Published online Nov 24, 2026. doi: 10.5306/wjco.126582
Revised: September 24, 2026
Accepted: November 2, 2026
Published online: November 24, 2026
Processing time: 101 Days and 18.1 Hours
Postoperative local-recurrence estimates could inform surveillance discussions after soft tissue sarcoma surgery, but machine-learning performance can be over
To develop and evaluate a model for binary 3-year local-recurrence status among patients with ascertainable status and compare it with prespecified logistic-regre
This retrospective two-institution study analyzed 1035 development patients with ascertainable binary recurrence status at 3 years (331 recurrences) and 212 patients in a second-institution cohort (67 recurrences). The estimand was recurrence-status probability conditional on 3-year status being ascertainable, not cumulative incidence in all postoperative patients. The locked analytical package did not re
CatBoost had a mean internal area under the receiver operating characteristic curve (AUC) of 0.8126 and Brier score of 0.1584. In the second institution, its AUC was 0.8050 [95% confidence interval (CI): 0.7433-0.8668], Brier score 0.1615 (95%CI: 0.1331-0.1912), log loss 0.4884 (95%CI: 0.4146-0.5678), calibration intercept 0.148 (95%CI: -0.222 to 0.490), and calibration slope 0.985 (95%CI: 0.703-1.357). The CatBoost-minus-six-variable-logistic differences were -0.0083 for AUC (95%CI: -0.0480 to 0.0299) and 0.0043 for Brier score (95%CI: -0.0112 to 0.0198). A secondary full 12-variable logistic model also showed no material difference from CatBoost.
In the protocol-frozen second-institution re-evaluation, CatBoost retained discrimination but did not demonstrate incremental performance over logistic comparators. The study was not designed or powered to establish equiva
Core Tip: We compared 243 configurations across 14 base-model families plus three prespecified ensembles by repeated nested cross-validation. The selected CatBoost model was frozen and applied without recalibration in a protocol-frozen second-institution re-evaluation. CatBoost retained an area under the receiver operating characteristic curve of 0.8050 but did not demonstrate incremental performance over either a six-variable or full 12-variable logistic model; this was not an equivalence or non-inferiority analysis. The browser interface is provided for reproducibility and independent evaluation, not for clinical decision-making.
- Citation: Li ZF, Yang BC, Liu YB, Liu SQ, Zhang Q, Yu SJ. Machine learning and logistic regression for ascertainable three-year recurrence status after soft tissue sarcoma surgery. World J Clin Oncol 2026; 17(11): 126582
- URL: https://www.wjgnet.com/2218-4333/full/v17/i11/126582.htm
- DOI: https://dx.doi.org/10.5306/wjco.126582
Soft tissue sarcomas (STS) are rare mesenchymal malignancies that account for approximately 1% of adult cancers and comprise more than 100 histological subtypes[1,2]. Curative-intent resection remains central to treatment, yet local recur
Existing tools address related but not identical outcomes. A postoperative nomogram estimates local recurrence after limb-sparing surgery[7]. Personalised sarcoma care was developed to support treatment decisions in high-grade extre
Machine-learning methods can represent nonlinear associations and interactions, but complexity does not establish added clinical value[13-16]. Model searches are optimistic when selection and performance estimation use the same data, and empirical comparisons have not shown a consistent advantage over logistic regression[17-20]. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD) + artificial intelligence (AI) recom
This study addressed two prespecified questions. First, what discrimination, prediction error, and calibration would a model selected from 243 configurations across 14 base-model families plus three ensemble strategies achieve in repeated nested evaluation and in a second institution after model freezing? Second, would the selected model improve on a fixed six-variable logistic benchmark? A secondary analysis compared CatBoost with a full 12-variable logistic model to sepa
This retrospective prediction-model study used a development cohort from the National Cancer Center of China and a second-institution cohort from Beijing Jishuitan Hospital. The development cohort included surgically treated patients assessed between January 2010 and December 2021. Clinical, pathological, treatment, and follow-up data were obtained from institutional records. The intended prediction landmark was completion of definitive tumor resection, after post
The second-institution cohort was assembled using the same eligibility, predictor, and binary outcome definitions. Before the current protocol was frozen, the same 212-patient cohort had been used in an earlier project iteration to evaluate a gradient-boosting model developed after Boruta feature selection performed in the development data. Second-institution outcomes were not used to perform that feature selection, to construct new predictors, or to tune the present model, but the investigators had seen the cohort’s case mix and earlier performance. The present assessment is therefore a protocol-frozen second-institution re-evaluation, not a fully blinded or previously unseen external validation.
For the current analysis, the statistical analysis plan and the 243-configuration registry were frozen on August 5, 2026. File-system timestamps and SHA-256 hashes document the sequence: Protocol and registry lock, development-only CatBoost selection and final fitting, model-artifact lock, and generation of second-institution predictions. No predictor, outcome, model definition, hyperparameter, recalibration step, or reported comparison was changed in response to the current re-evaluation results. The audit trail is summarized in Supplementary Table 1.
Eligible patients had histopathologically confirmed soft tissue sarcoma and underwent definitive surgery with curative intent. Exclusion criteria were non-sarcoma pathology, palliative or non-definitive local treatment, missing prespecified predictors or binary outcome, insufficient follow-up to establish 3-year status in patients without documented recurrence, or duplicate records. These rules produced complete-case, ascertainable-status analytical cohorts. An earlier cohort-assembly report documented 1651 development-institution records assessed and 1035 included; however, reason-specific exclusion counts and covariates for the 616 non-included records were not retained in the locked analytical package. For the second institution, the locked package retained only the 212 final analytical records, not a reproducible upstream screening denominator or excluded-patient dataset. Consequently, excluded-versus-included baseline comparisons could not be performed; the available participant-flow audit is reported in Supplementary Table 2.
