External Validation Research Articles

Predictive analysis of COVID-19 occurrence and vaccination impacts across the 50 US states

ObjectiveThis study aimed to outline a machine learning model to assess the effectiveness of vaccination in COVID-19 confirmed cases and fatalities. The proposed model was evaluated using external validation to ensure optimal protection of vaccinated populations, distinguishing between males and females. MethodsThe data from the Centers for Disease Control and Prevention (CDC) in the US, collected between 2021 and 2023, were preprocessed through merging and imputation. A deep learning long short-term memory (LSTM) model was developed to analyze the effectiveness of vaccination in predicting COVID-19 cases and fatalities. The model, which was validated internally and externally, examined the impact of vaccination according to sex. The performance was assessed against current state-of-the-art models, with the LSTM model exhibiting lower root mean square error (RMSE) values. ResultsWe performed intra-, inter-, and external-validation analyses. First, one- and two-dose vaccinations significantly reduced the number of COVID-19 cases and mortality in highly affected states. Second, in the inter-model analysis, the LSTM outperformed the autoregressive integrated moving average (ARIMA) model in predicting cases and deaths, yielding superior results for Texas, California, and Florida. Third, with external validation, our LSTM model effectively predicted vaccination impacts regardless of sex. ConclusionsOur study demonstrates the effectiveness of COVID-19 vaccination, showing that full vaccination significantly reduced the number of confirmed cases and deaths, influencing future public health policies.

Development and Validation of a Novel Model to Discriminate Idiosyncratic Drug-Induced Liver Injury and Autoimmune Hepatitis.

Discriminating between idiosyncratic drug-induced liver injury (DILI) and autoimmune hepatitis (AIH) is critical yet challenging. We aim to develop and validate a machine learning (ML)-based model to aid in this differentiation. This multicenter cohort study utilised a development set from Beijing Friendship Hospital, with retrospective and prospective validation sets from 10 tertiary hospitals across various regions of China spanning January 2009 to May 2023. Different ML algorithms were tested using 24 routine laboratory parameters. The Shapley Additive exPlanations (SHAP) analysis was used to evaluate the contribution of each parameter in the ML model. A total of 2554 patients (1750 for DILI and 804 for AIH) were included. Using Gradient Boost Decision Tree algorithm, five key parameters-aspartate transaminase, globulin, prealbumin, creatinine and platelet count-were selected to construct the ML model. Consequently, a web-based tool named Beijing-AID (BJ-AID) was developed (http://43.143.153.225:5000/). The BJ-AID model demonstrated excellent discrimination performance, with an area under the receiver operating characteristic curve (AUROC) of 0.94 (95% CI, 0.902-0.975) in the development set, 0.91 (95% CI, 0.900-0.928) in all external validation sets and 0.93 (95% CI, 0.889-0.974) in a prospective validation set. Notably, the BJ-AID model also effectively discriminated atypical cases, including drug-induced autoimmune-like hepatitis and AIH with the history of drug consumption, achieving an AUROC = 0.85 (95% CI, 0.742-0.949). We successfully developed and validated a machine learning-based model, BJ-AID, which exhibits a strong discrimination performance. BJ-AID can assist practitioners and hepatologists in diagnosing both typical and atypical cases of DILI and AIH. ClinicalTrials.gov identifier: NCT05532345.

Real-Time Assessment of Alcohol Reward, Stimulation, and Negative Affect in Individuals With and Without Alcohol Use Disorder and Depressive Disorders.

The allostasis theory states that, as addiction develops, alcohol is consumed to relieve negative affect rather than to produce positive effects. This study aimed to investigate the real-time subjective effects of alcohol in individuals with alcohol use disorder (AUD) and those prone to negative affect by virtue of having comorbid depressive disorder (DEP). Participants (N=221) completed high-resolution ecological momentary assessments during 3-hour monitoring of one alcohol drinking episode and one non-alcohol drinking episode in their natural environment. Participants also completed daily mood surveys and next-day surveys. Linear mixed-effect models were used to compare drinking behavior and subjective responses (stimulation, sedation, liking, wanting, negative affect) among 120 participants with AUD (AUD+; with depression [DEP+]: N=64, without depression [DEP-]: N=56) and 101 participants without AUD (AUD-; DEP+: N=45, DEP-: N=56). During the monitoring period, participants with AUD consumed an average of 8.5 standard alcohol drinks (estimated blood alcohol concentration [eBAC]=0.115 g/dl) versus 3.7 drinks (eBAC=0.040 g/dl) for non-AUD participants. The AUD group, regardless of comorbid DEP, reported increases in stimulation and rewarding effects that persisted throughout most of the alcohol episode relative to the non-alcohol episode. To a lesser extent, alcohol relieved negative affect but this was not specific to AUD or DEP groups. Contrary to the allostasis model of addiction's emphasis on negative reinforcement drinking, findings demonstrated that people with AUD prone to negative affect displayed positive alcohol reinforcement with pronounced and prolonged sensitivity to alcohol's pleasurable effects, akin to their noncomorbid counterparts. The findings provided critical testing of addiction theories in the natural environment to enhance external validity.

