Persistence of immune response to an adjuvanted varicella-zoster virus subunit vaccine for up to year nine in older adults

Background: In adults aged ≥60 years, two doses of the herpes zoster subunit vaccine (HZ/su; 50 µg varicella-zoster virus glycoprotein E [gE] and AS01_B Adjuvant System) elicited humoral and cell-mediated immune responses persisting for at least six years. We assessed immunogenicity nine years post-initial vaccination.

Methods: This open extension study (NCT02735915) followed 70 participants who received two HZ/su doses in the initial trial (NCT00434577). Blood samples to assess the cellular (intracellular cytokine staining) and humoral (ELISA) immunity were taken at year nine post-initial vaccination.

Results: Participants' mean age at dose 1 was 72.3 years. The fold increases over pre-vaccination in the mean frequency of gE-specific CD4+ T-cells expressing ≥2 activation markers plateaued from year four post-dose 1 until year nine. Anti-gE antibody geometric mean concentrations plateaued and remained above pre-vaccination levels from year four onwards. Immunogenicity at year nine was similar across age strata (60–69, ≥70 years) and confirmed statistical prediction model results using data for up to year six. Further modeling using all data up to year nine predicted immune responses would remain above the pre-vaccination level up to year 15.

Conclusion: In adults aged ≥60 years, HZ/su-induced immunogenicity remained above pre-vaccination levels for at least nine years post-initial vaccination.

Summary: After vaccination with HZ/su, both cell mediated and humoral immunity remained above pre-vaccination levels up to year 9 regardless of age group. Immune responses are predicted to remain above baseline up to 15 years post initial vaccination.

Single-Subject Studies in Translational Nutrition Research

There is a great deal of interest in personalized, individualized, or precision interventions for disease and health-risk mitigation. This is as true of nutrition-based intervention and prevention strategies as it is for pharmacotherapies and pharmaceutical-oriented prevention strategies. Essentially, technological breakthroughs have enabled researchers to probe an individual's unique genetic, biochemical, physiological, behavioral, and exposure profile, allowing them to identify very specific and often nuanced factors that an individual might possess, which may make it more or less likely that he or she responds favorably to a particular intervention (e.g., nutrient supplementation) or disease prevention strategy (e.g., specific diet). However, as compelling and intuitive as personalized nutrition might be in the current era in which data-intensive biomedical characterization of individuals is possible, appropriately and objectively vetting personalized nutrition strategies is not trivial and requires novel study designs and data analytical methods. These designs and methods must consider a very integrated use of the multiple contemporary biomedical assays and technologies that motivate them, which adds to their complexity. Single-subject or N-of-1 trials can be used to assess the utility of personalized interventions and, in addition, can be crafted in such a way as to accommodate the necessarily integrated use of many emerging biomedical technologies and assays. In this review, we consider the motivation, design, and implementation of N-of-1 trials in translational nutrition research that are meant to assess the utility of personalized nutritional strategies. We provide a number of example studies, discuss appropriate analytical methods given the complex data they generate and require, and consider how such studies could leverage integration of various biomarker assays and clinical end points. Importantly, we also consider the development of strategies and algorithms for matching nutritional needs to individual biomedical profiles and the issues surrounding them. Finally, we discuss the limitations of personalized nutrition studies, possible extensions of N-of-1 nutritional intervention studies, and areas of future research.

Early prediction of spontaneous Patent Ductus Arteriosus (PDA) closure and PDA-associated outcomes: a prospective cohort investigation

BACKGROUND

Patent ductus arteriosus (PDA), the most commonly diagnosed cardiovascular condition in preterm infants, is associated with increased mortality and harmful long-term outcomes (chronic lung disease, neurodevelopmental delay). Although pharmacologic and/or interventional treatments to close PDA likely benefit some infants, widespread routine treatment of all preterm infants with PDA may not improve outcomes. Most PDAs close spontaneously by 44-weeks postmenstrual age; treatment is increasingly controversial, varying markedly between institutions and providers. Because treatment detriments may outweigh benefits, especially in infants destined for early, spontaneous PDA closure, the relevant unanswered clinical question is not whether to treat all preterm infants with PDA, but whom to treat (and when). Clinicians cannot currently predict in the first month which infants are at highest risk for persistent PDA, nor which combination of clinical risk factors, echocardiographic measurements, and biomarkers best predict PDA-associated harm.

METHODS

Prospective cohort of untreated infants with PDA (n=450) will be used to predict spontaneous ductal closure timing. Clinical measures, serum (brain natriuretic peptide, N-terminal pro-brain natriuretic peptide) and urine (neutrophil gelatinase-associated lipocalin, heart-type fatty acid-binding protein) biomarkers, and echocardiographic variables collected during each of first 4 postnatal weeks will be analyzed to identify those associated with long-term impairment. Myocardial deformation imaging and tissue Doppler imaging, innovative echocardiographic techniques, will facilitate quantitative evaluation of myocardial performance. Aim1 will estimate probability of spontaneous PDA closure and predict timing of ductal closure using echocardiographic, biomarker, and clinical predictors. Aim2 will specify which echocardiographic predictors and biomarkers are associated with mortality and respiratory illness severity at 36-weeks postmenstrual age. Aim3 will identify which echocardiographic predictors and biomarkers are associated with 22 to 26-month neurodevelopmental delay. Models will be validated in a separate cohort of infants (n=225) enrolled subsequent to primary study cohort.