The target outcome was documented local recurrence within 3 years after definitive surgery. Local recurrence was relapse at or adjacent to the primary surgical site, ascertained from surveillance imaging reports, clinic documentation, pathological confirmation when available, and institutional follow-up records. The same binary definition and coding rules were used at both institutions. Surveillance modality and frequency were clinician-directed rather than protocol-mandated, and the locked extracts did not preserve visit-level schedules, adjudicator identity, recurrence time, censoring time, death, transfer, or competing-event time. Range checks, category checks, duplicate-identifier checks, event-count checks, and cross-institution coding harmonization were completed before analysis. The estimand was therefore recur
Both cohorts were fixed retrospective samples, and no prospective sample-size target was used. Precision was eva
Twelve predictors were prespecified on the basis of clinical relevance[23,24] and routine availability: Age, sex, anatomical location, tumor depth[25], maximum tumor size, French Federation of Cancer Centers Sarcoma Group grade, surgical margin[26-29], radiotherapy, chemotherapy, bone or neurovascular invasion, unplanned surgery[30-33], and symptom duration. Histological subtype was not included because numerous uncommon subtypes would create sparse categories. The full predictor dictionary, coding, source, and intended measurement time are provided in Supplementary Table 3.
Radiotherapy and chemotherapy were binary status fields in the retrospective analytical exports. The exports did not preserve treatment dates or distinguish neoadjuvant treatment, a plan for adjuvant treatment, and completed postope
After the eligibility and complete-case restrictions had been applied, neither locked analytical table contained missing predictor or outcome values. This statement describes the final exports and does not imply that the source clinical popu
The registry contained 243 configurations across 14 base-model families: Fixed six-variable logistic regression (1), fixed full 12-variable logistic regression (1), elastic-net logistic regression (35), random forest (18), extremely randomized trees (18), histogram gradient boosting (18), XGBoost (24), LightGBM (24), CatBoost (24), gradient boosting (18), radial-basis-function support vector machine (20), k-nearest neighbors (18), multilayer perceptron (18), and shrinkage linear discri
Internal performance was estimated with stratified five-fold outer cross-validation repeated five times, producing 25 outer test folds. Hyperparameters were selected by stratified five-fold inner cross-validation within each outer training set. All preprocessing, tuning, component selection, and ensemble-weight estimation remained inside the outer training data. Each outer test fold was used only to generate held-out predictions. Identical outer and inner splits were used across strategies.
Mean outer-cycle Brier score was the primary selection measure. Secondary measures were area under the receiver ope
CatBoost hyperparameters were re-tuned within the prespecified CatBoost registry by fixed stratified 10-fold cross-validation in the complete development cohort, again minimizing Brier score. The locked configuration (CFG-159) used 300 trees, depth 7, learning rate 0.03, L2 leaf regularization 5, and random strength 1. Both the six-variable and full 12-variable logistic models were fitted in the development cohort without use of second-institution outcomes. Model coefficients are provided in Supplementary Table 5.
All second-institution performance measures were calculated from predictions generated once by the development-fitted models. Bootstrap resampling did not retrain, recalibrate, or modify a model. AUC intervals used DeLong’s method for the primary two models. Other intervals and paired differences used 2000 patient-level bootstrap samples. Calibration was summarized by intercept, slope, observed-to-expected event (O/E) ratio, and a flexible cubic-spline curve (four quantile-positioned knots, degree 3) with pointwise bootstrap bands; eight quantile groups were superimposed descrip
Two post hoc analyses were added during revision. First, predictor distributions and CatBoost predicted probabilities were compared between institutions using standardized mean differences (SMDs). Second, to examine the treatment-timing concern without altering the primary model, CatBoost was refitted in the complete development cohort with the locked CFG-159 hyperparameters after excluding radiotherapy and chemotherapy; no external-data tuning or recalibration was performed. Its fixed second-institution predictions were evaluated with the same metrics and 2000 patient-level bootstrap samples. This sensitivity analysis was not used to replace or reselect the primary model.
Global feature attribution for the locked CatBoost model was quantified in the development cohort using SHAP. Mean absolute SHAP values ranked the average magnitude of each predictor’s contribution to the raw model output (log-odds scale)[42-44]. These post hoc attributions were not used for predictor or model selection and do not estimate causal or treatment effects.
A browser-based English-language research interface was made available at https://sts-recurrence-calculator.potatoli.chatgpt.site. It requires all 12 model inputs and returns the CatBoost estimate of binary 3-year recurrence-status probabi
Analyses used Python 3.9.6 with CatBoost 1.2.10, scikit-learn 1.6.1, XGBoost 2.1.4, LightGBM 4.6.0, SHAP 0.49.1, pandas 2.3.3, NumPy 2.0.2, and SciPy 1.13.1; primary second-institution confidence intervals were generated in R 4.6.1. Random seeds, fold assignments, candidate scores, held-out predictions, model files, data hashes, and software versions were retained. The report was structured using TRIPOD + AI for prediction-model studies and STROBE for the retrospective cohort description[21,45]. No patients or members of the public were involved in study design, conduct, or reporting.
