Good Model Calibration Research Articles

UroARC: A novel surgical risk calculator for older adults undergoing pelvic organ prolapse and stress urinary incontinence surgery.

Surgeries for pelvic organ prolapse (POP) and stress urinary incontinence (SUI) are commonly performed in older adults, many of whom are also frail. A surgical risk calculator for older adults undergoing POP/SUI surgeries that incorporates frailty, a factor known to increase the risk of surgical complications, would be helpful for preoperative counseling but currently does not exist. Medicare Carrier, Outpatient, and MedPAR files were examined for beneficiaries undergoing POP and SUI surgery between 2014 and 2016. A total of 15 POP/SUI categories were examined. The Claims-Based Frailty Index (CFI), a validated measure of frailty in Medicare data, and Charlson Comorbidity Index were deconstructed into their individual variables, and individual variables were entered into stepwise logistic regression models to determine which variables were most highly predictive of 30-day complications and 1-year mortality. To verify the prognostic accuracy for each model for surgical complications of interest, calibration curves and tests of model fit, including C-statistic, Brier scores, and Spiegelhalter p values, were determined. In total, 108 479 beneficiaries were included. Among these, 4.7% had CFI scores consistent with mild to severe frailty (CFI≥0.25). A total of 13 prognostic variable categories were found to be most highly predictive of postoperative complications. Calibration curves for each outcome of interest showed models were well-fit. Most models demonstrated high c-statistic values (≥0.7) and high Spiegelhalter p values (≥0.9), indicating good model calibration and excellent discrimination, and low Brier scores (<0.02), indicating high model accuracy. Urologic surgery for older Adults Risk Calculator serves as a novel surgical risk calculator that is readily accessible to both patients and clinicians that specifically factors in components of frailty. Furthermore, this calculator accounts for the heterogeneity of an aging population and can assist in individualized surgical decision-making for these common procedures.

Construction and validation of nomogram to predict surgical site infection after hysterectomy: a retrospective study

This study aimed to develop a predictive tool for surgical site infections (SSI) following hysterectomy and propose strategies for their prevention and control. We conducted a retrospective analysis at a tertiary maternity and child specialist hospital in Zhejiang Province, focusing on patients who underwent hysterectomy between January 2018 and December 2023 for gynecological malignancies or benign reproductive system diseases resistant to medical treatment. Risk factors associated with surgical site infections (SSI) following hysterectomy were identified using LASSO regression analysis on data from 2018 to 2022 as the training set. Independent risk factors were then used to develop a nomogram. The model was validated using data from 2023 as the validation set. Model performance was assessed using the area under the receiver operating characteristic curve (ROC), while calibration curves were employed to gauge model accuracy. Furthermore, clinical utility was evaluated through clinical decision curve analysis (DCA) and clinical impact curve analysis (CIC), providing insights into the practical application of the nomogram. Multivariate analysis identified six independent risk factors associated with SSI development after hysterectomy: BMI ≥ 24 kg/m2 (OR: 2.58; 95% CI 1.14–6.19; P < 0.05), hypoproteinaemia diagnosis (OR: 4.99; 95% CI 1.95–13.02; P < 0.05), postoperative antibiotic use for ≥ 3 days (OR: 49.53; 95% CI 9.73–91.01; P < 0.05), history of previous abdominal surgery (OR: 7.46; 95% CI 2.93–20.01; P < 0.05), hospital stay ≥ 10 days (OR: 9.67; 95% CI 2.06–76.46; P < 0.05), and malignant pathological type (OR: 4.62; 95% CI 1.78–12.76; P < 0.05). A nomogram model was constructed using these variables. ROC and calibration curves demonstrated good model calibration and discrimination in both training and validation sets. Analysis with DCA and CIC confirmed the clinical utility of the nomogram. Personalized nomogram mapping for SSI after hysterectomy enables early identification of high-risk patients, facilitating timely interventions to reduce SSI incidence post-surgery.

Nomogram for Predicting Risk of Cancer Therapy-Related Cardiac Dysfunction in Patients With Human Epidermal Growth Factor Receptor 2-Positive Breast Cancer.

