Machine Learning is the study of methods for programming computers to learn. Computers are applied to a wide range of tasks, and for most of these it is relatively easy for programmers to design and implement the necessary software. However, there are many tasks for which this is difficult or impossible. These can be divided into four general categories. First, there are problems for which there exist no human experts. For example, in modern automated manufacturing facilities, there is a need to predict machine failures before they occur by analyzing sensor readings. Because the machines are new, there are no human experts who can be interviewed by a programmer to provide the knowledge necessary to build a computer system. A machine learning system can study recorded data and subsequent machine failures and learn prediction rules. Second, there are problems where human experts exist, but where they are unable to explain their expertise. This is the case in many perceptual tasks, such as speech recognition, hand-writing recognition, and natural language understanding. Virtually all humans exhibit expert-level abilities on these tasks, but none of them can describe the detailed steps that they follow as they perform them. Fortunately, humans can provide machines with examples of the inputs and correct outputs for these tasks, so machine learning algorithms can learn to map the inputs to the outputs. Third, there are problems where phenomena are changing rapidly. In finance, for example, people would like to predict the future behavior of the stock market, of consumer purchases, or of exchange rates. These behaviors change frequently, so that even if a programmer could construct a good predictive computer program, it would need to be rewritten frequently. A learning program can relieve the programmer of this burden by constantly modifying and tuning a set of learned prediction rules. Fourth, there are applications that need to be customized for each computer user separately. Consider, for example, a program to filter unwanted electronic mail messages. Different users will need different filters. It is unreasonable to expect each user to program his or her own rules, and it is infeasible to provide every user with a software engineer to keep the rules up-to-date. A machine learning system can learn which mail messages the user rejects and maintain the filtering rules automatically. Machine learning addresses many of the same research questions as the fields of statistics, data mining, and psychology, but with differences of emphasis. Statistics focuses on understanding the phenomena that have generated the data, often with the goal of testing different hypotheses about those phenomena. Data mining seeks to find patterns in the data that are understandable by people. Psychological studies of human learning aspire to understand the mechanisms underlying the various learning behaviors exhibited by people (concept learning, skill acquisition, strategy change, etc.).

Machine learning

BACKGROUND: Somatoform disorders are characterised by chronic, medically unexplained physical symptoms (MUPS). Although different medications are part of treatment routines for people with somatoform disorders in clinics and private practices, there exists no systematic review or meta-analysis on the efficacy and tolerability of these medications. We aimed to synthesise to improve optimal treatment decisions.OBJECTIVES: To assess the effects of pharmacological interventions for somatoform disorders (specifically somatisation disorder, undifferentiated somatoform disorder, somatoform autonomic dysfunction, and pain disorder) in adults.SEARCH METHODS: We searched the Cochrane Depression, Anxiety and Neurosis Review Group's Specialised Register (CCDANCTR) (to 17 January 2014). This register includes relevant randomised controlled trials (RCTs) from The Cochrane Library (all years), MEDLINE (1950 to date), EMBASE (1974 to date), and PsycINFO (1967 to date). To identify ongoing trials, we searched ClinicalTrials.gov, Current Controlled Trials metaRegister, the World Health Organization International Clinical Trials Registry Platform, and the Chinese Clinical Trials Registry. For grey literature, we searched ProQuest Dissertation {\&} Theses Database, OpenGrey, and BIOSIS Previews. We handsearched conference proceedings and reference lists of potentially relevant papers and systematic reviews and contacted experts in the field.SELECTION CRITERIA: We selected RCTs or cluster RCTs of pharmacological interventions versus placebo, treatment as usual, another medication, or a combination of different medications for somatoform disorders in adults. We included people fulfilling standardised diagnostic criteria for somatisation disorder, undifferentiated somatoform disorder, somatoform autonomic dysfunction, or somatoform pain