The final complete-case, ascertainable-status development cohort comprised 1035 patients, of whom 331 (32.0%) had documented local recurrence within 3 years. The second-institution cohort comprised 212 patients, including 67 recur
| Characteristic | Development cohort (n = 1035) | Second-institution cohort (n = 212) |
| Age, years | 47.78 ± 13.68 | 48.88 ± 13.57 |
| Sex | ||
| Female | 457 (44.15) | 90 (42.45) |
| Male | 578 (55.85) | 122 (57.55) |
| Location | ||
| Upper extremity | 150 (14.49) | 31 (14.62) |
| Lower extremity | 504 (48.70) | 95 (44.81) |
| Trunk | 381 (36.81) | 86 (40.57) |
| Depth | ||
| Superficial | 307 (29.66) | 60 (28.30) |
| Deep | 728 (70.34) | 152 (71.70) |
| Tumor size, cm | 5.95 ± 4.23 | 6.39 ± 4.12 |
| FNCLCC grade | ||
| 1 | 464 (44.83) | 92 (43.40) |
| 2 | 246 (23.77) | 48 (22.64) |
| 3 | 325 (31.40) | 72 (33.96) |
| Surgical margin | ||
| Negative | 952 (91.98) | 190 (89.62) |
| Positive | 83 (8.02) | 22 (10.38) |
| Radiotherapy | ||
| No | 658 (63.57) | 110 (51.89) |
| Yes | 377 (36.43) | 102 (48.11) |
| Chemotherapy | ||
| No | 793 (76.62) | 154 (72.64) |
| Yes | 242 (23.38) | 58 (27.36) |
| Bone or neurovascular invasion | ||
| No | 802 (77.49) | 163 (76.89) |
| Yes | 233 (22.51) | 49 (23.11) |
| Unplanned surgery | ||
| No | 822 (79.42) | 178 (83.96) |
| Yes | 213 (20.58) | 34 (16.04) |
| Symptom duration, months | 24.66 ± 13.21 | 24.86 ± 13.84 |
| Local recurrence within 3 years | 331 (32.0) | 67 (31.60) |
Between-institution SMDs were below 0.12 for all baseline predictors except radiotherapy (absolute SMD: 0.238). The next largest differences were unplanned surgery (0.118) and tumor size (0.106). The descriptive SMD for CatBoost pre
All 14 base-model families and three ensemble strategies generated complete outer-fold predictions in each of the five repeats. The 243 prespecified configurations produced 6075 inner candidate scores across 25 outer folds without an un
| Strategy | AUC | Brier score | Calibration slope |
| CatBoost | 0.8126 ± 0.0052 | 0.1584 ± 0.0017 | 1.070 ± 0.080 |
| Equal-weight top-3 ensemble | 0.8134 ± 0.0043 | 0.1586 ± 0.0016 | 1.086 ± 0.044 |
| Brier-weighted top-3 ensemble | 0.8133 ± 0.0043 | 0.1586 ± 0.0016 | 1.086 ± 0.044 |
| Gradient boosting | 0.8144 ± 0.0031 | 0.1589 ± 0.0011 | 1.117 ± 0.019 |
| XGBoost | 0.8124 ± 0.0035 | 0.1595 ± 0.0011 | 1.054 ± 0.027 |
| Logistic top-3 stack | 0.8117 ± 0.0044 | 0.1608 ± 0.0023 | 0.962 ± 0.038 |
| Histogram gradient boosting | 0.8092 ± 0.0026 | 0.1613 ± 0.0012 | 0.912 ± 0.018 |
| Random forest | 0.8058 ± 0.0053 | 0.1620 ± 0.0024 | 1.134 ± 0.073 |
| RBF support vector machine | 0.8005 ± 0.0045 | 0.1634 ± 0.0022 | 0.976 ± 0.051 |
| Extra trees | 0.8015 ± 0.0056 | 0.1637 ± 0.0030 | 1.067 ± 0.090 |
| Elastic-net logistic | 0.7994 ± 0.0017 | 0.1640 ± 0.0009 | 1.042 ± 0.011 |
| LightGBM | 0.8023 ± 0.0048 | 0.1642 ± 0.0018 | 0.837 ± 0.035 |
| Multilayer perceptron | 0.7955 ± 0.0087 | 0.1647 ± 0.0026 | 0.867 ± 0.119 |
| Full 12-variable logistic | 0.7998 ± 0.0021 | 0.1647 ± 0.0011 | 0.907 ± 0.013 |
| Shrinkage LDA | 0.7974 ± 0.0021 | 0.1657 ± 0.0011 | 0.878 ± 0.015 |
| Six-variable logistic benchmark | 0.7948 ± 0.0013 | 0.1663 ± 0.0005 | 0.948 ± 0.009 |
| k-nearest neighbors | 0.7629 ± 0.0077 | 0.1802 ± 0.0018 | 1.012 ± 0.178 |
CatBoost had the lowest mean Brier score (0.1584; SD: 0.0017), with mean AUC: 0.8126 (SD: 0.0052), log loss: 0.4813 (SD: 0.0051), calibration intercept: 0.011 (SD: 0.030), calibration slope: 1.070 (SD: 0.080), and O/E ratio: 1.005 (SD: 0.015). The six-variable logistic benchmark yielded mean Brier score: 0.1663 (SD: 0.0005), AUC: 0.7948 (SD: 0.0013), and log loss: 0.4959 (SD: 0.0010). These SDs describe variability across correlated repeated cycles and are not confidence intervals. Complete internal metrics are given in Supplementary Table 8.
For the paired internal comparison, each patient’s five out-of-fold predictions were averaged before calculating model differences. CatBoost-minus-six-variable-logistic differences were 0.0232 for AUC [95% confidence interval (CI): 0.0084-0.0394] and -0.0099 for Brier score (95%CI: -0.0163 to -0.0039). These patient-level estimates differ from differences be
Fixed 10-fold tuning in the complete development cohort selected CFG-159. The final model artifact, predictor schema, data hash, package versions, and model hash were frozen. Apparent performance after fitting the complete cohort was not used as a validation estimate. The global SHAP analysis is shown in Figure 3 and Supplementary Table 9. French Federation of Cancer Centers Sarcoma Group grade had the largest mean absolute SHAP value (0.772), followed by bone or neurovascular invasion (0.478), unplanned surgery (0.476), tumor size (0.294), and anatomical location (0.229). These values summarize model attribution on the raw-output scale and do not indicate causal importance.
