Independent Validation Cohort Research Articles

Differentiating multiple sclerosis from non-specific white matter changes using a convolutional neural network image classification model

BackgroundThe diagnosis of multiple sclerosis (MS) relies heavily on neuroimaging with magnetic resonance imaging (MRI) and exclusion of mimics. This can be a challenging task due to radiological overlap in several disorders and may require ancillary testing or longitudinal follow up. One of the most common radiological MS mimickers is non-specific white matter disease (NSWMD). We aimed to develop and evaluate models leveraging machine learning algorithms to help distinguish MS and NSWMD. MethodsAll adult patients who underwent MRI brain using a demyelinating protocol with available electronic medical records between 2015 and 2019 at Cleveland Clinic affiliated facilities were included. Diagnosis of MS and NSWMD were assessed from clinical documentation. Those with a diagnosis of MS and NSWMD were matched using total T2 lesion volume (T2LV) and used to train models with logistic regression and convolutional neural networks (CNN). Performance metrices were reported for each model. ResultsA total of 250 NSWMD MRI scans were identified, and 250 unique MS MRI scans were matched on T2LV. Cross validated logistic regression model was able to use 20 variables (including spinal cord area, regional volumes, and fractions) to predict MS compared to NSWMD with 68.0% accuracy while the CNN model was able to classify MS compared to NSWMD in two independent validation and testing cohorts with 77% and 78% accuracy on average. ConclusionAutomated methods can be used to differentiate MS compared to NSWMD. These methods can be used to supplement currently available diagnostic tools for patients being evaluated for MS.

Risk Prediction Scores for Type 2 Diabetes Microvascular and Cardiovascular Complications Derived and Validated With Real-world Data From 2 Provinces: The DIabeteS COmplications (DISCO) Risk Scores

ObjectivesExisting tools to predict the risk of complications among people with type 2 diabetes poorly discriminate high- from low-risk patients. Our aim in this study was to develop risk prediction scores for major type 2 diabetes complications using real-world clinical care data, and to externally validate these risk scores in a different jurisdiction. MethodsUsing health-care administrative data and electronic medical records data, risk scores were derived using data from 25,088 people with type 2 diabetes from the Canadian province of Ontario, followed between 2002 and 2017. Scores were developed for major clinically important microvascular events (treatment for retinopathy, foot ulcer, incident end-stage renal disease), cardiovascular disease events (acute myocardial infarction, heart failure, stroke, amputation), and mortality (cardiovascular, noncardiovascular, all-cause). They were then externally validated using the independent data of 11,416 people with type 2 diabetes from the province of Manitoba. ResultsThe 10 derived risk scores had moderate to excellent discrimination in the independent validation cohort, ranging from 0.705 to 0.977. Their calibration to predict 5-year risk was excellent across most levels of predicted risk, albeit with some displaying underestimation at the highest levels of predicted risk. ConclusionsThe DIabeteS COmplications (DISCO) risk scores for major type 2 diabetes complications were derived and externally validated using contemporary real-world clinical data. As a result, they may be more accurate than other risk prediction scores derived using randomized trial data. The use of more accurate risk scores in clinical practice will help improve personalization of clinical care for patients with type 2 diabetes.

Discovery and verification of extracellular microRNA biomarkers for diagnostic and prognostic assessment of preeclampsia at triage.

We report on the identification of extracellular miRNA (ex-miRNA) biomarkers for early diagnosis and prognosis of preeclampsia (PE). Small RNA sequencing of maternal serum prospectively collected from participants undergoing evaluation for suspected PE revealed distinct patterns of ex-miRNA expression among different categories of hypertensive disorders in pregnancy. Applying an iterative machine learning method identified three bivariate miRNA biomarkers (miR-522-3p/miR-4732-5p, miR-516a-5p/miR-144-3p, and miR-27b-3p/let-7b-5p) that, when applied serially, distinguished between PE cases of different severity and differentiated cases from controls with a sensitivity of 93%, specificity of 79%, positive predictive value (PPV) of 55%, and negative predictive value (NPV) of 89%. In a small independent validation cohort, these ex-miRNA biomarkers had a sensitivity of 91% and specificity of 57%. Combining these ex-miRNA biomarkers with the established sFlt1:PlGF protein biomarker ratio performed better than either set of biomarkers alone (sensitivity of 89.4%, specificity of 91.3%, PPV of 95.5%, and NPV of 80.8%).

