Identification of potentially undiagnosed patients with nontuberculous mycobacterial lung disease using machine learning applied to primary care data in the UK.

Nontuberculous mycobacterial lung disease (NTMLD) is a rare lung disease often missed due to a low index of suspicion and unspecific clinical presentation. This retrospective study was designed to characterise the prediagnosis features of NTMLD patients in primary care and to assess the feasibility of using machine learning to identify undiagnosed NTMLD patients.IQVIA Medical Research Data (incorporating THIN, a Cegedim Database), a UK electronic medical records primary care database was used. NTMLD patients were identified between 2003 and 2017 by diagnosis in primary or secondary care or record of NTMLD treatment regimen. Risk factors and treatments were extracted in the prediagnosis period, guided by literature and expert clinical opinion. The control population was enriched to have at least one of these features.741 NTMLD and 112 784 control patients were selected. Annual prevalence rates of NTMLD from 2006 to 2016 increased from 2.7 to 5.1 per 100 000. The most common pre-existing diagnoses and treatments for NTMLD patients were COPD and asthma and penicillin, macrolides and inhaled corticosteroids. Compared to random testing, machine learning improved detection of patients with NTMLD by almost a thousand-fold with AUC of 0.94. The total prevalence of diagnosed and undiagnosed cases of NTMLD in 2016 was estimated to range between 9 and 16 per 100 000.This study supports the feasibility of machine learning applied to primary care data to screen for undiagnosed NTMLD patients, with results indicating that there may be a substantial number of undiagnosed cases of NTMLD in the UK.

CNVs conferring risk of autism or schizophrenia affect cognition in controls.

In a small fraction of patients with schizophrenia or autism, alleles of copy-number variants (CNVs) in their genomes are probably the strongest factors contributing to the pathogenesis of the disease. These CNVs may provide an entry point for investigations into the mechanisms of brain function and dysfunction alike. They are not fully penetrant and offer an opportunity to study their effects separate from that of manifest disease. Here we show in an Icelandic sample that a few of the CNVs clearly alter fecundity (measured as the number of children by age 45). Furthermore, we use various tests of cognitive function to demonstrate that control subjects carrying the CNVs perform at a level that is between that of schizophrenia patients and population controls. The CNVs do not all affect the same cognitive domains, hence the cognitive deficits that drive or accompany the pathogenesis vary from one CNV to another. Controls carrying the chromosome 15q11.2 deletion between breakpoints 1 and 2 (15q11.2(BP1-BP2) deletion) have a history of dyslexia and dyscalculia, even after adjusting for IQ in the analysis, and the CNV only confers modest effects on other cognitive traits. The 15q11.2(BP1-BP2) deletion affects brain structure in a pattern consistent with both that observed during first-episode psychosis in schizophrenia and that of structural correlates in dyslexia.

Parsing the Roles of the Frontal Lobes and Basal Ganglia in Task Control Using Multivoxel Pattern Analysis.

Cognitive control has traditionally been associated with pFC based on observations of deficits in patients with frontal lesions. However, evidence from patients with Parkinson disease indicates that subcortical regions also contribute to control under certain conditions. We scanned 17 healthy volunteers while they performed a task-switching paradigm that previously dissociated performance deficits arising from frontal lesions in comparison with Parkinson disease, as a function of the abstraction of the rules that are switched. From a multivoxel pattern analysis by Gaussian Process Classification, we then estimated the forward (generative) model to infer regional patterns of activity that predict Switch/Repeat behavior between rule conditions. At 1000 permutations, Switch/Repeat classification accuracy for concrete rules was significant in the BG, but at chance in the frontal lobe. The inverse pattern was obtained for abstract rules, whereby the conditions were successfully discriminated in the frontal lobe but not in the BG. This double dissociation highlights the difference between cortical and subcortical contributions to cognitive control and demonstrates the utility of multivariate approaches in investigations of functions that rely on distributed and overlapping neural substrates.

