Machine Learning to Build and Validate a Model for Radiation Pneumonitis Prediction in Patients with Non-Small Cell Lung Cancer

The causal association between serum metabolites and lung cancer based on multivariate Mendelian randomization

This study seeks to understand the causal association between serum metabolites and different lung cancer types, an area yet to be extensively studied. We Used a two-sample Mendelian randomization (TSMR) approach, utilizing 486 blood metabolites as exposures and 3 distinct lung cancer types genome-wide association studies datasets as outcomes. We employed inverse variance weighting, MR-Egger, weighted median, simple mode, and weighted mode to estimate causal effects. We performed sensitivity analyses using Cochran Q test, MR-Egger intercept test, and MR-pleiotropy residual sum and outlier (MR-PRESSO). Linkage disequilibrium score (LDSC) analysis was conducted on the selected metabolites, and common confounding single nucleotide polymorphisms were eliminated using the human genotype-phenotype association Database. Metabolic pathway analysis was performed with MetaboAnalyst 5.0 software. Subsequently, a multivariate Mendelian randomization analysis was conducted to ascertain independent risk exposures. Our findings suggest independent risk factors for specific types of lung cancer: 7-methylxanthine and isoleucine for lung adenocarcinoma, cysteine and 1-arachidonoylglycerophosphocholine are identified as independent protective and risk factors for squamous lung cancer. Undecanoate (11:0) with Linoleate (18:2n6) showed a protective effect for small cell lung cancer. Additionally, 11 metabolic pathways were associated with lung cancer. This novel perspective offers a multidimensional understanding of lung cancer phenotypes, providing valuable guidance for identifying and screening of diverse lung cancer phenotypes.

AI/ML advances in non-small cell lung cancer biomarker discovery

Lung cancer is the leading cause of cancer deaths among both men and women, representing approximately 25% of cancer fatalities each year. The treatment landscape for non-small cell lung cancer (NSCLC) is rapidly evolving due to the progress made in biomarker-driven targeted therapies. While advancements in targeted treatments have improved survival rates for NSCLC patients with actionable biomarkers, long-term survival remains low, with an overall 5-year relative survival rate below 20%. Artificial intelligence/machine learning (AI/ML) algorithms have shown promise in biomarker discovery, yet NSCLC-specific studies capturing the clinical challenges targeted and emerging patterns identified using AI/ML approaches are lacking. Here, we employed a text-mining approach and identified 215 studies that reported potential biomarkers of NSCLC using AI/ML algorithms. We catalogued these studies with respect to BEST (Biomarkers, EndpointS, and other Tools) biomarker sub-types and summarized emerging patterns and trends in AI/ML-driven NSCLC biomarker discovery. We anticipate that our comprehensive review will contribute to the current understanding of AI/ML advances in NSCLC biomarker research and provide an important catalogue that may facilitate clinical adoption of AI/ML-derived biomarkers.

STING signaling activation modulates macrophage polarization via CCL2 in radiation-induced lung injury

BACKGROUND

Radiation-induced lung injury (RILI) is a prevalent complication of thoracic radiotherapy in cancer patients. A comprehensive understanding of the underlying mechanisms of RILI is essential for the development of effective prevention and treatment strategies.

METHODS

To investigate RILI, we utilized a mouse model that received 12.5 Gy whole-thoracic irradiation. The evaluation of RILI was performed using a combination of quantitative real-time polymerase chain reaction (qRT-PCR), enzyme-linked immunosorbent assay (ELISA), histology, western blot, immunohistochemistry, RNA sequencing, and flow cytometry. Additionally, we established a co-culture system consisting of macrophages, lung epithelial cells, and fibroblasts for in vitro studies. In this system, lung epithelial cells were irradiated with a dose of 4 Gy, and we employed STING knockout macrophages. Translational examinations were conducted to explore the relationship between STING expression in pre-radiotherapy lung tissues, dynamic changes in circulating CCL2, and the development of RILI.