Applied without recalibration to the second institution, CatBoost achieved AUC: 0.8050 (95%CI: 0.7433-0.8668), Brier score: 0.1615 (95%CI: 0.1331-0.1912), log loss: 0.4884 (95%CI: 0.4146-0.5678), calibration intercept: 0.148 (95%CI: -0.222 to 0.490), calibration slope: 0.985 (95%CI: 0.703-1.357), and O/E ratio: 1.079 (95%CI: 0.885-1.276) (Table 3; Figure 4A and B). The slope point estimate was close to 1, but its interval and the intercept interval were wide.
| Model | AUC | Brier score | Log loss | Calibration intercept | Calibration slope | O/E ratio |
| CatBoost | 0.8050 (0.7433-0.8668) | 0.1615 (0.1331-0.1912) | 0.4884 (0.4146-0.5678) | 0.148 (-0.222 to 0.490) | 0.985 (0.703-1.357) | 1.079 (0.885-1.276) |
| Six-variable logistic | 0.8134 (0.7533-0.8734) | 0.1572 (0.1310-0.1866) | 0.4770 (0.4116-0.5509) | -0.044 (-0.408 to 0.305) | 0.978 (0.710-1.333) | 0.979 (0.811-1.151) |
| Full 12-variable logistic | 0.8104 (0.7488-0.8719) | 0.1580 (0.1297-0.1863) | 0.4800 (0.4066-0.5549) | -0.059 (-0.432 to 0.272) | 0.921 (0.691-1.259) | 0.973 (0.806-1.129) |
The six-variable logistic benchmark achieved AUC: 0.8134 (95%CI: 0.7533-0.8734), Brier score: 0.1572 (95%CI: 0.1310-0.1866), log loss: 0.4770 (95%CI: 0.4116-0.5509), calibration intercept: -0.044 (95%CI: -0.408 to 0.305), calibration slope: 0.978 (95%CI: 0.710-1.333), and O/E ratio: 0.979 (95%CI: 0.811-1.151). CatBoost-minus-logistic differences were -0.0083 for AUC (95%CI: -0.0480 to 0.0299; paired DeLong P = 0.677) and 0.0043 for Brier score (95%CI: -0.0112 to 0.0198). The data there
In the secondary same-predictor-set comparison, full 12-variable logistic regression achieved AUC: 0.8104 (95%CI: 0.7488-0.8719), Brier score: 0.1580 (bootstrap 95%CI: 0.1297-0.1863), and log loss: 0.4800 (95%CI: 0.4066-0.5549). CatBoost-minus-full-logistic differences were -0.0054 for AUC (bootstrap 95%CI: -0.0438 to 0.0307) and 0.0035 for Brier score (95%CI: -0.0121 to 0.0192). Thus, changing the algorithm while retaining all 12 predictors did not demonstrate an incre
The decision curves crossed across the examined range, and neither model showed a consistent descriptive net-benefit advantage (Figure 4C). Because no surveillance threshold was clinically established for this use case, these findings are exploratory. Figure 4D defines the plotted effects as CatBoost minus logistic for AUC and logistic minus CatBoost for Brier score, so positive values consistently favor CatBoost.
In the post hoc treatment-variable sensitivity analysis, CatBoost excluding radiotherapy and chemotherapy achieved an AUC of 0.7984 (bootstrap 95%CI: 0.7291-0.8595) and Brier score of 0.1640 (95%CI: 0.1358-0.1940). Relative to the locked 12-predictor CatBoost model, the AUC difference was -0.0067 (95%CI: -0.0193 to 0.0053) and the sensitivity-minus-locked Brier difference was 0.0026 (95%CI: -0.0019 to 0.0072). Thus, removing the two treatment-status variables did not mate
The frozen CatBoost model retained an AUC of 0.8050 in the second institution. Its calibration-slope point estimate was close to 1, but confidence intervals for slope and intercept were wide; the data were compatible with approximate cali
Repeated nested validation separated tuning from held-out evaluation. Each patient contributed held-out predictions in five outer cycles, while preprocessing, hyperparameter tuning, selection of ensemble components, and ensemble fitting remained within the corresponding training data. This design reduces selection-induced optimism relative to a single split[20]. The small SDs across repeated cycles should nevertheless not be read as independent-sample uncertainty because the same patients were reused. CatBoost was selected by the lowest mean Brier score in development data alone.
The protocol-frozen second-institution comparison placed the internal improvement in context. Point estimates for CatBoost-minus-six-variable-logistic AUC and Brier score changed direction, but their confidence intervals included no difference. The full 12-variable logistic comparison led to the same substantive conclusion, indicating that the finding was not explained only by the smaller benchmark predictor set. These results agree with empirical comparisons in which machine learning has not consistently outperformed logistic regression for clinical prediction. The study’s principal value is therefore the controlled assessment of whether additional algorithmic complexity produced incremental performance, not a claim of machine-learning superiority. Similar point estimates do not establish equivalence or non-inferiority.
SHAP described how the fitted CatBoost model distributed prediction across its inputs. These attributions cannot separate causal effects from correlation, treatment selection, measurement patterns, or unmeasured disease severity. Radiotherapy and chemotherapy are preferentially administered to selected patients, often those with aggressive biology, unfavorable anatomy, higher stage, or other high-risk features; consequently, any model attribution for these variables is vulnerable to confounding by indication. Clinicians must not infer from their SHAP values that either treatment promotes recurrence, prevents recurrence, or lacks efficacy. The post hoc model excluding both treatment fields produced similar overall performance but does not convert observational attribution into causal evidence.