Cancer therapy-related cardiac dysfunction (CTRCD) is an important treatment-limiting toxicity for patients with human epidermal growth factor receptor 2 (HER2)-positive breast cancer that adversely affects cancer and cardiovascular outcomes. Easy-to-use tools that incorporate readily accessible clinical variables for individual estimation of CTRCD risk are needed. From 2004 to 2013, 1440 patients with stage I to III HER2-positive breast cancer treated with trastuzumab-based therapy were identified. A multivariable Cox proportional hazards model was constructed to identify risk factors for CTRCD and included the 1377 patients in whom data were complete. Nine clinical variables, including age, race, body mass index, left ventricular ejection fraction, systolic blood pressure, coronary artery disease, diabetes, arrhythmia, and anthracycline exposure were built into a nomogram estimating risk of CTRCD at 1 year. The nomogram was validated for calibration and discrimination using bootstrap resampling. A total of 177 CTRCD events occurred within 1 year of HER2-targeted treatment. The nomogram for prediction of 1-year CTRCD probability demonstrated good discrimination, with a concordance index of 0.687. The predicted and observed probabilities of CTRCD were similar, demonstrating good model calibration. A nomogram composed of 9 readily accessible clinical variables provides an individualized 1-year risk estimate of CTRCD among women with HER2-positive breast cancer receiving HER2-targeted therapy. This nomogram represents a simple-to-use tool for clinicians and patients that can inform clinical decision-making on breast cancer treatment options, optimal frequency of cardiac surveillance, and role of cardioprotective strategies.

Prognostic Performance of the China Liver Cancer Staging System in Hepatocellular Carcinoma Following Transarterial Chemoembolization.

To validate prognostic performance of the China liver cancer (CNLC) staging system as well as to compare these parameters with those of the Barcelona Clinic Liver Cancer (BCLC) staging system for Chinese hepatocellular carcinoma (HCC) treated with transarterial chemoembolization (TACE). This multicenter retrospective study included 1,124 patients with HCC between January 2012 and December 2020 from six Chinese hospitals. Based on overall survival (OS), the prognostic performance outcomes for the CNLC and BCLC staging systems were compared by model discrimination [C statistic and Akaike information criterion (AIC)], monotonicity of the gradient (linear trend chi-square test), homogeneity (likelihood ratio chi-square test), and calibration (calibration plots). A prospective cohort of 44 patients receiving TACE-based therapy included between January 2021 and December 2022 was used to prospectively validate the outcomes. Median OS was 19.1 (18.2-20.0) months, with significant differences in OS between stages defined by the CNLC and BCLC observed (p<0.001). The CNLC performed better than the BCLC regarding model discrimination (C-index: 0.661 vs. 0.644; AIC: 10,583.28 vs. 10,583.72), model monotonicity of the gradient (linear trend chi-square test: 66.107 vs. 57.418; p<0.001), model homogeneity (159.2 vs. 158.7; p<0.001). Both staging systems had good model calibration. Similar results were observed in the prospective cohort. Combining model discrimination, gradient monotonicity, homogeneity, and calibration, the CNLC performed better than the BCLC for Chinese HCC patients receiving TACE.

A model incorporating histopathology and preoperative gadoxetic acid-enhanced MRI to predict early recurrence of hepatocellular carcinoma without microvascular invasion after curative hepatectomy.

To assess the predictive value of preoperative gadoxetic acid (GA)-enhanced magnetic resonance imaging (MRI) features and postoperative histopathological grading for early recurrence of hepatocellular carcinoma (HCC) without microvascular invasion (MVI) after curative hepatectomy. A total of 85 MVI-negative HCC cases were retrospectively analyzed. Cox analyses were used to identify the independent predictors of early recurrence (within a 24 months span). The clinical prediction Model-1 or Model-2 was established without or with postoperative pathological factor, respectively. Nomogram models were constructed and receiver operating characteristic (ROC) curve analysis was used to assess the models' predictive ability. Internal validation of the prediction models for early HCC recurrence was performed using a bootstrap re-sampling approach. In the multivariate cox regression analysis, Edmondson-Steiner grade, peritumoral hypointensity on hepatobiliary phase (HBP), and relative intensity ratio (RIR) in HBP were identified as independent variables associated with early recurrence. The C-index of the nomogram models and internal validation were both between 0.7 and 0.8, showing good model fitting and calibration effects. The area under the ROC curve (AUC) was 0.781 for Model-1 based on the two preoperative MRI factors. When a third factor, the Edmondson-Steiner grade, was included (Model-2), the AUC increased to 0.834, and the sensitivity increased from 71.4 to 96.4%. Edmondson-Steiner grade, peritumoral hypointensity on HBP, and RIR on HBP can help predict early recurrence of MVI-negative HCC. In comparison with Model-1 (only imaging features), Model-2 (imaging features + histopathological grades) increases the sensitivity in predicting early recurrence of HCC without MVI. Preoperative GA-enhanced MRI signs are of great value in predicting early postoperative recurrence of HCC without MVI, and a combined pathological model was established to evaluate the feasibility and effectiveness of this technique.

