Area Under The ROC Curve Values Research Articles

Predictors of Hematologic Responses in Patients with Non-Transfusion-Dependent β-Thalassemia Receiving Thalidomide Therapy

To explore the predictors of hematologic responses of non-transfusion-dependent β-thalassemia (NTDT) to thalidomide. 33 patients with NTDT who treated with thalidomide in the 923rd Hospital of the Joint Logistics Support Force of the People's Liberation Army from May 2016 to June 2019 were included in the study. The basic data, hematological indexes, degree of treatment response and genetic background of the patients were analyzed. The baseline fetal hemoglobin (HbF) level of main responders (MaR) was significantly higher than that of minor responders (MiR) and no responders (NR) (P=0.001). And the baseline HbF level was positively correlated with hemoglobin increment after treatment (r=0.601). Genetic background analysis showed that the frequencies of the genotype CT of HBG2 rs7482144 (P=0.031), the genotypes CT/CC (P=0.030) and the minor allele C (P=0.015) of HBS1L-MYB rs9399137, the genotypes AT/TT (P=0.030) and the minor allele T (P=0.028) of HBS1L-MYB rs4895440, the genotypes AG/GG (P=0.030) and the minor allele G (P=0.028) of HBS1L-MYB rs4895441 (P=0.030) in MaR group were significantly higher than those in MiR and NR groups. Comparing the area under the ROC curve (AUC) of the above indicators to predict the main response, the results demonstrated that the predictive value of baseline HbF level was significantly better than rs7482144 (0.91 vs 0.72, P=0.003), rs9399137 (0.91 vs 0.74, P=0.022), rs4895440 (0.91 vs 0.74, P=0.023) and rs4895441 (0.91 vs 0.74, P=0.023), but there was no significant difference in the predictive value between combined single nucleotide polymorphisms (SNPs) (0.91 vs 0.88, P=0.658）and baseline HbF combined SNPs (0.91 vs 0.97, P=0.132). The AUC value of baseline HbF predicting the efficacy of thalidomide as the main response was 0.91, the cut-off value was 27.4%, the sensitivity was 100%, and the specificity was 58.3% (P=0.001). The hematologic response of NTDT to thalidomide is variable and complex. Compared to genetic background, baseline HbF may be a simpler and more efficient tool to predict efficacy response.

Development of deep learning-assisted overscan decision algorithm in low-dose chest CT: Application to lung cancer screening in Korean National CT accreditation program.

We propose a deep learning-assisted overscan decision algorithm in chest low-dose computed tomography (LDCT) applicable to the lung cancer screening. The algorithm reflects the radiologists' subjective evaluation criteria according to the Korea institute for accreditation of medical imaging (KIAMI) guidelines, where it judges whether a scan range is beyond landmarks' criterion. The algorithm consists of three stages: deep learning-based landmark segmentation, rule-based logical operations, and overscan determination. A total of 210 cases from a single institution (internal data) and 50 cases from 47 institutions (external data) were utilized for performance evaluation. Area under the receiver operating characteristic (AUROC), accuracy, sensitivity, specificity, and Cohen's kappa were used as evaluation metrics. Fisher's exact test was performed to present statistical significance for the overscan detectability, and univariate logistic regression analyses were performed for validation. Furthermore, an excessive effective dose was estimated by employing the amount of overscan and the absorbed dose to effective dose conversion factor. The algorithm presented AUROC values of 0.976 (95% confidence interval [CI]: 0.925-0.987) and 0.997 (95% CI: 0.800-0.999) for internal and external dataset, respectively. All metrics showed average performance scores greater than 90% in each evaluation dataset. The AI-assisted overscan decision and the radiologist's manual evaluation showed a statistically significance showing a p-value less than 0.001 in Fisher's exact test. In the logistic regression analysis, demographics (age and sex), data source, CT vendor, and slice thickness showed no statistical significance on the algorithm (each p-value > 0.05). Furthermore, the estimated excessive effective doses were 0.02 ± 0.01 mSv and 0.03 ± 0.05 mSv for each dataset, not a concern within slight deviations from an acceptable scan range. We hope that our proposed overscan decision algorithm enables the retrospective scan range monitoring in LDCT for lung cancer screening program, and follows an as low as reasonably achievable (ALARA) principle.

Novel biomarkers identifying hypertrophic cardiomyopathy and its obstructive variant based on targeted amino acid metabolomics.

