Potential Predictive Features Research Articles

Bias in Prediction Models to Identify Patients With Colorectal Cancer at High Risk for Readmission After Resection

PURPOSE Machine learning (ML) algorithms are used for predictive modeling in medicine, but studies often do not evaluate or report on the potential biases of the models. Our purpose was to develop clinical prediction models for readmission after surgery in patients with colorectal cancer (CRC) and to examine their potential for racial bias. METHODS We used the 2012-2020 American College of Surgeons' National Surgical Quality Improvement Program Participant Use File and Targeted Colectomy File. Patients were categorized into four race groups—White, Black or African American, other, and unknown/not reported. Potential predictive features were identified from studies of risk factors of 30-day readmission in patients with CRC. We compared four ML-based methods—logistic regression, multilayer perceptron, random forest, and XGBoost (XGB). Model bias was assessed using false-negative rate (FNR) difference, false-positive rate (FPR) difference, and disparate impact (DI). RESULTS In all, 112,077 patients were included, 67.2% of whom were White, 9.2% Black, 5.6% other race, and 18% with race not recorded. There were significant differences in the area under the receiver operating characteristics curve, FPR, and FNR between race groups across all models. Notably, patients in the other race category had higher FNR compared with Black patients in all but the XGB model, while Black patients had higher FPR than White patients in some models. Patients in the other category consistently had the lowest FPR. Applying the 80% rule for DI, the models consistently met the threshold for unfairness for the other race category. CONCLUSION Predictive models for 30-day readmission after colorectal surgery may perform unequally for different race groups, potentially propagating to inequalities in delivery of care and patient outcomes if the predictions from these models are used to direct care.

A novel hybrid supervised and unsupervised hierarchical ensemble for COVID-19 cases and mortality prediction

Though COVID-19 is no longer a pandemic but rather an endemic, the epidemiological situation related to the SARS-CoV-2 virus is developing at an alarming rate, impacting every corner of the world. The rapid escalation of the coronavirus has led to the scientific community engagement, continually seeking solutions to ensure the comfort and safety of society. Understanding the joint impact of medical and non-medical interventions on COVID-19 spread is essential for making public health decisions that control the pandemic. This paper introduces two novel hybrid machine-learning ensembles that combine supervised and unsupervised learning for COVID-19 data classification and regression. The study utilizes publicly available COVID-19 outbreak and potential predictive features in the USA dataset, which provides information related to the outbreak of COVID-19 disease in the US, including data from each of 3142 US counties from the beginning of the epidemic (January 2020) until June 2021. The developed hybrid hierarchical classifiers outperform single classification algorithms. The best-achieved performance metrics for the classification task were Accuracy = 0.912, ROC-AUC = 0.916, and F1-score = 0.916. The proposed hybrid hierarchical ensemble combining both supervised and unsupervised learning allows us to increase the accuracy of the regression task by 11% in terms of MSE, 29% in terms of the area under the ROC, and 43% in terms of the MPP metric. Thus, using the proposed approach, it is possible to predict the number of COVID-19 cases and deaths based on demographic, geographic, climatic, traffic, public health, social-distancing-policy adherence, and political characteristics with sufficiently high accuracy. The study reveals that virus pressure is the most important feature in COVID-19 spread for classification and regression analysis. Five other significant features were identified to have the most influence on COVID-19 spread. The combined ensembling approach introduced in this study can help policymakers design prevention and control measures to avoid or minimize public health threats in the future.

Predicting heart failure outcomes by integrating breath-by-breath measurements from cardiopulmonary exercise testing and clinical data through a deep learning survival neural network.

Mathematical models previously developed to predict outcomes in patients with heart failure (HF) generally have limited performance and have yet to integrate complex data derived from cardiopulmonary exercise testing (CPET), including breath-by-breath data. We aimed to develop and validate a time-to-event prediction model using a deep learning framework using the DeepSurv algorithm to predict outcomes of HF. Inception cohort of 2490 adult patients with high-risk cardiac conditions or HF underwent CPET with breath-by-breath measurements. Potential predictive features included known clinical indicators, standard summary statistics from CPETs, and mathematical features extracted from the breath-by-breath time series of 13 measurements. The primary outcome was a composite of death, heart transplant, or mechanical circulatory support treated as a time-to-event outcomes. Predictive features ranked as most important included many of the features engineered from the breath-by-breath data in addition to traditional clinical risk factors. The prediction model showed excellent performance in predicting the composite outcome with an area under the curve of 0.93 in the training and 0.87 in the validation data sets. Both the predicted vs. actual freedom from the composite outcome and the calibration of the prediction model were excellent. Model performance remained stable in multiple subgroups of patients. Using a combined deep learning and survival algorithm, integrating breath-by-breath data from CPETs resulted in improved predictive accuracy for long-term (up to 10 years) outcomes in HF. DeepSurv opens the door for future prediction models that are both highly performing and can more fully use the large and complex quantity of data generated during the care of patients with HF.