Linking household members and defining relational networks using routine health data.

The growth of rich electronic health record (EHR) data and large health databases has introduced new opportunities to link individuals together into households and relational networks. These 'linked relational networks' hold promise for providing family-level care and studying intergenerational epidemiology and clinical outcomes. However, as linked relational networks become more commonly available in EHRs and research databases, it is critical to understand their challenges and limitations. Matching algorithms are being used to create linked relational networks in EHR and health databases. Clinically, these algorithms have been most useful to provide dyadic maternal-newborn care. In research, studies using these algorithms investigate topics ranging from the pharmacoepidemiology of parental drug exposure on childhood health outcomes, to heritability of chronic conditions, to associations between parental and child healthcare access and service delivery. However, ethical and technical challenges continue to limit use of these algorithms. There is also a critical research gap in the external validity of these matching algorithms. Linked relational networks are in widespread use in pediatric clinical care and research. More research is needed to understand the scope, limitations, and biases inherent in existing matching strategies.

Predicting mortality in upper gastrointestinal tract cancer using long short-term memory neural networks.

340 Background: Upper gastrointestinal (GI) tract cancers represent a significant global health burden, characterized by high morbidity and mortality rates, with an increasing incidence among aging populations and varying outcomes based on gender and ethnicity. Early prediction of mortality rates in upper gastrointestinal cancer can significantly enhance treatment strategies and patient outcomes. This study describes current trends in upper gastrointestinal cancer and proposes a predictive model, developed using Long Short-Term Memory (LSTM) neural networks to project future trends using demographic data. Methods: We extracted mortality data from the Center for Disease Control and Prevention (CDC) Wide-Ranging Online Data for Epidemiological Research (WONDER) Database spanning 1999, normalized it and encoded categorical variables such as age, sex, ethnicity, and race numerically. Using Studio R to analyze mortality trends, we developed an LSTM model known for its ability to capture and utilize long term data patterns. The model, trained with a sequence length of 20, predicted future data points based on the previous 20. We then assessed its accuracy and generalization with a test set. Ultimately, our model is able to utilize demographic variables to forecast annual mortality and survival rates. Results: The mortality rates for upper GI tract cancers are notably higher among elderly individuals, particularly those aged 80 and above, with males and certain ethnic groups, such as Hispanics. The accuracy of the trained model was evaluated by calculating the Mean Squared Error (MSE), a measure of the difference between the predicted and actual value, with zero indicating no difference between the two. Our LSTM model achieved a training MSE of 0.00085 and a validation MSE of 0.0016, indicating good reliability and accuracy. Conclusions: Our analysis highlighted clear mortality trends and demonstrated the model’s effectiveness in improving mortality predictions in upper gastrointestinal cancer. However, limitations due to the dataset’s lack of clinical variables (such as comorbidities, cancer stage, and prior treatments) and its reliance on a single dataset may affect external validity. Future research is needed to determine if the model can be applied for clinical prognosis or strategic planning. We ultimately created and validated an LSTM neural network model with the CDC WONDER dataset, showing promising results. Future development will aim to include clinical characteristics to create a more personalized predictive tool.