Development of novel prediction models for nodal and distant metastasis in G1 and G2 colorectal neuroendocrine tumors

We aimed to construct novel nomograms for nodal and distant metastasis for patients with G1 and G2 colorectal neuroendocrine tumors (NETs). A training set and validation set were obtained by including G1 and G2 colorectal NET patients from the Surveillance, Epidemiology, and End Results database and the Cancer Hospital Chinese Academy of Medical Science, respectively. The area under curve (AUC) values under receiver operating characteristic (ROC) curves, calibration plots, and the Hosmer‒Lemeshow tests were used to evaluate the discriminability and calibration of nomograms. In total, 3690 and 172 patients were included in the training set and validation set, respectively. Tumor size, location, and T stage were included in the nomogram predicting nodal metastasis. The AUC values of the nomogram were 0.972 (95% confidence interval (CI): 0.964-0.980) and 0.897 (95% CI: 0.846-0.948) in the training set and validation set, respectively. The calibration plots and Hosmer‒Lemeshow test for the training set (P = 0.999) and validation set (P = 0.537) showed good model calibration. Tumor size, T stage, and N stage were incorporated into the nomogram predicting distant metastasis. The ROC curves demonstrated desirable discrimination both in the training set (AUC = 0.938 (95% CI: 0.921-0.954)) and validation set (AUC = 0.938 (95% CI: 0.890-0.988)). The calibration curves and Hosmer‒Lemeshow test indicated acceptable model calibration both in the training set (P = 0.908) and validation set (P = 0.722). The proposed nomograms may be used as a reliable tool to predict the nodal and distant metastasis in G1 and G2 colorectal NETs.

Development and validation of a predictive model of the impact of single nucleotide polymorphisms in the ICAM-1 gene on the risk of ischemic cardiomyopathy

Previous research has linked single nucleotide polymorphisms (SNPs) in the ICAM-1 gene to an increased risk of developing ischemic cardiomyopathy (ICM); however, a diagnostic model of ICM according to the ICAM-1 variant has not yet been developed. Therefore, this study aimed to explore the correlation between SNPs in ICAM-1 and the presence of ICM, along with developing a diagnostic model for ICM based on the variants of the ICAM-1 gene. This study recruited a total of 252 patients with ICM and 280 healthy controls. In addition, all the participants were genotyped for SNPs in the ICAM-1 gene by polymerase chain reaction-restriction fragment length polymorphism (PCR-RFLP). Using the training dataset of 371 people, we constructed a nomogram model based on ICAM-1 gene variants and clinical variables. To optimize the feature choice for the ICM risk model, a least absolute shrinkage and selection operator (LASSO) regression model was adopted. We also employed multivariable logistic regression analysis to build a prediction model by integrating the clinical characteristics chosen in the LASSO regression model. Following the receiver operating characteristic (ROC), a calibration plot and decision curve analysis (DCA) were used to evaluate the discrimination, calibration, and clinical usefulness of the predictive model. The predictors involved in the prediction nomogram included age, smoking, diabetes, low-density lipoprotein-cholesterol, hemoglobin, N-terminal pro-B-type natriuretic peptide, ejection fraction, and the rs5491 SNP. The nomogram model exhibited good discrimination ability, with the AUC value of ROC of 0.978 (95%CI: 0.967-0.989, P < 0.001) in the training group and 0.983 (95% CI: 0.969-0.998, P < 0.001) in the validation group. The Hosmer-Lemeshow test demonstrated good model calibration with consistency (P training group = 0.937; P validation group = 0.910). The DCA showed that the ICM nomogram was clinically beneficial, with the threshold probabilities ranging from 0.0 to 1.0. The AT genotype in rs5491 of the ICAM-1 gene was associated with having a higher frequency of ICM. Individuals carrying the mutant AT genotype showed a 5.816-fold higher frequency of ICM compared with those with the AA genotype. ICM patients with the AT genotype also had a higher rate of cardiogenic death. We, therefore, developed a nomogram model that could offer an individualized prediction of ICM risk factors.

Optimization of Dry Weight Assessment in Hemodialysis Patients via Reinforcement Learning.

