. This is largely attributed to the relative radio-resistance of these tumors. Radiation therapy is often administered for palliative purposes in cases where tumors are inoperable. Consequently, patients undergoing radiation therapy typically exhibit a poorer prognosis. This outcome is partly because radiation therapy is not the preferred modality for treating osteosarcoma and is generally reserved for scenarios where surgical intervention is not feasible [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eSurgical treatment type significantly impacts the prognosis of osteosarcoma patients [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]. Although osteosarcoma is a rare tumor, surgery remains its primary effective treatment modality [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]. Our study found that limb-salvage surgery (LSS) is more effective in reducing risk scores compared to amputation. LSS not only preserves limb function and reduces psychological burden, aiding in social integration, but also due to its less invasive nature and lesser blood loss, it reduces the risk of postoperative complications, thus aiding in quicker patient recovery, decreasing the likelihood of postoperative complications, and enhancing treatment success rates. These advantages make LSS beneficial in terms of physical, psychological, and social aspects, improving patient rehabilitation and quality of life [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eTreating axial or the skull/mandible osteosarcomas is very hard, mostly because they have a high chance of spreading to other parts of the body and their locations are complex. These lumps usually live near big blood channels, making it more possible for cancer cells to spread through the blood [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]. Also, symptoms linked to spine tumors are often misunderstood. This can cause late diagnoses and increased chances of spreading the cancer throughout the body [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]. The confusing structure of the axial part makes major surgery harder. This can cause problems during the operation like too much blood loss, implant failure or paralysis which all make things bad for patients and their future health outlook. This is particularly the case when the tumor infiltrates crucial regions like the upper cervical area, vertebral artery foramen, or odontoid process. Such invasion significantly escalates the complexity of margin resection and complicates the treatment process [\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eGender differences matter a lot in how bad osteosarcoma is, with males usually getting worse results [\u003cspan additionalcitationids=\"CR40\" citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e]. This disparity might arise from several factors: firstly, male patients might exhibit lower adherence to recommended treatment protocols [\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e], and secondly, they may demonstrate less favorable responses to neoadjuvant chemotherapy compared to female patients [\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e]. Next, natural things like sex hormones inside the body could help make osteosarcoma cells grow or shrink in different ways between male and female patients [\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e, \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e]. Also, some research shows that poor habits like smoking and drinking in men can raise cancer risk and make the condition worse. This then affects how well they are expected to get better [\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eChemotherapy is commonly considered a fundamental component of osteosarcoma treatment, with numerous studies highlighting its potential advantages in enhancing prognosis [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e]. Nonetheless, our investigation proposes that chemotherapy may not consistently confer benefits. While chemotherapy aids in controlling tumor growth and dissemination, it can also compromise the immune system, increasing vulnerability to infections and other illnesses, thereby exerting adverse effects on patients [\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e, \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e]. Moreover, chemotherapy medications frequently entail a spectrum of side effects, including cardiotoxicity, nephrotoxicity, and the risks of infertility and secondary tumors, which can significantly impact patients' quality of life and adherence to treatment [\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eOur study also presents an alternative interpretation: the challenge of ascertaining the specific chemotherapy regimens and patients' adherence levels. Additionally, the introduction of a 'No/Unknown' category in the chemotherapy variable introduced ambiguity, impeding our ability to confirm whether patients indeed underwent chemotherapy. This limitation potentially impacted the statistical robustness of our findings.\u003c/p\u003e \u003cp\u003eUltimately, the grade of the tumor stands as an independent prognostic determinant [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e]. Elevated tumor grades are intrinsically linked to an augmented propensity for lung metastasis [\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e], underscoring the pivotal role of tumor grade in shaping both prognosis and strategic treatment considerations.