RESULTS

Our findings revealed significant activation of the cGAS-STING pathway and M1 polarization of macrophages in the lungs of irradiated mice. In vitro studies demonstrated that the deficiency of cGAS-STING signaling led to impaired macrophage polarization and RILI. Through RNA sequencing, cytokine profiling, and rescue experiments using a CCL2 inhibitor called Bindarit, we identified the involvement of CCL2 in the regulation of macrophage polarization and the development of RILI. Moreover, translational investigations using patient samples collected before and after thoracic radiotherapy provided additional evidence supporting the association between cGAS-STING signaling activity, CCL2 upregulation, and the development of radiation pneumonitis.

CONCLUSIONS

The cGAS-STING signaling pathway plays a crucial role in regulating the recruitment and polarization of macrophages, partly through CCL2, during the pathogenesis of RILI.

Radiomics and Dosiomics Signature From Whole Lung Predicts Radiation Pneumonitis: A Model Development Study With Prospective External Validation and Decision-curve Analysis

PURPOSE

Radiation pneumonitis (RP) is one of the common side effects of radiation therapy in the thoracic region. Radiomics and dosiomics quantify information implicit within medical images and radiation therapy dose distributions. In this study we demonstrate the prognostic potential of radiomics, dosiomics, and clinical features for RP prediction.

METHODS AND MATERIALS

Radiomics, dosiomics, dose-volume histogram (DVH) metrics, and clinical parameters were obtained on 314 retrospectively collected and 35 prospectively enrolled patients diagnosed with lung cancer between 2013 to 2019. A radiomics risk score (R score) and dosiomics risk score (D score), as well as a DVH-score, were calculated based on logistic regression after feature selection. Six models were built using different combinations of R score, D score, DVH score, and clinical parameters to evaluate their added prognostic power. Overoptimism was evaluated by bootstrap resampling from the training set, and the prospectively collected cohort was used as the external test set. Model calibration and decision-curve characteristics of the best-performing models were evaluated. For ease of further evaluation, nomograms were constructed for selected models.

RESULTS

A model built by integrating all of the R score, D score, and clinical parameters had the best discriminative ability with areas under the curve of 0.793 (95% confidence interval [CI], 0.735-0.851), 0.774 (95% CI, 0.762-0.786), and 0.855 (95% CI, 0.719-0.990) in the training, bootstrapping, and external test sets, respectively. The calibration curve image showed good agreement between the predicted and actual values, with a slope of 1.21 and intercept of -0.04. The decision curve image showed a positive net benefit for the final model based on the nomogram.

CONCLUSIONS

Radiomic and dosiomic features have the potential to assist with the prediction of RP, and the combination of radiomics, dosiomics, and clinical parameters led to the best prognostic model in the present study.

Identifying an optimal machine learning model generated circulating biomarker to predict chronic postoperative pain in patients undergoing hepatectomy

Chronic postsurgical pain (CPSP) after hepatectomy is highly prevalent and challenging to treat. Several risk factors have been unmasked for CPSP after hepatectomy, such as acute postoperative pain. The current secondary analysis of a clinical study sought to extend previous research by investigating more clinical variables and inflammatory biomarkers as risk factors for CPSP after hepatectomy and sifting those strongly related to CPSP to build a reliable machine learning model to predict CPSP occurring. Participants included 91 adults undergoing hepatectomy who was followed 3 months postoperatively. Twenty-four hours after surgery, participants completed numerical rating scale (NRS) grading and blood sample collecting. Three months after surgery, participants also reported whether CPSP occurred through follow-up. The Random Forest and Support Vector Machine models were conducted to predict pain outcomes 3 months after surgery. The results showed that the SVM model had better performance in predicting CPSP which consists of acute postoperative pain (evaluated by NRS) and matrix metalloprotease 3 (MMP3) level. What's more, besides traditional cytokines, several novel inflammatory biomarkers like C-X-C motif chemokine ligand 10 (CXCL10) and MMP2 levels were found to be closely related to CPSP and a novel spectrum of inflammatory biomarkers was created. These findings demonstrate that the SVM model consisting of acute postoperative pain and MMP3 level predicts greater chronic pain intensity 3 months after hepatectomy and with this model, intervention administration before CPSP occurs may prevent or minimize CPSP intensity successfully.