DISCUSSION

The current study will make significant contributions to scientific knowledge and effective PDA management. Study results will reduce unnecessary and harmful overtreatment of infants with a high probability of early spontaneous PDA closure and facilitate development of outcomes-focused trials to examine effectiveness of PDA closure in "high-risk" infants most likely to receive benefit.

TRIAL REGISTRATION

ClinicalTrials.gov NCT03782610. Registered 20 December 2018.

Risk prediction scores for acute on chronic liver failure development and mortality

BACKGROUND & AIMS

Acute on chronic liver failure (ACLF) causes high short-term mortality in patients with previously stable chronic liver disease. To date there are no models to predict which patients are likely to develop ACLF, and existing models to predict ACLF mortality are based on limited cohorts. We sought to create novel risk prediction scores using a large cohort of patients with cirrhosis.

METHODS

We performed a retrospective cohort study of 74 790 patients with incident cirrhosis in the Veterans Health Administration database using randomized 70% derivation/30% validation sets. ACLF events were identified per the European ACLF criteria. Multivariable logistic regression was used to derive prediction models for developing ACLF at 3, 6 and 12 months, and ACLF mortality at 28 and 90 days. Mortality models were compared to model for end-stage liver disease (MELD), MELD-sodium and the Chronic Liver Failure Consortium (CLIF-C) ACLF score.

RESULTS

Models for the developing ACLF had very good discrimination (concordance [C] statistics 0.83-0.87) at all timepoints. Models for ACLF mortality also had good discrimination at 28 and 90 days (C-statistics 0.79-0.82), and were superior to MELD, MELD-sodium and the CLIF-C ACLF score. The calibration of the novel models was excellent at all timepoints.

CONCLUSION

We have obtained highly-predictive models for developing ACLF, as well as for ACLF short-term mortality in a diverse United States cohort. These may be used to identify outpatients at significant risk of ACLF, which may prompt closer follow-up or early transplant referral, and facilitate decision making for patients with diagnosed ACLF, including escalation of care, expedited transplant evaluation or palliation.

Predicting Poor Outcome Before Endovascular Treatment in Patients With Acute Ischemic Stroke

Background: Although endovascular treatment (EVT) has greatly improved outcomes in acute ischemic stroke, still one third of patients die or remain severely disabled after stroke. If we could select patients with poor clinical outcome despite EVT, we could prevent futile treatment, avoid treatment complications, and further improve stroke care. We aimed to determine the accuracy of poor functional outcome prediction, defined as 90-day modified Rankin Scale (mRS) score ≥5, despite EVT treatment. Methods: We included 1,526 patients from the MR CLEAN Registry, a prospective, observational, multicenter registry of ischemic stroke patients treated with EVT. We developed machine learning prediction models using all variables available at baseline before treatment. We optimized the models for both maximizing the area under the curve (AUC), reducing the number of false positives. Results: From 1,526 patients included, 480 (31%) of patients showed poor outcome. The highest AUC was 0.81 for random forest. The highest area under the precision recall curve was 0.69 for the support vector machine. The highest achieved specificity was 95% with a sensitivity of 34% for neural networks, indicating that all models contained false positives in their predictions. From 921 mRS 0-4 patients, 27-61 (3-6%) were incorrectly classified as poor outcome. From 480 poor outcome patients in the registry, 99-163 (21-34%) were correctly identified by the models. Conclusions: All prediction models showed a high AUC. The best-performing models correctly identified 34% of the poor outcome patients at a cost of misclassifying 4% of non-poor outcome patients. Further studies are necessary to determine whether these accuracies are reproducible before implementation in clinical practice.

Classifying Patients with Amyotrophic Lateral Sclerosis by Changes in FVC. A Group-based Trajectory Analysis

Rationale: A model for stratifying progression of respiratory muscle weakness in amyotrophic lateral sclerosis (ALS) would identify disease mechanisms and phenotypes suitable for future investigations. This study sought to categorize progression of FVC after presentation to an outpatient ALS clinic.Objectives: To identify clinical phenotypes of ALS respiratory progression based on FVC trajectories over time.Methods: We derived a group-based trajectory model from a single-center cohort of 837 patients with ALS who presented between 2006 and 2015. We applied our model to the Pooled Resource Open-Access ALS Clinical Trials (PRO-ACT) database with 7,461 patients with ALS. Baseline characteristics at first visit were used as predictors of trajectory group membership. The primary outcome was trajectory of FVC over time in months.Measurements and Main Results: We found three trajectories of FVC over time, termed "stable low," "rapid progressor," and "slow progressor." Compared with the slow progressors, the rapid progressors had shorter diagnosis delay, more bulbar-onset disease, and a lower ALS Functional Rating Scale-Revised (ALSFRS-R) total score at baseline. The stable low group had a shorter diagnosis delay, lower body mass index, more bulbar-onset disease, lower ALSFRS-R total score, and were more likely to have an ALSFRS-R orthopnea score lower than 4 compared with the slow progressors. We found that projected group membership predicted respiratory insufficiency in the PRO-ACT cohort (concordance statistic = 0.78, 95% CI, 0.76-0.79).Conclusions: We derived a group-based trajectory model for FVC progression in ALS, which validated against the outcome of respiratory insufficiency in an external cohort. Future studies may focus on patients predicted to be rapid progressors.