Bayesian multi-task learning for decoding multi-subject neuroimaging data.

Decoding models based on pattern recognition (PR) are becoming increasingly important tools for neuroimaging data analysis. In contrast to alternative (mass-univariate) encoding approaches that use hierarchical models to capture inter-subject variability, inter-subject differences are not typically handled efficiently in PR. In this work, we propose to overcome this problem by recasting the decoding problem in a multi-task learning (MTL) framework. In MTL, a single PR model is used to learn different but related "tasks" simultaneously. The primary advantage of MTL is that it makes more efficient use of the data available and leads to more accurate models by making use of the relationships between tasks. In this work, we construct MTL models where each subject is modelled by a separate task. We use a flexible covariance structure to model the relationships between tasks and induce coupling between them using Gaussian process priors. We present an MTL method for classification problems and demonstrate a novel mapping method suitable for PR models. We apply these MTL approaches to classifying many different contrasts in a publicly available fMRI dataset and show that the proposed MTL methods produce higher decoding accuracy and more consistent discriminative activity patterns than currently used techniques. Our results demonstrate that MTL provides a promising method for multi-subject decoding studies by focusing on the commonalities between a group of subjects rather than the idiosyncratic properties of different subjects.

Finding undiagnosed patients with hepatitis C infection: an application of artificial intelligence to patient claims data.

Hepatitis C virus (HCV) remains a significant public health challenge with approximately half of the infected population untreated and undiagnosed. In this retrospective study, predictive models were developed to identify undiagnosed HCV patients using longitudinal medical claims linked to prescription data from approximately ten million patients in the United States (US) between 2010 and 2016. Features capturing information on demographics, risk factors, symptoms, treatments and procedures relevant to HCV were extracted from patients' medical history. Predictive algorithms were developed based on logistic regression, random forests, gradient boosted trees and a stacked ensemble. Descriptive analysis indicated that patients exhibited known symptoms of HCV on average 2-3 years prior to their diagnosis. The precision was at least 95% for all algorithms at low levels of recall (10%). For recall levels >50%, the stacked ensemble performed best with a precision of 97% compared with 87% for the gradient boosted trees and just 31% for the logistic regression. For context, the Center for Disease Control recommends screening in an at-risk sub-population with an estimated HCV prevalence of 2.23%. The artificial intelligence (AI) algorithm presented here has a precision which is substantially higher than the screening rates associated with recommended clinical guidelines, suggesting that AI algorithms have the potential to provide a step change in the effectiveness of HCV screening.

Personalized Medication Response Prediction for Attention-Deficit Hyperactivity Disorder: Learning in the Model Space vs. Learning in the Data Space.

Attention-Deficit Hyperactive Disorder (ADHD) is one of the most common mental health disorders amongst school-aged children with an estimated prevalence of 5% in the global population (American Psychiatric Association, 2013). Stimulants, particularly methylphenidate (MPH), are the first-line option in the treatment of ADHD (Reeves and Schweitzer, 2004; Dopheide and Pliszka, 2009) and are prescribed to an increasing number of children and adolescents in the US and the UK every year (Safer et al., 1996; McCarthy et al., 2009), though recent studies suggest that this is tailing off, e.g., Holden et al. (2013). Around 70% of children demonstrate a clinically significant treatment response to stimulant medication (Spencer et al., 1996; Schachter et al., 2001; Swanson et al., 2001; Barbaresi et al., 2006). However, it is unclear which patient characteristics may moderate treatment effectiveness. As such, most existing research has focused on investigating univariate or multivariate correlations between a set of patient characteristics and the treatment outcome, with respect to dosage of one or several types of medication. The results of such studies are often contradictory and inconclusive due to a combination of small sample sizes, low-quality data, or a lack of available information on covariates. In this paper, feature extraction techniques such as latent trait analysis were applied to reduce the dimension of on a large dataset of patient characteristics, including the responses to symptom-based questionnaires, developmental health factors, demographic variables such as age and gender, and socioeconomic factors such as parental income. We introduce a Bayesian modeling approach in a "learning in the model space" framework that combines existing knowledge in the literature on factors that may potentially affect treatment response, with constraints imposed by a treatment response model. The model is personalized such that the variability among subjects is accounted for by a set of subject-specific parameters. For remission classification, this approach compares favorably with conventional methods such as support vector machines and mixed effect models on a range of performance measures. For instance, the proposed approach achieved an area under receiver operator characteristic curve of 82-84%, compared to 75-77% obtained from conventional regression or machine learning ("learning in the data space") methods.