Circulating levels of IL-6 and TGF-β1 in patients with prostate cancer undergoing radiotherapy: associations with acute radiotoxicity and fatigue symptoms

Background

The goal of research was to investigate the possible relations between serum concentrations of IL-6 and TGF-β1, individual and clinical characteristics, and adverse effects of radiotherapy in patients with prostate cancer: acute and late genitourinary and gastrointestinal toxicity, and fatigue.

Methods

Thirty-nine patients with localized or locally advanced prostate cancer who were treated with radiotherapy were enrolled in this study. The acute radiotoxicity grades and fatigue levels were assessed during the radiotherapy and 1 month after the radiotherapy. Estimation of the late radiotoxicity was performed every three months in the first year, every four months in the second year, and then every six months. Serum levels of IL-6 and TGF-β1 were determined before radiotherapy and after the 25th radiotherapy fraction by ELISA.

Results

The significant positive association between diabetes mellitus and changes in acute genitourinary toxicity grades during the radiotherapy was observed in prostate cancer patients. In addition, patients who were smokers had significantly higher maximum fatigue levels in comparison with patients who were non-smokers. The circulating IL-6 levels were significantly higher after the 25th radiotherapy fraction in comparison with levels determined before radiotherapy. The significant positive correlations between pretreatment TGF-β1 levels and maximum genitourinary toxicity grades and between TGF-β1 levels after the 25th fraction and genitourinary toxicity grades after the 25th fraction, were found. The pretreatment IL-6 concentrations and TGF-β1 concentrations after the 25th fraction were positively correlated with maximum genitourinary toxicity grades. The IL-6 levels after the 25th fraction were positively associated with genitourinary toxicity grades after this fraction. The pretreatment IL-6 concentrations were significantly positively correlated with maximum fatigue scores. The significant positive correlation between IL-6 concentrations and fatigue scores after the 25th fraction was determined. The positive correlations between IL-6 and TGF-β1 concentrations measured after the 25th fraction and maximum fatigue scores were observed.

Conclusions

Our results suggest that serum levels of IL-6 and TGF-β1 might influence the severity of acute genitourinary radiotoxicity and fatigue in patients with prostate cancer. Combining clinical parameters and circulating cytokine levels might be useful for the prediction of adverse reactions to radiotherapy.

Why not try to predict autism spectrum disorder with crucial biomarkers in cuproptosis signaling pathway?

The exact pathogenesis of autism spectrum disorder (ASD) is still unclear, yet some potential mechanisms may not have been evaluated before. Cuproptosis is a novel form of regulated cell death reported this year, and no study has reported the relationship between ASD and cuproptosis. This study aimed to identify ASD in suspected patients early using machine learning models based on biomarkers of the cuproptosis pathway. We collected gene expression profiles from brain samples from ASD model mice and blood samples from humans with ASD, selected crucial genes in the cuproptosis signaling pathway, and then analysed these genes with different machine learning models. The accuracy, sensitivity, specificity, and areas under the receiver operating characteristic curves of the machine learning models were estimated in the training, internal validation, and external validation cohorts. Differences between models were determined with Bonferroni's test. The results of screening with the Boruta algorithm showed that FDX1, DLAT, LIAS, and ATP7B were crucial genes in the cuproptosis signaling pathway for ASD. All selected genes and corresponding proteins were also expressed in the human brain. The k-nearest neighbor, support vector machine and random forest models could identify approximately 72% of patients with ASD. The artificial neural network (ANN) model was the most suitable for the present data because the accuracy, sensitivity, and specificity were 0.90, 1.00, and 0.80, respectively, in the external validation cohort. Thus, we first report the prediction of ASD in suspected patients with machine learning methods based on crucial biomarkers in the cuproptosis signaling pathway, and these findings may contribute to investigations of the potential pathogenesis and early identification of ASD.