External validation of postnatal gestational age estimation using newborn metabolic profiles in Matlab, Bangladesh

This study sought to evaluate the performance of metabolic gestational age estimation models developed in Ontario, Canada in infants born in Bangladesh. Cord and heel prick blood spots were collected in Bangladesh and analyzed at a newborn screening facility in Ottawa, Canada. Algorithm-derived estimates of gestational age and preterm birth were compared to ultrasound-validated estimates. 1036 cord blood and 487 heel prick samples were collected from 1069 unique newborns. The majority of samples (93.2% of heel prick and 89.9% of cord blood) were collected from term infants. When applied to heel prick data, algorithms correctly estimated gestational age to within an average deviation of 1 week overall (root mean square error = 1.07 weeks). Metabolic gestational age estimation provides accurate population-level estimates of gestational age in this data set. Models were effective on data obtained from both heel prick and cord blood, the latter being a more feasible option in low-resource settings.

Algal Bloom Prediction Using Extreme Learning Machine Models at Artificial Weirs in the Nakdong River, Korea

In this study, we design an intelligent model to predict chlorophyll-a concentration, which is the primary indicator of algal blooms, using extreme learning machine (ELM) models. Modeling algal blooms is important for environmental management and ecological risk assessment. For this purpose, the performance of the designed models was evaluated for four artificial weirs in the Nakdong River, Korea. The Nakdong River has harmful annual algal blooms that can affect health due to exposure to toxins. In contrast to conventional neural network (NN) that use backpropagation (BP) learning methods, ELMs are fast learning, feedforward neural networks that use least square estimates (LSE) for regression. The weights connecting the input layer to the hidden nodes are randomly assigned and are never updated. The dataset used in this study includes air temperature, rainfall, solar radiation, total nitrogen, total phosphorus, N/P ratio, and chlorophyll-a concentration, which were collected on a weekly basis from January 2013 to December 2016. Here, upstream chlorophyll-a concentration data was used in our ELM2 model to improve algal bloom prediction performance. In contrast, the ELM1 model only uses downstream chlorophyll-a concentration data. The experimental results revealed that the ELM2 model showed better performance in comparison to the ELM1 model. Furthermore, the ELM2 model showed good prediction and generalization performance compared to multiple linear regression (LR), conventional neural network with backpropagation (NN-BP), and adaptive neuro-fuzzy inference system (ANFIS).

Predicting High Imaging Utilization Based on Initial Radiology Reports: A Feasibility Study of Machine Learning

RATIONALE AND OBJECTIVES

Imaging utilization has significantly increased over the last two decades, and is only recently showing signs of moderating. To help healthcare providers identify patients at risk for high imaging utilization, we developed a prediction model to recognize high imaging utilizers based on their initial imaging reports.

MATERIALS AND METHODS

The prediction model uses a machine learning text classification framework. In this study, we used radiology reports from 18,384 patients with at least one abdomen computed tomography study in their imaging record at Stanford Health Care as the training set. We modeled the radiology reports in a vector space and trained a support vector machine classifier for this prediction task. We evaluated our model on a separate test set of 4791 patients. In addition to high prediction accuracy, in our method, we aimed at achieving high specificity to identify patients at high risk for high imaging utilization.

RESULTS

Our results (accuracy: 94.0%, sensitivity: 74.4%, specificity: 97.9%, positive predictive value: 87.3%, negative predictive value: 95.1%) show that a prediction model can enable healthcare providers to identify in advance patients who are likely to be high utilizers of imaging services.

CONCLUSIONS

Machine learning classifiers developed from narrative radiology reports are feasible methods to predict imaging utilization. Such systems can be used to identify high utilizers, inform future image ordering behavior, and encourage judicious use of imaging.

Weak grip strength does not predict upper extremity musculoskeletal symptoms or injuries among new workers

PURPOSE

Grip strength is often tested during post-offer pre-placement screening for workers in hand-intensive jobs. The purpose of this study was to evaluate the association between grip strength and upper extremity symptoms, work disability, and upper extremity musculoskeletal disorders (UE MSDs) in a group of workers newly employed in both high and low hand intensive work.

METHODS

1,107 recently-hired workers completed physical examinations including grip strength measurements. Repeated surveys obtained over 3 years described the presence of upper extremity symptoms, report of physician-diagnosed musculoskeletal disorders (MSDs), and job titles. Baseline measured grip values were used in analytic models as continuous and categorized values to predict upper extremity symptoms, work disability, or UE MSD diagnosis.

RESULTS

Twenty-six percent of males and 20 % of females had low baseline hand strength compared to normative data. Multivariate logistic regression analyses showed no consistent associations between grip strength and three health outcomes (UE symptoms, work disability, and MSDs) in this young cohort (mean age 30 years). Past MSD and work type were significant predictors of these outcomes.

CONCLUSIONS

Physical hand strength testing was not useful for identifying workers at risk for developing UE MSDs, and may be an inappropriate measure for post-offer job screens.

Development and validation of a 5-year mortality prediction model using regularized regression and Medicare data

PURPOSE

De-implementation of low-value services among patients with limited life expectancy is challenging. Robust mortality prediction models using routinely collected health care data can enhance health care stakeholders' ability to identify populations with limited life expectancy. We developed and validated a claims-based prediction model for 5-year mortality using regularized regression methods.