Abstract TP167: Segmentation of leukoaraiosis on noncontrast brain using CT-MRI paired data without human annotation

Purpose: White matter hyperintensities (WMH), also known as leukoaraiosis (LA), are common brain abnormalities in elderly individuals. Evaluating LA on CT is challenging due to the less distinguishable hypoattenuation against white matter. We aimed to develop and validate a segmenting algorithm of LA using CT-MRI paired data. Methods: We included 744 CT-MRI paired dataset of patients with ischemic stroke from 4 stroke centers. An external test set comprised 411 patients from non-overlapping hospitals. WMH on MRI was segmented using validated software and the segmentation mask was registered onto NCCT space. We compared predicted LA versus ground-truth registered LA and WMH on MRI using Dice similarity coefficient (DSC) and concordance correlation coefficient (ρ). Results: Mean age (SD) for training and external test datasets were 68.1 (SD 12.7) and 69.2 (SD 13.5) years and 33.2% and 47.9% were female, respectively. In the internal validation dataset, the algorithm achieved a DSC of 0.53, with a volumetric correlation (ρ) of 0.848 with registered LA volume on CT. External validation showed a DSC of 0.527, with ρ values of 0.919 and 0.760 for predicted LA volumes compared to registered LA and WMH volumes on MRI, respectively. Subgroup analysis demonstrated consistent performance across different CT vendors and infarct volumes. Conclusion: Our deep learning algorithm offers a significant advancement in LA segmentation on CT, bridging the gap between CT and MRI assessments. It improves patient care by providing a consistent, accessible, and accurate method for evaluating LA, supporting both clinical practice and large-scale research.

Prediction of relapse-free survival in stage III gastric cancer patients treated with postoperative adjuvant chemotherapy: Random survival forest analysis from the START-2 trial.

446 Background: The START-2 trial demonstrated that docetaxel plus S-1 (DS) was superior to S-1 alone (S-1) in terms of recurrence-free survival (RFS) and overall survival (OS) as the adjuvant chemotherapy for stage III gastric cancer (GC) patients (pts) (J Clin Oncol 2019, Gastric Cancer 2022). We conducted the START-2 AR study as a retrospective analysis using data from the START-2 trial to develop a prognostic model for stage III GC pts treated with postoperative chemotherapy. Methods: Among 912 pts in the START-2 trial, 599 pts signed the consent form to participate in the START-2 AR study. There were 542 pts (265 pts in DS group and 277 pts in S-1 group) without missing values of prognostic candidates. Pts were randomly divided into a training set and a validation set with a ratio of 7:3. Prognostic model for RFS was developed using random survival forest (RSF) with the training set. The RSF model was applied to the validation set to assess its performance, evaluated using Harrell's C-index. We calculated tertiles of risk scores from the RSF model to divide the training set into low-, middle-, and high-risk subgroups. These cut-off values were used to classify the validation set. A Cox proportional hazard model was applied to compare the prognosis among the risk groups. Results: There were no statistically significant differences in patient characteristicsbetween the 599 pts in the START-2 AR study and the 912 pts in the STRAT-2 trial. RFS was comparable between the two groups. The RSF model for the training set (n = 380) yielded a C-index of 0.873, identifying the number of metastatic lymph nodes, serum albumin, CEA, tumor diameter, the primary tumor location, platelet count, age as prognostic factors in order of importance. In the validation set (n = 162), the C-index was 0.648. Both the low-risk group (n = 35) (hazard ratio [HR]: 0.34, 95% confidence interval [CI]: 0.17 – 0.68, p < 0.01) and the middle-risk group (n = 74) (HR: 0.55, 95% CI: 0.33 – 0.89, p = 0.02) in the validation set had significantly better RFS compared to the high-risk group (n = 53). Among the 76 pts in the low-risk group treated with DS, 44 were categorized into the low-risk group when assuming they had been treated with S-1 alone. The analysis of 599 pts after multiple imputation yielded results consistent with those from the complete data. Conclusions: We developed the RSF-based predictive model for stage III GC pts, which demonstrated good performance. The number of metastatic lymph modes was identified as the most important prognostic factor. Risk stratification based on this model effectively differentiated prognostic outcomes. Future external validation of the model is required. Clinical trial information: UMIN000011438.

Disparities in clinical trial enrollment among LGBTQ+ individuals and its impact on cancer treatment and management: A systematic review.