11-deoxycortisol positively correlates with T cell immune traits in physiological conditions

Endogenous steroid hormones have significant effects on inflammatory and immune processes, but the immunological activities of steroidogenesis precursors remain largely unexplored. We conducted a systematic approach to examine the association between steroid hormones profile and immune traits in a cohort of 534 healthy volunteers. Serum concentrations of steroid hormones and their precursors (cortisol, progesterone, testosterone, androstenedione, 11-deoxycortisol and 17-OH progesterone) were determined by liquid chromatography-tandem mass spectrometry. Immune traits were evaluated by quantifying cellular composition of the circulating immune system and exvivo cytokine responses elicited by major human pathogens and microbial ligands. An independent cohort of 321 individuals was used for validation, followed by invitro validation experiments. We observed a positive association between 11-deoxycortisol and lymphoid cellular subsets numbers and function (especially IL-17 response). The association with lymphoid cellularity was validated in an independent validation cohort. Invitro experiments showed that, as compared to androstenedione and 17-OH progesterone, 11-deoxycortisol promoted T cell proliferation and Candida-induced Th17 polarization at physiologically relevant concentrations. Functionally, 11-deoxycortisol-treated T cells displayed a more activated phenotype (PD-L1high CD25high CD62Llow CD127low) in response to CD3/CD28 co-stimulation, and downregulated expression of T-bet nuclear transcription factor. Our findings suggest a positive association between 11-deoxycortisol and T-cell function under physiological conditions. Further investigation is needed to explore the potential mechanisms and clinical implications. Found in acknowledgements.

Method for the extraction of circulating nucleic acids based on MOF reveals cell-free RNA signatures in liver cancer.

Cell-free RNA (cfRNA) allows assessment of health, status, and phenotype of a variety of human organs and is a potential biomarker to non-invasively diagnose numerous diseases. Nevertheless, there is a lack of highly efficient and bias-free cfRNA isolation technologies due to the low abundance and instability of cfRNA. Here, we developed a reproducible and high-efficiency isolation technology for different types of cell-free nucleic acids (containing cfRNA and viral RNA) in serum/plasma based on the inclusion of nucleic acids by metal-organic framework (MOF) materials, which greatly improved the isolation efficiency and was able to preserve RNA integrity compared with the most widely used research kit method. Importantly, the quality of cfRNA extracted by the MOF method is about 10-fold that of the kit method, and the MOF method isolates more than three times as many different RNA types as the kit method. The whole transcriptome mapping characteristics of cfRNA in serum from patients with liver cancer was described and a cfRNA signature with six cfRNAs was identified to diagnose liver cancer with high diagnostic efficiency (area under curve=0.905 in the independent validation cohort) using this MOF method. Thus, this new MOF isolation technique will advance the field of liquid biopsy, with the potential to diagnose liver cancer.

Proteomics-driven noninvasive screening of circulating serum protein panels for the early diagnosis of hepatocellular carcinoma

Early diagnosis of hepatocellular carcinoma (HCC) lacks highly sensitive and specific protein biomarkers. Here, we describe a staged mass spectrometry (MS)-based discovery-verification-validation proteomics workflow to explore serum proteomic biomarkers for HCC early diagnosis in 1002 individuals. Machine learning model determined as P4 panel (HABP2, CD163, AFP and PIVKA-II) clearly distinguish HCC from liver cirrhosis (LC, AUC 0.979, sensitivity 0.925, specificity 0.915) and healthy individuals (HC, AUC 0.992, sensitivity 0.975, specificity 1.000) in an independent validation cohort, outperforming existing clinical prediction strategies. Furthermore, the P4 panel can accurately predict LC to HCC conversion (AUC 0.890, sensitivity 0.909, specificity 0.877) with predicting HCC at a median of 11.4 months prior to imaging in prospective external validation cohorts (No.: Keshen 2018_005_02 and NCT03588442). These results suggest that proteomics-driven serum biomarker discovery provides a valuable reference for the liquid biopsy, and has great potential to improve early diagnosis of HCC.

Leveraging a gene signature associated with disulfidptosis identified by machine learning to forecast clinical outcomes, immunological heterogeneities, and potential therapeutic targets within lower-grade glioma.

Disulfidptosis, a newly defined type of programmed cell death, has emerged as a significant regulatory process in the development and advancement of malignant tumors, such as lower-grade glioma (LGG). Nevertheless, the precise biological mechanisms behind disulfidptosis in LGG are yet to be revealed, considering the limited research conducted in this field. We obtained LGG data from the TCGA and CGGA databases and performed comprehensive weighted co-expression network analysis, single-sample gene set enrichment analysis, and transcriptome differential expression analyses. We discovered nine genes associated with disulfidptosis by employing machine learning methods like Cox regression, LASSO regression, and SVM-RFE. These were later used to build a predictive model for patients with LGG. To confirm the expression level, functional role, and impact on disulfidptosis of ABI3, the pivotal gene of the model, validation experiments were carried out in vitro. The developed prognostic model successfully categorized LGG patients into two distinct risk groups: high and low. There was a noticeable difference in the time the groups survived, which was statistically significant. The model's predictive accuracy was substantiated through two independent external validation cohorts. Additional evaluations of the immune microenvironment and the potential for immunotherapy indicated that this risk classification could function as a practical roadmap for LGG treatment using immune-based therapies. Cellular experiments demonstrated that suppressing the crucial ABI3 gene in the predictive model significantly reduced the migratory and invasive abilities of both SHG44 and U251 cell lines while also triggering cytoskeletal retraction and increased cell pseudopodia. The research suggests that the prognostic pattern relying on genes linked to disulfidptosis can provide valuable insights into the clinical outcomes, tumor characteristics, and immune alterations in patients with LGG. This could pave the way for early interventions and suggests that ABI3 might be a potential therapeutic target for disulfidptosis.