METHODS

Medicare beneficiaries age 66 or older with an office visit and at least 12 months of pre-visit continuous Medicare A/B enrollment were identified in 2008. Five-year mortality was assessed through 2013. Secondary outcomes included 30-, 90-, and 180-day and 1-year mortality. Claims-based predictors, including comorbidities and indicators of disability, frailty, and functional impairment, were selected using regularized logistic regression, applying the least absolute shrinkage and selection operator (LASSO) in a random 80% training sample. Model performance was assessed and compared with the Gagne comorbidity score in the 20% validation sample.

RESULTS

Overall, 183 204 (24%) individuals died. In addition to demographics, 161 indicators of comorbidity and function were included in the final model. In the validation sample, the c-statistic was 0.825 (0.823-0.828). Median-predicted probability of 5-year mortality was 14%; almost 4% of the cohort had a predicted probability greater than 80%. Compared with the Gagne score, the LASSO model led to improved 5-year mortality classification (net reclassification index = 9.9%; integrated discrimination index = 5.2%).

CONCLUSIONS

Our claims-based model predicting 5-year mortality showed excellent discrimination and calibration, similar to the Gagne score model, but resulted in improved mortality classification. Regularized regression is a feasible approach for developing prediction tools that could enhance health care research and evaluation of care quality.

Machine Learning for Outcome Prediction in First-Line Surgery of Prolactinomas

BACKGROUND

First-line surgery for prolactinomas has gained increasing acceptance, but the indication still remains controversial. Thus, accurate prediction of unfavorable outcomes after upfront surgery in prolactinoma patients is critical for the triage of therapy and for interdisciplinary decision-making.

OBJECTIVE

To evaluate whether contemporary machine learning (ML) methods can facilitate this crucial prediction task in a large cohort of prolactinoma patients with first-line surgery, we investigated the performance of various classes of supervised classification algorithms. The primary endpoint was ML-applied risk prediction of long-term dopamine agonist (DA) dependency. The secondary outcome was the prediction of the early and long-term control of hyperprolactinemia.

METHODS

By jointly examining two independent performance metrics - the area under the receiver operating characteristic (AUROC) and the Matthews correlation coefficient (MCC) - in combination with a stacked super learner, we present a novel perspective on how to assess and compare the discrimination capacity of a set of binary classifiers.

RESULTS

We demonstrate that for upfront surgery in prolactinoma patients there are not a one-algorithm-fits-all solution in outcome prediction: different algorithms perform best for different time points and different outcomes parameters. In addition, ML classifiers outperform logistic regression in both performance metrics in our cohort when predicting the primary outcome at long-term follow-up and secondary outcome at early follow-up, thus provide an added benefit in risk prediction modeling. In such a setting, the stacking framework of combining the predictions of individual base learners in a so-called super learner offers great potential: the super learner exhibits very good prediction skill for the primary outcome (AUROC: mean 0.9, 95% CI: 0.92 - 1.00; MCC: 0.85, 95% CI: 0.60 - 1.00). In contrast, predicting control of hyperprolactinemia is challenging, in particular in terms of early follow-up (AUROC: 0.69, 95% CI: 0.50 - 0.83) vs. long-term follow-up (AUROC: 0.80, 95% CI: 0.58 - 0.97). It is of clinical importance that baseline prolactin levels are by far the most important outcome predictor at early follow-up, whereas remissions at 30 days dominate the ML prediction skill for DA-dependency over the long-term.

CONCLUSIONS

This study highlights the performance benefits of combining a diverse set of classification algorithms to predict the outcome of first-line surgery in prolactinoma patients. We demonstrate the added benefit of considering two performance metrics jointly to assess the discrimination capacity of a diverse set of classifiers.

Corticospinal Tract Microstructure Predicts Distal Arm Motor Improvements in Chronic Stroke

BACKGROUND AND PURPOSE

The corticospinal tract (CST) is a crucial brain pathway for distal arm and hand motor control. We aimed to determine whether a diffusion tensor imaging (DTI)-derived CST metric predicts distal upper extremity (UE) motor improvements in chronic stroke survivors.

METHODS

We analyzed clinical and neuroimaging data from a randomized controlled rehabilitation trial. Participants completed clinical assessments and neuroimaging at baseline and clinical assessments 4 months later, postintervention. Using univariate linear regression analysis, we determined the linear relationship between the DTI-derived CST fractional anisotropy asymmetry (FAasym) and the percentage of baseline change in log-transformed average Wolf Motor Function Test time for distal items (ΔlnWMFT-distal_%). The least absolute shrinkage and selection operator (LASSO) linear regressions with cross-validation and bootstrapping were used to determine the relative weighting of CST FAasym, other brain metrics, clinical outcomes, and demographics on distal motor improvement. Logistic regression analyses were performed to test whether the CST FAasym can predict clinically significant UE motor improvement.

RESULTS

lnWMFT-distal significantly improved at the group level. Baseline CST FAasym explained 26% of the variance in ΔlnWMFT-distal_%. A multivariate LASSO model including baseline CST FAasym, age, and UE Fugl-Meyer explained 39% of the variance in ΔlnWMFT-distal_%. Further, CST FAasym explained more variance in ΔlnWMFT-distal_% than the other significant predictors in the LASSO model.