\u003c/p\u003e \u003cp\u003eOur research addresses a critical gap in predictive modeling for fibroblastic osteosarcoma. Through the comprehensive collection of disease-related predictors, encompassing baseline patient data, clinical diagnoses, and therapeutic interventions, including both medical and surgical treatments, we harnessed the formidable capabilities of Machine Learning (ML). In recent years, ML has demonstrated remarkable proficiency in handling multifaceted variables and has found extensive application in cancer detection and prognostication [\u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e51\u003c/span\u003e, \u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e52\u003c/span\u003e]. Within this study, we meticulously crafted Cox regression and ML models, and an evaluation of their performance, employing five metrics (C-index, accuracy, recall, F1 score, and time-dependent AUC), was conducted. Our findings unequivocally indicate the impressive predictive capabilities of ML algorithms in the context of cancer-specific survival for fibroblastic osteosarcoma.\u003c/p\u003e \u003cp\u003eTo conduct a comprehensive assessment of ML model performance, we specifically chose to focus on a detailed analysis of the Extreme Survival Tree model, which demonstrated the best overall performance across various metrics. Within our analysis, we focused our attention on two fundamental metrics: the Calibration Curve and Decision Curve Analysis (DCA). The Calibration Curve was employed to gauge the degree of alignment between the model's predictions and actual events, shedding light on the model's predictive consistency. Simultaneously, DCA was utilized to quantify the clinical utility of the model across varying decision thresholds, aiding in the evaluation of its practical significance in clinical settings. The examination of these metrics brought to the forefront the strengths of the Extreme Survival Tree model. Notably, it showcased a high level of accuracy, closely mirroring real-world outcomes, and also demonstrated substantial clinical effectiveness, thereby offering robust decision support for clinicians.\u003c/p\u003e \u003cp\u003eMoreover, through the amalgamation of SHAP values, the utilization of machine learning (ML) techniques effectively identified pivotal predictive factors, ultimately culminating in the development of an exceedingly precise survival prediction model. This groundbreaking approach furnishes a visual and intuitive means for discerning both protective and risk factors, offering invaluable guidance for the realm of clinical judgment and decision-making. Our ML models consistently exhibited exceptional performance metrics, underscoring their formidable generalization capabilities and profound clinical relevance. These models are characterized by their provision of lucid explanations, thereby facilitating the prediction of survival rates and consequently enhancing the comprehension of the decision-making process pertaining to the assessment of disease severity.\u003c/p\u003e \u003cp\u003eAs is widely acknowledged, the Cox Proportional Hazards regression model stands as one of the most extensively employed clinical prediction models, playing a pivotal role in assisting clinicians in making informed decisions [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e53\u003c/span\u003e]. Nonetheless, a fundamental assumption underpinning the Cox model is the proportionality of hazards, signifying that the relative hazard remains consistent over time across various predictor or covariate levels. It is imperative for this assumption to hold true for the effective utilization of the Cox Proportional Hazards regression model. Nevertheless, in real-world practice, hazard ratios frequently exhibit temporal variability [\u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e54\u003c/span\u003e], which has the potential to diminish the predictive accuracy of the Cox regression model under specific circumstances.\u003c/p\u003e \u003cp\u003eIn the age of big data, machine learning (ML) algorithms enhance the quantification of individual patient risks and play a pivotal role in assessing disease prognosis. When compared to traditional statistical methods, ML models offer a key advantage \u0026ndash; their accurate applicability at both the individual and population levels, coupled with a high potential for clinical application. ML algorithms do not take into account factors such as non-proportionality, multicollinearity, or nonlinearity, thereby reducing prediction bias resulting from modeling uncertainty. However, their adoption in clinical practice faces a hurdle due to a lack of interpretability. Consequently, the integration of SHAP (SHapley Additive exPlanations) aims to elucidate how machine models process outputs in an easily comprehensible manner, mitigating the aforementioned limitations. To date, there has been no targeted application of machine learning algorithms for predicting the survival of patients with fibroblastic osteosarcoma.