805 Background: There is a significant gap in oncologic research, which arises from the exclusion of LGBTQ+ individuals from cancer clinical trials, compromising the development of personalized cancer therapies and limiting the generalizability of findings. Despite known health disparities in this population, including higher cancer risk factors and unique psychosocial stressors, data on clinical trial participation remains sparse. This study aims to elucidate the extent of underrepresentation of LGBTQ+ individuals in cancer clinical trials and examine its impact on treatment efficacy, safety, and overall management. Methods: We conducted a systematic review of literature and clinical trial databases, including PubMed, ClinicalTrials.gov, and other registries, focusing on cancer trials from 2010 to 2024. We extracted data on LGBTQ+ enrollment, analyzed demographic inclusivity criteria, and identified barriers to participation. Subgroup analyses evaluated the correlation between LGBTQ+ representation and treatment outcomes, considering variables such as trial design, recruitment strategies, and data collection on sexual orientation and gender identity. The review also examined policy changes and inclusivity initiatives aimed at improving trial access for this community. Results: The review identified a significant underrepresentation of LGBTQ+ individuals in cancer trials, with sexual orientation and gender identity data reported in only 6.2% of trials. Structural and sociocultural barriers, including heteronormative trial designs, lack of inclusive recruitment strategies, and pervasive discrimination, were frequently cited as impediments. Trials that included LGBTQ+ participants rarely conducted stratified analyses to assess differential treatment responses, leading to a paucity of data on drug safety, efficacy, and toxicity profiles specific to this population. The absence of these data may contribute to suboptimal therapeutic decision-making and potential adverse outcomes in clinical practice. Conclusions: The substantial underrepresentation of LGBTQ+ individuals in cancer clinical trials impairs the ability to develop inclusive, evidence-based oncology care. Addressing this gap requires systematic changes, including the integration of sexual and gender minority data collection, implementation of LGBTQ+-specific recruitment frameworks, and stratified analyses of treatment outcomes. Enhancing trial inclusivity will not only improve the external validity of oncologic research but also support the development of tailored therapeutic approaches that address the unique needs of LGBTQ+ cancer patients, ultimately advancing health equity in cancer care.

Deep learning for automated hip fracture detection and classification : achieving superior accuracy.

The aim of this study was to develop and evaluate a deep learning-based model for classification of hip fractures to enhance diagnostic accuracy. A retrospective study used 5,168 hip anteroposterior radiographs, with 4,493 radiographs from two institutes (internal dataset) for training and 675 radiographs from another institute for validation. A convolutional neural network (CNN)-based classification model was trained on four types of hip fractures (Displaced, Valgus-impacted, Stable, and Unstable), using DAMO-YOLO for data processing and augmentation. The model's accuracy, sensitivity, specificity, Intersection over Union (IoU), and Dice coefficient were evaluated. Orthopaedic surgeons' diagnoses served as the reference standard, with comparisons made before and after artificial intelligence assistance. The accuracy, sensitivity, specificity, IoU, and Dice coefficients of the model for the four fracture categories in the internal dataset were as follows: Displaced (1.0, 0.79, 1.0, 0.70, 0.82), Valgus-impacted (1.0, 0.80, 1.0, 0.70, 0.82), Stable (0.99, 0.95, 0.99, 0.83, 0.89), and Unstable (1.0, 0.98, 0.99, 0.86, 0.92), respectively. For the external validation dataset, the sensitivity and specificity were as follows: Displaced (0.83, 0.94), Valgus-impacted (0.89, 0.90), Stable (0.88, 0.95), and Unstable (0.85, 0.99), respectively. The overall means (Micro AVG and Macro AVG) for the external dataset were Micro AVG (0.83 (SD 0.05), 0.96 (SD 0.01)) and Macro AVG (0.69 (SD 0.02), 0.95 (SD 0.02)), respectively. Compared to human diagnosis alone, our study demonstrates that the developed model significantly improves the accuracy of detecting and classifying hip fractures. Our model has shown great potential in assisting clinicians with the accurate diagnosis and classification of hip fractures.