Comprehensive analysis of the role of cuproptosis-related genes in the prognosis and immune infiltration of adrenocortical Carcinoma

BackgroundCuproptosis is a recently discovered form of nonapoptotic programmed cell death. However, no research on cuproptosis in the context of adrenocortical carcinoma has been conducted, and the prognostic value of assessing cuproptosis remains unclear. MethodsIn this study, we established comprehensive models to assess gene expression changes, mutation status, and prognosis prediction and developed a prognostic nomogram for cuproptosis-related genes. Using data from The Cancer Genome Atlas (TCGA), Gene Expression Omnibus (GEO), and Genotype-Tissue Expression (GTEx) databases, an analysis of 11 cuproptosis-related genes was performed. Additionally, a risk scoring method and nomogram were used to assess the relationships among cuproptosis-associated genes, transcript expression, clinical characteristics, and prognosis. The connections among tumors, immune checkpoints, and immune infiltration were also analyzed. ResultsThe patterns observed in patients with adrenocortical carcinoma who were assessed using cuproptosis-associated risk scores provide useful information for understanding gene mutations, clinical outcomes, immune cell infiltration, and immune checkpoint analysis results. FDX1, LIPT1, MTF1, COX11, CYP2D6, DLAT, ATP7Band CDKN2A were differentially expressed in patients with adrenocortical carcinoma and normal controls. In addition, higher risk scores were significantly associated with poor overall survival and progression-free interval. The nomogram model subsequently developed to facilitate the clinical application of the analysis showed good predictive and calibration capabilities. GSE10927 and GSE33371 were used for independent cohort validation. Moreover, CDKN2A, FDX1, and other cuproptosis-related genes were significantly associated with immune infiltration and checkpoints. ConclusionWe confirmed that our model had excellent predictive ability in patients with adrenocortical carcinoma. Therefore, an in-depth evaluation of patients using cuproptosis-related risk scores is clinically essential and can assist in therapy in the future.

Specific gut microbiome signature predicts hepatitis B virus-related hepatocellular carcinoma patients with microvascular invasion

Background We aimed to assess differences in intestinal microflora between patients with operable hepatitis B virus-related hepatocellular carcinoma (HBV-HCC) with microvascular invasion (MVI) and those without MVI. Additionally, we investigated the potential of the microbiome as a non-invasive biomarker for patients with MVI. Methods We analyzed the preoperative gut microbiomes (GMs) of two groups, the MVI (n = 46) and non-MVI (n = 56) groups, using 16S ribosomal RNA gene sequencing data. At the operational taxonomic unit level, we employed random forest models to predict MVI risk and validated the results in independent validation cohorts [MVI group (n = 17) and non-MVI group (n = 15)]. Results β diversity analysis, utilizing weighted UniFrac distances, revealed a significant difference between the MVI and non-MVI groups, as indicated by non-metric multidimensional scaling and principal coordinate analysis. We also observed a significant correlation between the characteristic intestinal microbial communities at the genus level and their main functions. Nine optimal microbial markers were identified, with an area under the curve of 79.76% between 46 MVI and 56 non-MVI samples and 79.80% in the independent verification group. Conclusion This pioneering analysis of the GM in patients with operable HBV-HCC with and without MVI opens new avenues for treating HBV-HCC with MVI. We successfully established a diagnostic model and independently verified microbial markers for patients with MVI. As preoperative targeted biomarkers, GM holds potential as a non-invasive tool for patients with HBV-HCC with MVI.

Integrative Multianalytical Model Based on Novel Plasma Protein Biomarkers for Distinguishing Lung Adenocarcinoma and Benign Pulmonary Nodules.

Given the pressing clinical problem of making a decision in diagnosis for subjects with pulmonary nodules, we aimed to discover novel plasma protein biomarkers for lung adenocarcinoma (LUAD) and benign pulmonary nodules (BPNs) and then develop an integrative multianalytical model to guide the clinical management of LUAD and BPN patients. Through label-free quantitative plasma proteomic analysis (data are available via ProteomeXchange with identifier PXD046731), 12 differentially expressed proteins (DEPs) in LUAD and BPN were screened. The diagnostic abilities of DEPs were validated in two independent validation cohorts. The results showed that the levels of three candidate proteins (PRDX2, PON1, and APOC3) were lower in the plasma of LUAD than in BPN. The three candidate proteins were combined with three promising computed tomography indicators (spiculation, vascular notch sign, and lobulation) and three traditional markers (CEA, CA125, and CYFRA21-1) to construct an integrative multianalytical model, which was effective in distinguishing LUAD from BPN, with an AUC of 0.904, a sensitivity of 81.44%, and a specificity of 90.14%. Moreover, the model possessed impressive diagnostic performance between early LUADs and BPNs, with the AUC, sensitivity, specificity, and accuracy of 0.868, 65.63%, 90.14%, and 82.52%, respectively. This model may be a useful auxiliary diagnostic tool for LUAD and BPN by achieving a better balance of sensitivity and specificity.