\u003c/p\u003e \u003cp\u003eIn the era characterized by the ubiquity of big data, machine learning (ML) algorithms play a pivotal role in significantly enhancing the precision of quantifying individual patient risks and facilitating the assessment of disease prognosis. In contrast to conventional statistical methods, ML models offer a substantial advantage due to their precise applicability at both the individual and population levels, coupled with their immense potential for clinical utility [\u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e55\u003c/span\u003e]. These ML algorithms adeptly tackle issues such as non-proportionality, multicollinearity, and nonlinearity within the data, thereby minimizing prediction bias arising from modeling uncertainties. Nonetheless, their widespread adoption in clinical practice often encounters hindrances stemming from challenges in interpretability. To address this limitation, the adoption of SHAP (SHapley Additive exPlanation) seeks to elucidate the decision-making process of machine models in a comprehensible format, effectively mitigating the aforementioned obstacles. As of the current state of knowledge, the targeted utilization of ML algorithms for predicting survival outcomes in patients afflicted with fibroblastic osteosarcoma remains an uncharted territory.\u003c/p\u003e \u003cp\u003eWhile our study offers valuable insights into the prediction of survival in fibroblastic osteosarcoma using both Cox regression and machine learning algorithms, it is essential to acknowledge several limitations. One notable constraint is the relatively small sample size, which not only limits the generalizability of our findings but also impacts the robustness of our statistical analysis. Furthermore, the absence of external validation raises concerns regarding the applicability of our results to diverse populations. Although the incorporation of SHAP (SHapley Additive exPlanation) values enhances the interpretability of our models, the inherent complexity of these machine learning approaches may still present challenges when translating them into practical clinical settings. Additionally, the presence of ambiguity introduced by the inclusion of a 'No/Unknown' option in chemotherapy data complicates the assessment of its influence on patient outcomes. It is important to note that limitations related to data inherent in the SEER database constrain the depth of our analysis.\u003c/p\u003e \u003cp\u003eIn anticipation of the future, it becomes imperative to conduct extensive studies involving larger and more diverse patient cohorts to validate and refine our predictive models. Subsequent research endeavors should be geared towards augmenting the interpretability and applicability of machine learning models within clinical contexts, ensuring their effective integration into healthcare practice. Moreover, addressing the intricacies associated with treatment data, particularly pertaining to chemotherapy regimens and patient adherence, will be pivotal for a more precise evaluation of treatment outcomes. By surmounting these challenges, we can advance our comprehension and management of fibroblastic osteosarcoma, thereby making a substantial contribution to the enhancement of patient care and ultimate clinical outcomes.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eIn short, we did a study that compared methods of cox regression and machine learning (ML) to predict the survival outcomes of patients with fibroblastic osteosarcoma. Importantly, this research looks at a part that has never been studied before in predicting results for fibroblastic osteosarcoma. Our findings indicate that ML algorithms, when coupled with SHapley Additive exPlanation (SHAP) values, demonstrate a promising predictive capability for survival in fibroblastic osteosarcoma. We need to be careful when looking at these results because a small sample size limits them and there's no outside checking. By adding to what's not well studied, our study sets up for more work and use in this area.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eMedical Scientific Research Foundation of Guangdong Province (A2021074).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting Interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors have no relevant financial or non-financial interests to disclose.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll authors significantly contributed to the conception and design of the study. Longteng Chao was responsible for data analysis and drafting the manuscript. Xinmiao Ye and Junyuan Chen played a key role in data collection. Guorong She and Zhengang Zha made substantial contributions to the study\u0026apos;s conception and design. All authors were actively involved in revising earlier versions of the manuscript and have read and approved the final version for publication.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets generated and analyzed during this study are accessible in the SEER database, which can be found at seer.cancer.gov.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003ePhilip T, Blay JY, Brunat-Mentigny M, Carrie C, Chauvot P, Farsi F, Fervers B, Gentet JC, Giammarile F, Kohler R, Mathoulin S, Patricot LM, Thiesse P (2001) French National Federation of Cancer (FNCLCC). Osteosarcoma. Br J Cancer. 2001;84 Suppl 2(Suppl 2):78\u0026ndash;80. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1054/bjoc.2000.1770\u003c/span\u003e\u003cspan address=\"10.1054/bjoc.2000.1770\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWHO Classification of Tumours Editorial Board (2020). WHO Classification of Tumours. Soft Tissue and Bone Tumors. 5th ed. Volume 3. International Agency for Research on Cancer; Lyon, France: 2020.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBacci G, Bertoni F, Longhi A, Ferrari S, Forni C, Biagini R, Bacchini P, Donati D, Manfrini M, Bernini G, Lari S (2003) Neoadjuvant chemotherapy for high-grade central osteosarcoma of the extremity. Histologic response to preoperative chemotherapy correlates with histologic subtype of the tumor. Cancer. 2003;97(12):3068-75. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1002/cncr.11456\u003c/span\u003e\u003cspan address=\"10.1002/cncr.11456\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim MS, Lee SY, Lee TR, Cho WH, Song WS, Koh JS, Lee JA, Yoo JY, Jeon DG (2009) Prognostic nomogram for predicting the 5-year probability of developing metastasis after neo-adjuvant chemotherapy and definitive surgery for AJCC stage II extremity osteosarcoma. Ann Oncol. 2009;20(5):955\u0026thinsp;\u0026ndash;\u0026thinsp;60. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/annonc/mdn723\u003c/span\u003e\u003cspan address=\"10.1093/annonc/mdn723\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Epub 2009 Jan 19.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBalachandran VP, Gonen M, Smith JJ, DeMatteo RP (2015) Nomograms in oncology: more than meets the eye. Lancet Oncol. 2015;16(4):e173-80. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/S1470-2045(14)71116-7\u003c/span\u003e\u003cspan address=\"10.1016/S1470-2045(14)71116-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhong W, Luo W, Lin Z, Wu Z, Yuan Y, He Y (2023) Prognostic analysis of telangiectatic osteosarcoma of the extremities. Front Oncol. 2023;12:1105054. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fonc.2022.1105054\u003c/span\u003e\u003cspan address=\"10.3389/fonc.2022.1105054\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen B, Zeng Y, Liu B, Lu G, Xiang Z, Chen J, Yu Y, Zuo Z, Lin Y, Ma J (2021) Risk Factors, Prognostic Factors, and Nomograms for Distant Metastasis in Patients With Newly Diagnosed Osteosarcoma: A Population-Based Study. Front Endocrinol (Lausanne). 2021;12:672024. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fendo.2021.672024\u003c/span\u003e\u003cspan address=\"10.3389/fendo.2021.672024\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee RJ, Arshi A, Schwartz HC, Christensen RE (2015) Characteristics and prognostic factors of osteosarcoma of the jaws: a retrospective cohort study. JAMA Otolaryngol Head Neck Surg. 2015;141(5):470-7. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1001/jamaoto.2015.0340\u003c/span\u003e\u003cspan address=\"10.1001/jamaoto.2015.0340\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu Z, Yin J, Zhou Q, Yang J, Zeng B, Yeung SJ, Shen J, Cheng C (2020) Survival after pulmonary metastasectomy for relapsed osteosarcoma. J Thorac Cardiovasc Surg. 2022;163(2):469\u0026ndash;479.e8. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.jtcvs.2020.10.137\u003c/span\u003e\u003cspan address=\"10.1016/j.jtcvs.2020.10.137\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Epub 2020 Nov 24.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang J, Zhanghuang C, Tan X, Mi T, Liu J, Jin L, Li M, Zhang Z, He D (2022) A Nomogram for Predicting Cancer-Specific Survival of Osteosarcoma and Ewing's Sarcoma in Children: A SEER Database Analysis. Front Public Health. 2022;10:837506. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fpubh.2022.837506\u003c/span\u003e\u003cspan address=\"10.3389/fpubh.2022.837506\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXue W, Zhang Z, Yu H, Li C, Sun Y, An J, Qi L, Zhang J, Liu Q (2023) Development of nomogram and discussion of radiotherapy effect for osteosarcoma survival. Sci Rep. 2023;13(1):223. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41598-023-27476-9\u003c/span\u003e\u003cspan address=\"10.1038/s41598-023-27476-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJiang F, Jiang Y, Zhi H, Dong Y, Li H, Ma S, Wang Y, Dong Q, Shen H, Wang Y (2017) Artificial intelligence in healthcare: past, present and future. Stroke Vasc Neurol. 2017;2(4):230\u0026ndash;243. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1136/svn-2017-000101\u003c/span\u003e\u003cspan address=\"10.1136/svn-2017-000101\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKhan K, Ahmad W, Amin MN, Ahmad A, Nazar S, Alabdullah AA (2022) Compressive Strength Estimation of Steel-Fiber-Reinforced Concrete and Raw Material Interactions Using Advanced Algorithms. Polymers (Basel). 2022;14(15):3065. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/polym14153065\u003c/span\u003e\u003cspan address=\"10.3390/polym14153065\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXu L, Cai L, Zhu Z, Chen G (2023) Comparison of the cox regression to machine learning in predicting the survival of anaplastic thyroid carcinoma. BMC Endocr Disord. 2023;23(1):129. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12902-023-01368-5\u003c/span\u003e\u003cspan address=\"10.1186/s12902-023-01368-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Erratum in: BMC Endocr Disord. 