Abstract WP222: External Validation of an Automated Hemorrhage Detection and Segmentation Algorithm on Follow-up CT scans in the AcT trial

Abstract WP222: External Validation of an Automated Hemorrhage Detection and Segmentation Algorithm on Follow-up CT scans in the AcT trial

Identification of multiple complications as independent risk factors associated with 1-, 3-, and 5-year mortality in hepatitis B-associated cirrhosis patients

BackgroundHepatitis B-associated cirrhosis (HBC) is associated with severe complications and adverse clinical outcomes. This study aimed to develop and validate a predictive model for the occurrence of multiple complications (three or more) in patients with HBC and to explore the effects of multiple complications on HBC prognosis.MethodsIn this retrospective cohort study, data from 121 HBC patients treated at Nanjing Second Hospital from February 2009 to November 2019 were analysed. The maximum follow-up period was 10.75 years, with a median of 5.75 years. Eight machine learning techniques were employed to construct predictive models, including C5.0, linear discriminant analysis (LDA), least absolute shrinkage and selection operator (LASSO), k-nearest neighbour (KNN), gradient boosting decision tree (GBDT), support vector machine (SVM), generalised linear model (GLM) and naive Bayes (NB), utilising variables such as medical history, demographics, clinical signs, and laboratory test results. Model performance was evaluated via receiver operating characteristic (ROC) curve analysis, residual analysis, calibration curve analysis, and decision curve analysis (DCA). The influence of multiple complications on HBC survival time was assessed via Kaplan‒Meier curve analysis. Furthermore, LASSO and univariable and multivariable Cox regression analyses were conducted to identify independent prognostic factors for overall survival (OS) in patients with HBC, followed by ROC, C-index, calibration curve, and DCA curve analyses of the constructed prognostic nomogram model. This study utilized bootstrap resampling for internal validation and employed the Medical Information Mart for Intensive Care IV (MIMIC-IV) database for external validation.ResultsThe GBDT model exhibited the highest area under the curve (AUC) and emerged as the optimal model for predicting the occurrence of multiple complications. The key predictive factors included posthospitalisation fever (PHF), body mass index (BMI), retinol binding protein (RBP), total bilirubin (TB) levels, and eosinophils (EOS). Kaplan–Meier analysis revealed that patients with multiple complications had significantly worse OS than those with fewer complications. Additionally, multivariable Cox regression analysis, informed by least absolute shrinkage and LASSO selection, identified hepatocellular carcinoma (HCC), multiple complications, and lactate dehydrogenase (LDH) levels as independent prognostic factors for OS. The prognostic model demonstrated 1-year, 3-year, and 5-year OS ROC AUCs of 0.802, 0.793, and 0.817, respectively. For the internal validation cohort, the corresponding AUC values were 0.797, 0.832, and 0.835. In contrast, the external validation cohort yielded a 1-year ROC AUC of 0.707. Calibration curves indicated good consistency of the model, and DCA demonstrated the model’s clinical utility, showing high net benefits within certain threshold ranges. Compared with the univariable models, the multivariable ROC curves indicated higher AUC values for this prognostic model, and the model also possessed the best c-index.ConclusionThe GBDT prediction model provides a reliable tool for the early identification of high-risk HBC patients prone to developing multiple complications. The concurrent occurrence of multiple complications is an independent prognostic factor for OS in patients with HBC. The constructed prognostic model demonstrated remarkable predictive performance and clinical applicability, indicating its crucial role in enhancing patient outcomes through timely and targeted interventions.

Internal and External Validation of a Deep Learning-Based Early Warning System of Cardiac Arrest with Variable-Length and Irregularly Measured Time Series Data

Internal and External Validation of a Deep Learning-Based Early Warning System of Cardiac Arrest with Variable-Length and Irregularly Measured Time Series Data

Fostering Advocacy in Student Pharmacists: Strategies for Curricular Integration and Assessment.

This study aimed to discuss teaching and assessment of advocacy skills in student pharmacists as called for in the American Association of Colleges of Pharmacy 2022 Curriculum Outcomes and Entrustable Professional Activities. The American Association of Colleges of Pharmacy 2022 Curriculum Outcomes and Entrustable Professional Activities categorize advocacy as a skill that should be developed to advance patient care and the profession. This is a change from the 2013 Center for Advancement of Pharmacy Education outcomes, in which the focus was on advocating for patients. Several thought papers have been released calling for an increase in advocacy training, but there has been limited published work on how advocacy is taught and assessed. In this review of published literature, examples of how pharmacy programs have incorporated and assessed advocacy are provided. Historically, teaching advocacy and its principles has been placed in the core curriculum content, elective courses, and/or cocurriculum, with limited assessment of skills related to advocacy. To support the greater focus on developing advocacy skills within the profession and among patient populations, it is recommended to incorporate the teaching of advocacy-related models of change into education curricula. Examples of 3 existing models are detailed. Advocacy is a core skill set for advancing patient care and shaping the direction of the profession. There are currently limited external validation tools to test if students understand what advocacy is and how to perform advocacy. Further development of teaching and evaluation tools should be explored to help provide a common foundation of this skill set in all student pharmacists.