Hybrid deep multi-task learning radiomics approach for predicting EGFR mutation status of non-small cell lung cancer in CT images

Objective. Epidermal growth factor receptor (EGFR) mutation genotyping plays a pivotal role in targeted therapy for non-small cell lung cancer (NSCLC). We aimed to develop a computed tomography (CT) image-based hybrid deep radiomics model to predict EGFR mutation status in NSCLC and investigate the correlations between deep image and quantitative radiomics features. Approach. First, we retrospectively enrolled 818 patients from our centre and 131 patients from The Cancer Imaging Archive database to establish a training cohort (N = 654), an independent internal validation cohort (N = 164) and an external validation cohort (N = 131). Second, to predict EGFR mutation status, we developed three CT image-based models, namely, a multi-task deep neural network (DNN), a radiomics model and a feature fusion model. Third, we proposed a hybrid loss function to train the DNN model. Finally, to evaluate the model performance, we computed the areas under the receiver operating characteristic curves (AUCs) and decision curve analysis curves of the models. Main results. For the two validation cohorts, the feature fusion model achieved AUC values of 0.86 ± 0.03 and 0.80 ± 0.05, which were significantly higher than those of the single-task DNN and radiomics models (all P < 0.05). There was no significant difference between the feature fusion and the multi-task DNN models (P > 0.8). The binary prediction scores showed excellent prognostic value in predicting disease-free survival (P = 0.02) and overall survival (P < 0.005) for validation cohort 2. Significance. The results demonstrate that (1) the feature fusion and multi-task DNN models achieve significantly higher performance than that of the conventional radiomics and single-task DNN models, (2) the feature fusion model can decode the imaging phenotypes representing NSCLC heterogeneity related to both EGFR mutation and patient NSCLC prognosis, and (3) high correlations exist between some deep image and radiomics features.

Single-cell transcriptomics reveals variations in monocytes and Tregs between gout flare and remission.

Gout commonly manifests as a painful, self-limiting inflammatory arthritis. Nevertheless, the understanding of the inflammatory and immune responses underlying gout flares and remission remains ambiguous. Here, based on single-cell RNA-Seq and an independent validation cohort, we identified the potential mechanism of gout flare, which likely involves the upregulation of HLA-DQA1+ nonclassical monocytes and is related to antigen processing and presentation. Furthermore, Tregs also play an essential role in the suppressive capacity during gout remission. Cell communication analysis suggested the existence of altered crosstalk between monocytes and other T cell types, such as Tregs. Moreover, we observed the systemic upregulation of inflammatory and cytokine genes, primarily in classical monocytes, during gout flares. All monocyte subtypes showed increased arachidonic acid metabolic activity along with upregulation of prostaglandin-endoperoxide synthase 2 (PTGS2). We also detected a decrease in blood arachidonic acid and an increase in leukotriene B4 levels during gout flares. In summary, our study illustrates the distinctive immune cell responses and systemic inflammation patterns that characterize the transition from gout flares to remission, and it suggests that blood monocyte subtypes and Tregs are potential intervention targets for preventing recurrent gout attacks and progression.

Clinical application of a population-based input function (PBIF) for a shortened dynamic whole-body FDG-PET/CT protocol in patients with metastatic melanoma treated by immunotherapy

BackgroundThe aim was to investigate the feasibility of a shortened dynamic whole-body (dWB) FDG-PET/CT protocol and Patlak imaging using a population-based input function (PBIF), instead of an image-derived input function (IDIF) across the 60-min post-injection period, and study its effect on the FDG influx rate (Ki) quantification in patients with metastatic melanoma (MM) undergoing immunotherapy.MethodsThirty-seven patients were enrolled, including a PBIF modeling group (n = 17) and an independent validation cohort (n = 20) of MM from the ongoing prospective IMMUNOPET2 trial. All dWB-PET data were acquired on Vision 600 PET/CT systems. The PBIF was fitted using a Feng’s 4-compartments model and scaled to the individual IDIF tail’s section within the shortened acquisition time. The area under the curve (AUC) of PBIFs was compared to respective IDIFs AUC within 9 shortened time windows (TW) in terms of linear correlation (R2) and Bland–Altman tests. Ki metrics calculated with PBIF vs IDIF on 8 organs with physiological tracer uptake, 44 tumoral lesions of MM and 11 immune-induced inflammatory sites of pseudo-progression disease were also compared (Mann–Whitney test).ResultsThe mean ± SD relative AUC bias was calculated at 0.5 ± 3.8% (R2 = 0.961, AUCPBIF = 1.007 × AUCIDIF). In terms of optimal use in routine practice and statistical results, the 5th–7th pass (R2 = 0.999 for both Ki mean and Ki max) and 5th–8th pass (mean ± SD bias = − 4.9 ± 6.5% for Ki mean and − 4.8% ± 5.6% for Ki max) windows were selected. There was no significant difference in Ki values from PBIF5_7 vs IDIF5_7 for physiological uptakes (p > 0.05) as well as for tumor lesions (mean ± SD Ki IDIF5_7 3.07 ± 3.27 vs Ki PBIF5_7 2.86 ± 2.96 100ml/ml/min, p = 0.586) and for inflammatory sites (mean ± SD Ki IDIF5_7 1.13 ± 0.59 vs Ki PBIF5_7 1.13 ± 0.55 100ml/ml/min, p = 0.98).ConclusionOur study showed the feasibility of a shortened dWB-PET imaging protocol with a PBIF approach, allowing to reduce acquisition duration from 70 to 20 min with reasonable bias. These findings open perspectives for its clinical use in routine practice such as treatment response assessment in oncology.

