Artificial Neural Networks and Deep Learning for Genomic Prediction of Binary, Ordinal, and Mixed Outcomes

In this chapter, we provide the main elements for implementing deep neural networks in Keras for binary, categorical, and mixed outcomes under feedforward networks as well as the main practical issues involved in implementing deep learning models with binary response variables. The same practical issues are provided for implementing deep neural networks with categorical and count traits under a univariate framework. We follow with a detailed assessment of information for implementing multivariate deep learning models for continuous, binary, categorical, count, and mixed outcomes. In all the examples given, the data came from plant breeding experiments including genomic data. The training process for binary, ordinal, count, and multivariate outcomes is similar to fitting DNN models with univariate continuous outcomes, since once we have the data to be trained, we need to (a) define the DNN model in Keras, (b) configure and compile the model, (c) fit the model, and finally, (d) evaluate the prediction performance in the testing set. In the next section, we provide illustrative examples of training DNN for binary outcomes in Keras R (Chollet and Allaire, Deep learning with R. Manning Publications, Manning Early Access Program (MEA), 2017; Allaire and Chollet, Keras: R interface to Keras’, 2019).

Mean corrected generalized estimating equations for longitudinal binary outcomes with report bias

Cocaine addiction is an important public health problem worldwide. Cognitive-behavioral therapy is a counseling intervention for supporting cocaine-dependent individuals through recovery and relapse prevention. It may reduce patients’ cocaine uses by improving their motivations and enabling them to recognize risky situations. To study the effect of cognitive behavioral therapy on cocaine dependence, the self-reported cocaine use with urine test data were collected at the Primary Care Center of Yale-New Haven Hospital. Its outcomes are binary, including both the daily self-reported drug uses and weekly urine test results. To date, the generalized estimating equations are widely used to analyze binary data with repeated measures. However, due to the existence of significant self-report bias in the self-reported cocaine use with urine test data, a direct application of the generalized estimating equations approach may not be valid. In this paper, we proposed a novel mean corrected generalized estimating equations approach for analyzing longitudinal binary outcomes subject to reporting bias. The mean corrected generalized estimating equations can provide consistently and asymptotically normally distributed estimators under true contamination probabilities. In the self-reported cocaine use with urine test study, accurate weekly urine test results are used to detect contamination. The superior performances of the proposed method are illustrated by both simulation studies and real data analysis.

The role of teleoaffective profiles in practice adaptation

Consumer research is interested in the way consumers navigate consumption in the face of disruption, often using practice theory to focus on how practitioners creatively realign practice elements in order to carry on. Although recognising their significance, this research undertheorizes the significance, role and characteristics of 'meanings' in practice adaptation, presenting them as constraining and yet easy to adapt. We explore and theorize meanings in practice adaptation by mobilising the theoretical leverage of Schatzki’s (2002) concept of ‘teleoaffective structures’. Through our empirical material, we illuminate how multifaceted teleoaffective components constituent of teleoaffective structures are integrated differently into routinised practice performances in relatively stable ways; incorporated via ‘teleoaffective profiles’ that are unique to practitioners but properties of practices. Furthermore, we propose that teleoaffective profiles have different characteristics that condition practice adaptation, as teleological orientations and affective engagements afford different pathways towards integration with available materials and competences. We use our empirical material, based on interviews with loyal gym-based resistance training practitioners during COVID-19 gym closures, to illuminate our argument that practitioners can have ‘rigid’, ‘elastic’ or ‘fluid’ teleoaffective profiles. The characteristics of these profiles, which are unique but remain the properties of the practice, mean that adaptation processes and experiences unfold differently. This perspective advances from accounts of adaptation that are centred on binary outcomes of success or failure. Furthermore, our theorization advances from practice-oriented consumption adaptation research that foregrounds practitioner creativity and fails to adequately incorporate understandings of how practice elements condition adaptation processes. Yet, we retain practitioner experiences in our analysis. Teleoaffective components, profiles and properties provides further theoretical leverage to the practice turn in consumption research and advances the burgeoning focus on the significance of teleoaffective structures in the topographies of practices