Abstract WMP39: First External Validation of the Predictive Swallow Score

Abstract WMP39: First External Validation of the Predictive Swallow Score

Development of machine learning-based models for vault prediction in implantable collamer lens surgery according to implant orientation.

The main objective was to develop a prediction model based on machine learning to calculate the postoperative vault as well as the ideal implantable collamer lens (ICL) size, considering for the first time the implantation orientation in a Caucasian population. Arruzafa Ophthalmological Hospital (Cordoba, Spain) and Barraquer Ophthalmology Center (Barcelona, Spain). Multicenter, randomized, retrospective study. Anterior segment biometric data from 235 eyes of patients who underwent ICL lens implantation surgery were collected using the anterior segment optical coherence tomography (AS-OCT) CASIA II, to train and validate five types of multiple regression models based on advanced machine learning techniques. To perform an external validation a dataset of 45 observations was used. The Pearson correlation coefficient between observed and predicted values was similar in the five models in the external validation, with least absolute shrinkage and selection operator (LASSO) regression being the highest (r = 0.62, p < 0.001), followed by random forest regression model (r = 0.60, p < 0.001) and backward stepwise regression (r = 0.58, ρ < 0.001). In addition, the predictions generated by the different models showed closer agreement with the actual vault compared with the Nakamura formulas. Using the new methods, about 70% of the observations had a prediction error below 150 µm. Advanced forms of regressions models based on machine learning allow satisfactory calculation of the ideal lens size, offering greater precision to surgeons customizing surgery according to implant orientation.

The Vienna CATScore for predicting cancer-associated venous thromboembolism: an external validation across multiple time points.

Patients with cancer undergoing systemic therapies have a high risk for venous thromboembolism (VTE). Risk assessment models were developed to select high-risk subgroups that might benefit from primary thromboprophylaxis, yet currently available models reportedly underperform in contemporary cancer treatment populations and risk models across multiple time points throughout therapy are not available. We, therefore, aimed to validate the Vienna CATScore, a nomogram-based model including tumor type and continuous D-dimer levels, in a prospective cohort study of patients initiating contemporary systemic anticancer therapies. The validity of the model was tested at study inclusion, 3 weeks, and 3 months after start of therapy. Overall, 598 patients were included [49% women, median age 62 years (interquartile range 53-70 years)]. Most patients had stage IV disease (68.2%). The 6-month cumulative incidence of VTE was 9.2% [95% confidence interval (CI) 6.8% to 11.5%]. The Vienna CATScore demonstrated good discriminatory ability (c-statistics: 0.69, 95% CI 0.61-0.76) at study baseline and across all evaluated time points (c-statistics: 0.68, 95% CI 0.63-0.73). Applying a 6-month predicted VTE risk threshold of 8%, the CATScore effectively distinguished between low- and high-risk groups at study inclusion (7.1% versus 15.1% observed VTE risk, P= 0.004) and across all three time points (6.3% versus 13.6% observed VTE risk, P < 0.001). Assuming a 50% risk reduction with thromboprophylaxis, this threshold resulted in a number needed to treat (NNT) of 13 and 15, respectively, in the high-risk group, while the NNT was 28 and 32, respectively, in the low-risk group. This external validation of the Vienna CATScore confirms its effectiveness in predicting VTE risk in the initial months of state-of-the-art systemic anticancer therapies and across multiple time points.

Genomic biomarkers of survival in patients with metastatic hormone-sensitive prostate cancer undergoing intensified androgen deprivation therapy.