2023;23(1):174.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJiang F, Jiang Y, Zhi H, Dong Y, Li H, Ma S, Wang Y, Dong Q, Shen H, Wang Y (2017) Artificial intelligence in healthcare: past, present and future. Stroke Vasc Neurol. 2017;2(4):230\u0026ndash;243. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1136/svn-2017-000101\u003c/span\u003e\u003cspan address=\"10.1136/svn-2017-000101\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJanssen KJ, Vergouwe Y, Donders AR, Harrell FE Jr, Chen Q, Grobbee DE, Moons KG (2009) Dealing with missing predictor values when applying clinical prediction models. Clin Chem. 2009;55(5):994\u0026ndash;1001. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1373/clinchem.2008.115345\u003c/span\u003e\u003cspan address=\"10.1373/clinchem.2008.115345\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Epub 2009 Mar 12.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTu YK, Kellett M, Clerehugh V, Gilthorpe MS (2005) Problems of correlations between explanatory variables in multiple regression analyses in the dental literature. Br Dent J. 2005;199(7):457\u0026thinsp;\u0026ndash;\u0026thinsp;61. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/sj.bdj.4812743\u003c/span\u003e\u003cspan address=\"10.1038/sj.bdj.4812743\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYue S, Li S, Huang X, Liu J, Hou X, Zhao Y, Niu D, Wang Y, Tan W, Wu J (2022) Machine learning for the prediction of acute kidney injury in patients with sepsis. J Transl Med. 2022;20(1):215. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12967-022-03364-0\u003c/span\u003e\u003cspan address=\"10.1186/s12967-022-03364-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLundberg SM, Lee S-I (2017) A unified approach to interpreting model predictions. Adv Neural Inf Process Syst 2017, 30.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHeus P, Reitsma JB, Collins GS, Damen JAAG, Scholten RJPM, Altman DG, Moons KGM, Hooft L (2020) Transparent Reporting of Multivariable Prediction Models in Journal and Conference Abstracts: TRIPOD for Abstracts. Ann Intern Med. 2020 Jun 2. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.7326/M20-0193\u003c/span\u003e\u003cspan address=\"10.7326/M20-0193\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Epub ahead of print.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChe WQ, Li YJ, Tsang CK, Wang YJ, Chen Z, Wang XY, Xu AD, Lyu J (2023) How to use the Surveillance, Epidemiology, and End Results (SEER) data: research design and methodology. Mil Med Res. 2023;10(1):50. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s40779-023-00488-2\u003c/span\u003e\u003cspan address=\"10.1186/s40779-023-00488-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXin S, Wei G (2020) Prognostic factors in osteosarcoma: A study level meta-analysis and systematic review of current practice. J Bone Oncol. 2020;21:100281. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.jbo.2020.100281\u003c/span\u003e\u003cspan address=\"10.1016/j.jbo.2020.100281\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIwata S, Ishii T, Kawai A, Hiruma T, Yonemoto T, Kamoda H, Asano N, Takeyama M (2014) Prognostic factors in elderly osteosarcoma patients: a multi-institutional retrospective study of 86 cases. Ann Surg Oncol. 2014;21(1):263-8. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1245/s10434-013-3210-4\u003c/span\u003e\u003cspan address=\"10.1245/s10434-013-3210-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Epub 2013 Aug 23.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTsuda Y, Ogura K, Shinoda Y, Kobayashi H, Tanaka S, Kawai A (2018) The outcomes and prognostic factors in patients with osteosarcoma according to age: a Japanese nationwide study with focusing on the age differences. BMC Cancer. 2018;18(1):614. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12885-018-4487-2\u003c/span\u003e\u003cspan address=\"10.1186/s12885-018-4487-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTsuchie H, Emori M, Nagasawa H, Miyakoshi N, Murahashi Y, Shimizu J, Mizushima E, Yamashita T, Shimada Y (2019) Prognosis of Primary Osteosarcoma in Elderly Patients: A Comparison between Young and Elderly Patients. Med Princ Pract. 2019;28(5):425\u0026ndash;431. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1159/000500404\u003c/span\u003e\u003cspan address=\"10.1159/000500404\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Epub 2019 Apr 17.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang J, Zhanghuang C, Tan X, Mi T, Liu J, Jin L, Li M, Zhang Z, He D (2022) A Nomogram for Predicting Cancer-Specific Survival of Osteosarcoma and Ewing's Sarcoma in Children: A SEER Database Analysis. Front Public Health. 2022;10:837506. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fpubh.2022.837506\u003c/span\u003e\u003cspan address=\"10.3389/fpubh.2022.837506\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBeird HC, Bielack SS, Flanagan AM, Gill J, Heymann D, Janeway KA, Livingston JA, Roberts RD, Strauss SJ, Gorlick R (2022) Osteosarcoma. Nat Rev Dis Primers. 2022;8(1):77. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41572-022-00409-y\u003c/span\u003e\u003cspan address=\"10.1038/s41572-022-00409-y\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Erratum in: Nat Rev Dis Primers. 