Dry weight (DW), defined as the lowest tolerated postdialysis weight following the ultrafiltration (UF) of excess fluid volume, is essential for any dialysis prescription for hemodialysis (HD) patients. However, there is no gold standard for DW assessment, and the difficulty of its accurate assessment increases given individual variations and the dynamic changes caused by the uncertainty of patients' condition. Therefore, the current empirical evaluation process is often crude, imprecise, experience-dependent, and energy-consuming. Here, we highlight the personalized dynamic changes in DW over time rather than the more accurate DW assessments at some point in time and formulate the DW evaluation problem into a sequential decision-making process using the Markov decision process (MDP) framework. A reinforcement learning (RL) algorithm based on a dueling double deep Q-network (Duel-DDQN) is proposed to optimize the DW assessment policy, and a multifaceted inspection is applied to assess policy effectiveness and safety. We utilize ten years of data from the Kidney Disease Center, enrolling 750 HD patients and 243,287 dialysis sessions. Good model calibration is confirmed, and off-policy evaluation demonstrates that our policy outperforms other policies, suggesting a decrease of 7.71% in the expected 5-year mortality rate and of 13.44% in the incidence of intradialytic symptoms compared with those of clinicians' strategy. The RL policy adjusts DW more frequently, responds to DW changes more actively, and observes a larger feature space. It is hoped that the proposed solution will help clinicians assess and monitor DW dynamically, making the estimation process more refined, personalized, and intelligent.

Modelling water fluxes to improve banana irrigation scheduling and management in Magdalena, Colombia

In this paper, an irrigation scheduling model for banana (Musa sp.) was developed to simulate crop growth and water fluxes under typical commercial plantation conditions. Whilst generic models exist for scheduling irrigation for many crops, their suitability for bananas are limited because of the asynchronous nature of crop growth. Individual fields on banana plantations typically contain trees at varying stages in their development cycle, so it is important for scheduling to account for this heterogeneity in simulating crop production. A crop modelling approach was developed using field data from Magdalena, an economically important region of banana production in Colombia. Following model development and calibration, irrigation water demand was estimated and weekly irrigation scheduling advice then transmitted by SMS to individual farmers in the region. The model also takes into account farmer feedback on actual irrigation practices to compare against estimated irrigation demands and to train model performance. Despite good model calibration, analysis of irrigation practices from farmer feedback showed only moderate to poor correlation between actual irrigation applications and the scheduling guidance. This implies a reluctance of farmers to change long-established traditional irrigation management practices, despite awareness of the impacts of systematic over-irrigation on yields and increased nutrient leaching risks. Significant ongoing research efforts will be needed to support improved knowledge and practical water management for key plantation crops.

Using CT-based body composition metrics and frailty index in predicting survival among older adults with cancer.

12057 Background: Older adults with cancer are at an increased risk of treatment related toxicities and excess mortality during cancer treatment. While altered body composition and frailty are associated with worse survival among older adults with cancer, no prior study has examined their combined influence on survival prediction. Methods: Prospective study of older adults (≥60 years) undergoing geriatric assessment (GA) at initial visit with a medical oncologist at UAB from 9/2017-07/2021 with available abdominal computed tomography (CT) within 60 days of GA. Using multi-slice CT images from T12 to L5 level, volumetric skeletal muscle (SMV), visceral (VATV) and subcutaneous (SATV) adipose tissues, and skeletal muscle density (SMD), were derived. Sex-specific z-scores for each measure were determined. A 44-item frailty index was obtained, using the deficit accumulation model. Overall survival (OS) was defined as time from GA to death or last follow-up (11/8/2021). Kaplan-Meier estimates of survival rates were compared using log-rank statistics. Multivariable cox regression models were used to predict OS in a random sub-sample (1:1 split of training:validation set), sequentially adding frailty and each body composition measure and assessing improvement with likelihood ratio tests and Harrel’s C statistic. Results: 815 patients were included (median age 68 years, 61% men, and 75% non-Hispanic Whites. 73% had gastrointestinal malignancies (stage III, 25%, stage, IV 48%). 32% were frail, 31% pre-frail. There was a weak negative correlation between height-adjusted SMV and frailty (r = -0.16), particularly among men (r -0.24). Over a median follow-up of 25.7 months (range 0.3-49.6 months), 268 patients (33%) died. The 2-year survival rate was 75.9%, 68.5% and 52.2% among robust, pre-frail and frail (log-rank p &lt;.001), respectively. In multivariable models adjusted for age, sex, race, cancer type and cancer stage, being frail (vs robust) (Hazards Ratio, HR = 2.32; 95%CI: 1.69-3.2; p &lt;.001) and higher skeletal muscle volume (HR = 0.85; 95%CI: 0.72-0.99; p= 0.04, per SD increment) were independently associated with OS. Adding body composition and frailty to clinical variables led to significant improvement in prediction (Harrel’s C increased from 0.69 to 0.74). In the validation set, discrimination was similar (Harrel’s C = 0.72) and plots suggested good model calibration. Conclusions: CT-based body composition metrics and frailty are independent predictors of OS among older adults with cancer and improve survival prediction compared to routine clinical risk factors.