Performance measures answered different questions. AUC measures ranking; Brier score and log loss quantify pre
The browser interface facilitates reproducibility and independent evaluation of the locked model. A model hash, feature schema, fixed test set, and patient-level concordance test provide checks against silent model drift. Public accessibility does not confer clinical validity. The interface has not undergone prospective impact, usability, security, or regu
Methodological strengths include a prespecified search space, identical resampling splits across 246 strategies, com
Several limitations constrain interpretation. Both cohorts were retrospective and drawn from high-volume tertiary sarcoma centers. The same second-institution cohort had been examined in earlier project work, including evaluation of a prior gradient-boosting model, so investigators had prior knowledge of its case mix and performance; protocol freezing prevents post-lock changes but cannot erase bias from historical exposure. Complete-case and ascertainable-status re
Future work should prioritize prospective multicenter evaluation with time-to-event outcomes, explicit handling of censoring, missingness and competing risks, prespecified recalibration rules, subgroup assessment, and decision-impact analysis. Clinical use would require a clearly defined user, decision, prediction time, and action threshold. The relevant question is whether a model improves decisions or outcomes beyond a simpler tool, not whether a larger algorithm search raises internal AUC.
In a protocol-frozen second-institution re-evaluation among patients with ascertainable binary 3-year recurrence status, the locked CatBoost model retained discrimination but did not demonstrate improvement over either the prespecified six-variable benchmark or a full 12-variable logistic model. This study was not designed or powered to establish equi
| 1. | Gamboa AC, Gronchi A, Cardona K. Soft-tissue sarcoma in adults: An update on the current state of histiotype-specific management in an era of personalized medicine. CA Cancer J Clin. 2020;70:200-229. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 172] [Cited by in RCA: 396] [Article Influence: 66.0] [Reference Citation Analysis (2)] |
| 2. | von Mehren M, Kane JM, Agulnik M, Bui MM, Carr-Ascher J, Choy E, Connelly M, Dry S, Ganjoo KN, Gonzalez RJ, Holder A, Homsi J, Keedy V, Kelly CM, Kim E, Liebner D, McCarter M, McGarry SV, Mesko NW, Meyer C, Pappo AS, Parkes AM, Petersen IA, Pollack SM, Poppe M, Riedel RF, Schuetze S, Shabason J, Sicklick JK, Spraker MB, Zimel M, Hang LE, Sundar H, Bergman MA. Soft Tissue Sarcoma, Version 2.2022, NCCN Clinical Practice Guidelines in Oncology. J Natl Compr Canc Netw. 2022;20:815-833. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 401] [Cited by in RCA: 302] [Article Influence: 75.5] [Reference Citation Analysis (2)] |
| 3. | Gronchi A, Miah AB, Dei Tos AP, Abecassis N, Bajpai J, Bauer S, Biagini R, Bielack S, Blay JY, Bolle S, Bonvalot S, Boukovinas I, Bovee JVMG, Boye K, Brennan B, Brodowicz T, Buonadonna A, De Álava E, Del Muro XG, Dufresne A, Eriksson M, Fagioli F, Fedenko A, Ferraresi V, Ferrari A, Frezza AM, Gasperoni S, Gelderblom H, Gouin F, Grignani G, Haas R, Hassan AB, Hecker-Nolting S, Hindi N, Hohenberger P, Joensuu H, Jones RL, Jungels C, Jutte P, Kager L, Kasper B, Kawai A, Kopeckova K, Krákorová DA, Le Cesne A, Le Grange F, Legius E, Leithner A, Lopez-Pousa A, Martin-Broto J, Merimsky O, Messiou C, Mir O, Montemurro M, Morland B, Morosi C, Palmerini E, Pantaleo MA, Piana R, Piperno-Neumann S, Reichardt P, Rutkowski P, Safwat AA, Sangalli C, Sbaraglia M, Scheipl S, Schöffski P, Sleijfer S, Strauss D, Strauss S, Sundby Hall K, Trama A, Unk M, van de Sande MAJ, van der Graaf WTA, van Houdt WJ, Frebourg T, Casali PG, Stacchiotti S; ESMO Guidelines Committee, EURACAN and GENTURIS. Soft tissue and visceral sarcomas: ESMO-EURACAN-GENTURIS Clinical Practice Guidelines for diagnosis, treatment and follow-up(☆). Ann Oncol. 2021;32:1348-1365. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 855] [Cited by in RCA: 820] [Article Influence: 164.0] [Reference Citation Analysis (3)] |
| 4. | Issels RD, Lindner LH, Verweij J, Wessalowski R, Reichardt P, Wust P, Ghadjar P, Hohenberger P, Angele M, Salat C, Vujaskovic Z, Daugaard S, Mella O, Mansmann U, Dürr HR, Knösel T, Abdel-Rahman S, Schmidt M, Hiddemann W, Jauch KW, Belka C, Gronchi A; European Organization for the Research and Treatment of Cancer-Soft Tissue and Bone Sarcoma Group and the European Society for Hyperthermic Oncology. Effect of Neoadjuvant Chemotherapy Plus Regional Hyperthermia on Long-term Outcomes Among Patients With Localized High-Risk Soft Tissue Sarcoma: The EORTC 62961-ESHO 95 Randomized Clinical Trial. JAMA Oncol. 2018;4:483-492. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 301] [Cited by in RCA: 243] [Article Influence: 30.4] [Reference Citation Analysis (0)] |
| 5. | Liang Y, Guo T, Hong D, Xiao W, Zhou Z, Zhang X. Time to Local Recurrence as a Predictor of Survival in Patients With Soft Tissue Sarcoma of the Extremity and Abdominothoracic Wall. Front Oncol. 2020;10:599097. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 2] [Cited by in RCA: 12] [Article Influence: 2.0] [Reference Citation Analysis (0)] |