Hypertrophic cardiomyopathy (HCM) is an underdiagnosed genetic heart disease worldwide. The management and prognosis of obstructive HCM (HOCM) and non-obstructive HCM (HNCM) are quite different, but it also remains challenging to discriminate these two subtypes. HCM is characterized by dysmetabolism, and myocardial amino acid (AA) metabolism is robustly changed. The present study aimed to delineate plasma AA and derivatives profiles, and identify potential biomarkers for HCM. Plasma samples from 166 participants, including 57 cases of HOCM, 52 cases of HNCM, and 57 normal controls (NCs), who first visited the International Cooperation Center for HCM, Xijing Hospital between December 2019 and September 2020, were collected and analyzed by high-performance liquid chromatography-mass spectrometry based on targeted AA metabolomics. Three separate classification algorithms, including random forest, support vector machine, and logistic regression, were applied for the identification of specific AA and derivatives compositions for HCM and the development of screening models to discriminate HCM from NC as well as HOCM from HNCM. The univariate analysis showed that the serine, glycine, proline, citrulline, glutamine, cystine, creatinine, cysteine, choline, and aminoadipic acid levels in the HCM group were significantly different from those in the NC group. Four AAs and derivatives (Panel A; proline, glycine, cysteine, and choline) were screened out by multiple feature selection algorithms for discriminating HCM patients from NCs. The receiver operating characteristic (ROC) analysis in Panel A yielded an area under the ROC curve (AUC) of 0.83 (0.75-0.91) in the training set and 0.79 (0.65-0.94) in the validation set. Moreover, among 10 AAs and derivatives (arginine, phenylalanine, tyrosine, proline, alanine, asparagine, creatine, tryptophan, ornithine, and choline) with statistical significance between HOCM and HNCM, 3 AAs (Panel B; arginine, proline, and ornithine) were selected to differentiate the two subgroups. The AUC values in the training and validation sets for Panel B were 0.83 (0.74-0.93) and 0.82 (0.66-0.98), respectively. The plasma AA and derivatives profiles were distinct between the HCM and NC groups. Based on the differential profiles, the two established screening models have potential value in assisting HCM screening and identifying whether it is obstructive.

Familial Hypercholesterolemia Identification by Machine Learning Using Lipid Profile Data Performs as Well as Clinical Diagnostic Criteria.

Familial hypercholesterolemia (FH) is a common genetic disorder and, if not diagnosed and treated early, results in premature cardiovascular disease. Most individuals with FH are undiagnosed and machine learning offers a new prospect to improve FH identification. Our objective was to create a machine learning model from basic lipid profile data with better screening performance than LDL-C (low-density lipoprotein cholesterol) cutoff levels and diagnostic performance comparable to the Dutch Lipid Clinic Network criteria. The model was developed combining logistic regression, deep learning, and random forest classification and trained on a 70% split of a data set of individuals clinically suspected of having FH. Model performance, as well as that of the LDL-C cutoff and Dutch Lipid Clinic Network criteria, were assessed on the internal 30% testing data set and an external data set by comparing the area under the receiver operator characteristic (AUROC) curves. All methodologies were measured against the gold standard of FH diagnosis by mutation identification. Furthermore, the model was also tested on 2 lower prevalence data sets. The machine learning model achieved an AUROC curve of 0.711 on the external data set (n=1376; FH prevalence=64%), which was superior to the LDL-C cutoff (AUROC=0.642) and comparable to the Dutch Lipid Clinic Network criteria (AUROC=0.705). The model performed even better when tested on the medium-prevalence (n=2655; FH prevalence=20%) and low-prevalence (n=1616; FH prevalence=1%) data sets, with AUROC curve values of 0.801 and 0.856, respectively. Despite absence of clinical information, the model better identified genetically confirmed FH in a cohort of individuals suspected of having FH than LDL-C cutoff values and was comparable to the Dutch Lipid Clinic Network criteria. The model achieved higher accuracy when tested on 2 cohorts with lower FH prevalence. The application of machine learning is, therefore, a promising tool in both the screening for, and diagnosis of, individuals with FH.

Quantitative analysis for detection and grading of hepatocellular carcinoma: Comparison of diffusion kurtosis imaging, intravoxel incoherent motion and conventional diffusion-weighted imaging.

The aim of the present study was to compare the diagnostic performance of the main parameters derived from diffusion kurtosis imaging (DKI), intravoxel incoherent motion (IVIM) and diffusion-weighted imaging (DWI) regarding the detection and grading of hepatocellular carcinoma (HCC). A total of 78 patients diagnosed with HCC by biopsy were prospectively enrolled in the present study, and underwent routine magnetic resonance imaging (MRI), DWI, IVIM, DKI and contrast-enhanced MRI prior to surgery. Measurements, including mean diffusivity (MD), mean diffusional kurtosis (MK), true diffusion coefficient (D), pseudo-diffusion coefficient (D*), perfusion fraction (f) and apparent diffusion coefficient (ADC), were compared with grading HCC using one-way ANOVA followed by the Student-Neuman-Keuls-q post-hoc test. Spearman's correlation coefficient was used to analyze the correlation between each parameter and pathological grade, while the diagnostic efficiency was evaluated using a receiver operating characteristic (ROC) curve. The 78 patients enrolled in the present study were grouped into highly (n=22), moderately (n=41) or poorly (n=15) differentiated HCC groups according to the criteria of Pathology and Genetics Tumors of the Digestive System. MK values differed significantly between different grades and decreased gradually with the degree of tumor differentiation. The MD, D and ADC values in the highly differentiated HCC group were significantly higher than those in the moderately or poorly differentiated HCC groups (all P<0.001), whereas no significant differences were observed in D* or f (P=0.502 and P=0.853, respectively). A significant correlation was observed between MK, MD, D and ADC, and HCC grades (r=0.705, r=0.570, r=0.423 and r=0.687, respectively). The comparison of the ROC curves of MK, MD, D, ADC, D* and f values for predicting highly differentiated HCC suggested that MK and D were the best indicators for predicting highly differentiated HCC, as the area under the ROC curve (AUC) of MK and D was significantly higher than that of ADC (Z=2.247 and 2.428, P=0.025 and 0.016, respectively), whereas non-statistically significant differences were observed in the AUC values between MK and D (Z=0.072; P=0.942). The DKI-derived MK and IVIM-derived D values had a similar diagnostic performance and were superior to ADC in discriminating the histological grade of HCC. In addition, the combination of MK and D values exhibited an improved diagnostic performance.