Assessment of the predictive value of the NTISS scale for neonatal outcomes

Introduction. Improving the disease severity scoring systems at the stages of inter-hospital transportation remains an actual in neonatal intensive care. Therapeutic scales remain poorly studied and their predictive value and practical applicability. The aim of the work is to determine the predictive value of the NTISS scale at the stage of pre-transport preparation in relation to the treatment outcomes of newborns.Materials and methods. The cohort study included data from 604 visits of the resuscitation and consultation center transport team. The evaluation was performed on the NTISS scale, and the outcomes were studied. The AUC ROC curve of the NTISS scale was calculated in relation to the binary outcomes. The correlation analysis of the quantitative data was performed by Spearman's criterion.Results. AUC greater than 0.8 was observed for the risk of death (AUC=0,823 (0,758-0,888)), 7-day mortality (AUC=0,827 (0,752-0,901)), late onset sepsis (AUC=0,808 (0,737-0,879)), bronchopulmonary dysplasia (AUC=0,810 (0,763-0,856)), severe intraventricular hemorrhage (AUC=0,847 (0,804-0,889)) иocclusivehydrocephalus(AUC=0,830 (0,757-0,904)). Similarresultswereobtained analyzing the outcomes among the surviving patients. For other binary outcomes, the scale shows an AUC of less than 0.8. The analysis of outcomes among the surviving patients showed a weak correlation between the NTISS score and the duration of intensive care, r=0.492, p<0.0001, and the duration of hospitalization, r=0.498, p<0.0001.Discussion. The NTISS scale demonstrated an acceptable level of accuracy (AUC>0.8) in predicting hospital mortality, late neonatal sepsis, bronchopulmonary dysplasia, severe intraventricular hemorrhage, and the formation of occlusive hydrocephalus, among both surviving patients and general sample. The observed results are comparable with the information content of other neonatal scales of various types.Conclusion. The predictive value of NTISS in relation to the outcomes of the hospital stage is comparable to the physiological scales described in the literature.

A Critical Review of Statistical Methods for Twin Studies Relating Exposure to Early Life Health Conditions

When investigating disease etiology, twin data provide a unique opportunity to control for confounding and disentangling the role of the human genome and exposome. However, using appropriate statistical methods is fundamental for exploiting such potential. We aimed to critically review the statistical approaches used in twin studies relating exposure to early life health conditions. We searched PubMed, Scopus, Web of Science, and Embase (2011–2021). We identified 32 studies and nine classes of methods. Five were conditional approaches (within-pair analyses): additive-common-erratic (ACE) models (11 studies), generalized linear mixed models (GLMMs, five studies), generalized linear models (GLMs) with fixed pair effects (four studies), within-pair difference analyses (three studies), and paired-sample tests (two studies). Four were marginal approaches (unpaired analyses): generalized estimating equations (GEE) models (five studies), GLMs with cluster-robust standard errors (six studies), GLMs (one study), and independent-sample tests (one study). ACE models are suitable for assessing heritability but require adaptations for binary outcomes and repeated measurements. Conditional models can adjust by design for shared confounders, and GLMMs are suitable for repeated measurements. Marginal models may lead to invalid inference. By highlighting the strengths and limitations of commonly applied statistical methods, this review may be helpful for researchers using twin designs.

Are Drugs that Cause Dysbiosis Longitudinally Associated with Cognitive Scores, Cognitive Impairment, & Dementia?

Recent research has examined how the microbiome may influence cognitive outcomes; however, there is a paucity of research understanding how medication associated with dysbiosis may be associated with cognitive changes. This study used data from the Health and Retirement Study and the Prescription Drug Study subset for adults 51 and older (n=3,898). Continuous (0-27) and categorical (cognitively normal=12-27; cognitive impairment=7-11; and dementia=0-6) cognitive outcomes were used. Prescriptions utilized were proton pump inhibitors, antibiotics, selective serotonin reuptake inhibitors, tricyclic antidepressants, antipsychotics, antihistamines, and a summed dose-response measure. Linear mixed models (LMM) and generalized linear mixed models (GLMM) were used for continuous and binary outcomes. For the LMM model, the main effect for those taking one medication was insignificant; however, the interaction with time showed a significant decrease over time (β: -0.07; 95% confidence interval (CI): -0.14, 0.01). The mean cognitive score was lower for those taking two or more medications (β: -1.48; 95% CI: -2.70, -0.25), although the interaction with time was insignificant. GLMM results showed those taking two or medications had odds that are 612% larger (odds ratio (OR): 7.12; 95% CI: 3.03, 16.71) of going from cognitively healthy to dementia but the interaction with time showed decreased odds over time (OR: 0.92; 95% CI 0.86, 0.97). For cognitive impairment, those who took two or more medications had odds that were 45% larger (OR: 1.45; 95% CI: 1.05, 2.00) of going from cognitively healthy to cognitively impaired. This study indicated a dose-response aspect to taking medications on cognitive outcomes.

The epidemiology of soil-transmitted helminth infections in children up to 8 years of age: Findings from an Ecuadorian birth cohort

Background There are few prospective longitudinal studies of soil-transmitted helminth (STH) infections during early childhood. We studied the epidemiology of and risk factors for soil-transmitted helminth infections from birth to 8 years of age in tropical Ecuador. Methods 2,404 newborns were followed to 8 years of age with periodic stool sample collections. Stool samples were collected also from household members at the time of the child’s birth and examined by microscopy. Data on social, environmental, and demographic characteristics were collected by maternal questionnaire. Associations between potential risk factors and STH infections were estimated using generalized estimated equations applied to longitudinal binary outcomes for presence or absence of infections at collection times. Results Of 2,404 children, 1,120 (46.6%) were infected with at least one STH infection during the first 8 years of life. The risk of A. lumbricoides (16.2%) was greatest at 3 years, while risks of any STH (25.1%) and T. trichiura (16.5%) peaked at 5 years. Factors significantly associated with any STH infection in multivariable analyses included age, day-care (OR 1.34, 95% CI 1.03–1.73), maternal Afro-Ecuadorian ethnicity (non-Afro vs. Afro, OR 0.55, 95% CI 0.43–0.70) and lower educational level (secondary vs. illiterate, OR 0.31, 95% CI 0.22–0.45)), household overcrowding (OR 1.53, 95% CI 1.21–1.94)), having a latrine rather than a water closet (WC vs. latrine, OR 0.77, 95% CI 0.62–0.95)), and STH infections among household members (OR 2.03, 95% CI 1.59–2.58)). T. trichiura was more associated with poverty (high vs. low socioeconomic status, OR, 0.63, 95% CI 0.40–0.99)] and presence of infected siblings in the household (OR 3.42, 95% CI 2.24–5.22). Conclusion STH infections, principally with A. lumbricoides and T. trichiura, peaked between 3 and 5 years in this cohort of children in tropical Ecuador. STH infections among household members were an important determinant of infection risk and could be targeted for control and elimination strategies.