Using deep belief network modelling to characterize differences in brain morphometry in schizophrenia.

Neuroimaging-based models contribute to increasing our understanding of schizophrenia pathophysiology and can reveal the underlying characteristics of this and other clinical conditions. However, the considerable variability in reported neuroimaging results mirrors the heterogeneity of the disorder. Machine learning methods capable of representing invariant features could circumvent this problem. In this structural MRI study, we trained a deep learning model known as deep belief network (DBN) to extract features from brain morphometry data and investigated its performance in discriminating between healthy controls (N = 83) and patients with schizophrenia (N = 143). We further analysed performance in classifying patients with a first-episode psychosis (N = 32). The DBN highlighted differences between classes, especially in the frontal, temporal, parietal, and insular cortices, and in some subcortical regions, including the corpus callosum, putamen, and cerebellum. The DBN was slightly more accurate as a classifier (accuracy = 73.6%) than the support vector machine (accuracy = 68.1%). Finally, the error rate of the DBN in classifying first-episode patients was 56.3%, indicating that the representations learned from patients with schizophrenia and healthy controls were not suitable to define these patients. Our data suggest that deep learning could improve our understanding of psychiatric disorders such as schizophrenia by improving neuromorphometric analyses.

A Spatiotemporal Profile of In Vivo Cerebral Blood Flow Changes Following Intranasal Oxytocin in Humans.

BACKGROUND: Animal and human studies highlight the role of oxytocin in social cognition and behavior and the potential of intranasal oxytocin (IN-OT) to treat social impairment in individuals with neuropsychiatric disorders such as autism. However, extensive efforts to evaluate the central actions and therapeutic efficacy of IN-OT may be marred by the absence of data regarding its temporal dynamics and sites of action in the living human brain. METHODS: In a placebo-controlled study, we used arterial spin labeling to measure IN-OT-induced changes in resting regional cerebral blood flow (rCBF) in 32 healthy men. Volunteers were blinded regarding the nature of the compound they received. The rCBF data were acquired 15 min before and up to 78 min after onset of treatment onset (40 IU of IN-OT or placebo). The data were analyzed using mass univariate and multivariate pattern recognition techniques. RESULTS: We obtained robust evidence delineating an oxytocinergic network comprising regions expected to express oxytocin receptors, based on histologic evidence, and including core regions of the brain circuitry underpinning social cognition and emotion processing. Pattern recognition on rCBF maps indicated that IN-OT-induced changes were sustained over the entire posttreatment observation interval (25-78 min) and consistent with a pharmacodynamic profile showing a peak response at 39-51 min. CONCLUSIONS: Our study provides the first visualization and quantification of IN-OT-induced changes in rCBF in the living human brain unaffected by cognitive, affective, or social manipulations. Our findings can inform theoretical and mechanistic models regarding IN-OT effects on typical and atypical social behavior and guide future experiments (e.g., regarding the timing of experimental manipulations).

The role of machine learning in neuroimaging for drug discovery and development.