Dengue risk zone mapping of Thiruvananthapuram district, India: a comparison of the AHP and F-AHP methods.

Dengue fever, which is spread by Aedes mosquitoes, has claimed many lives in Kerala, with the Thiruvananthapuram district bearing the brunt of the toll. This study aims to demarcate the dengue risk zones in Thiruvananthapuram district using the analytical hierarchy process (AHP) and the fuzzy-AHP (F-AHP) methods. For the risk modelling, geo-environmental factors (normalized difference vegetation index, land surface temperature, topographic wetness index, land use/land cover types, elevation, normalized difference built-up index) and demographic factors (household density, population density) have been utilized. The ArcGIS 10.8 and ERDAS Imagine 8.4 software tools have been used to derive the risk zone maps. The area of the risk maps is classified into five zones. The dengue risk zone maps were validated using dengue case data collected from the Integrated Disease Surveillance Programme portal. From the receiver operating characteristic (ROC) curve analysis and the area under the ROC curve (AUC) values, it is proved that the F-AHP method (AUC value of 0.971) has comparatively more prediction capability than the AHP method (AUC value of 0.954) in demarcating the dengue risk zones. Also, based on the comparison of the risk zone map with actual case data, it was confirmed that around 82.87% of the dengue cases occurred in the very high and high-risk zones, thus proving the efficacy of the model. According to the dengue risk map prepared using the F-AHP model, 9.09% of the area of Thiruvananthapuram district is categorized as very high risk. The prepared dengue risk maps will be helpful for decision-makers, staff with the health, and disaster management departments in adopting effective measures to prevent the risks of dengue spread and thereby minimize loss of life.

Predicting lymph node metastasis and recurrence in patients with early stage colorectal cancer.

Tumor budding (TB), a powerful, independent predictor of colorectal cancer (CRC), is important for making appropriate treatment decisions. Currently, TB is assessed only using the tumor bud count (TBC). In this study, we aimed to develop a novel prediction model, which includes different TB features, for lymph node metastasis (LNM) and local recurrence in patients with pT1 CRC. Enrolled patients (n = 354) were stratified into training and validation cohorts. Independent predictors of LNM and recurrence were identified to generate predictive nomograms that were assessed using the area under the receiver operating characteristic (AUROC) and decision curve analysis (DCA). Seven LNM predictors [gross type, histological grade, lymphovascular invasion (LVI), stroma type, TBC, TB mitosis, and TB CDX2 expression] were identified in the training cohort. LNM, histology grade, LVI, TBC, stroma type, and TB mitosis were independent predictors of recurrence. We constructed an LNM predictive nomogram with a high clinical application value using the DCA. Additionally, a nomogram predicting recurrence-free survival (RFS) was constructed. It presented an AUROC value of 0.944 for the training cohort. These models may assist surgeons in making treatment decisions. In the high-risk group, radical surgery with a postoperative adjuvant chemotherapy was associated with RFS. Postoperative chemotherapy can be better for high-risk patients with pT1 CRC. We showed that TB features besides TBC play important roles in CRC pathogenesis, and our study provides prognostic information to guide the clinical management of patients with early stage CRC.

A radiogenomics biomarker based on immunological heterogeneity for non-invasive prognosis of renal clear cell carcinoma.