On Small-Sample Inference in Multiphase Stepped Wedge Cluster Randomized Trial (MSW-CRT) With Binary Outcomes

BackgroundThe stepped wedge cluster randomized trial (SW-CRT) design is now preferred for many health- related trials because of its flexibility on resource allocation and clinical ethics concerns. However, as a necessary extension of studying multiple interventions, multiphase stepped wedge designs (MSW-CRT) have not been studied adequately. Since estimated intervention effect from Generalized estimating equations (GEE) has a population-average interpretation, valid inference methods for binary outcomes based on GEE are preferred by public health policy makers.MethodsWe form hypothesis testing of add-on effect of a second treatment based on GEE analysis in an MSW-CRT design with limited number of clusters. Four variance-correction estimators are used to adjust the bias of the sandwich estimator. Simulation studies have been used to compare the statistical power and type I error rate of these methods under different correlation matrices.Results We demonstrate that an average estimator with t(I-3) can stably maintain type I error close to the nominal level with limited sample sizes in our settings. We show that power of testing the add-on effect depends on the baseline event rate, effect sizes of two interventions and the number of clusters. Moreover, by changing the design with including more sequences, power benefit can be achieved. ConclusionsFor designing the MSW-CRT, we suggest using more sequences and checking event rate after initiating the first intervention via interim analysis. When the number of clusters is not very large in MSW-CRTs, inference can be conduct using GEE analysis with an average estimator with t(I-3) sampling distribution.

Diversity Forests: Using Split Sampling to Enable Innovative Complex Split Procedures in Random Forests

The diversity forest algorithm is an alternative candidate node split sampling scheme that makes innovative complex split procedures in random forests possible. While conventional univariable, binary splitting suffices for obtaining strong predictive performance, new complex split procedures can help tackling practically important issues. For example, interactions between features can be exploited effectively by bivariable splitting. With diversity forests, each split is selected from a candidate split set that is sampled in the following way: for $$l = 1, \dots , {nsplits}$$ l = 1 , ⋯ , nsplits : (1) sample one split problem; (2) sample a single or few splits from the split problem sampled in (1) and add this or these splits to the candidate split set. The split problems are specifically structured collections of splits that depend on the respective split procedure considered. This sampling scheme makes innovative complex split procedures computationally tangible while avoiding overfitting. Important general properties of the diversity forest algorithm are evaluated empirically using univariable, binary splitting. Based on 220 data sets with binary outcomes, diversity forests are compared with conventional random forests and random forests using extremely randomized trees. It is seen that the split sampling scheme of diversity forests does not impair the predictive performance of random forests and that the performance is quite robust with regard to the specified nsplits value. The recently developed interaction forests are the first diversity forest method that uses a complex split procedure. Interaction forests allow modeling and detecting interactions between features effectively. Further potential complex split procedures are discussed as an outlook.

Statistical reanalysis of vascular event outcomes in primary and secondary vascular prevention trials

Background Vascular prevention trials typically use dichotomous event outcomes although this may be inefficient statistically and gives no indication of event severity. We assessed whether ordinal outcomes would be more efficient and how to best analyse them. Methods Chief investigators of vascular prevention randomised controlled trials that showed evidence of either benefit or harm, or were included in a systematic review that overall showed benefit or harm, shared individual participant data from their trials. Ordered categorical versions of vascular event outcomes (such as stroke and myocardial infarction) were analysed using 15 statistical techniques and their results then ranked, with the result with the smallest p-value given the smallest rank. Friedman and Duncan’s multiple range tests were performed to assess differences between tests by comparing the average ranks for each statistical test. Results Data from 35 trials (254,223 participants) were shared with the collaboration. 13 trials had more than two treatment arms, resulting in 59 comparisons. Analysis approaches (Mann Whitney U, ordinal logistic regression, multiple regression, bootstrapping) that used ordinal outcome data had a smaller average rank and therefore appeared to be more efficient statistically than those that analysed the original binary outcomes. Conclusions Ordinal vascular outcome measures appear to be more efficient statistically than binary outcomes and provide information on the severity of event. We suggest a potential role for using ordinal outcomes in vascular prevention trials.