Promising Biomarkers of Radiation-Induced Lung Injury: A Review

Radiation-induced lung injury (RILI) is one of the main dose-limiting side effects in patients with thoracic cancer during radiotherapy. No reliable predictors or accurate risk models are currently available in clinical practice. Severe radiation pneumonitis (RP) or pulmonary fibrosis (PF) will reduce the quality of life, even when the anti-tumor treatment is effective for patients. Thus, precise prediction and early diagnosis of lung toxicity are critical to overcome this longstanding problem. This review summarizes the primary mechanisms and preclinical animal models of RILI reported in recent decades, and analyzes the most promising biomarkers for the early detection of lung complications. In general, ideal integrated models considering individual genetic susceptibility, clinical background parameters, and biological variations are encouraged to be built up, and more prospective investigations are still required to disclose the molecular mechanisms of RILI as well as to discover valuable intervention strategies.

Radiation Versus Immune Checkpoint Inhibitor Associated Pneumonitis: Distinct Radiologic Morphologies

BACKGROUND

Patients with non-small cell lung cancer may develop pneumonitis after thoracic radiotherapy (RT) and immune checkpoint inhibitors (ICIs). We hypothesized that distinct morphologic features are associated with different pneumonitis etiologies.

MATERIALS AND METHODS

We systematically compared computed tomography (CT) features of RT- versus ICI-pneumonitis. Clinical and imaging features were tested for association with pneumonitis severity. Lastly, we constructed an exploratory radiomics-based machine learning (ML) model to discern pneumonitis etiology.

RESULTS

Between 2009 and 2019, 82 patients developed pneumonitis: 29 after thoracic RT, 23 after ICI, and 30 after RT + ICI. Fifty patients had grade 2 pneumonitis, 22 grade 3, and 7 grade 4. ICI-pneumonitis was more likely bilateral (65% vs. 28%; p = .01) and involved more lobes (66% vs. 45% involving at least three lobes) and was less likely to have sharp border (17% vs. 59%; p = .004) compared with RT-pneumonitis. Pneumonitis morphology after RT + ICI was heterogeneous, with 47% bilateral, 37% involving at least three lobes, and 40% sharp borders. Among all patients, risk factors for severe pneumonitis included poor performance status, smoking history, worse lung function, and bilateral and multifocal involvement on CT. An ML model based on seven radiomic features alone could distinguish ICI- from RT-pneumonitis with an area under the receiver-operating curve of 0.76 and identified the predominant etiology after RT + ICI concordant with multidisciplinary consensus.

CONCLUSION

RT- and ICI-pneumonitis exhibit distinct spatial features on CT. Bilateral and multifocal lung involvement is associated with severe pneumonitis. Integrating these morphologic features in the clinical management of patients who develop pneumonitis after RT and ICIs may improve treatment decision-making.

IMPLICATIONS FOR PRACTICE

Patients with non-small cell lung cancer often receive thoracic radiation and immune checkpoint inhibitors (ICIs), both of which can cause pneumonitis. This study identified similarities and differences in pneumonitis morphology on computed tomography (CT) scans among pneumonitis due to radiotherapy (RT) alone, ICI alone, and the combination of both. Patients who have bilateral CT changes involving at least three lobes are more likely to have ICI-pneumonitis, whereas those with unilateral CT changes with sharp borders are more likely to have radiation pneumonitis. After RT and/or ICI, severe pneumonitis is associated with bilateral and multifocal CT changes. These results can help guide clinicians in triaging patients who develop pneumonitis after radiation and during ICI treatment.

Efficacy and Safety Aiming at the Combined-Modality Therapy of External Beam Radiotherapy (40Gy) and Iodine-125 Seed Implantation for Locally Advanced NSCLC in the Elderly

Purpose

To evaluate the efficacy and safety of combined-modality therapy for elderly patients with locally advanced non-small-cell lung cancer (NSCLC) invading the chest wall.

Patients and Methods

We retrospectively enrolled 21 elderly patients (aged ≥60 years) with locally advanced NSCLC invading the chest wall. For external beam radiotherapy (EBRT) of the primary tumor, 40Gy was applied and supplemented with iodine-125 seed implantation while 60Gy was applied to the lymph nodes of the mediastinum. Follow-up was conducted every 3 months postoperatively. The related analytic parameters were change in tumor size, the objective response rate (ORR), the disease control rate (DCR), the degree of pain relief, the improvement of physical status, and toxicity.