BackgroundTumor immunological heterogeneity potentially influences the prognostic disparities among patients with clear cell renal cell carcinoma (ccRCC); however, there is a lack of macroscopic imaging tools that can be used to predict immune-related gene expression in ccRCC.MethodsA novel non-invasive radiogenomics biomarker was constructed for immune-related gene expression in ccRCC. First, 520 ccRCC transcriptomic datasets from The Cancer Genome Atlas (TCGA) were analyzed using a non-negative matrix decomposition (NMF) clustering to identify immune-related molecular subtypes. Immune-related prognostic genes were analyzed through Cox regression and Gene Set Enrichment Analysis (GSEA). We then built a risk model based on an immune-related gene subset to predict prognosis in patients with ccRCC. CT images corresponding to the ccRCC patients in The Cancer Imaging Archive (TCIA) database were used to extract radiomic features. To stratify immune-related gene expression levels, extracted radiogenomics features were identified according to standard consecutive steps. A nomogram was built to combine radiogenomics and clinicopathological information through multivariate logistic regression to further enhance the radiogenomics model. Mann–Whitney U test and ROC curves were used to assess the effectiveness of the radiogenomics marker.ResultsNMF methods successfully clustered patients into diverse subtypes according to gene expression levels in the tumor microenvironment (TME). The relative abundance of 10 immune cell populations in each tissue was also analyzed. The immune-related genomic signature (consisting of eight genes) of the tumor was shown to be significantly associated with survival in patients with ccRCC in TCGA database. The immune-related genomic signature was delineated by grouping the signature expression as either low- or high-risk. Using TCIA database, we constructed a radiogenomics biomarker consisting of 11 radiomic features that were optimal predictors of immune-related gene signature expression levels, which demonstrated AUC (area under the ROC curve) values of 0.76 and 0.72 in the training and validation groups, respectively. The nomogram built by combining radiomics and clinical pathological information could further improve the predictive efficacy of the radiogenomics model (AUC = 0.81, 074).ConclusionsThe novel prognostic radiogenomics biomarker achieved excellent correlation with the immune-related gene expression status of patients with ccRCC and could successfully stratify the survival status of patients in TCGA database. It is anticipated that this work will assist in selecting precise clinical treatment strategies. This study may also lead to precise theranostics for patients with ccRCC in the future.

Diagnostic values of plasma matrix metalloproteinase-7, interleukin-8, and gamma-glutamyl transferase in biliary atresia.

Biliary atresia (BA) is a severe cholestatic liver disease in children featuring cholestasis and liver fibrosis. The early diagnosis of BA is still challenging. This study aimed to evaluate the diagnostic values of matrix metalloprotease-7 (MMP-7), interleukin-8 (IL-8), and gamma-glutamyl transferase (GGT) in BA. Infants diagnosed with BA and non-BA between 2013 and 2018 were retrospectively analyzed. Plasma levels of MMP-7, IL-8, and GGT were measured in these infants. The receiver operating characteristic (ROC) curves and area under the ROC curve (AUC) were used to assess the diagnostic values of MMP-7, IL-8, and GGT. The expression of MMP-7 and IL-8 in the livers was detected by immunofluorescence staining. A total of 229 infants were enrolled in this study: 156 BA infants and 73 non-BA infants including 16 ones with infantile hepatitis syndrome. The plasma levels of MMP-7, IL-8, and GGT in BA infants had a median of 11.8ng/mL (interquartile range, IQR: 5.3-57.5), 1.5ng/mL (IQR: 1.0-2.8), and 381.0 U/L (IQR: 197.0-749.0), respectively, which were higher than non-BA subjects [MMP-7, 4.4ng/mL (IQR: 3.3-6.1); IL-8, 0.7ng/mL (IQR: 0.5-1.0); GGT, 59.0 U/L (IQR: 26.0-124.0)]. The AUC values of MMP-7, IL-8, and GGT for the diagnosis of BA were 0.8035, 0.8083, and 0.9126, respectively. The AUC values of MMP-7 + IL-8, MMP-7 + GGT, IL-8 + GGT, and MMP-7 + IL-8 + GGT for the diagnosis of BA were 0.8248, 0.9382, 0.9168, and 0.9392, respectively. The AUC values of MMP-7, IL-8, and GGT for differentiating BA infants with cholic stool from non-BA infants with cholic stool were 0.8006, 0.8258, and 0.9141, respectively. The expression of MMP-7 and IL-8 was increased in the cholangiocytes in BA livers. Conclusion: Plasma MMP-7, IL-8, and GGT alone or a combination of them has good accuracy to differentiate BA from non-BA and may be reliable biomarkers for BA. What is Known: • Biliary atresia (BA) is a severe cholestatic liver disease in children featuring cholestasis and progressive liver fibrosis. • Although early diagnosis of BA is crucial for good outcomes, it remains a clinical challenge. What is New: • Plasma MMP-7, IL-8, and GGT alone or a combination of them has good accuracy to differentiate BA from non-BA. • Plasma MMP-7, IL-8, and GGT have good accuracy for differentiating BA infants with cholic stool from non-BA infants with cholic stool.

Classification of expert-level therapeutic decisions for degenerative cervical myelopathy using ensemble machine learning algorithms.

BackgroundTherapeutic decisions for degenerative cervical myelopathy (DCM) are complex and should consider various factors. We aimed to develop machine learning (ML) models for classifying expert-level therapeutic decisions in patients with DCM.MethodsThis retrospective cross-sectional study included patients diagnosed with DCM, and the diagnosis of DCM was confirmed clinically and radiologically. The target outcomes were defined as conservative treatment, anterior surgical approaches (ASA), and posterior surgical approaches (PSA). We performed the following classifications using ML algorithms: multiclass, one-versus-rest, and one-versus-one. Two ensemble ML algorithms were used: random forest (RF) and extreme gradient boosting (XGB). The area under the receiver operating characteristic curve (AUC-ROC) was the primary metric. We also identified the variable importance for each classification.ResultsIn total, 304 patients were included (109 conservative, 66 ASA, 125 PSA, and 4 combined surgeries). For multiclass classification, the AUC-ROC of RF and XGB models were 0.91 and 0.92, respectively. In addition, ML models showed AUC-ROC values of >0.9 for all types of binary classifications. Variable importance analysis revealed that the modified Japanese Orthopaedic Association score and central motor conduction time were the two most important variables for distinguishing between conservative and surgical treatments. When classifying ASA and PSA, the number of involved levels, age, and body mass index were important contributing factors.ConclusionML-based classification of DCM therapeutic options is valid and feasible. This study can be a basis for establishing generalizable ML-based surgical decision models for DCM. Further studies are needed with a large multicenter database.