Adaptive Confidence Calibration

Confidence calibration is a quickly growing area of research in deep learning, including computer vision applications. New model architectures and loss functions are being introduced to improve model calibration, an important topic for safety critical AI applications. In order to evaluate these new methods, they are frequently compared to simple baselines, which serve as a comparison and help measure new methods’ effectiveness. Popular baselines for evaluating model confidence calibration include label smoothing, mixup and dropout. Despite these methods frequent use and simple implementations, the parameter values that define them are rarely validated, and are usually a default value. This paper demonstrates the danger in using these default values for calibration benchmarks on common datasets; poor model calibration, specifically, a model that is purely over- or underconfident. We present an adaptive framework that can adjust the parameter value of these baseline methods during training based on validation accuracy and confidence to maintain good model calibration and balance over- and underconfidence. A disadvantage of binning-based calibration metrics is that they are not differentiable and therefore cannot be added to the training loss to penalize calibration error. Our adaptive framework solves this problem and provides a solution for optimizing calibration methods. Experiments with the CIFAR-10 and CIFAR-100 image classification datasets show that our approach improves model calibration compared to using these popular methods with default values and we achieve good model calibration regardless of model architecture and dataset. Further, our analysis provides a comparison between these baseline methods that can be considered for future confidence calibration research

Derivation and Validation of a Risk Score for Abdominal Compartment Syndrome after Endovascular Aneurysm Repair for Ruptured Abdominal Aortic Aneurysms

Abdominal compartment syndrome (ACoS) is a devastating complication after endovascular aneurysm repair for ruptured abdominal aortic aneurysms (rEVAR). This study sought to develop a risk score for ACoS to identify patients who would benefit from early decompressive laparotomy. Model derivation was performed with Vascular Quality Initiative data for rEVAR from 2013 to 2020. The primary outcome was evacuation of abdominal hematoma. A multivariable logistic regression was used to create and validate a scoring system to predict ACoS. The model was validated using institutional data for rEVAR from 1998 to2019. The derivation cohort included 2,310 patients with rEVAR. Abdominal hematoma evacuation occurred in 265 patients (11.5%). Factors associated with abdominal hematoma evacuation on a multivariable analysis included transfer from an outside hospital, preoperative creatinine ≥1.4 mg/dL, preoperative systolic blood pressure ≤85 mmHg, preoperative altered mental status, ≥3.0 liters intraoperative crystalloid, and ≥4 units of red blood cells transfused intraoperatively. The validation cohort consisted of 67 rEVAR; ACoS occurred in 8 patients (11.9%). The c-statistic was 0.84 in the derivation and 0.87 in the validation cohort, whereas Hosmer-Lemeshow was P = 0.15 in the derivation and 0.84 in the validation cohorts, suggesting good model discrimination and calibration. Points were applied based on β-coefficients to produce a risk score ranging from -1 to 13. A cutoff of risk score ≥8 resulted in a sensitivity and specificity of 87.5% and 83.1% for detecting patients with ACoS, respectively. ACoS conveyed a significantly higher mortality in both the derivation (ACoS: 49.8% vs. No ACoS: 17.8%; P < 0.001) and validation cohorts (ACoS: 75.0% vs. No ACoS: 15.2%; P < 0.001). In patients with equivocal signs/symptoms of ACoS, this scoring system can be used to guide surgeons on when to perform decompressive laparotomy prior to leaving the operating room for rEVAR. Patients with a risk score ≥8 would benefit from decompressive laparotomy at index rEVAR.

Development of a Prognostic App (iCanPredict) to Predict Survival for Chinese Women With Breast Cancer: Retrospective Study.