| 6. | Rothermundt C, Whelan JS, Dileo P, Strauss SJ, Coleman J, Briggs TW, Haile SR, Seddon BM. What is the role of routine follow-up for localised limb soft tissue sarcomas? A retrospective analysis of 174 patients. Br J Cancer. 2014;110:2420-2426. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 57] [Cited by in RCA: 81] [Article Influence: 6.8] [Reference Citation Analysis (0)] |
| 7. | Cahlon O, Brennan MF, Jia X, Qin LX, Singer S, Alektiar KM. A postoperative nomogram for local recurrence risk in extremity soft tissue sarcomas after limb-sparing surgery without adjuvant radiation. Ann Surg. 2012;255:343-347. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 107] [Cited by in RCA: 139] [Article Influence: 9.9] [Reference Citation Analysis (1)] |
| 8. | van Praag VM, Rueten-Budde AJ, Jeys LM, Laitinen MK, Pollock R, Aston W, van der Hage JA, Dijkstra PDS, Ferguson PC, Griffin AM, Willeumier JJ, Wunder JS, van de Sande MAJ, Fiocco M. A prediction model for treatment decisions in high-grade extremity soft-tissue sarcomas: Personalised sarcoma care (PERSARC). Eur J Cancer. 2017;83:313-323. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 44] [Cited by in RCA: 97] [Article Influence: 10.8] [Reference Citation Analysis (0)] |
| 9. | De Sanctis R, Zelic R, Santoro A. Nomograms predicting local and distant recurrence and disease-specific mortality for R0/R1 soft tissue sarcomas of the extremities. Front Oncol. 2022;12:941896. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 1] [Cited by in RCA: 3] [Article Influence: 0.8] [Reference Citation Analysis (0)] |
| 10. | Callegaro D, Miceli R, Bonvalot S, Ferguson P, Strauss DC, Levy A, Griffin A, Hayes AJ, Stacchiotti S, Pechoux CL, Smith MJ, Fiore M, Dei Tos AP, Smith HG, Mariani L, Wunder JS, Pollock RE, Casali PG, Gronchi A. Development and external validation of two nomograms to predict overall survival and occurrence of distant metastases in adults after surgical resection of localised soft-tissue sarcomas of the extremities: a retrospective analysis. Lancet Oncol. 2016;17:671-680. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 211] [Cited by in RCA: 378] [Article Influence: 37.8] [Reference Citation Analysis (4)] |
| 11. | Callegaro D, Miceli R, Bonvalot S, Ferguson PC, Strauss DC, van Praag VVM, Levy A, Griffin AM, Hayes AJ, Stacchiotti S, Pèchoux CL, Smith MJ, Fiore M, Tos APD, Smith HG, Catton C, Szkandera J, Leithner A, van de Sande MAJ, Casali PG, Wunder JS, Gronchi A. Development and external validation of a dynamic prognostic nomogram for primary extremity soft tissue sarcoma survivors. EClinicalMedicine. 2019;17:100215. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 30] [Cited by in RCA: 59] [Article Influence: 8.4] [Reference Citation Analysis (0)] |
| 12. | Kobbeltvedt MR, Lobmaier I, Spreafico M, Callegaro D, Miceli R, Kizilaslan F, Swanson D, Hompland I, Pasquali S, Fiocco M, van de Sande MAJ, Gronchi A, Boye K. A comparison of the risk prediction models PERSARC and Sarculator in patients with localized soft tissue sarcoma of the extremities and trunk wall. ESMO Open. 2025;10:105517. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in RCA: 6] [Reference Citation Analysis (0)] |
| 13. | Topol EJ. High-performance medicine: the convergence of human and artificial intelligence. Nat Med. 2019;25:44-56. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 6739] [Cited by in RCA: 4727] [Article Influence: 675.3] [Reference Citation Analysis (16)] |
| 14. | Crombé A, Spinnato P, Italiano A, Brisse HJ, Feydy A, Fadli D, Kind M. Radiomics and artificial intelligence for soft-tissue sarcomas: Current status and perspectives. Diagn Interv Imaging. 2023;104:567-583. [RCA] [PubMed] [DOI] [Full Text] [Cited by in RCA: 28] [Reference Citation Analysis (0)] |
| 15. | Zhang Z, Zhao Y, Canes A, Steinberg D, Lyashevska O; written on behalf of AME Big-Data Clinical Trial Collaborative Group. Predictive analytics with gradient boosting in clinical medicine. Ann Transl Med. 2019;7:152. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 56] [Cited by in RCA: 184] [Article Influence: 26.3] [Reference Citation Analysis (2)] |
| 16. | Ji GW, Jiao CY, Xu ZG, Li XC, Wang K, Wang XH. Development and validation of a gradient boosting machine to predict prognosis after liver resection for intrahepatic cholangiocarcinoma. BMC Cancer. 2022;22:258. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 32] [Cited by in RCA: 31] [Article Influence: 7.8] [Reference Citation Analysis (0)] |
| 17. | Barreñada L, Dhiman P, Timmerman D, Boulesteix AL, Van Calster B. Understanding overfitting in random forest for probability estimation: a visualization and simulation study. Diagn Progn Res. 2024;8:14. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 13] [Cited by in RCA: 45] [Article Influence: 22.5] [Reference Citation Analysis (0)] |
| 18. | Christodoulou E, Ma J, Collins GS, Steyerberg EW, Verbakel JY, Van Calster B. A systematic review shows no performance benefit of machine learning over logistic regression for clinical prediction models. J Clin Epidemiol. 2019;110:12-22. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 1585] [Cited by in RCA: 1352] [Article Influence: 193.1] [Reference Citation Analysis (7)] |