DISCUSSION AND CONCLUSIONS

CST microstructure is a significant predictor of improvement in distal UE motor function in the context of an UE rehabilitation trial in chronic stroke survivors with mild-to-moderate motor impairment.Video Abstract available for more insight from the authors (see the Video, Supplemental Digital Content 1, available at: http://links.lww.com/JNPT/A350).

A Machine Learning-Based Model to Predict Acute Traumatic Coagulopathy in Trauma Patients Upon Emergency Hospitalization

Acute traumatic coagulopathy (ATC) is an extremely common but silent murderer; this condition presents early after trauma and impacts approximately 30% of severely injured patients who are admitted to emergency departments (EDs). Given that conventional coagulation indicators usually require more than 1 hour after admission to yield results—a limitation that frequently prevents the ability for clinicians to make appropriate interventions during the optimal therapeutic window—it is clearly of vital importance to develop prediction models that can rapidly identify ATC; such models would also facilitate ancillary resource management and clinical decision support. Using the critical care Emergency Rescue Database and further collected data in ED, a total of 1385 patients were analyzed and cases with initial international normalized ratio (INR) values >1.5 upon admission to the ED met the defined diagnostic criteria for ATC; nontraumatic conditions with potentially disordered coagulation systems were excluded. A total of 818 individuals were collected from Emergency Rescue Database as derivation cohorts, then were split 7:3 into training and test data sets. A Pearson correlation matrix was used to initially identify likely key clinical features associated with ATC, and analysis of data distributions was undertaken prior to the selection of suitable modeling tools. Both machine learning (random forest) and traditional logistic regression were deployed for prediction modeling of ATC. After the model was built, another 587 patients were further collected in ED as validation cohorts. The ATC prediction models incorporated red blood cell count, Shock Index, base excess, lactate, diastolic blood pressure, and potential of hydrogen. Of 818 trauma patients filtered from the database, 747 (91.3%) patients did not present ATC (INR ≤ 1.5) and 71 (8.7%) patients had ATC (INR > 1.5) upon admission to the ED. Compared to the logistic regression model, the model based on the random forest algorithm showed better accuracy (94.0%, 95% confidence interval [CI]: 0.922-0.954 to 93.5%, 95% CI: 0.916-0.95), precision (93.3%, 95% CI: 0.914-0.948 to 93.1%, 95% CI: 0.912-0.946), F1 score (93.4%, 95% CI: 0.915-0.949 to 92%, 95% CI: 0.9-0.937), and recall score (94.0%, 95% CI: 0.922-0.954 to 93.5%, 95% CI: 0.916-0.95) but yielded lower area under the receiver operating characteristic curve (AU-ROC) (0.810, 95% CI: 0.673-0.918 to 0.849, 95% CI: 0.732-0.944) for predicting ATC in the trauma patients. The result is similar in the validation cohort. The values for classification accuracy, precision, F1 score, and recall score of random forest model were 0.916, 0.907, 0.901, and 0.917, while the AU-ROC was 0.830. The values for classification accuracy, precision, F1 score, and recall score of logistic regression model were 0.905, 0.887, 0.883, and 0.905, while the AU-ROC was 0.858. We developed and validated a prediction model based on objective and rapidly accessible clinical data that very confidently identify trauma patients at risk for ATC upon their arrival to the ED. Beyond highlighting the value of ED initial laboratory tests and vital signs when used in combination with data analysis and modeling, our study illustrates a practical method that should greatly facilitates both warning and guided target intervention for ATC.

Proteomics of Primary Uveal Melanoma: Insights into Metastasis and Protein Biomarkers

Uveal melanoma metastases are lethal and remain incurable. A quantitative proteomic analysis of 53 metastasizing and 47 non-metastasizing primary uveal melanoma (pUM) was pursued for insights into UM metastasis and protein biomarkers. The metastatic status of the pUM specimens was defined based on clinical data, survival histories, prognostic analyses, and liver histopathology. LC MS/MS iTRAQ technology, the Mascot search engine, and the UniProt human database were used to identify and quantify pUM proteins relative to the normal choroid excised from UM donor eyes. The determined proteomes of all 100 tumors were very similar, encompassing a total of 3935 pUM proteins. Proteins differentially expressed (DE) between metastasizing and non-metastasizing pUM (n = 402) were employed in bioinformatic analyses that predicted significant differences in the immune system between metastasizing and non-metastasizing pUM. The immune proteins (n = 778) identified in this study support the immune-suppressive nature and low abundance of immune checkpoint regulators in pUM, and suggest CDH1, HLA-DPA1, and several DE immune kinases and phosphatases as possible candidates for immune therapy checkpoint blockade. Prediction modeling identified 32 proteins capable of predicting metastasizing versus non-metastasizing pUM with 93% discriminatory accuracy, supporting the potential for protein-based prognostic methods for detecting UM metastasis.

A Machine Learning Model for Predicting Mortality within 90 Days of Dialysis Initiation

Background

The first 90 days after dialysis initiation are associated with high morbidity and mortality in end-stage kidney disease (ESKD) patients. A machine learning-based tool for predicting mortality could inform patient-clinician shared decision making on whether to initiate dialysis or pursue medical management. We used the eXtreme Gradient Boosting (XGBoost) algorithm to predict mortality in the first 90 days after dialysis initiation in a nationally representative population from the United States Renal Data System.