BackgroundAccurate prediction of survival is crucial for both physicians and women with breast cancer to enable clinical decision making on appropriate treatments. The currently available survival prediction tools were developed based on demographic and clinical data obtained from specific populations and may underestimate or overestimate the survival of women with breast cancer in China.ObjectiveThis study aims to develop and validate a prognostic app to predict the overall survival of women with breast cancer in China.MethodsNine-year (January 2009-December 2017) clinical data of women with breast cancer who received surgery and adjuvant therapy from 2 hospitals in Xiamen were collected and matched against the death data from the Xiamen Center of Disease Control and Prevention. All samples were randomly divided (7:3 ratio) into a training set for model construction and a test set for model external validation. Multivariable Cox regression analysis was used to construct a survival prediction model. The model performance was evaluated by receiver operating characteristic (ROC) curve and Brier score. Finally, by running the survival prediction model in the app background thread, the prognostic app, called iCanPredict, was developed for women with breast cancer in China.ResultsA total of 1592 samples were included for data analysis. The training set comprised 1114 individuals and the test set comprised 478 individuals. Age at diagnosis, clinical stage, molecular classification, operative type, axillary lymph node dissection, chemotherapy, and endocrine therapy were incorporated into the model, where age at diagnosis (hazard ratio [HR] 1.031, 95% CI 1.011-1.051; P=.002), clinical stage (HR 3.044, 95% CI 2.347-3.928; P<.001), and endocrine therapy (HR 0.592, 95% CI 0.384-0.914; P=.02) significantly influenced the survival of women with breast cancer. The operative type (P=.81) and the other 4 variables (molecular classification [P=.91], breast reconstruction [P=.36], axillary lymph node dissection [P=.32], and chemotherapy [P=.84]) were not significant. The ROC curve of the training set showed that the model exhibited good discrimination for predicting 1- (area under the curve [AUC] 0.802, 95% CI 0.713-0.892), 5- (AUC 0.813, 95% CI 0.760-0.865), and 10-year (AUC 0.740, 95% CI 0.672-0.808) overall survival. The Brier scores at 1, 5, and 10 years after diagnosis were 0.005, 0.055, and 0.103 in the training set, respectively, and were less than 0.25, indicating good predictive ability. The test set externally validated model discrimination and calibration. In the iCanPredict app, when physicians or women input women’s clinical information and their choice of surgery and adjuvant therapy, the corresponding 10-year survival prediction will be presented.ConclusionsThis survival prediction model provided good model discrimination and calibration. iCanPredict is the first tool of its kind in China to provide survival predictions to women with breast cancer. iCanPredict will increase women’s awareness of the similar survival rate of different surgeries and the importance of adherence to endocrine therapy, ultimately helping women to make informed decisions regarding treatment for breast cancer.

Development and validation using NHANES data of a predictive model for depression risk in myocardial infarction survivors

BackgroundDepression after myocardial infarction (MI) is associated with poor prognosis. This study aimed to develop and validate a nomogram to predict the risk of depression in patients with MI. MethodsThis retrospective study included 1615 survivors of MI aged >20 years who were selected from the 2005–2018 National Health and Nutrition Examination Survey database. The 899 subjects from the 2005–2012 survey comprised the development group, and the remaining 716 subjects comprised the validation group. Univariate and multivariate analyses identified variables significantly associated with depression. The least absolute shrinkage and selection operator (LASSO) binomial regression model was used to select the best predictive variables. ResultsA full predictive model and a simplified model were developed using multivariate analysis and LASSO binomial regression results, respectively, and validated using data from the validation group. The receiver operator characteristic curve and Hosmer–Lemeshow goodness of fit test were used to assess the nomogram's performance. The full nomogram model included 8 items: age, BMI, smoking, drinking, diabetes, exercise, insomnia, and PIR. The area under the curve for the development group was 0.799 and for the validation group was 0.731, indicating that our model has good stability and predictive accuracy. The goodness of fit test showed a good model calibration for both groups. The simplified model includes age, smoking, PIR, and insomnia. The AUC of the simplified model was 0.772 and 0.711 in the development and validation groups, respectively, indicating that the simplified model still possessed good predictive accuracy. ConclusionOur nomogram helped assess the individual probability of depression after MI and can be used as a complement to existing depression screening scales to help physicians make better treatment decisions.

Mortality-Risk Prediction Model from Road-Traffic Injury in Drunk Drivers: Machine Learning Approach.