Androgen deprivation therapy intensification (ADTi) with androgen receptor pathway inhibitors (ARPI), docetaxel or both has been shown to improvesurvival outcomes in patients with metastatic hormone-sensitive prostate cancer (mHSPC). Currently, baseline tumor genomic markers have no role in clinical decision-making in patients with mHSPC. In this IRB-approved retrospective study, patients diagnosed with mHSPC who underwent comprehensive genomic profiling from primary tissue or metastatic sites and treated with ADTi were included. Genomic alterations with an incidence ≥5% were included in the analysis. A total of 276 patients were eligible and included in the study. In the multivariable analysis, TP53 (HR 1.71, 95% CI 1.17-2.49, p = 0.006), RB1 (HR 2.32, 95% CI 1.28-4.18, p = 0.006), PTEN (HR 1.74, 95% CI 1.12-2.7, p = 0.014), and BRCA2 (HR 2.64, 95% CI 1.42-4.92, p = 0.003) were associated with significantly shorter PFS, while TP53 (HR 1.63, 95% CI 1.00-2.64, p = 0.049), RB1 (HR 4.5, 95% CI 2.32-8.70, p < 0.001), and PTEN (HR 2.4, 95% CI 1.38-4.2, p = 0.003) were associated with significantly worse OS. This is one of the largest studies to show the association of baseline tumor genomic markers with survival in patients with mHSPC treated with ADTi. Upon external validation, these results may aid in developing a clinical-genomic risk stratification model, patient counseling, and prognostication.

A nomogram for predicting prognosis for young cervical neuroendocrine carcinoma: A SEER-based study and external validation

A nomogram for predicting prognosis for young cervical neuroendocrine carcinoma: A SEER-based study and external validation

A Composite Endpoint of Liver Surgery (CELS): Development and Validation of a Clinically Relevant Endpoint Requiring a Smaller Sample Size.

The feasibility of trials in liver surgery using a single-component clinical endpoint is low because single endpoints require large samples due to their low incidence. The current study sought to develop and validate a novel composite endpoint of liver surgery (CELS) to facilitate the generation of more feasible and robust high-level evidence in the field of liver surgery. Patients who underwent curative-intent hepatectomy for hepatocellular carcinoma, intrahepatic cholangiocarcinoma, or colorectal liver metastasis were identified using a multi-institutional database. Components of CELS were selected based on perioperative liver surgery-specific complications using univariable logistic regression models. The association of CELS with prolonged length of stay (LOS) and surgery-related death was evaluated and externally validated. Sample sizes were calculated for both individual outcomes and CELS. Among 1958 patients, 377 (19.3%) met CELS criteria based on postoperative bile leak (n = 221, 11.3%), post-hepatectomy liver failure (n = 71, 3.6%), post-hepatectomy hemorrhage (n = 38, 1.9%), or intraoperative blood loss of 2000 ml or greater (n = 101, 5.2%). CELS demonstrated favorable discriminative accuracy of surgery-related death (analytic cohort: area under the curve [AUC], 0.79 vs external validation cohort: AUC, 0.85). In addition LOS was longer among the patients with a positive CELS (analytic cohort: 14 vs. 9 days [p < 0.001] vs. the validation cohort: 10 vs. 6 days [p < 0.001]). Relative to individual endpoints, CELS allowed a 45.8-91.6% reduction in sample size. CELS effectively predicted surgery-related death and can be used as a standardized, clinically relevant endpoint in prospective trials, facilitating smaller sample sizes and enhancing feasibility compared with single quality outcomemetrics.

DICOM LUT is a Key Step in Medical Image Preprocessing Towards AI Generalizability.

Image pre-processing has significant impact on performance of deep learning models in medicine; yet, there is no standardized method for DICOM pre-processing. In this study, we investigate the impact of two commonly used image preprocessing techniques, histogram equalization (HE) and values-of-interest look-up-table (VOI-LUT) transformations on the performance deep learning classifiers for chest X-rays (CXR). We generated two baseline datasets (raw pixel and standard DICOM processed) from our internal CXR dataset and then enhanced both with HE to create four distinct datasets. Four independent deep learning models for diagnosis of pneumothorax were trained and evaluated on two external datasets. Results reveal that HE enhancement significantly affects model performance, particularly in terms of generalizability. Models trained solely on HE-enhanced datasets exhibit poorer performance on external validation sets, suggesting potential overfitting and information loss. These models also exhibit shortcut learning, relying on spurious correlations in the training data for their prediction. This study highlights the importance of machine learning practitioners being aware of preprocessing techniques applied to datasets and their potential impacts on model performance, as well as need for including preprocessing information when sharing datasets. Additionally, this research underscores the necessity of using pixel values closer to clinical standards during dataset curation to improve model robustness and mitigate the risk of information loss.