Supersonic shear wave imaging of the tibial nerve for diagnosis of diabetic peripheral neuropathy: A meta-analysis.

BackgroundDiabetic peripheral neuropathy (DPN) is the most common diabetes-associated complication and imposes a significant burden to healthcare systems. Thus, early diagnosis of DPN is extremely critical for management and outcome of diabetic patients. Supersonic Shear Wave Imaging (SSI) enables the noninvasive measurement of nerve stiffness. However, previous studies on SSI in the diagnosis of DPN were limited in sample sizes and reported various results. In this meta-analysis, we aimed to obtain comprehensive evidence on the value of tibial nerve stiffness measurement by SSI in the diagnosis of DPN.MethodsA comprehensive literature search in English and Chinese electronic database was conducted for studies (published until January 25, 2022) that investigated the diagnostic performance of tibial nerve stiffness measurement by SSI for detecting DPN. Summary receiver operating characteristics (SROC) modelling was constructed to conduct the meta-analysis of diagnostic accuracy of SSI for detecting DPN.ResultsFinally, a total of 12 eligible studies with 1325 subjects were included for evaluation, and a meta-analysis was conducted to evaluate the diagnostic performance of tibial nerve stiffness measurement by SSI for detecting DPN. For tibial nerve stiffness measurement by SSI, the summary sensitivity and specificity for the diagnosis of DPN were 80% (95% confidence interval [CI]: 73%–86%) and 86% (95% CI: 82%–89%), respectively. The summary area under the ROC curve (AUROC) value of the SROC was 0.90 (95% CI: 0.87–0.92), for diagnosing DPN. A subgroup analysis of 11 SSI studies from China revealed similar diagnostic performance, with a summary sensitivity of 79% (95% CI: 72%–85%), specificity of 86% (95% CI: 82%–89%) and summary AUROC value of the SROC of 0.90 (95% CI: 0.87–0.92) for diagnosing DPN.ConclusionsOur meta-analysis suggests that a tibial nerve stiffness measurement by SSI shows good performance in diagnosing DPN and has considerable potential as a noninvasive tool for detecting DPN.

Comparison of survival prediction values of different scoring models for patients undergoing transjugular intrahepatic portal shunt: A multicenter retrospective study

AbstractAimThe aim of this study is to compare the prognostic values of the Child–Pugh, integrated model for end‐stage liver disease (iMELD), albumin–bilirubin (ALBI), and Freiburg index of postsurvival (FIPS) scores in patients undergoing transjugular intrahepatic portosystemic shunt (TIPS).MethodsWe conducted a multicenter retrospective study including patients who underwent TIPS by collecting data from several hospitals in southwest China between January 2014 and February 2021. We compared the performance of different scoring models for survival prediction in these patients. The performance of each scoring model was assessed via area under the receiver‐operating characteristic (AUROC) curve analysis.ResultsThe study included 378 TIPS patients (268 men, 110 women; median age 52 [interquartile range, 45–60] years). Age; cirrhosis etiology; ascites severity; albumin levels; international normalized ratio; total bilirubin levels; sodium levels; and Child–Pugh, iMELD, ALBI, and FIPS scores were significant prognostic factors in cirrhotic patients who underwent TIPS. The Child–Pugh, iMELD, ALBI, and FIPS scores were all independent predictors of survival in TIPS patients. Survival analysis showed that all scoring models effectively stratified the prognostic risk of these patients. The Child–Pugh score was the best predictor of postoperative survival, followed by the ALBI and FIPS scores. The iMELD score was the worst predictor. The Child–Pugh, iMELD, ALBI, and FIPS scores predicted the 1‐year postoperative survival, with AUROC values of 0.832, 0.677, 0.761, and 0.745, respectively, and the 3‐year postoperative survival, with AUROC values of 0.710, 0.668, 0.721, and 0.658, respectively. The calibration curve showed that the Child–Pugh, ALBI, and FIPS models performed well in predicting 1‐ and 3‐year survival, whereas the iMELD model was a poor predictor.ConclusionsThe four scoring models can predict survival in cirrhotic patients after TIPS and can effectively stratify prognostic risk. The Child–Pugh score may be more suitable for predicting survival after TIPS in patients with liver cirrhosis.

Validation of Hepatocellular Carcinoma Risk Prediction Models in Patients with Hepatitis B-Related Cirrhosis.