Results

The combined-modality therapy significantly inhibited local growth of the tumor (from 7.84±1.20 to 4.69±1.90 cm) (P <0.0001), with 71.4% ORR and 90.5% DCR at 1 year. The cancer-related pain was significantly relieved (P <0.05) and physical status was significantly improved (P <0.05). No procedure-associated death or grade > 2 irradiation-related adverse effects were reported in this study.

Conclusion

The combined-modality therapy of EBRT with 40Gy and permanent iodine-125 seed implantation is an efficacious and safe treatment option for elderly patients with locally advanced NSCLC invading the chest wall.

A situational awareness Bayesian network approach for accurate and credible personalized adaptive radiotherapy outcomes prediction in lung cancer patients

PURPOSE

A situational awareness Bayesian network (SA-BN) approach is developed to improve physicians' trust in the prediction of radiation outcomes and evaluate its performance for personalized adaptive radiotherapy (pART).

METHODS

118 non-small-cell lung cancer patients with their biophysical features were employed for discovery (n = 68) and validation (n = 50) of radiation outcomes prediction modeling. Patients' important characteristics identified by radiation experts to predict individual's tumor local control (LC) or radiation pneumonitis with grade ≥ 2 (RP2) were incorporated as expert knowledge (EK). Besides generating an EK-based naïve BN (EK-NBN), an SA-BN was developed by incorporating the EK features into pure data-driven BN (PD-BN) methods to improve the credibility of LC or / and RP2 prediction. After using area under the free-response receiver operating characteristics curve (AU-FROC) to assess the joint prediction of these outcomes, their prediction performances were compared with a regression approach based on the expert yielded estimates (EYE) penalty and its variants.

RESULTS

In addition to improving the credibility of radiation outcomes prediction, the SA-BN approach outperformed the EYE penalty and its variants in terms of the joint prediction of LC and RP2. The value of AU-FROC improves from 0.70 (95% CI: 0.54-0.76) using EK-NBN, to 0.75 (0.65-0.82) using a variant of EYE penalty, to 0.83 (0.75-0.93) using PD-BN and 0.83 (0.77-0.90) using SA-BN; with similar trends in the validation cohort.

CONCLUSIONS

The SA-BN approach can provide an accurate and credible human-machine interface to gain physicians' trust in clinical decision-making, which has the potential to be an important component of pART.

A novel nomogram containing acute radiation esophagitis predicting radiation pneumonitis in thoracic cancer receiving radiotherapy

Background

Radiation-induced pneumonitis (RP) is a non-negligible and sometimes life-threatening complication among patients with thoracic radiation. We initially aimed to ascertain the predictive value of acute radiation-induced esophagitis (SARE, grade ≥ 2) to symptomatic RP (SRP, grade ≥ 2) among thoracic cancer patients receiving radiotherapy. Based on that, we established a novel nomogram model to provide individualized risk assessment for SRP.

Methods

Thoracic cancer patients who were treated with thoracic radiation from Jan 2018 to Jan 2019 in Shandong Cancer Hospital and Institute were enrolled prospectively. All patients were followed up during and after radiotherapy (RT) to observe the development of esophagitis as well as pneumonitis. Variables were analyzed by univariate and multivariate analysis using the logistic regression model, and a nomogram model was established to predict SRP by “R” version 3.6.0.

Results

A total of 123 patients were enrolled (64 esophageal cancer, 57 lung cancer and 2 mediastinal cancer) in this study prospectively. RP grades of 0, 1, 2, 3, 4 and 5 occurred in 29, 57, 31, 0, 3 and 3 patients, respectively. SRP appeared in 37 patients (30.1%). In univariate analysis, SARE was shown to be a significant predictive factor for SRP (P < 0.001), with the sensitivity 91.9% and the negative predictive value 93.5%. The incidence of SRP in different grades of ARE were as follows: Grade 0–1: 6.5%; Grade 2: 36.9%; Grade 3: 80.0%; Grade 4: 100%. Besides that, the dosimetric factors considering total lung mean dose, total lung V5, V20, ipsilateral lung mean dose, ipsilateral lung V5, and mean esophagus dose were correlated with SRP (all P < 0.05) by univariate analysis. The incidence of SRP was significantly higher in patients whose symptoms of RP appeared early. SARE, mean esophagus dose and ipsilateral mean lung dose were still significant in multivariate analysis, and they were included to build a predictive nomogram model for SRP.