Neuroimaging has been identified as a potentially powerful probe for the in vivo study of drug effects on the brain with utility across several phases of drug development spanning preclinical and clinical investigations. Specifically, neuroimaging can provide insight into drug penetration and distribution, target engagement, pharmacodynamics, mechanistic action and potential indicators of clinical efficacy. In this review, we focus on machine learning approaches for neuroimaging which enable us to make predictions at the individual level based on the distributed effects across the whole brain. Crucially, these approaches can be trained on data from one study and applied to an independent study and, unlike group-level statistics, can be readily use to assess the generalisability to unseen data. In this review, we present examples and suggestions for how machine learning could help answer fundamental questions spanning the drug discovery pipeline: (1) Who should I recruit for this study? (2) What should I measure and when should I measure it? (3) How does the pharmacological agent behave using an experimental medicine model?, and (4) How does a compound differ from and/or resemble existing compounds? Specifically, we present studies from the literature and we suggest areas for the focus of future development. Further refinement and tailoring of machine learning techniques may help realise their tremendous potential for drug discovery and drug validation.

Estimating multivariate similarity between neuroimaging datasets with sparse canonical correlation analysis: an application to perfusion imaging.

An increasing number of neuroimaging studies are based on either combining more than one data modality (inter-modal) or combining more than one measurement from the same modality (intra-modal). To date, most intra-modal studies using multivariate statistics have focused on differences between datasets, for instance relying on classifiers to differentiate between effects in the data. However, to fully characterize these effects, multivariate methods able to measure similarities between datasets are needed. One classical technique for estimating the relationship between two datasets is canonical correlation analysis (CCA). However, in the context of high-dimensional data the application of CCA is extremely challenging. A recent extension of CCA, sparse CCA (SCCA), overcomes this limitation, by regularizing the model parameters while yielding a sparse solution. In this work, we modify SCCA with the aim of facilitating its application to high-dimensional neuroimaging data and finding meaningful multivariate image-to-image correspondences in intra-modal studies. In particular, we show how the optimal subset of variables can be estimated independently and we look at the information encoded in more than one set of SCCA transformations. We illustrate our framework using Arterial Spin Labeling data to investigate multivariate similarities between the effects of two antipsychotic drugs on cerebral blood flow.

The cortical thickness phenotype of individuals with DISC1 translocation resembles schizophrenia.

BACKGROUND: The disrupted in schizophrenia 1 (DISC1) gene locus was originally identified in a Scottish pedigree with a high incidence of psychiatric disorders that is associated with a balanced t(1;11)(q42.1;q14.3) chromosomal translocation. Here, we investigated whether members of this family carrying the t(1;11)(q42.1;q14.3) translocation have a common brain-related phenotype and whether this phenotype is similar to that observed in schizophrenia (SCZ), using multivariate pattern recognition techniques. METHODS: We measured cortical thickness, cortical surface area, subcortical volumes, and regional cerebral blood flow (rCBF) in healthy controls (HC) (n = 24), patients diagnosed with SCZ (n = 24), patients diagnosed with bipolar disorder (BP) (n = 19), and members of the original Scottish family (n = 30) who were either carriers (T+) or noncarriers (T-) of the DISC1 translocation. Binary classification models were developed to assess the differences and similarities across groups. RESULTS: Based on cortical thickness, 72% of the T- group were assigned to the HC group, 83% of the T+ group were assigned to the SCZ group, and 45% of the BP group were classified as belonging to the SCZ group, suggesting high specificity of this measurement in predicting brain-related phenotypes. Shared brain-related phenotypes between SCZ and T+ individuals were found for cortical thickness only. Finally, a classification accuracy of 73% was achieved when directly comparing the pattern of cortical thickness of T+ and T- individuals. CONCLUSION: Together, the results of this study suggest that the DISC1 translocation may increase the risk of psychiatric disorders in this pedigree by affecting neurostructural phenotypes such as cortical thickness. FUNDING: This work was supported by the National Health Service Research Scotland, the Scottish Translational Medicine Research Collaboration, the Innovative Medicines Initiative (IMI), the Engineering and Physical Sciences Research Council (EPSRC), The Wellcome Trust, the National Institute of Health Research (NIHR), and Pfizer.

Brain shaving: adaptive detection for brain PET data.