Background: Alcohol-related road-traffic injury is the leading cause of premature death in middle- and lower-income countries, including Thailand. Applying machine-learning algorithms can improve the effectiveness of driver-impairment screening strategies by legal limits. Methods: Using 4794 RTI drivers from secondary cross-sectional data from the Thai Governmental Road Safety Evaluation project in 2002–2004, the machine-learning models (Gradient Boosting Classifier: GBC, Multi-Layers Perceptrons: MLP, Random Forest: RF, K-Nearest Neighbor: KNN) and a parsimonious logistic regression (Logit) were developed for predicting the mortality risk from road-traffic injury in drunk drivers. The predictors included alcohol concentration level in blood or breath, driver characteristics and environmental factors. Results: Of 4974 drivers in the derived dataset, 4365 (92%) were surviving drivers and 429 (8%) were dead drivers. The class imbalance was rebalanced by the Synthetic Minority Oversampling Technique (SMOTE) into a 1:1 ratio. All models obtained good-to-excellent discrimination performance. The AUC of GBC, RF, KNN, MLP, and Logit models were 0.95 (95% CI 0.90 to 1.00), 0.92 (95% CI 0.87 to 0.97), 0.86 (95% CI 0.83 to 0.89), 0.83 (95% CI 0.78 to 0.88), and 0.81 (95% CI 0.75 to 0.87), respectively. MLP and GBC also had a good model calibration, visualized by the calibration plot. Conclusions: Our machine-learning models can predict road-traffic mortality risk with good model discrimination and calibration. External validation using current data is recommended for future implementation.

Diagnostic significance of clinical and laboratory indices in predicting non-alcoholic fatty liver disease during screening studies

To study the significance of clinical and laboratory non-invasive indexes along with the insulin resistance index when carrying out diagnostic assessment of non-alcoholic fatty liver disease (NAFLD) during screening examinations. The study involved 348 employees working at oil-production enterprises. An ultrasound scanning of the liver was carried out to assess the criteria of NAFLD. The following indexes were calculated: fatty liver index (FLI), hepatic steatosis index (HSI), lipid accumulation products (LAP), and homeostasis model assessment of insulin resistance (HOMA1-IR). The prognostic significance of these indexes in relation to the probability of NAFLD diagnosis based on ultrasound data was studied using single-factor and multi-factor logistic regression models followed by ROC-analysis. The FLI, HSI, and HOMA1-IR indexes in single-factor logistic regression models showed a high statistical significance when carrying out diagnostic assessment the NAFLD with good model calibration capability. The percentage of correct binary classification regards the presence/absence of NAFLD amounted to 82.4% for FLI, 79.7% for HSI, and 72.7% for HOMA1-IR (p0.001). According to the ROC-analysis, the area under the curve (AUC) by the NAFLD diagnostic assessment was 0.917 (95% CI 0.8890.945); 0.880 (95% CI 0.8460.915) and 0.849 (95% CI 0.7640.934), respectively. The multi-factor logistic regression model with the inclusion of FLI and HOMA1-IR 72.7% enabled us to achieve the correct binary classification in terms of NAFLD in 84.2% of cases. When it comes to the ROC-analysis, considering the probabilities predicted in the multi-factor logistic model as the test variable and NAFLD in ultrasound examination as the state variable, it was possible to set the value of AUC 0.933 (95% CI 0.8820.985). The studied clinical and laboratory indexes (FLI, HSI, HOMA1-IR) have a high diagnostic significance regarding NAFLD diagnosed using ultrasonographic criteria. The application of the proposed two-factor logistics model makes it possible to predict the presence of NAFLD when examining a large number of patients, without involving additional ultrasound diagnostics specialists in order to use medical resources rationally.

External validation of the Toronto hepatocellular carcinoma risk index in a Swedish population

Background & AimsThe Toronto hepatocellular carcinoma (HCC) risk index (THRI) is a predictive model to determine the risk of HCC in patients with cirrhosis. This study aimed to externally validate the THRI in a Swedish setting to investigate whether it could identify patients not requiring HCC surveillance.MethodsFrom 2004-2017, 2,491 patients with cirrhosis at the Karolinska University Hospital were evaluated. Patients were classified into low-, intermediate- and high-risk groups for future HCC according to the THRI. Harrell’s C-index, calibration-in-the-large, calibration slope and goodness-of-fit estimates were calculated to assess model discrimination and calibration. Cox proportional hazards regression was used to determine the risk of HCC.ResultsMost patients were male (n = 1,638, 66%). The most common etiologies of cirrhosis were steatohepatitis (n = 1,182, 48%) followed by viral hepatitis (n = 987, 40%). In all, 131 patients (5.3%) were designated as low risk for HCC. Harrell’s C-index was 0.69. Calibration-in-the-large (0.11), calibration slope (1.24, not different from 1, p = 0.66) and goodness-of-fit showed good model calibration. Patients in the high-risk group had a 7.1-fold (95% CI 2.9–17.2) higher risk of HCC and patients in the intermediate-risk group had a 2.5-fold (95% CI 1.0–6.3) higher risk compared to the low-risk group.ConclusionsIn a Swedish setting, the THRI could differentiate between low- and high-risk of HCC development. However, because the low-risk group was relatively small (5.3%), the clinical applicability of the THRI could be limited.Lay summaryThe Toronto hepatocellular carcinoma (HCC) risk index (THRI) is a novel prediction model used to stratify patients with cirrhosis based on future risk of HCC. In this study, the THRI was validated in an external cohort using the TRIPOD guidance. Few patients were identified as low-risk, and the THRI had a modest discriminative ability, limiting its clinical applicability.