Conclusions

As an early index that can reflect the tissue’s radiosensitivity visually, SARE can be used as a predictor for SRP in patients receiving thoracic radiation. And the nomogram containing SARE may be fully applied in future’s clinical work.

Establishment and Verification of a Prediction Model for Symptomatic Radiation Pneumonitis in Patients with Esophageal Cancer Receiving Radiotherapy

BACKGROUND This study aimed to determine the value of the significant index in predicting symptomatic radiation pneumonitis (RP) in esophageal cancer patients, establish a nomogram prediction model, and verify the model. MATERIAL AND METHODS The patients enrolled were divided into 2 groups: a model group and a validation group. According to the logistic regression analysis, the independent predictors for symptomatic RP were obtained, and the nomogram prediction model was established according to these independent predictors. The consistency index (C-index) and calibration curve were used to evaluate the accuracy of the model, and the prediction ability of the model was verified in the validation group. Recursive partitioning analysis (RPA) was used for the risk stratification analysis. RESULTS The ratio of change regarding the pre-albumin at the end of treatment (P=0.001), platelet-to-lymphocyte ratio during treatment (P=0.027), and neutrophil-to-lymphocyte ratio at the end of treatment (P=0.001) were the independent predictors for symptomatic RP. The C-index of the nomogram model was 0.811. According to the risk stratification of RPA, the whole group was divided into 3 groups: a low-risk group, a medium-risk group, and a high-risk group. The incidence of symptomatic RP was 0%, 16.9%, and 57.6%, respectively. The receiver operating characteristic curve also revealed that the nomogram model has good accuracy in the validation group. CONCLUSIONS The developed nomogram and corresponding risk classification system have superior prediction ability for symptomatic RP and can predict the occurrence of RP in the early stage.

Dosimetric Factors and Radiomics Features Within Different Regions of Interest in Planning CT Images for Improving the Prediction of Radiation Pneumonitis

PURPOSE

This study aimed to establish machine learning models using dosimetric factors and radiomics features within 5 regions of interest (ROIs) in treatment planning computed tomography images to improve the prediction of symptomatic radiation pneumonitis (RP) (grade ≥2).

METHODS AND MATERIALS

This study retrospectively collected data on 79 patients with lung cancer (25 RP ≥2) who underwent chemoradiotherapy between 2015 and 2018. We defined 5 ROIs in planning computed tomography images: gross tumor volume (GTV), planning tumor volume (PTV), PTV-GTV, total lung (TL)-GTV, and TL-PTV. We calculated the mean dose, V5, V10, V20, and V30 within TL-GTV and TL-PTV and the mean dose within the other ROIs. A total of 1924 radiomics features were extracted from all 5 ROIs. We selected the best predictors for classifying 2 groups of patients using a sequential backward elimination support vector machine model. A permutation test was used to assess its statistical significance (P < .05).

RESULTS

The best predictors for symptomatic RP were the combination of 11 radiomics features, 5 dosimetric factors, age, and T stage, achieving an area under the curve (AUC) of 0.94 (95% confidence interval [CI], 0.85-1) (accuracy, 90%; sensitivity, 80% [95% CI, 44%-96%]; specificity, 95% [95% CI, 73%-100%]; P = 8 × 10^-4). The clinical characteristics, dosimetric factors, and their combination showed limited predictive power (accuracy, 63.3%, 70%, and 70%; AUC [95% CI]: 0.73 [0.54-0.92], 0.53 [0.31-0.75], and 0.72 [0.51-0.92], respectively). The radiomics features of PTV-GTV and TL-PTV outperformed those of the other ROIs (accuracy, 76.7% and 76.7%; AUC [95% CI]: 0.82 [0.65-0.99] and 0.80 [0.59-1], respectively).