Hierarchical classification-based pan-cancer methylation analysis to classify primary cancer

Hierarchical classification offers a more specific categorization of data and breaks down large classification problems into subproblems, providing improved prediction accuracy and predictive power for undefined categories, while also mitigating the impact of poor-quality data. Despite these advantages, its application in predicting primary cancer is rare. To leverage the similarity of cancers and the specificity of methylation patterns among them, we developed the Cancer Hierarchy Classification Tool (CHCT) using the idea of hierarchical classification, with methylation data from 30 cancer types and 8239 methylome samples downloaded from publicly available databases (The Cancer Genome Atlas (TCGA) and the Gene Expression Omnibus (GEO)). We used unsupervised clustering to divide the classification subproblems and screened differentially methylated sites using Analysis of variance (ANOVA) test, Tukey-kramer test, and Boruta algorithms to construct models for each classifier module. After validation, CHCT accurately classified 1568 out of 1660 cases in the test set, with an average accuracy of 94.46%. We further curated an independent validation cohort of 677 cancer samples from GEO and assigned a diagnosis using CHCT, which showed high diagnostic potential with generally high accuracies (an average accuracy of 91.40%). Moreover, CHCT demonstrates predictive capability for additional cancer types beyond its original classifier scope as demonstrated in the medulloblastoma and pituitary tumor datasets. In summary, CHCT can hierarchically classify primary cancer by methylation profile, by splitting a large-scale classification of 30 cancer types into ten smaller classification problems. These results indicate that cancer hierarchical classification has the potential to be an accurate and robust cancer classification method.

Proteomic characterization of the colorectal cancer response to chemoradiation and targeted therapies reveals potential therapeutic strategies

Chemoradiation and targeted therapies are the major treatments for colorectal cancer (CRC); however, molecular properties associated with therapy resistance are incompletely characterized. Here, we profile the proteome of 254 tumor tissues from patients with CRC undergoing chemotherapy, chemoradiation, or chemotherapy combined with targeted therapy. Proteome-based classification reveals four subtypes featured with distinct biological and therapeutic characteristics. The integrative analysis of CRC cell lines and clinical samples indicates that immune regulation is significantly associated with drug sensitivity. HSF1 can increase DNA damage repair and cell cycle, thus inducing resistance to radiation, while high expression of HDAC6 is negatively associated with response of cetuximab. Furthermore, we develop prognostic models with high accuracy to predict the therapeutic response, further validated by parallel reaction monitoring (PRM) assay in an independent validation cohort. This study provides a rich resource for investigating the mechanisms and indicators of chemoradiation and targeted therapy in CRC.

White matter changes in the cholinergic system in Alzheimer’s disease and dementia with Lewy bodies

AbstractBackgroundThe early degeneration of the cholinergic nucleus basalis of Meynert (NBM) and its cortical projections is a hallmark of both dementia with Lewy bodies (DLB) and Alzheimer’s disease (AD). In addition to the NBM, the other major cholinergic projection system originates in the pedunculopontine nucleus (PPN) and provides innervation to the thalamus. Given that visual hallucinations are a hallmark of DLB, but not AD, and have recently been attributed to thalamic degeneration, we tested whether the PPN projection system is differentially altered in DLB compared to AD.MethodsNBM‐cortical and PPN‐thalamus white matter pathways were tracked using diffusion‐weighted imaging data from 46 patients with AD, 48 with DLB, 35 with mild cognitive impairment (MCI) with AD, 38 with MCI with Lewy bodies, and 71 control participants. A free‐water corrected DTI model was estimated. Free water fraction along the cholinergic pathways was compared between groups controlling for age, sex, and free water fraction from a white matter control mask. Multiple linear regression was applied to test associations with cognitive function, core Lewy body symptoms with a focus on visual hallucinations, and changes in cognition using longitudinal follow‐up data.ResultsFree water fraction in the NBM lateral pathway was increased in both dementia and MCI groups compared to controls and this was associated with performance on overall cognition and phonemic fluency tests as well as with longitudinal changes in cognition (Figure 1). Free water fraction along the PPN‐thalamus tract was increased only in DLB patients while it appeared relatively spared in AD and this increase in the Lewy body patients was related to the presence of visual hallucinations (Figure 2). These results were largely replicated in an independent validation cohort (34 AD, 34 DLB, 35 controls).ConclusionOur cross‐validated findings show that the cholinergic input to the thalamus from the PPN is selectively affected in DLB and contributes to visual hallucinations.

Predicting the pathological invasiveness in patients with a solitary pulmonary nodule via Shapley additive explanations interpretation of a tree-based machine learning radiomics model: a multicenter study.