PurposeSeveral risk models have been developed to predict the hepatocellular carcinoma (HCC) risk in patients with chronic hepatitis B (CHB); however, it remains unclear whether these models are useful for risk assessment in patients with hepatitis B virus (HBV)-related cirrhosis undergoing antiviral therapy.Patients and MethodsA total of 252 treatment-naive cirrhosis patients with no history of HCC who underwent treatment with nucleos(t)ide analogues between January 2010 and July 2014 were enrolled. Cox proportional hazards model was used to analyze the risk factors for HCC. “TimeROC” and “survival ROC” package, written for R, were used to compare the time-dependent area under the receiver operating characteristic (AUROC) curves for the predictability of the HCC risk scores.ResultsDuring the mean follow-up period of 56.96 months, 48 (19.0%) patients developed HCC. Cox multivariate stepwise regression analysis revealed that international normalized ratio (hazard ratio [HR] 2.771, 95% confidence interval [CI] 1.462–5.254; P=0.002), alpha-fetoprotein (HR 1.001, 95% CI 1.000–1.003; P=0.035), diabetes mellitus (HR 3.061, 95% CI 1.542–6.077; P=0.001), and alcohol intake (HR 2.250, 95% CI 1.042–4.856; P=0.039) were independent indicators of the HCC risk. AUROC at 3 (0.739) and 5 years (0.695) for the REAL-B score were consistently higher than those of the other risk models except RWS-HCC. The time-dependent AUROC value at 1 year for the REAL-B score was similar to those of the other risk models. According to REAL-B score stratification (0–3, low; 4–7, moderate; and 8–13, high), the HCC risk rates at 1, 3, and 5 years were 2.4%, 5.6%, and 9.0% in the intermediate-risk group, and 7.2%, 21.1%, and 26.3% in the high-risk group, respectively (all P<0.001 between each pair).ConclusionREAL-B score showed a persistently high prognostic capability in predicting the HCC risk in HBV-related cirrhosis patients undergoing antiviral therapy.

P06-04 Agreement between single item physical activity measures in population surveillance in New Zealand

BackgroundSport New Zealand conducts continuous physical activity and sport “Active NZ” surveys, conducted since 2017; these survey around 20,000 representatively sampled adults annually (response rate 29-32%, data weighted to NZ Census population). There has been international interest in “single item” physical activity questions to estimate and monitor physical activity levels. Validated questions for adults have asked about the number of days in the past week people were active for > =30 mins for leisure or transport (SI-days). The Active NZ surveys also ask IPAQ-long form questions and a new single item question on the number of hours people were active in sport/recreation in the past 7 days (SI-hours). The public health problem is that the validated SI-days question cannot directly estimate the WHO recommended threshold of 150+ mins of PA/week. This study describes the prevalence of these questions, and examines the relationships between SI-days, SI-hours and IPAQ-long form [using 600 total PA met-mins as the threshold approximating 150 mins of moderate PA, and also using the leisure time domain only in IPAQ).MethodsAnalyses were descriptive, and inter-method comparisons using Spearman's correlations, and the best fit with the PA thresholds were estimated using Area under the ROC curve (AUC) and Youden's Index.ResultsIPAQ and SI-days were only collected mid 2019 to early 2020 (n = 15044). SI-days showed a mean of 3.2 days (SD=2.2), SI-hours 5.3 (SD=6.2), and SI-hours 150 mins+ was reported by 60.6%. 88.2% reached the IPAQ-total met threshold, and 41.4% met the IPAQ leisure only threshold. Correlations between SI measures and IPAQ were 0.22 for IPAQ-total, and ∼ 0.45 for IPAQ-leisure. SI-days and SI-hours were correlated rho=0.51. ROC curve analyses showed AUC values with IPAQ measures were between 0.63 to 0.76, but the SI-days showed a good AUC of 0.82 (0.81-0.83) with the SI-hours 150 mins threshold. Youdens index suggested the best fit was at 3+ days/week were most likely to meet the SI-hours threshold.ConclusionThese data show SI questions reflect health enhancing thresholds well, and that those reporting >3 SI days showed best fit with the 150mins threshold based on SI hours, indicating good surveillance utility.

Artificial intelligence for predicting survival following deceased donor liver transplantation: Retrospective multi-center study