CONCLUSIONS

Combining dosimetric factors and radiomics features within different ROIs can improve the prediction of symptomatic RP. Our results can help physicians adjust the radiation dose distribution of the dose-sensitive lungs and target volumes based on personalized RP estimates.

Weighted-Support Vector Machine Learning Classifier of Circulating Cytokine Biomarkers to Predict Radiation-Induced Lung Fibrosis in Non-Small-Cell Lung Cancer Patients

Background

Radiation-induced lung fibrosis (RILF) is an important late toxicity in patients with non-small-cell lung cancer (NSCLC) after radiotherapy (RT). Clinically significant RILF can impact quality of life and/or cause non-cancer related death. This study aimed to determine whether pre-treatment plasma cytokine levels have a significant effect on the risk of RILF and investigate the abilities of machine learning algorithms for risk prediction.

Methods

This is a secondary analysis of prospective studies from two academic cancer centers. The primary endpoint was grade≥2 (RILF2), classified according to a system consistent with the consensus recommendation of an expert panel of the AAPM task for normal tissue toxicity. Eligible patients must have at least 6 months’ follow-up after radiotherapy commencement. Baseline levels of 30 cytokines, dosimetric, and clinical characteristics were analyzed. Support vector machine (SVM) algorithm was applied for model development. Data from one center was used for model training and development; and data of another center was applied as an independent external validation.

Results

There were 57 and 37 eligible patients in training and validation datasets, with 14 and 16.2% RILF2, respectively. Of the 30 plasma cytokines evaluated, SVM identified baseline circulating CCL4 as the most significant cytokine associated with RILF2 risk in both datasets (P = 0.003 and 0.07, for training and test sets, respectively). An SVM classifier predictive of RILF2 was generated in Cohort 1 with CCL4, mean lung dose (MLD) and chemotherapy as key model features. This classifier was validated in Cohort 2 with accuracy of 0.757 and area under the curve (AUC) of 0.855.

Conclusions

Using machine learning, this study constructed and validated a weighted-SVM classifier incorporating circulating CCL4 levels with significant dosimetric and clinical parameters which predicts RILF2 risk with a reasonable accuracy. Further study with larger sample size is needed to validate the role of CCL4, and this SVM classifier in RILF2.

Development and assessment of machine learning algorithms for predicting remission after transsphenoidal surgery among patients with acromegaly

PURPOSE

Preoperative prediction of transsphenoidal surgical (TSS) response is important for determining individual treatment strategies for acromegaly. There is currently no accurate predictive model for TSS response for acromegaly. The current study sought to develop and validate machine learning (ML)-based models for preoperative prediction of TSS response for acromegaly.

METHODS

Six hundred sixty-eight patients with acromegaly were enrolled and divided into training (n = 534) and text datasets (n = 134) in this retrospective, data mining and ML study. The forward search algorithm was used to select features, and six ML algorithms were applied to construct TSS response prediction models. The performance of these ML models was validated using receiver operating characteristics analysis. Model calibration, discrimination ability, and clinical usefulness were also assessed.

RESULTS

Three hundred forty-nine (52.2%) patients achieved postoperative remission criteria and exhibited good TSS response. A univariate analysis was conducted and eight features, including age, hypertension, ophthalmic disorders, GH, IGF-1, nadir GH, maximal tumor diameter, and Knosp grade, were significantly associated with the TSS response in patients with acromegaly. After feature selection, the gradient boosting decision tree (GBDT), which was constructed with the eight significant features showed the best favorable discriminatory ability both the training (AUC = 0.8555) and validation (AUC = 0.8178) cohorts. The GBDT model showed good discrimination ability and calibration, with the highest levels of accuracy and specificity, and provided better estimates of TTS responses of patients with acromegaly compared with using only the Knosp grade. Decision curve analysis confirmed that the model was clinically useful.

CONCLUSIONS

ML-based models could aid neurosurgeons in the preoperative prediction of TTS response for patients with acromegaly, and could contribute to determining individual treatment strategies.