2022;8(1):82.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHristov B, Shokek O, Frassica DA (2007) The role of radiation treatment in the contemporary management of bone tumors. J Natl Compr Canc Netw. 2007;5(4):456\u0026thinsp;\u0026ndash;\u0026thinsp;66. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.6004/jnccn.2007.0040\u003c/span\u003e\u003cspan address=\"10.6004/jnccn.2007.0040\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchwarz R, Bruland O, Cassoni A, Schomberg P, Bielack S (2009) The role of radiotherapy in oseosarcoma. Cancer Treat Res. 2009;152:147\u0026thinsp;\u0026ndash;\u0026thinsp;64.doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/978-1-4419-0284-9_7\u003c/span\u003e\u003cspan address=\"10.1007/978-1-4419-0284-9_7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMavrogenis AF, Abati CN, Romagnoli C, Ruggieri P (2012) Similar survival but better function for patients after limb salvage versus amputation for distal tibia osteosarcoma. Clin Orthop Relat Res. 2012;470(6):1735-48. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s11999-011-2238-7\u003c/span\u003e\u003cspan address=\"10.1007/s11999-011-2238-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Epub 2012 Jan 24.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMason GE, Aung L, Gall S, Meyers PA, Butler R, Kr\u0026uuml;g S, Kim M, Healey JH, Gorlick R (2013) Quality of life following amputation or limb preservation in patients with lower extremity bone sarcoma. Front Oncol. 2013;3:210. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fonc.2013.00210\u003c/span\u003e\u003cspan address=\"10.3389/fonc.2013.00210\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRitter J, Bielack SS (2010) Osteosarcoma. Ann Oncol. 2010;21 Suppl 7:vii320-5. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/annonc/mdq276\u003c/span\u003e\u003cspan address=\"10.1093/annonc/mdq276\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEvans DR, Lazarides AL, Visgauss JD, Somarelli JA, Blazer DG 3rd, Brigman BE, Eward WC (2020) Limb salvage versus amputation in patients with osteosarcoma of the extremities: an update in the modern era using the National Cancer Database. BMC Cancer. 2020;20(1):995. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12885-020-07502-z\u003c/span\u003e\u003cspan address=\"10.1186/s12885-020-07502-z\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eQi L, Ren X, Liu Z, Li S, Zhang W, Chen R, Chen C, Tu C, Li Z (2020) Predictors and Survival of Patients with Osteosarcoma After Limb Salvage versus Amputation: A Population-Based Analysis with Propensity Score Matching. World J Surg. 2020;44(7):2201\u0026ndash;2210. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s00268-020-05471-9\u003c/span\u003e\u003cspan address=\"10.1007/s00268-020-05471-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFahey M, Spanier SS, Vander Griend RA (1992) Osteosarcoma of the pelvis. A clinical and histopathological study of twenty-five patients. J Bone Joint Surg Am. 1992;74(3):321\u0026thinsp;\u0026ndash;\u0026thinsp;30.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSundaresan N, Rosen G, Boriani S (2008) Primary malignant tumors of the spine. Orthop Clin North Am. 2009;40(1):21\u0026ndash;36, v. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ocl.2008.10.004\u003c/span\u003e\u003cspan address=\"10.1016/j.ocl.2008.10.004\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGrimer RJ, Taminiau AM, Cannon SR (2002) Surgical Subcommitte of the European Osteosarcoma Intergroup. Surgical outcomes in osteosarcoma. J Bone Joint Surg Br. 2002;84(3):395\u0026ndash;400. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1302/0301-620x.84b3.12019\u003c/span\u003e\u003cspan address=\"10.1302/0301-620x.84b3.12019\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWeeden S, Grimer RJ, Cannon SR, Taminiau AH, Uscinska BM (2001) European Osteosarcoma Intergroup. The effect of local recurrence on survival in resected osteosarcoma. Eur J Cancer. 2001;37(1):39\u0026ndash;46. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/s0959-8049(00)00362-2\u003c/span\u003e\u003cspan address=\"10.1016/s0959-8049(00)00362-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWilliams LA, Richardson M, Marcotte EL, Poynter JN, Spector LG (2019) Sex ratio among childhood cancers by single year of age. Pediatr Blood Cancer. 2019;66(6):e27620. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1002/pbc.27620\u003c/span\u003e\u003cspan address=\"10.1002/pbc.27620\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Epub 2019 Feb 28.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDuchman KR, Gao Y, Miller BJ (2015) Prognostic factors for survival in patients with high-grade osteosarcoma using the Surveillance, Epidemiology, and End Results (SEER) Program database. Cancer Epidemiol. 2015;39(4):593-9. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.canep.2015.05.001\u003c/span\u003e\u003cspan address=\"10.1016/j.canep.2015.05.001\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Epub 2015 May 20.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePetrilli AS, Gentil FC, Epelman S, Lopes LF, Bianchi A, Lopes A, Figueiredo MT, Marques E, De Bellis N, Consentino E, et al (1991) Increased survival, limb preservation, and prognostic factors for osteosarcoma. Cancer. 