One toolkit to bring them all, and in silico analyze them.

One toolkit to bring them all, and in silico analyze them.

Distinct trajectories of response to prefrontal tDCS in major depression: results from a 3-arm randomized controlled trial.

Transcranial direct current stimulation (tDCS) is a safe, effective treatment for major depressive disorder (MDD). While antidepressant effects are heterogeneous, no studies have investigated trajectories of tDCS response. We characterized distinct improvement trajectories and associated baseline characteristics for patients treated with prefrontal tDCS, an active pharmacotherapy (escitalopram), and placebo. This is a secondary analysis of a randomized, non-inferiority, double-blinded trial (ELECT-TDCS, N = 245). Participants were diagnosed with an acute unipolar, nonpsychotic, depressive episode, and presented Hamilton Depression Rating Scale (17-items, HAM-D) scores ≥17. Latent trajectory modeling was used to identify HAM-D response trajectories over a 10-week treatment. Top-down (hypothesis-driven) and bottom-up (data-driven) methods were employed to explore potential predictive features using, respectively, conservatively corrected regression models and a cross-validated stability ranking procedure combined with elastic net regularization. Three trajectory classes that were distinct in response speed and intensity (rapid, slow, and no/minimal improvement) were identified for escitalopram, tDCS, and placebo. Differences in response and remission rates were significant early for all groups. Depression severity, use of benzodiazepines, and age were associated with no/minimal improvement. No significant differences in trajectory assignment were found in tDCS vs. placebo comparisons (38.3, 34, and 27.6%; vs. 23.3, 43.3, and 33.3% for rapid, slow, and no/minimal trajectories, respectively). Additional features are suggested in bottom-up analyses. Summarily, groups treated with tDCS, escitalopram, and placebo differed in trajectory class distributions and baseline predictors of response. Our results might be relevant for designing further studies.

Feature Selection for UK Disabled Students’ Engagement Post Higher Education: A Machine Learning Approach for a Predictive Employment Model

While only 4.2 million people out of a population of 7.9 million disabled people are working, a considerable contribution is still required from universities and industries to increase employability among the disabled, in particular, by providing adequate career guidance post higher education. This study aims to identify the potential predictive features, which will improve the chances of engaging disabled school leavers in employment about 6 months after graduation. MALSEND is an analytical platform that consists of information about UK Destinations Leavers from Higher Education (DLHE) survey results from 2012 to 2017. The dataset of 270,934 student records with a known disability provides anonymised information about students' age range, year of study, disability type, results of the first degree, among others. Using both qualitative and quantitative approaches, characteristics of disabled candidates during and after school years were investigated to identify their engagement patterns. This article builds on constructing and selecting subsets of features useful to build a good predictor regarding the engagement of disabled students 6 months after graduation using the big data approach with machine learning principles. Features such as age, institution, disability type, among others were found to be essential predictors of the proposed employment model. A pilot was developed, which shows that the Decision Tree Classifier and Logistic Regression models provided the best results for predicting the Standard Occupation Classification (SOC) of a disabled school leaver in the UK with an accuracy of 96%.

Incidence of Dural Venous Sinus Thrombosis in Patients with Glioblastoma and Its Implications

Glioblastoma (GBM) is associated with increased risk of developing dural venous sinus thrombosis (DVST), which often goes undiagnosed as symptoms are readily attributed to tumor. The purpose of this study was to investigate the incidence of DVST, potential predictive features on imaging, complications, its effect on survival, and time of greatest risk for developing DVST. A retrospective search of patients with GBM who had surgery followed by chemotherapy and/or radiation therapy between 2009 and 2015 at our institution was performed. Magnetic resonance imaging studies of thebrain were reviewed on volumetric postgadolinium T1-weighted sequences for DVST. Tumors were characterized using the Visually Accessible REMBRANDT (Repository for Molecular Brain Neoplasia Data) Images classification, and identified thromboses were tracked for propagation, regression, or resolution. Statistical analyses were directed at identifying clinical predictors and survival differences between the DVST and no-DVST groups. In total, 163 cases totaling 1637 scans, were reviewed; 12 patients (7.4%) developed DVST, of whom 11 presented with thrombus before any treatment. Tumor invasion of dural sinuses and greater T1/fluid-attenuated inversion recovery ratios were significantly associated with thrombus development (P= 0.02 and P= 0.02, respectively). In patients who developed DVST, thrombosis was more likely to develop ipsilateral to tumor side (P=0.01) and was associated with a greater likelihood of developing extracranial venous thromboembolism (P= 0.012). There were no venous infarcts and no significant difference in survival between groups (P= 0.83). Patients with GBM have increased risk of developing DVST, independent of surgical treatment or chemoradiation. DVST presence does not affect survival. Tumor invasion of dural sinuses and greater T1/fluid-attenuated inversion recovery ratio on preoperative imaging were the most significant predictors of DVST development.