| 19. | Iwagami M, Inokuchi R, Kawakami E, Yamada T, Goto A, Kuno T, Hashimoto Y, Michihata N, Goto T, Shinozaki T, Sun Y, Taniguchi Y, Komiyama J, Uda K, Abe T, Tamiya N. Comparison of machine-learning and logistic regression models for prediction of 30-day unplanned readmission in electronic health records: A development and validation study. PLOS Digit Health. 2024;3:e0000578. [RCA] [PubMed] [DOI] [Full Text] [Cited by in RCA: 15] [Reference Citation Analysis (0)] |
| 20. | Varma S, Simon R. Bias in error estimation when using cross-validation for model selection. BMC Bioinformatics. 2006;7:91. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 869] [Cited by in RCA: 1093] [Article Influence: 54.7] [Reference Citation Analysis (0)] |
| 21. | Collins GS, Moons KGM, Dhiman P, Riley RD, Beam AL, Van Calster B, Ghassemi M, Liu X, Reitsma JB, van Smeden M, Boulesteix AL, Camaradou JC, Celi LA, Denaxas S, Denniston AK, Glocker B, Golub RM, Harvey H, Heinze G, Hoffman MM, Kengne AP, Lam E, Lee N, Loder EW, Maier-Hein L, Mateen BA, McCradden MD, Oakden-Rayner L, Ordish J, Parnell R, Rose S, Singh K, Wynants L, Logullo P. TRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods. BMJ. 2024;385:e078378. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 1587] [Cited by in RCA: 2634] [Article Influence: 1317.0] [Reference Citation Analysis (17)] |
| 22. | Wolff RF, Moons KGM, Riley RD, Whiting PF, Westwood M, Collins GS, Reitsma JB, Kleijnen J, Mallett S; PROBAST Group†. PROBAST: A Tool to Assess the Risk of Bias and Applicability of Prediction Model Studies. Ann Intern Med. 2019;170:51-58. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 2040] [Cited by in RCA: 2020] [Article Influence: 288.6] [Reference Citation Analysis (9)] |
| 23. | Weskamp P, Ufton D, Drysch M, Wagner JM, Dadras M, Lehnhardt M, Behr B, Wallner C. Risk Factors for Occurrence and Relapse of Soft Tissue Sarcoma. Cancers (Basel). 2022;14:1273. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 3] [Cited by in RCA: 19] [Article Influence: 4.8] [Reference Citation Analysis (0)] |
| 24. | Sambri A, Caldari E, Montanari A, Fiore M, Cevolani L, Ponti F, D'Agostino V, Bianchi G, Miceli M, Spinnato P, De Paolis M, Donati DM. Vascular Proximity Increases the Risk of Local Recurrence in Soft-Tissue Sarcomas of the Thigh-A Retrospective MRI Study. Cancers (Basel). 2021;13:6325. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 5] [Cited by in RCA: 11] [Article Influence: 2.2] [Reference Citation Analysis (0)] |
| 25. | Lin JS, Coleman L, Voskuil RT, Malik A, Mayerson JL, Scharschmidt TJ. Local recurrence rates of superficial versus deep soft tissue sarcoma. Arch Orthop Trauma Surg. 2024;144:2967-2973. [RCA] [PubMed] [DOI] [Full Text] [Cited by in RCA: 2] [Reference Citation Analysis (0)] |
| 26. | Gundle KR, Kafchinski L, Gupta S, Griffin AM, Dickson BC, Chung PW, Catton CN, O'Sullivan B, Wunder JS, Ferguson PC. Analysis of Margin Classification Systems for Assessing the Risk of Local Recurrence After Soft Tissue Sarcoma Resection. J Clin Oncol. 2018;36:704-709. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 189] [Cited by in RCA: 170] [Article Influence: 21.3] [Reference Citation Analysis (0)] |
| 27. | Bilgeri A, Klein A, Lindner LH, Nachbichler S, Knösel T, Birkenmaier C, Jansson V, Baur-Melnyk A, Dürr HR. The Effect of Resection Margin on Local Recurrence and Survival in High Grade Soft Tissue Sarcoma of the Extremities: How Far Is Far Enough? Cancers (Basel). 2020;12:2560. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 18] [Cited by in RCA: 40] [Article Influence: 6.7] [Reference Citation Analysis (0)] |
| 28. | Wittenberg S, Paraskevaidis M, Jarosch A, Flörcken A, Brandes F, Striefler J, Kaul D, Roohani S, Khakzad T, Märdian S, Rau D. Surgical Margins in Soft Tissue Sarcoma Management and Corresponding Local and Systemic Recurrence Rates: A Retrospective Study Covering 11 Years and 169 Patients in a Single Institution. Life (Basel). 2022;12:1694. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 20] [Cited by in RCA: 11] [Article Influence: 2.8] [Reference Citation Analysis (0)] |
| 29. | Yurtbay A, Aydın Şimşek Ş, Cengiz T, Bariş YS, Say F, Dabak N. The Impact of Surgical Margin Distance on Local Recurrence and Survival in Patients with Soft Tissue Sarcoma. Medicina (Kaunas). 2025;61:289. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 1] [Cited by in RCA: 6] [Article Influence: 6.0] [Reference Citation Analysis (1)] |
| 30. | Larios F, Gonzalez MR, Ruiz-Arellanos K, Aquilino E Silva G, Pretell-Mazzini J. Is Unplanned Excision of Soft Tissue Sarcomas Associated with Worse Oncological Outcomes?-A Systematic Review and Meta-Analysis. Cancers (Basel). 2024;16:443. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 24] [Cited by in RCA: 20] [Article Influence: 10.0] [Reference Citation Analysis (0)] |
| 31. | Sacchetti F, Alsina AC, Morganti R, Innocenti M, Andreani L, Muratori F, Scoccianti G, Totti F, Campanacci DA, Capanna R. Re-excision after unplanned excision of soft tissue sarcoma: A systematic review and metanalysis. The rationale of systematic re-excision. J Orthop. 2021;25:244-251. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 33] [Cited by in RCA: 30] [Article Influence: 6.0] [Reference Citation Analysis (0)] |
| 32. | Scoccianti G, Innocenti M, Frenos F, Muratori F, Sacchetti F, Beltrami G, Capanna R, Campanacci DA. Re-excision after unplanned excision of soft tissue sarcomas: Long-term results. Surg Oncol. 2020;34:212-217. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 5] [Cited by in RCA: 14] [Article Influence: 2.3] [Reference Citation Analysis (0)] |