1991;68(4):733-7. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1002/1097-0142(19910815)68:4\u0026lt;733::aid-cncr2820680412\u0026gt;3.0.co;2-0\u003c/span\u003e\u003cspan address=\"10.1002/1097-0142(19910815)68:4%3C733::aid-cncr2820680412%3E3.0.co;2-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBielack SS, Kempf-Bielack B, Delling G, Exner GU, Flege S, Helmke K, Kotz R, Salzer-Kuntschik M, Werner M, Winkelmann W, Zoubek A, J\u0026uuml;rgens H, Winkler K (2023) Prognostic factors in high-grade osteosarcoma of the extremities or trunk: an analysis of 1,702 patients treated on neoadjuvant cooperative osteosarcoma study group protocols. J Clin Oncol. 2002;20(3):776\u0026thinsp;\u0026ndash;\u0026thinsp;90. doi: 10.1200/JCO.2002.20.3.776. Corrected and republished in: J Clin Oncol. 2023;41(27):4323\u0026ndash;4337.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePan Y, Chen D, Hu T, Lv G, Dai Z (2019 ) Characteristics and Prognostic Factors of Patients With Osteosarcoma Older Than 60 Years From the SEER Database. Cancer Control. 2019 Jan-Dec;26(1):1073274819888893. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1177/1073274819888893\u003c/span\u003e\u003cspan address=\"10.1177/1073274819888893\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWagle S, Park SH, Kim KM, Moon YJ, Bae JS, Kwon KS, Park HS, Lee H, Moon WS, Kim JR, Jang KY (2015) DBC1/CCAR2 is involved in the stabilization of androgen receptor and the progression of osteosarcoma. Sci Rep. 2015;5:13144. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/srep13144\u003c/span\u003e\u003cspan address=\"10.1038/srep13144\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKatzke VA, Kaaks R, K\u0026uuml;hn T (2015) Lifestyle and cancer risk. Cancer J. 2015 Mar-Apr;21(2):104\u0026ndash;10. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1097/PPO.0000000000000101\u003c/span\u003e\u003cspan address=\"10.1097/PPO.0000000000000101\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSanford NN, Martin AM, Brunner AM, Cote GM, Choy E, DeLaney TF, Aizer AA, Chen YL (2018) Secondary Acute Leukemia in Sarcoma Patients: A Population-Based Study. Int J Radiat Oncol Biol Phys. 2018;100(3):687\u0026ndash;694. doi: 10.1016/j.ijrobp.2017.11.011. Epub 2017 Nov 14.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAnderson PM (2017) Immune Therapy for Sarcomas. Adv Exp Med Biol. 2017;995:127\u0026ndash;140. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/978-3-319-53156-4_6\u003c/span\u003e\u003cspan address=\"10.1007/978-3-319-53156-4_6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFerrari S, Palmerini E (2007) Adjuvant and neoadjuvant combination chemotherapy for osteogenic sarcoma. Curr Opin Oncol. 2007;19(4):341-6. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1097/CCO.0b013e328122d73f\u003c/span\u003e\u003cspan address=\"10.1097/CCO.0b013e328122d73f\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDeng G, Chen P (2023) Characteristics and prognostic factors of adult patients with osteosarcoma from the SEER database. Medicine (Baltimore). 2023;102(37):e33653. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1097/MD.0000000000033653\u003c/span\u003e\u003cspan address=\"10.1097/MD.0000000000033653\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXie L, Huang W, Wang H, Zheng C, Jiang J (2020) Risk factors for lung metastasis at presentation with malignant primary osseous neoplasms: a population-based study. J Orthop Surg Res. 2020;15(1):32. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s13018-020-1571-5\u003c/span\u003e\u003cspan address=\"10.1186/s13018-020-1571-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee C, Light A, Alaa A, Thurtle D, van der Schaar M, Gnanapragasam VJ (2021) Application of a novel machine learning framework for predicting non-metastatic prostate cancer-specific mortality in men using the Surveillance, Epidemiology, and End Results (SEER) database. Lancet Digit Health. 2021;3(3):e158-e165. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/S2589-7500(20)30314-9\u003c/span\u003e\u003cspan address=\"10.1016/S2589-7500(20)30314-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Epub 2021 Feb 3.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi C, Liu M, Zhang Y, Wang Y, Li J, Sun S, Liu X, Wu H, Feng C, Yao P, Jia Y, Zhang Y, Wei X, Wu F, Du C, Zhao X, Zhang S, Qu J (2023) Novel models by machine learning to predict prognosis of breast cancer brain metastases. J Transl Med. 2023;21(1):404. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12967-023-04277-2\u003c/span\u003e\u003cspan address=\"10.1186/s12967-023-04277-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXu G, Wu H, Xu Y, Zhang Y, Lin F, Baklaushev VP, Chekhonin VP, Peltzer K, Wang X, Mao M, Wang G, Cui P, Zhang C (2021) Homogenous and Heterogenous Prognostic Factors for Patients with Bone Sarcoma. Orthop Surg. 2021;13(1):134\u0026ndash;144. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1111/os.12851\u003c/span\u003e\u003cspan address=\"10.1111/os.12851\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Epub 2020 Dec 10.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGrant SW, Hickey GL, Head SJ (2019) Statistical primer: multivariable regression considerations and pitfalls. Eur J Cardiothorac Surg. 2019;55(2):179\u0026ndash;185. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1093/ejcts/ezy403\u003c/span\u003e\u003cspan address=\"10.1093/ejcts/ezy403\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDeo RC (2015) Machine Learning in Medicine. Circulation. 2015;132(20):1920-30. doi: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1161/CIRCULATIONAHA.115.001593\u003c/span\u003e\u003cspan address=\"10.1161/CIRCULATIONAHA.115.001593\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Fibroblastic Osteosarcoma, Cancer-Specific Survival, Machine Learning, Cox Regression, SHAP Analysis","lastPublishedDoi":"10.21203/rs.3.rs-3839137/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-3839137/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eBone cancer called osteosarcoma (OS), especially its fibroblastic type, makes things very hard in the world of bone diseases. This happens because of its fierce character and the complexity involved in deciding outcomes. Current prognostic models, like the American Joint Committee on Cancer (AJCC) system and Tumor Node Metastasis (TNM) Staging System, don't always fully include important individual patient factors such as age, sex and race. These things are very important for making a correct prediction.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eA total of 394 patients with fibroblastic osteosarcoma were included in the study, adhering to specified inclusion and exclusion criteria. The cohort was subsequently segregated into training and validation sets at a 7:3 ratio. X-tile software facilitated the determination of optimal age and tumor size cutoffs. Missing data were managed using multiple imputation and K-Nearest Neighbor (KNN) methods. The primary endpoint was cancer-specific survival (CSS), categorized into binary data (survival status at 3 and 5 years) and time-to-event data. Independent prognostic factors were ascertained using the Boruta algorithm, which informed the construction of predictive models employing Cox regression and diverse machine learning algorithms such as Survival Tree, Extra Survival Trees, Random Survival Forest, Gradient Boosting Survival Analysis, Fast Kernel Survival SVM, and Minlip Survival Analysis. Model performance metrics included the concordance index (C-index), accuracy, recall, F1 score, and time-dependent Area Under the Curve (AUC). A calibration plot was generated to validate the accuracy of the most proficient machine learning model. Decision curve analysis (DCA) was implemented to ascertain the model's clinical utility. Additionally, we used the SHapley Additive exPlanations (SHAP) method to show how important our model found key things that can predict outcomes.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eFor age, the determined optimal cutoff points were established at 40 and 57 years. Regarding tumor size, these points were set at 60mm and 103mm. Our study identified nine significant independent prognostic factors impacting the cancer-specific survival in patients with fibroblastic osteosarcoma. These included age group, tumor stage, tumor size group, radiation, surgery type, primary site, sex, chemotherapy, and grade group. Comparative analysis of different algorithms, utilizing metrics such as accuracy, recall, F1 score, C-index, and time-dependent AUC, highlighted the Extra Survival Trees model as the superior predictive tool for machine learning. This model demonstrated high efficiency (3-year CSS accuracy: 0.91, 5-year CSS accuracy: 0.89), notable recall rates (3-year: 0.81, 5-year: 0.74), and robust F1 scores (3-year: 0.83, 5-year: 0.80), along with an average AUC of 0.89 and a C-index of 0.92 for training and 0.80 for validation. The calibration curve for this model indicated high predictive accuracy, and its clinical usefulness was further corroborated by decision curve analysis (DCA). SHAP analysis identified 'age group', 'tumor stage', and 'tumor size group' as the three most influential variables impacting cancer-specific survival predictions in fibroblastic osteosarcoma. Our study suggested otherwise than previous ones. It showed that radiation and chemotherapy may not work for treating this type of bone cancer called fibroblastic osteosarcoma.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eResearch indicates that predictive analysis using machine learning outperforms traditional methods in forecasting outcomes for patients with fibroblastic osteosarcoma. This development offers considerable promise for enhancing tailored therapeutic approaches and prognostic outcomes in fibroblastic osteosarcoma.\u003c/p\u003e","manuscriptTitle":"Comparison of Cox Regression to Machine Learning in Predicting Cancer-Specific Survival of Fibroblastic Osteosarcoma","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-01-09 19:29:42","doi":"10.21203/rs.3.rs-3839137/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"84e55c7a-92a4-4f2e-8b5f-cc661fdd9049","owner":[],"postedDate":"January 9th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2024-01-20T00:59:11+00:00","versionOfRecord":[],"versionCreatedAt":"2024-01-09 19:29:42","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-3839137","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-3839137","identity":"rs-3839137","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}