Radiomics models could help assess the benign and malignant invasiveness and prognosis of pulmonary nodules. However, the lack of interpretability limits application of these models. We thus aimed to construct and validate an interpretable and generalized computed tomography (CT) radiomics model to evaluate the pathological invasiveness in patients with a solitary pulmonary nodule in order to improve the management of these patients. We retrospectively enrolled 248 patients with CT-diagnosed solitary pulmonary nodules. Radiomic features were extracted from nodular region and perinodular regions of 3 and 5 mm. After coarse-to-fine feature selection, the radiomics score (radscore) was calculated using the least absolute shrinkage and selection operator logistic method. Univariate and multivariate logistic regression analyses were performed to determine the invasiveness-related clinicoradiological factors. The clinical-radiomics model was then constructed using the logistic and extreme gradient boosting (XGBoost) algorithms. The Shapley additive explanations (SHAP) method was then used to explain the contributions of the features. After removing batch effects with the ComBat algorithm, we assessed the generalization of the explainable clinical-radiomics model in two independent external validation cohorts (n=147 and n=149). The clinical-radiomic XGBoost model integrating the radscore, CT value, nodule length, and crescent sign demonstrated better predictive performance than did the clinical-radiomics logistic model in assessing pulmonary nodule invasiveness, with an area under the receiver operating characteristic (ROC) curve (AUC) of 0.889 [95% confidence interval (CI), 0.848-0.927] in the training cohort. The SHAP algorithm illustrates the contribution of each feature in the final model. The specific model decision process was visualized using a tree-based decision heatmap. Satisfactory generalization performance was shown with AUCs of 0.889 (95% CI, 0.823-0.942) and 0.915 (95% CI, 0.851-0.963) in the two external validation cohorts. An interpretable and generalized clinical-radiomics model for predicting pulmonary nodule invasibility was constructed to help clinicians determine the invasiveness of pulmonary nodules and devise assessment strategies in an easily understandable manner.

CD49d Expression Is Included in a Revised 4-Factor Model Predicting Outcome in Patients with Chronic Lymphocytic Leukemia Treated with Ibrutinib: A Multi-Center Real-World Experience

Introduction. Expression of the CD49d integrin chain has prognostic impact on ibrutinib (IB)-treated CLL patients (Tissino et al, J Exp Med, 2018; Tissino et al, Blood 2020). In addition, a 4-factor (4-f) score, based on TP53 aberration, β2-microglobulin (β2M), lactate dehydrogenase (LDH) and previous therapy lines was proposed to identify patients at a higher risk of treatment failure or death during IB therapy (Ahn et al, J Clin Oncol, 2020; Morabito et al, Am J Hematol, 2021). The clinical impact of TP53 disruption was recently refined by demonstrating that only the co-presence of TP53 deletion and mutations, and not the single aberrations, has prognostic value in the IB setting (Bomben et al, Leukemia 2022). The aim here is to integrate these observations in a comprehensive scoring for the management of IB-treated patients. Methods. This study is a retrospective/multicenter analysis of 410 CLL patients treated with IB in the current clinical practice (December/2013-March/2021). Outcome data were updated as of March/2023. Median follow-up from IB treatment was 29.5 months (95% CI 26.7-33.1 months); 57 patients were treatment naïve (0 previous lines), while 353 had ≥1 previous lines. CLL patients were characterized for CD49d expression and for the 4-f variables, TP53 disruption was evaluated by FISH and TP53 mutational status by NGS (low-VAF mutations included). OS or PFS were measured from the date of IB treatment to the date of death (OS), or progression/death (PFS) or last follow-up. Results. 1) The canonical 4-f was applied as previously described, identifying 111 low-, 163 intermediate (int)-, and 136 high-risk patients. Patients in the low-risk group had a longer OS than patients in the int-, and high-risk groups ( P=0.0115 and P=0.0011), while no difference was found comparing patients in the int- vs. high-risk groups (P=0.3036). 2) We generated a modified version of the 4-f score by considering: i) as TP53 disrupted only patients presenting a concomitant TP53 deletion and mutation (n=93); ii) since no difference in terms of OS was found comparing previously untreated patients (n=57) and patients treated with 1 previous line (n=157; P=0.78), these two groups were combined (n=214) and kept separated from patients who had received &amp;gt;1 previous lines of therapy (n=196). Accordingly, the 4-f-modified identified 218 low-, 109 int-, and 83 high-risk patients, the 3 groups presented different OS, the latter model outperforming the canonical 4-f score (C-indices 0.616 vs 0.639; P&amp;lt;0.0001). 3) CD49d-high CLL cases (i.e. CLL with CD49d expression ≥30% and/or with CD49d-bimodal expression; n=268) had shorter OS and PFS intervals ( P=0.0005 and 0.0194, respectively) than CD49d-low cases (n=142). By combining these observations, CD49d remained an independent prognostic factor for OS ( P=0.004) in a multivariable model, which included the 4-f-modified int-risk ( P=0.0585) and high-risk ( P=0.0011) groups. Driven by these findings, we re-analyzed the single parameters of the 4-f score along with CD49d expression. According to a new multivariable analysis (Fig. A), LDH was not included in the final model, which in turn included CD49d expression as independent OS predictor ( P=0.0011). 4) We then redesigned a novel 4-f model excluding LDH data in favor of CD49d expression data (4-factor-CD49d). Each of the four factors contributed one point and stratified patients (n=407/410) into three groups with a significantly different risk for OS probability, merged on a one-to-one risk group comparison basis: score 0-1, low-risk (n=185); score 2-3, int-risk (n=211); score 4, high-risk (n=11; Fig. B). The inclusion of CD49d into this novel 4-f model improved its prediction capability (C-index=0.663; P&amp;lt;0.0001 versus both the original 4-f and the 4-f-modified). Similar results were obtained when considering PFS as clinical readout. Conclusions. Evidence is provided that in the context of CLL patients treated with IB in a real-world setting, variables like B2M, the number of previous lines of therapy and assessment of TP53 disruption remain important tools for identifying patients at a higher risk of relapse/death. Here we emphasize the need of a comprehensive assessment of TP53 aberrations considering TP53 deletions and mutations simultaneously, and propose to add the assessment CD49d expression, integrating these four variables into a novel 4-f-CD49d scoring system. Further validation in independent cohorts is needed.