Justification of indicators list for modeling the integral evaluation of customer service quality at a pharmacy

The problem of multicollinearity in complex indicator modeling of customer service quality at a pharmacy has been solved by reducing the sufficient number of potential predictors by identifying individual diagnostic features and generating aggregate indicators using factor analysis. Aim. To create an informative and non-redundant set of indicators by reducing the size of potential predictive features space for (further) modeling an integrated quality assessment of customer service at pharmacy. Results. Correlation between 26 potential predictive features that characterize the quality of service at the pharmacy has been investigated. As a result of the formation of synthetic indicators and the allocation of individual diagnostic features, the dimension of the input space is reduced by 2 times. It was identified one representative sign from related pairs of indicators assessments that characterize the availability of goods in the pharmacy, as well as the internal arrangement of the pharmacy for customer service. Using the factor analysis of the input structure, two characteristics of the additional conditions of customer service and related services were synthesized into a single indicator of the «factor of additional services». The aggregate indicator «factor of work of the personnel» has been constructed, taking into account in the complex of 11 professional and personal characteristics of the pharmacists serving visitors. Conclusions. As a result of the research using the methods of correlation and factor analysis, a list of predictive indicators (11 individual indicators and 2 aggregates) was obtained, which laid the basis for further development of the model of the integral indicator of the service quality at the pharmacy.

Association of Clinical and Radiographic Features with Perinephric “Sticky” Fat

The discovery of thick, adherent, perinephric sticky fat (PSF) is relatively common during open or laparoscopic retroperitoneal surgery. To our knowledge, however, there has been no previous analysis of clinical or radiographic features associated with the development of PSF or of perioperative outcomes for those patients in whom it is found. Our objective is to analyze potential predictive features and determine whether there is any effect on clinical or pathologic outcomes for patients with perinephric sticky fat. Patients undergoing partial nephrectomy or laparoscopic cryoablation with available preoperative imaging were identified from 2005 to 2011. Operative records were reviewed to identify patients with and without PSF. Preoperative images and medical records were examined to obtain patient data regarding potential predictors as well as clinical and pathologic outcomes. A total of 29 patients were identified-16 with PSF and 13 controls. Statistically significant factors associated with PSF included sex, tumor size, presence of perinephric stranding, tumor >50% exophytic, and thickness of perinephric fat (P<0.05). Median total operative time for patients with sticky fat was nearly 40 minutes longer than the control group (228 min vs 190 min, P<0.05). All four (17%) patients with Fuhrman grade 3 or 4 renal-cell carcinoma were from the sticky fat group (P=0.09). Despite the small sample size, multiple possible factors associated with perinephric sticky fat were identified and may provide guidance for future investigation of this phenomenon.

The predictive factors and patient-selection associated with the survival benefits of gefitinib in patients with advanced non-small cell lung cancer (NSCLC)