BackgroundPrevious studies have indicated that the model for end-stage liver disease (MELD) score may fail to predict post-transplantation patient survival. Similarly, other scores (donor MELD score, balance of risk score) that have been developed to predict transplant outcomes have not gained widespread use. These scores are typically derived using linear statistical models. This study aimed to compare the performance of traditional statistical models with machine learning approaches for predicting survival following liver transplantation. Materials and methodsData were obtained from 785 deceased donor liver transplant recipients enrolled in the Korean Organ Transplant Registry (2014–2019). Five machine learning methods (random forest, artificial neural networks, decision tree, naïve Bayes, and support vector machine) and four traditional statistical models (Cox regression, MELD score, donor MELD score and balance of risk score) were compared to predict survival. ResultsAmong the machine learning methods, the random forest yielded the highest area under the receiver operating characteristic curve (AUC-ROC) values (1-month = 0.80; 3-month = 0.85; and 12-month = 0.81) for predicting survival. The AUC-ROC values of the Cox regression analysis were 0.75, 0.86, and 0.77 for 1-month, 3-month, and 12-month post-transplant survival, respectively. However, the AUC-ROC values of the MELD, donor MELD, and balance of risk scores were all below 0.70. Based on the variable importance of the random forest analysis in this study, the major predictors associated with survival were cold ischemia time, donor ICU stay, recipient weight, recipient BMI, recipient age, recipient INR, and recipient albumin level. As with the Cox regression analysis, donor ICU stay, donor bilirubin level, BAR score, and recipient albumin levels were also important factors associated with post-transplant survival in the RF model. The coefficients of these variables were also statistically significant in the Cox model (p < 0.05). The SHAP ranges for selected predictors for the 12-month survival were (−0.02,0.10) for recipient albumin, (−0.05,0.07) for donor bilirubin and (−0.02,0.25) for recipient height. Surprisingly, although not statistically significant in the Cox model, recipient weight, recipient BMI, recipient age, or recipient INR were important factors in our random forest model for predicting post-transplantation survival. ConclusionMachine learning algorithms such as the random forest were superior to conventional Cox regression and previously reported survival scores for predicting 1-month, 3-month, and 12-month survival following liver transplantation. Therefore, artificial intelligence may have significant potential in aiding clinical decision-making during liver transplantation, including matching donors and recipients.

Predicting ACL Injury Using Machine Learning on Data From an Extensive Screening Test Battery of 880 Female Elite Athletes

Background: Injury risk prediction is an emerging field in which more research is needed to recognize the best practices for accurate injury risk assessment. Important issues related to predictive machine learning need to be considered, for example, to avoid overinterpreting the observed prediction performance. Purpose: To carefully investigate the predictive potential of multiple predictive machine learning methods on a large set of risk factor data for anterior cruciate ligament (ACL) injury; the proposed approach takes into account the effect of chance and random variations in prediction performance. Study Design: Case-control study; Level of evidence, 3. Methods: The authors used 3-dimensional motion analysis and physical data collected from 791 female elite handball and soccer players. Four common classifiers were used to predict ACL injuries (n = 60). Area under the receiver operating characteristic curve (AUC-ROC) averaged across 100 cross-validation runs (mean AUC-ROC) was used as a performance metric. Results were confirmed with repeated permutation tests (paired Wilcoxon signed-rank-test; P < .05). Additionally, the effect of the most common class imbalance handling techniques was evaluated. Results: For the best classifier (linear support vector machine), the mean AUC-ROC was 0.63. Regardless of the classifier, the results were significantly better than chance, confirming the predictive ability of the data and methods used. AUC-ROC values varied substantially across repetitions and methods (0.51-0.69). Class imbalance handling did not improve the results. Conclusion: The authors’ approach and data showed statistically significant predictive ability, indicating that there exists information in this prospective data set that may be valuable for understanding injury causation. However, the predictive ability remained low from the perspective of clinical assessment, suggesting that included variables cannot be used for ACL prediction in practice.

Proteome Multimarker Panel for the Early Detection of Hepatocellular Carcinoma: Multicenter Derivation, Validation, and Comparison.

Conventional methods for the surveillance of hepatocellular carcinoma (HCC) by imaging, with and without serum tumor markers, are suboptimal with regard to accuracy. We aimed to develop and validate a reliable serum biomarker panel for the early detection of HCC using a proteomic technique. This multicenter case–control study comprised 727 patients with HCC and patients with risk factors but no HCC. We developed a multiple reaction monitoring–mass spectrometry (MRM-MS) multimarker panel using 17 proteins from the sera of 398 patients. Area under the receiver operating characteristics curve (AUROC) values of this MRM-MS panel with and without α-fetoprotein (AFP) and protein induced by vitamin K absence or antagonist-II (PIVKA-II) were compared. The combination and standalone MRM-MS panels had higher AUROC values than AFP in the training (0.940 and 0.929 vs 0.775, both P < 0.05), test (0.894 and 0.893 vs 0.593, both P < 0.05), and confirmation sets (0.961 and 0.937 vs 0.806, both P < 0.05) in detecting small single HCC. The combination and standalone MRM-MS panels had significantly higher AUROC values than the GALAD score (0.945 and 0.931 vs 0.829, both P < 0.05). Our proteome 17-protein multimarker panel distinguished HCC patients from high-risk controls and had high accuracy in the early detection of HCC.

Predicting prolonged postoperative length of stay risk in patients undergoing lumbar fusion surgery: Development and assessment of a novel predictive nomogram.