Risk Stratification of Patients with RUNX1-mutated Acute Myeloid Leukemia

Mutations in RUNX1 ( RUNX1mut) occur in ~15% of intensively treated AML cases. RUNX1mut have no specific hotspot and various types of alteration are observed. The European LeukemiaNet (ELN) risk stratification assigns adverse prognosis to RUNX1mut if they do not co-occur with favorable-risk genotypes. Considering the biological complexity of RUNX1 it seems implausible that all alterations have similar consequences. Using clinical and genetic variables, we developed a prognostic risk stratification model for ELN adverse-risk RUNX1mut AML patients. We combined data from five groups, totaling 609 patients with intensively treated RUNX1mut AML, to develop the model. Our training set included 448 patients treated on trials of the AML Cooperative Group (AMLCG; Herold et al, Leukemia, 2020; (n=178)), AML Study Group (Gerstung et al, NEJM, 2016; (n=116)) and Study Alliance Leukemia (n=154). Patients from the Munich Leukemia Laboratory (MLL; (n=107)) and of the Alliance group (trials NCT00048958, NCT00899223, NCT00900224; Support: U10CA180821, U10CA180882, U24CA196171; https://acknowledgments.alliancefound.org; (n=54)) served as independent validation cohorts. Additionally, 955 patients without RUNX1mut treated on AMLCG trials served as controls. Patients with t(15;17), prior treatment, or RUNX1mut with co-occurring favorable-risk genotypes according to ELN 2017 were excluded. Differences between RUNX1mut patients and controls were investigated using univariate logistic regression. Testing was performed using likelihood ratio tests and adjusted for study group. Univariate analyses were adjusted for multiple testing using the Benjamini-Hochberg procedure. We obtained risk prediction models using multivariate Cox regression. Missing values were imputed using the missForest approach. Model selection was performed using forward selection based on the Bayesian information criterion. Cut-offs were based on the 25 th- 50 th- and 75 th-percentile score values obtained in the training data. Performance was evaluated using Kaplan-Meier curves. For internal validation Harrell's C index was estimated using cross-study validation. For external validation, the final risk prediction models were separately applied to the external datasets. RUNX1mut were more common in older, male patients and sAML (table). White blood cells, lactate dehydrogenase and bone marrow blasts were lower in RUNX1mut patients. Mutations in several myelodysplasia-related genes were enriched in RUNX1mut patients ( ASXL1, BCOR, BCORL1, EZH2, KMT2A, PHF6, and STAG2, SF3B1, SRSF2, and U2AF1), whereas DNMT3A, NPM1 and FLT3 were more frequently altered in controls. A strong association with mutations of the splicing factor complex was identified (49% vs. 13%, p&amp;lt;0.0001). Contradicting previous reports, we found no association with IDH mutations. The risk prediction score we obtained for OS is as follows: 0.03054 x age (y) + 0.74996 x adverse MRC + 0.43779 x FLT3-ITD + 0.00317 x WBC count (10^9/L) - 0.00158 x platelet count (10^9/L) + 0.37401 x NRAS-mutation, where the obtained cut-off values are: &amp;lt;1.592 (low risk); 1.592 to 2.303 (moderate risk) &amp;gt;2.303 (high risk). Binary variables are coded as 0 or 1. Harrell's C index estimated using internal validation was 0.6. Kaplan-Meier curves estimated using the training set suggest strong differences in survival between the risk categories. Median overall survival (OS) was 2.5, 1.0 and 0.6 years for low-, intermediate-, and high-risk. External validation using the Alliance cohort shows similar results (Figure). Results obtained for the MLL cohort show smaller differences. However, we still observe clear separation of risk groups with a significant difference between low- and high-risk groups (adjusted p-value: 0.0266). Scores for relapse-free survival (RFS) as well as for OS and RFS censored for allogeneic transplant show similar results. We analyzed a large collection of intensively treated RUNX1mut AML patients and observed heterogenous outcomes that could be predicted by applying few variables (age, MRC-score, FLT3/NRAS mutation-status, and WBC/platelet count). The OS of RUNX1mut high-risk patients is discouraging, highlighting the unmet need of these patients. In addition, our work demonstrates that ELN risk groups can be further stratified and that integrated approaches using routinely available variables can further advance risk prediction.