17020 Background: Gefitinib, an inhibitor of epidermal growth factor receptor (EGFR) tyrosine kinase, has shown meaningful antitumor activity for advanced NSCLC. Survival benefit of treatment with gefitinib has not been shown in unselected patients, however heterogeneity in survival outcomes between sub-group (histological types, gender, smoking status and ethnicity) of patients has been observed. The aim of this study was to identify the potential predictive features associated with the survival benefit of treatment with gefitinib. Methods: All NSCLC patients who began taking gefitinib during the period from July 2002 until July 2005 in our institute were retrospectively reviewed. Potential factors analyzed included age, gender, smoking history, performance status (PS), histology, stage, and prior chemotherapy. Results: Overall 221 patients were analyzed. The overall response rate of gefitinib was 26.7%. Median survival time (MST) was 8.0 months. MST of patients whose tumor response was partial response (PR) was 34.5 months. Univariate analysis revealed that the predictive factors of favorable survival were female (p = 0.001), never-smoker (p &lt; 0.0001), good PS (p &lt; 0.0001) and adenocarcinoma (p &lt; 0.0001). Multivariate analysis showed that never-smoking history (p = 0.005), adenocarcinoma (p = 0.004), and better PS (P &lt; 0.001) were significant independent predictors of long survival. The Brinkman index (BI), number of cigarettes per day multiplied by number of years smoked, was associated with survival benefit. MST was 17.6 months in the BI = 0 group (never-smokers), 9.7 months in 1 ≤BI &lt;500, 6.9 months in 500 ≤BI &lt;1000, 4.4 months in 1000 ≤BI &lt;1500, and 4.0 months in BI ≥1500. Conclusion: Never-smokers and adenocarcinoma are predictive factors for prolonged survival. As for smoking status, lower BI is associated with better prognosis after treatment with gefitinib. Therefore, histological factors and smoking status must be considered when treating with gefitinib. No significant financial relationships to disclose.

Retrospective analysis of the predictive factors associated with the response and survival benefit of gefitinib in patients with advanced non-small-cell lung cancer

Background: The purpose of the study was to identify the potential predictive features associated with the response and survival benefit of gefitinib administration. We have retrospectively reviewed data of all patients who received a single regimen of gefitinib in our institution from August 1998 until July 2003. Methods: Overall 101 patients with non-small-cell lung cancer (NSCLC) who have received a single use of gefitinib were analyzed. Potential factors associated with the response of gefitinib included smoking index, gender, histology, performance status (PS), number of pre-treatments, age and stage. Univariate analysis was performed for these strata by Fisher’s exact test and multivariate analysis was then performed using the logistic regression model. Results: The overall response rate was 19.8%. Univariate analysis revealed that significant predictive factors were associated with the response for ‘adenocarcinoma’, ‘female’, ‘good PS’ (0–l) and ‘non-smoker’ categories. Multivariate analysis limited the predictive factors associated with the response for ‘female’ ( P = 0.0032), ‘good PS’ ( P < 0.02) and ‘non-smoker’ ( P = 0.0417). In survival analyses, ‘female’ ( P < 0.005), ‘good PS’ ( P < 0.0001), and a low level of the smoking index ( P < 0.05) indicated significantly prolonged survival. Response and survival data in elderly patients were equivalent to those in younger patients. Adverse events (AEs) were generally mild and were almost always skin reactions and diarrhea. Interstitial lung disease (ILD) occurred in 4% of the group under observation. Conclusions: Gefitinib provided clinical benefit for the following factors ‘female’, ‘good PS’ and ‘non-smoker’. A low smoking index is reported as a novel predictive prognostic factor following a single regimen of gefitinib.

Bronchioloalveolar pathologic subtype and smoking history predict sensitivity to gefitinib in advanced non-small-cell lung cancer.

Gefitinib, an inhibitor of the epidermal growth factor receptor tyrosine kinase, induces radiographic regressions and symptomatic improvement in patients with non-small-cell lung cancer (NSCLC). Phase II trials suggested female sex and adenocarcinoma were associated with response. We undertook this analysis to identify additional clinical and pathologic features associated with sensitivity to gefitinib. We reviewed medical records, pathologic material, and imaging studies of all 139 NSCLC patients treated on one of three consecutive studies of gefitinib monotherapy performed at our institution. We identified patients experiencing a major objective response and compared their clinical and pathologic features with the others. Univariate and multivariable analyses were performed on potential predictive features associated with sensitivity to gefitinib. Of 139 patients, 21 (15%; 95% CI, 9% to 21%), experienced a partial radiographic response. Variables identified as significant in univariate analysis included adenocarcinoma versus other NSCLC (19% v 0%; P=.004), adenocarcinoma with bronchioloalveolar features versus other adenocarcinomas (38% v 14%; P<.001), never smoker status versus former/current (36% v 8%; P<.001), and Karnofsky performance status > or =80% versus < or =70% (22% v 8%; P=.03). Multivariable analysis revealed the presence of adenocarcinoma with any bronchioloalveolar features (P=.004) and being a never smoker (P=.006) were independent predictors of response. Our data suggest that individuals in whom gefitinib is efficacious are more likely to have adenocarcinomas of the bronchioloalveolar subtype and to be never smokers. These observations may provide clues to mechanisms determining sensitivity to this agent and suggest that NSCLC has a different biology in patients who never smoked and those with bronchioloalveolar carcinoma.