Methods

A cohort of adults initiating dialysis between 2008-2017 were studied for outcome of death within 90 days of dialysis initiation. The study dataset included 188 candidate predictors prognostic of early mortality that were known on or before the first day of dialysis and was partitioned into training (70%) and testing (30%) subsets. XGBoost modeling used a complete-case set and a dataset obtained from multiple imputation. Model performance was evaluated by c-statistics overall and stratified by subgroups of age, sex, race, and dialysis modality.

Results

The analysis included 1,150,195 patients with ESKD, of whom 86,083 (8%) died in the first 90 days after dialysis initiation. The XGBoost models discriminated mortality risk in the nonimputed (c=0.826, 95% CI, 0.823 to 0.828) and imputed (c=0.827, 95% CI, 0.823 to 0.827) models and performed well across nearly every subgroup (race, age, sex, and dialysis modality) evaluated (c>0.75). Across predicted risk thresholds of 10%-50%, higher risk thresholds showed declining sensitivity (0.69-0.04) with improving specificity (0.79-0.99); similarly, positive likelihood ratio was highest at the 40% threshold, whereas the negative likelihood ratio was lowest at the 10% threshold. After calibration using isotonic regression, the model accurately estimated the probability of mortality across all ranges of predicted risk.

Conclusions

The XGBoost-based model developed in this study discriminated risk of early mortality after dialysis initiation with excellent calibration and performed well across key subgroups.

Non-CYP2D6 Variants Selected by a GWAS Improve the Prediction of Impaired Tamoxifen Metabolism in Patients with Breast Cancer

A certain minimum plasma concentration of (Z)-endoxifen is presumably required for breast cancer patients to benefit from tamoxifen therapy. In this study, we searched for DNA variants that could aid in the prediction of risk for insufficient (Z)-endoxifen exposure. A metabolic ratio (MR) corresponding to the (Z)-endoxifen efficacy threshold level was adopted as a cutoff value for a genome-wide association study comprised of 287 breast cancer patients. Multivariate regression was used to preselect variables exhibiting an independent impact on the MR and develop models to predict below-threshold MR values. In total, 15 single-nucleotide polymorphisms (SNPs) were significantly associated with below-threshold MR values. The strongest association was with rs8138080 (WBP2NL). Two alternative models for MR prediction were developed. The predictive accuracy of Model 1, including rs7245, rs6950784, rs1320308, and the CYP2D6 genotype, was considerably higher than that of the CYP2D6 genotype alone (AUC 0.879 vs 0.758). Model 2, which was developed using the same three SNPs as for Model 1 plus rs8138080, appeared as an interesting alternative to the full CYP2D6 genotype testing. In conclusion, the four novel SNPs, tested alone or in combination with the CYP2D6 genotype, improved the prediction of impaired tamoxifen-to-endoxifen metabolism, potentially allowing for treatment optimization.

Modeling residential indoor concentrations of PM2.5 , NO2 , NOx , and secondhand smoke in the Subpopulations and Intermediate Outcome Measures in COPD (SPIROMICS) Air study

Increased outdoor concentrations of fine particulate matter (PM_2.5 ) and oxides of nitrogen (NO₂ , NO_x ) are associated with respiratory and cardiovascular morbidity in adults and children. However, people spend most of their time indoors and this is particularly true for individuals with chronic obstructive pulmonary disease (COPD). Both outdoor and indoor air pollution may accelerate lung function loss in individuals with COPD, but it is not feasible to measure indoor pollutant concentrations in all participants in large cohort studies. We aimed to understand indoor exposures in a cohort of adults (SPIROMICS Air, the SubPopulations and Intermediate Outcome Measures in COPD Study of Air pollution). We developed models for the entire cohort based on monitoring in a subset of homes, to predict mean 2-week-measured concentrations of PM_2.5 , NO₂ , NO_x , and nicotine, using home and behavioral questionnaire responses available in the full cohort. Models incorporating socioeconomic, meteorological, behavioral, and residential information together explained about 60% of the variation in indoor concentration of each pollutant. Cross-validated R² for best indoor prediction models ranged from 0.43 (NO_x ) to 0.51 (NO₂ ). Models based on questionnaire responses and estimated outdoor concentrations successfully explained most variation in indoor PM_2.5 , NO₂ , NO_x , and nicotine concentrations.

Landscape of Human Immunodeficiency Virus Neutralization Susceptibilities Across Tissue Reservoirs

BACKGROUND

Human immunodeficiency virus type 1 (HIV-1) sequence diversity and the presence of archived epitope muta-tions in antibody binding sites are a major obstacle for the clinical application of broadly neutralizing antibodies (bNAbs) against HIV-1. Specifically, it is unclear to what degree the viral reservoir is compartmentalized and if virus susceptibility to antibody neutralization differs across tissues.

METHODS

The Last Gift cohort enrolled 7 people with HIV diagnosed with a terminal illness and collected antemortem blood and postmortem tissues across 33 anatomical compartments for near full-length env HIV genome sequencing. Using these data, we applied a Bayesian machine-learning model (Markov chain Monte Carlo-support vector machine) that uses HIV-1 envelope sequences and approximated glycan-occupancy information to quantitatively predict the half-maximal inhib-itory concentrations (IC50) of bNAbs, allowing us to map neutralization resistance pattern across tissue reservoirs.