The intricacy of brain biology is such that the variation of imaging end-points in health and disease exhibits an unpredictable range of spatial distributions from the extremely localized to the very diffuse. This represents a challenge for the two standard approaches to analysis, the mass univariate and the multivariate that exhibit either strong specificity but not as good sensitivity (the former) or poor specificity and comparatively better sensitivity (the latter). In this work, we develop an analytical methodology for positron emission tomography that operates an extraction ('shaving') of coherent patterns of signal variation while maintaining control of the type I error. The methodology operates two rotations on the image data, one local using the wavelet transform and one global using the singular value decomposition. The control of specificity is obtained by using the gap statistic that selects, within each eigenvector, a subset of significantly coherent elements. Face-validity of the algorithm is demonstrated using a paradigmatic data-set with two radiotracers, [(11)C]-raclopride and [(11)C]-(R)-PK11195, measured on the same Huntington's disease patients, a disorder with a genetic based diagnosis. The algorithm is able to detect the two well-known separate but connected processes of dopamine neuronal loss (localized in the basal ganglia) and neuroinflammation (diffusive around the whole brain). These processes are at the two extremes of the distributional envelope, one being very sparse and the latter being perfectly Gaussian and they are not adequately detected by the univariate and the multivariate approaches.

Predicting progression of Alzheimer's disease using ordinal regression.

We propose a novel approach to predicting disease progression in Alzheimer's disease (AD)--multivariate ordinal regression--which inherently models the ordered nature of brain atrophy spanning normal aging (CTL) to mild cognitive impairment (MCI) to AD. Ordinal regression provides probabilistic class predictions as well as a continuous index of disease progression--the ORCHID (Ordinal Regression Characteristic Index of Dementia) score. We applied ordinal regression to 1023 baseline structural MRI scans from two studies: the US-based Alzheimer's Disease Neuroimaging Initiative (ADNI) and the European based AddNeuroMed program. Here, the acquired AddNeuroMed dataset was used as a completely independent test set for the ordinal regression model trained on the ADNI cohort providing an optimal assessment of model generalizability. Distinguishing CTL-like (CTL and stable MCI) from AD-like (MCI converters and AD) resulted in balanced accuracies of 82% (cross-validation) for ADNI and 79% (independent test set) for AddNeuroMed. For prediction of conversion from MCI to AD, balanced accuracies of 70% (AUC of 0.75) and 75% (AUC of 0.81) were achieved. The ORCHID score was computed for all subjects. We showed that this measure significantly correlated with MMSE at 12 months (ρ =  -0.64, ADNI and ρ =  -0.59, AddNeuroMed). Additionally, the ORCHID score can help fractionate subjects with unstable diagnoses (e.g. reverters and healthy controls who later progressed to MCI), moderately late converters (12-24 months) and late converters (24-36 months). A comparison with results in the literature and direct comparison with a binary classifier suggests that the performance of this framework is highly competitive.

Bridging paradigms: hybrid mechanistic-discriminative predictive models.

Many disease processes are extremely complex and characterized by multiple stochastic processes interacting simultaneously. Current analytical approaches have included mechanistic models and machine learning (ML), which are often treated as orthogonal viewpoints. However, to facilitate truly personalized medicine, new perspectives may be required. This paper reviews the use of both mechanistic models and ML in healthcare as well as emerging hybrid methods, which are an exciting and promising approach for biologically based, yet data-driven advanced intelligent systems.

"... Take up thy bed, and vote." Measuring the relationship between voting behaviour and indicators of health.

In most countries health policy is an important part of the political agenda. Yet few studies have examined the relationship between the two. This study investigates the association between health and voter turnout in Britain using the National Child Development Study. Self-rated general health, the Malaise Inventory score and indicators of smoking and alcohol consumption, as measured at ages 23, 33 and 42, are regressed on voter turnout in the 1979, 1987 and 1997 general elections. The results indicate that individuals with poor general and mental health and smokers are less likely to vote at election time.