Molecular Subtypes in BCR::ABL1-Positive ALL Are Defined By Cellular Origins and Cooperating Genomic Events

Two developmental trajectories are acknowledged in BCR::ABL1-positive acute lymphoblastic leukemia (ALL) as distinct diagnostic entities by the ICC classification: ‘lymphoid-only’ with BCR::ABL1 restricted to the leukemic B precursor compartment vs. ‘multilineage’ with BCR::ABL1 involvement also in other hematopoietic lineages. Diagnostic standards for establishing this distinction are lacking and associated biological and clinical features are insufficiently characterized. To establish developmental trajectories, biological phenotypes, and clinical impact we analyzed n=277 BCR::ABL1-positive ALL patients (age: 2-84 years, median: 46) including our GMALL reference data set (n=113) and two independent validation cohorts (MLL: n=61; St Jude's: n=103; Gu Z, et al. Nat Genet. 2019). Unsupervised gene expression analysis (RNA-Seq: n=277) identified two major gene expression clusters with two further sub-clusters each (Figure A). This clustering was confirmed by a machine learning classifier trained on the GMALL data set which achieved similar grouping of samples in the external validation cohorts. BCR:: ABL1-FISH on FACS-sorted bone marrow/peripheral blood samples revealed the presence of BCR::ABL1 in leukemic as well as myeloid cells in n=18/18 samples from the 1 st major cluster (termed 'multilineage'). In the 2 nd major cluster ('lymphoid'), BCR::ABL1 was restricted to the lymphoid lineage in n=13/16 samples (p&amp;lt;0.001), with n=3/16 cases harboring BCR::ABL1-positive myeloid cells at lower frequencies compared to the multilineage cluster. In each cluster, two patients also harbored BCR::ABL1-positive mature B cells. T cells remained BCR::ABL1-negative. Diagnostic FACS data and analysis of proximity to normal human lymphoid gene expression confirmed more frequent myeloid co-expression and higher proximity to normal pro-B cells in the multilineage cluster, whereas lymphoid cases had increased lymphoid surface marker expression and a stronger proximity to normal pre-B I cells. Genomic profiling using WGS/SNParrays (n=160) revealed significant enrichment for genomic events in the four gene expression sub-clusters: focal deletions comprising exons 1 and 2 of HBS1-like translational GTPase ( HBS1L) or monosomy 7 were strongly enriched in the two multilineage sub-clusters ('delHBS1L'; n=20/27 vs. n=3/133 in remaining cohort or 'del7'; n=16/25 vs. n=10/135; p&amp;lt;0.001). Remarkably, a novel alternative HBS1L transcript with a putative TSS in intron 3 was highly expressed in both multilineage sub-clusters but not in BCR::ABL1 negative ALL cases or healthy B lymphoid progenitors, suggesting this transcript as novel cooperating event in multilineage BCR::ABL1-positive ALL. One lymphoid sub-cluster was enriched for homozygous deletions in IKZF1 ('IKZF1'; n=11/55 vs. n=4/105; p=0.008), whereas the other was enriched for homozygous CDKN2A/B deletions (n=21/53 vs. n=3/107; p&amp;lt;0.001) and PAX5 deletions (n=33/53 vs n=26/107; p&amp;lt;0.001; 'CDKN2A/PAX5'). Hyperdiploid karyotypes were exclusive to the lymphoid main cluster, mostly in CDKN2A/PAX5. We analyzed the clinical implications of these newly established BCR::ABL1-positive ALL subtypes in our homogenously treated GMALL adult patient cohort (n=98, first diagnosis 2014-2021) including TKI treatment combined with age-adapted chemotherapy, MRD monitoring and allo-SCT in 1 st CR (n=84/98, 86%). Overall survival (OS) at 3 years was uniformly high in multilineage and lymphoid cases (70%±6% vs. 70%±8%; p=0.890; Figure B). However, analysis of the sub-clusters revealed an inferior 3-years OS for the IKZF1 -/- enriched cluster in contrast to excellent outcome in hyperdiploid cases and intermediate outcomes in the remaining sub-clusters (Figure B). These data highlight that transcriptomic signatures in BCR::ABL1-positive ALL are driven by developmental disease origins (‘lymphoid’ vs. ‘multilineage‘) and specific patterns of corresponding genomic events. Novel molecular subtypes of BCR::ABL1-positive ALL have distinct outcomes in the context of current GMALL treatment protocols. Gene-expression based subtype definitions enable classification of BCR::ABL1-positive ALL according to ICC definitions based on RNA-Seq data alone. These definitions have been implemented in ALLCatchR - our freely available tool for ALL subtype allocation - to facilitate validation and application in clinical diagnostics.