RESULTS

Predicted mean susceptibilities across tissues within participants were relatively homogenous, and the susceptibility pattern observed in blood often matched what was predicted for tissues. However, selected tissues, such as the brain, showed ev-idence of compartmentalized viral populations with distinct neutralization susceptibilities in some participants. Additionally, we found substantial heterogeneity in the range of neutralization susceptibilities across tissues within and between indi-viduals, and between bNAbs within individuals (standard deviation of log2(IC50) >3.4).

CONCLUSIONS

Blood-based screening methods to determine viral susceptibility to bNAbs might underestimate the presence of resistant viral variants in tissues. The extent to which these resistant viruses are clinically relevant, that is, lead to bNAb therapeutic failure, needs to be further explored.

Optimizing an existing prediction model for quality of life one-year post-intensive care unit: An exploratory analysis

BACKGROUND

This study aimed to improve the PREPARE model, an existing linear regression prediction model for long-term quality of life (QoL) of intensive care unit (ICU) survivors by incorporating additional ICU data from patients' electronic health record (EHR) and bedside monitors.

METHODS

The 1308 adult ICU patients, aged ≥16, admitted between July 2016 and January 2019 were included. Several regression-based machine learning models were fitted on a combination of patient-reported data and expert-selected EHR variables and bedside monitor data to predict change in QoL 1 year after ICU admission. Predictive performance was compared to a five-feature linear regression prediction model using only 24-hour data (R²  = 0.54, mean square error (MSE) = 0.031, mean absolute error (MAE) = 0.128).

RESULTS

The 67.9% of the included ICU survivors was male and the median age was 65.0 [IQR: 57.0-71.0]. Median length of stay (LOS) was 1 day [IQR 1.0-2.0]. The incorporation of the additional data pertaining to the entire ICU stay did not improve the predictive performance of the original linear regression model. The best performing machine learning model used seven features (R²  = 0.52, MSE = 0.032, MAE = 0.125). Pre-ICU QoL, the presence of a cerebro vascular accident (CVA) upon admission and the highest temperature measured during the ICU stay were the most important contributors to predictive performance. Pre-ICU QoL's contribution to predictive performance far exceeded that of the other predictors.

CONCLUSION

Pre-ICU QoL was by far the most important predictor for change in QoL 1 year after ICU admission. The incorporation of the numerous additional features pertaining to the entire ICU stay did not improve predictive performance although the patients' LOS was relatively short.

MiSurv: an Integrative Web Cloud Platform for User-Friendly Microbiome Data Analysis with Survival Responses

Investigators have studied the treatment effects on human health or disease, the treatment effects on human microbiome, and the roles of the microbiome on human health or disease. Especially, in a clinical trial, investigators commonly trace disease status over a lengthy period to survey the sequential disease progression for different treatment groups (e.g., treatment versus placebo, new treatment versus old treatment). Hence, disease responses are often available in the form of survival (i.e., time-to-event) responses stratified by treatment groups. While the recent web cloud platforms have enabled user-friendly microbiome data processing and analytics, there is currently no web cloud platform to analyze microbiome data with survival responses. Therefore, we introduce here an integrative web cloud platform, called MiSurv, for comprehensive microbiome data analysis with survival responses. IMPORTANCE MiSurv consists of a data processing module and its following four data analytic modules: (i) Module 1: Comparative survival analysis between treatment groups, (ii) Module 2: Comparative analysis in microbial composition between treatment groups, (iii) Module 3: Association testing between microbial composition and survival responses, (iv) Module 4: Prediction modeling using microbial taxa on survival responses. We demonstrate its use through an example trial on the effects of antibiotic use on the survival rate against type 1 diabetes (T1D) onset and gut microbiome composition, respectively, and the effects of the gut microbiome on the survival rate against T1D onset. MiSurv is freely available on our web server (http://misurv.micloud.kr) or can alternatively run on the user's local computer (https://github.com/wg99526/MiSurvGit).

Development and Internal Validation of a Model Predicting Premature Cardiovascular Disease Among Women With Hypertensive Disorders of Pregnancy: A Population-Based Study in Quebec, Canada

Background

Hypertensive disorders of pregnancy (HDP) are associated with an increased risk of premature cardiovascular disease (CVD), but existing cardiovascular prediction models do not adequately capture risks in young women. We developed a model to predict the 10‐year risk of premature CVD and mortality among women who have HDP.

Methods and Results

Using a population‐based cohort of women with HDP who delivered between April 1989 and March 2017 in Quebec, Canada, we developed a 10‐year CVD risk model using Cox proportional hazards regression. Women aged 18 to 45 years were followed from their first HDP‐complicated delivery until March 2018. We assessed performance of the model based on discrimination, calibration, and risk stratification ability. Internal validity was assessed using the bootstrap method. The cohort included 95 537 women who contributed 1 401 084 person‐years follow‐up. In total, 4024 (4.2%) of women were hospitalized for CVD, of which 1585 events (1.6%) occurred within 10 years of follow‐up. The final model had modest discriminatory performance (area under the receiver operating characteristic curve, 0.66; 95% CI, 0.65–0.67) and good calibration with slope of 0.95 and intercept of −0.19. There was moderate classification accuracy (likelihood ratio+: 5.90; 95% CI, 5.01–6.95) in the highest‐risk group upon risk stratification.