ObjectiveThe purpose of this study was to develop and internally validate a prediction nomogram model in patients undergoing lumbar fusion surgery.MethodsA total of 310 patients undergoing lumbar fusion surgery were reviewed, and the median and quartile interval were used to describe postoperative length of stay (PLOS). Patients with PLOS > P75 were defined as prolonged PLOS. The least absolute shrinkage and selection operator (LASSO) regression was used to filter variables for building the prolonged PLOS risk model. Multivariable logistic regression analysis was applied to build a predictive model using the variables selected in the LASSO regression model. The area under the ROC curve (AUC) of the predicting model was calculated and significant test was performed. The Kappa consistency test between the predictive model and the actual diagnosis was performed. Discrimination, calibration, and the clinical usefulness of the predicting model were assessed using the C-index, calibration plot, and decision curve analysis. Internal validation was assessed using the bootstrapping validation.ResultsAccording to the interquartile range of PLOS in a total of 310 patients, the PLOS of 235 patients was ≤P75 (7 days) (normal PLOS), and the PLOS of 75 patients was > P75 (prolonged PLOS). The LASSO selected predictors that were used to build the prediction nomogram included BMI, diabetes, hypertension, duration of surgery, duration of anesthesia, anesthesia type, intraoperative blood loss, sufentanil for postoperative analgesia, and postoperative complication. The model displayed good discrimination with an AUC value of 0.807 (95% CI: 0.758–0.849, P < 0.001), a Kappa value of 0.5186 (cutoff value, 0.2445, P < 0.001), and good calibration. A high C-index value of 0.776 could still be reached in the interval validation. Decision curve analysis showed that the prolonged PLOS nomogram was clinically useful when intervention was decided at the prolonged PLOS possibility threshold of 3%.ConclusionsThis study developed a novel nomogram with a relatively good accuracy to help clinicians access the risk of prolonged PLOS in lumbar fusion surgery patients. By an estimate of individual risk, surgeons and anesthesiologists may shorten PLOS and accelerate postoperative recovery of lumbar fusion surgery through more accurate individualized treatment.

Development and validation of a prediction model of catheter-related thrombosis in patients with cancer undergoing chemotherapy based on ultrasonography results and clinical information

Central venous catheters can be used conveniently to deliver medications and improve comfort in patients with cancer. However, they can cause major complications. The current study aimed to develop and validate an individualized nomogram for early prediction of the risk of catheter-related thrombosis (CRT) in patients with cancer receiving chemotherapy. In total, 647 patients were included in the analysis. They were randomly assigned to the training (n = 431) and validation (n = 216) cohorts. A nomogram for predicting the risk of CRT in the training cohort was developed based on logistic regression analysis results. The accuracy and discriminatory ability of the model were determined using area under the receiver operating characteristic curve (AUROC) values and calibration plots. Multivariate logistic regression analysis showed that body mass index, risk of cancer-related thrombosis, d-dimer level, and blood flow velocity were independent risk factors of CRT. The calibration plot showed an acceptable agreement between the predicted and actual probabilities of CRT. The AUROC values of the nomogram were 0.757 (95% confidence interval: 0.717–0.809) and 0.761 (95% confidence interval: 0.701–0.821) for the training and validation cohorts, respectively. Our model presents a novel, user-friendly tool for predicting the risk of CRT in patients with cancer receiving chemotherapy. Moreover, it can contribute to clinical decision-making.

Diagnostic and Prognostic Nomograms for Lung Metastasis in Triple-Negative Breast Cancer.

Background The lungs are one of the common sites of metastasis of triple-negative breast cancer (TNBC). Patients with lung metastases (LM) have a shorter duration of survival. This study is aimed at determining the prognostic factors of patients with TNBC with LM and constructing two nomograms to assess the risk of LM and the prognosis of patients with TNBC with LM. Methods Clinicopathological and follow-up data of patients with TNBC between 2010 and 2015 were retrieved from the Surveillance, Epidemiology, and End Results (SEER) database. Univariate and multivariate Cox regression analyses were used to screen for independent predictors of LM in patients with TNBC and identify the independent prognostic factors of patients with TNBC with LM. The two nomograms were appraised using calibration curves, receiver operating characteristic (ROC) curves, and decision curve analysis (DCA). Results A total of 27,048 patients with TNBC were included in this study. Age, tumour size, T stage, and N stage were identified as independent risk factors for LM in patients with TNBC. Histological type, marital status, prior surgery, chemotherapy, bone metastases, brain metastases, and LM were confirmed as independent prognostic factors for patients with TNBC with LM. The area under the ROC curve (AUC) of the diagnostic nomogram was 0.838 (95% confidence interval 0.817-0.860) in the training cohort and 0.894 (95% confidence interval 0.875-0.917) in the verification cohort. The AUC values of the 6-, 12-, and 18-month prognostic nomograms in the training cohort were 0.809 (95% confidence interval 0.771-0.868), 0.779 (95% confidence interval 0.737-0.834), and 0.735 (95% confidence interval 0.699-0.811), respectively, and the corresponding AUC values in the validation cohort were 0.735(95% confidence interval 0.642-0.820), 0.672 (95% confidence interval 0.575-0.758), and 0.705 (95% confidence interval 0.598-0.782), respectively. According to the calibration curves and data analysis, both nomograms exhibited good performance. Conclusion We successfully constructed and verified two valuable nomograms for predicting the incidence of LM and prognosis of patients TNBC with LM.