| 33. | Fromm J, Klein A, Mentrup F, Lindner LH, Nachbichler S, Holzapfel BM, Goller SS, Knösel T, Dürr HR. Unplanned Resections of Soft Tissue Sarcomas-Necessity of Re-Resection? Cancers (Basel). 2024;16:1851. [RCA] [PubMed] [DOI] [Full Text] [Cited by in RCA: 6] [Reference Citation Analysis (0)] |
| 34. | Prokhorenkova L, Gusev G, Vorobev A, Dorogush AV, Gulin A. CatBoost: Unbiased boosting with categorical features. Adv Neural Inf Process Syst. 2018;31:6639-6649. [DOI] [Full Text] |
| 35. | Zhang Z. Introduction to machine learning: k-nearest neighbors. Ann Transl Med. 2016;4:218. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 798] [Cited by in RCA: 389] [Article Influence: 38.9] [Reference Citation Analysis (0)] |
| 36. | Chen TQ, Guestrin C. XGBoost: A Scalable Tree Boosting System. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD ‘16). Assoc Comput Mach. 2016;785-794. [RCA] [DOI] [Full Text] [Cited by in Crossref: 12755] [Cited by in RCA: 11091] [Article Influence: 1109.1] [Reference Citation Analysis (4)] |
| 37. | Ke G, Meng Q, Finley T, Wang T, Chen W, Ma W, Ye Q, Liu TY. LightGBM: A highly efficient gradient boosting decision tree. Adv Neural Inf Process Syst. 2017;30:3146-3154. |
| 38. | Alba AC, Agoritsas T, Walsh M, Hanna S, Iorio A, Devereaux PJ, McGinn T, Guyatt G. Discrimination and Calibration of Clinical Prediction Models: Users' Guides to the Medical Literature. JAMA. 2017;318:1377-1384. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 1323] [Cited by in RCA: 1272] [Article Influence: 141.3] [Reference Citation Analysis (7)] |
| 39. | Steyerberg EW, Vickers AJ, Cook NR, Gerds T, Gonen M, Obuchowski N, Pencina MJ, Kattan MW. Assessing the performance of prediction models: a framework for traditional and novel measures. Epidemiology. 2010;21:128-138. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 4068] [Cited by in RCA: 3896] [Article Influence: 243.5] [Reference Citation Analysis (9)] |
| 40. | Vickers AJ, Elkin EB. Decision curve analysis: a novel method for evaluating prediction models. Med Decis Making. 2006;26:565-574. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 4224] [Cited by in RCA: 4577] [Article Influence: 228.9] [Reference Citation Analysis (12)] |
| 41. | Vickers AJ, van Calster B, Steyerberg EW. A simple, step-by-step guide to interpreting decision curve analysis. Diagn Progn Res. 2019;3:18. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 901] [Cited by in RCA: 893] [Article Influence: 127.6] [Reference Citation Analysis (1)] |
| 42. | Ponce-Bobadilla AV, Schmitt V, Maier CS, Mensing S, Stodtmann S. Practical guide to SHAP analysis: Explaining supervised machine learning model predictions in drug development. Clin Transl Sci. 2024;17:e70056. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 675] [Cited by in RCA: 420] [Article Influence: 210.0] [Reference Citation Analysis (2)] |
| 43. | Loh HW, Ooi CP, Seoni S, Barua PD, Molinari F, Acharya UR. Application of explainable artificial intelligence for healthcare: A systematic review of the last decade (2011-2022). Comput Methods Programs Biomed. 2022;226:107161. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 55] [Cited by in RCA: 275] [Article Influence: 68.8] [Reference Citation Analysis (3)] |
| 44. | Nohara Y, Matsumoto K, Soejima H, Nakashima N. Explanation of machine learning models using shapley additive explanation and application for real data in hospital. Comput Methods Programs Biomed. 2022;214:106584. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 37] [Cited by in RCA: 459] [Article Influence: 114.8] [Reference Citation Analysis (3)] |
| 45. | von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, Vandenbroucke JP; STROBE Initiative. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Lancet. 2007;370:1453-1457. [RCA] [PubMed] [DOI] [Full Text] [Cited by in Crossref: 15623] [Cited by in RCA: 16097] [Article Influence: 847.2] [Reference Citation Analysis (12)] |
| 46. | Rutkowski P, Lugowska I. Follow-up in soft tissue sarcomas. Memo. 2014;7:92-96. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 13] [Cited by in RCA: 21] [Article Influence: 1.8] [Reference Citation Analysis (0)] |
| 47. | Amann J, Blasimme A, Vayena E, Frey D, Madai VI; Precise4Q consortium. Explainability for artificial intelligence in healthcare: a multidisciplinary perspective. BMC Med Inform Decis Mak. 2020;20:310. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 1509] [Cited by in RCA: 895] [Article Influence: 149.2] [Reference Citation Analysis (7)] |
| 48. | Collins GS, Dhiman P, Ma J, Schlussel MM, Archer L, Van Calster B, Harrell FE Jr, Martin GP, Moons KGM, van Smeden M, Sperrin M, Bullock GS, Riley RD. Evaluation of clinical prediction models (part 1): from development to external validation. BMJ. 2024;384:e074819. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 435] [Cited by in RCA: 510] [Article Influence: 255.0] [Reference Citation Analysis (11)] |
| 49. | Efthimiou O, Seo M, Chalkou K, Debray T, Egger M, Salanti G. Developing clinical prediction models: a step-by-step guide. BMJ. 2024;386:e078276. [RCA] [PubMed] [DOI] [Full Text] [Full Text (PDF)] [Cited by in Crossref: 265] [Cited by in RCA: 355] [Article Influence: 177.5] [Reference Citation Analysis (2)] |