Conclusions

Overall, our model had modest performance in predicting the 10‐year risk of premature CVD for women with HDP. We recommend the addition of clinical variables, and external validation, before consideration for clinical use.

A Novel Metric for Developing Easy-to-Use and Accurate Clinical Prediction Models: The Time-cost Information Criterion

BACKGROUND

Guidelines recommend that clinicians use clinical prediction models to estimate future risk to guide decisions. For example, predicted fracture risk is a major factor in the decision to initiate bisphosphonate medications. However, current methods for developing prediction models often lead to models that are accurate but difficult to use in clinical settings.

OBJECTIVE

The objective of this study was to develop and test whether a new metric that explicitly balances model accuracy with clinical usability leads to accurate, easier-to-use prediction models.

METHODS

We propose a new metric called the Time-cost Information Criterion (TCIC) that will penalize potential predictor variables that take a long time to obtain in clinical settings. To demonstrate how the TCIC can be used to develop models that are easier-to-use in clinical settings, we use data from the 2000 wave of the Health and Retirement Study (n=6311) to develop and compare time to mortality prediction models using a traditional metric (Bayesian Information Criterion or BIC) and the TCIC.

RESULTS

We found that the TCIC models utilized predictors that could be obtained more quickly than BIC models while achieving similar discrimination. For example, the TCIC identified a 7-predictor model with a total time-cost of 44 seconds, while the BIC identified a 7-predictor model with a time-cost of 119 seconds. The Harrell C-statistic of the TCIC and BIC 7-predictor models did not differ (0.7065 vs. 0.7088, P=0.11).

CONCLUSION

Accounting for the time-costs of potential predictor variables through the use of the TCIC led to the development of an easier-to-use mortality prediction model with similar discrimination.

Development and External Validation of a Multivariable Prediction Model to Identify Nondiabetic Hyperglycemia and Undiagnosed Type 2 Diabetes: Diabetes Risk Assessment in Dentistry Score (DDS)

The aim of this study was to develop and externally validate a score for use in dental settings to identify those at risk of undiagnosed nondiabetic hyperglycemia (NDH) or type 2 diabetes (T2D). The Studies of Health in Pomerania (SHIP) project comprises 2 representative population-based cohort studies conducted in northeast Germany. SHIP-TREND-0, 2008 to 2012 (the development data set) had 3,339 eligible participants, with 329 having undiagnosed NDH or T2D. Missing data were replaced using multiple imputation. Potential covariates were selected for inclusion in the model using backward elimination. Heuristic shrinkage was used to reduce overfitting, and the final model was adjusted for optimism. We report the full model and a simplified paper-based point-score system. External validation of the model and score employed an independent data set comprising 2,359 participants with 357 events. Predictive performance, discrimination, calibration, and clinical utility were assessed. The final model included age, sex, body mass index, smoking status, first-degree relative with diabetes, presence of a dental prosthesis, presence of mobile teeth, history of periodontal treatment, and probing pocket depths ≥5 mm as well as prespecified interaction terms. In SHIP-TREND-0, the model area under the curve (AUC) was 0.72 (95% confidence interval [CI] 0.69, 0.75), calibration in the large was -0.025. The point score AUC was 0.69 (95% CI 0.65, 0.72), with sensitivity of 77.0 (95% CI 76.8, 77.2), specificity of 51.5 (95% CI 51.4, 51.7), negative predictive value of 94.5 (95% CI 94.5, 94.6), and positive predictive value of 17.0 (95% CI 17.0, 17.1). External validation of the point score gave an AUC of 0.69 (95% CI 0.66, 0.71), sensitivity of 79.2 (95% CI 79.0, 79.4), specificity of 49.9 (95% CI 49.8, 50.00), negative predictive value 91.5 (95% CI 91.5, 91.6), and positive predictive value of 25.9 (95% CI 25.8, 26.0). A validated prediction model involving dental variables can identify NDH or undiagnosed T2DM. Further studies are required to validate the model for different European populations.

Advancing Scenario Planning through Integrating Urban Growth Prediction with Future Flood Risk Models

High uncertainty about future urbanization and flood risk conditions limits the ability to increase resiliency in traditional scenario-based urban planning. While scenario planning integrating urban growth prediction modeling is becoming more common, these models have not been effectively linked with future flood plain changes due to sea level rise. This study advances scenario planning by integrating urban growth prediction models with flood risk scenarios. The Land Transformation Model, a land change prediction model using a GIS based artificial neural network, is used to predict future urban growth scenarios for Tampa, Florida, USA, and future flood risks are then delineated based on the current 100-year floodplain using NOAA level rise scenarios. A multi-level evaluation using three urban prediction scenarios (business as usual, growth as planned, and resilient growth) and three sea level rise scenarios (low, high, and extreme) is conducted to determine how prepared Tampa's current land use plan is in handling increasing resilient development in lieu of sea level rise. Results show that the current land use plan (growth as planned) decreases flood risk at the city scale but not always at the neighborhood scale, when compared to no growth regulations (business as usual). However, flood risk when growing according to the current plan is significantly higher when compared to all future growth residing outside of the 100-year floodplain (resilient growth). Understanding the potential effects of sea level rise depends on understanding the probabilities of future development options and extreme climate conditions.