A Numerical Investigation of Laser Beam Welding of Stainless Steel Sheets with a Gap

Keyhole laser beam welding (LBW) of 304L stainless steel sheets with a gap in between was numerically simulated with a three-dimensional, transient, multi-physical model for laser material processing based on the finite volume method (FVM). First, the model’s ability to reproduce experimental results on a relatively coarse computational mesh within reasonable computing time, so as to serve as process optimization tool, is presented. An example of process optimization is shown, wherein a given set of weld seam quality criteria is fulfilled by iteratively optimizing a secondary laser beam. The relatively coarse mesh, in combination with a good model calibration for the experimental conditions, allows for sufficiently fast simulations to use this approach for optimization tasks. Finally, using a finer spatial and temporal discretization, the dynamic processes in the vicinity of the keyhole leading to the formation of pores are investigated. The physical phenomena predicted by the simulation are coherent with experimental observations found in literature.

A new concept based on ensemble strategy and derivative for the quantitative analysis of infrared data

AbstractPreprocessing and variable selection are the most widely used strategies to develop accurate predictive models based on infrared spectroscopy. In our study, a new conception that the derivative combined with ensemble strategy based on competitive adaptive reweighted sampling (CARS), stability competitive adaptive reweighted sampling (SCARS), Monte Carlo uninformative variables elimination (MCUVE), and bootstrapping soft shrinkage (BOSS) is put forward. The proposed concept makes the best of the derivative spectra information and successfully combines the strengths of derivative spectra, CARS, SCARS, MCUVE, BOSS, and ensemble submodels. Compared with other methods in this study, this new method can establish good calibration models without increasing the complexity from the perspective of an end user. Also, overfitting issues can be prevented. Derivative1st‐ECARS and Derivative1st‐ESCARS have shown significant improvements in partial least regression calibration based on the experiments of three datasets. The proposed concept shows great potential of the chemometrics approaches applied to infrared data in multivariate calibration.

Dosimetry of Submandibular Glands on Xerostomia for Nasopharyngeal Carcinoma.

PurposeTo investigate dosimetry of submandibular glands on xerostomia after intensity-modulated radiotherapy for nasopharyngeal carcinoma (NPC).MethodsFrom September 2015 to March 2016, 195 NPC patients were investigated. Xerostomia was evaluated at 12 months after treatment via the RTOG/EORTC system. The least absolute shrinkage and selection operator regression model was used to optimize feature selection for grades 2–3 xerostomia. Multivariable logistic regression analysis was applied to build a predicting model incorporating the feature selected in the least absolute shrinkage and selection operator regression model. Discrimination, calibration, and clinical usefulness of the predicting model were assessed using the C-index, calibration plot, and decision curve analysis.ResultsThe V30 of the parotid glands was selected based on the least absolute shrinkage and selection operator regression. The nomogram displayed good discrimination with a C-index of 0.698 (95% confidence interval [CI]: 0.626–0.771) and good calibration (model 1). Addition of the dosimetric parameters including the mean dose to the submandibular glands, V50 of the submandibular glands, and volume of the submandibular glands to the model 1 failed to show incremental prognostic value (model 2). The model 2 showed a C-index of 0.704 (95% CI: 0.632–0.776). Decision curve analysis demonstrated that the model 1 was clinically useful when intervention was decided at the possibility threshold of > 20%. Within this range, net benefit was comparable between the model 1 and model 2.ConclusionPGv30 was a major predictive factor of grades 2–3 xerostomia for NPC. In contrast, the mean dose to the submandibular glands, V50 of the submandibular glands, and volume of the submandibular glands were not independent predictive factors.