disorder.DATA COLLECTION AND ANALYSIS: One review author and one research assistant independently extracted data and assessed risk of bias. Primary outcomes included the severity of MUPS on a continuous measure, and acceptability of treatment.MAIN RESULTS: We included 26 RCTs (33 reports), with 2159 participants, in the review. They examined the efficacy of different types of antidepressants, the combination of an antidepressant and an antipsychotic, antipsychotics alone, or natural products (NPs). The duration of the studies ranged between two and 12 weeks.One meta-analysis of placebo-controlled studies showed no clear evidence of a significant difference between tricyclic antidepressants (TCAs) and placebo for the outcome severity of MUPS (SMD -0.13; 95{\%} CI -0.39 to 0.13; 2 studies, 239 participants; I(2) = 2{\%}; low-quality evidence). For new-generation antidepressants (NGAs), there was very low-quality evidence showing they were effective in reducing the severity of MUPS (SMD -0.91; 95{\%} CI -1.36 to -0.46; 3 studies, 243 participants; I(2) = 63{\%}). For NPs there was low-quality evidence that they were effective in reducing the severity of MUPS (SMD -0.74; 95{\%} CI -0.97 to -0.51; 2 studies, 322 participants; I(2) = 0{\%}).One meta-analysis showed no clear evidence of a difference between TCAs and NGAs for severity of MUPS (SMD -0.16; 95{\%} CI -0.55 to 0.23; 3 studies, 177 participants; I(2) = 42{\%}; low-quality evidence). There was also no difference between NGAs and other NGAs for severity of MUPS (SMD -0.16; 95{\%} CI -0.45 to 0.14; 4 studies, 182 participants; I(2) = 0{\%}).Finally, one meta-analysis comparing selective serotonin reuptake inhibitors (SSRIs) with a combination of SSRIs and antipsychotics showed low-quality evidence in favour of combined treatment for severity of MUPS (SMD 0.77; 95{\%} CI 0.32 to 1.22; 2 studies, 107 participants; I(2) = 23{\%}).Differences regarding the acceptability of the treatment (rate of all-cause drop-outs) were neither found between NGAs and placebo (RR 1.01, 95{\%} CI 0.64 to 1.61; 2 studies, 163 participants; I(2) = 0{\%}; low-quality evidence) or NPs and placebo (RR 0.85, 95{\%} CI 0.40 to 1.78; 3 studies, 506 participants; I(2) = 0{\%}; low-quality evidence); nor between TCAs and other medication (RR 1.48, 95{\%} CI 0.59 to 3.72; 8 studies, 556 participants; I(2) =14{\%}; low-quality evidence); nor between antidepressants and the combination of an antidepressant and an antipsychotic (RR 0.80, 95{\%} CI 0.25 to 2.52; 2 studies, 118 participants; I(2) = 0{\%}; low-quality evidence). Percental attrition rates due to adverse effects were high in all antidepressant treatments (0{\%} to 32{\%}), but low for NPs (0{\%} to 1.7{\%}).The risk of bias was high in many domains across studies. Seventeen trials (65.4{\%}) gave no information about random sequence generation and only two (7.7{\%}) provided information about allocation concealment. Eighteen studies (69.2{\%}) revealed a high or unclear risk in blinding participants and study personnel; 23 studies had high risk of bias relating to blinding assessors. For the comparison NGA versus placebo, there was relatively high imprecision and heterogeneity due to one outlier study. Although we identified 26 studies, each comparison only contained a few studies and small numbers of participants so the results were imprecise.AUTHORS' CONCLUSIONS: The current review found very low-quality evidence for NGAs and low-quality evidence for NPs being effective in treating somatoform symptoms in adults when compared with placebo. There was some evidence that different classes of antidepressants did not differ in efficacy; however, this was limited and of low to very low quality. These results had serious shortcomings such as the high risk of bias, strong heterogeneity in the data, and small sample sizes. Furthermore, the significant effects of antidepressant treatment have to be balanced against the relatively high rates of adverse effects. Adverse effects produced by medication can have amplifying effects on symptom perceptions, particularly in people focusing on somatic symptoms without medical causes. We can only draw conclusions about short-term efficacy of the pharmacological interventions because no trial included follow-up assessments. For each of the comparisons where there were available data on acceptability rates (NGAs versus placebo, NPs versus placebo, TCAs versus other medication, and antidepressants versus a combination of an antidepressant and an antipsychotic), no clear differences between the intervention and comparator were found.Future high-quality research should be carried out to determine the effectiveness of medications other than antidepressants, to compare antidepressants more thoroughly, and to follow-up participants over longer periods (the longest follow up was just 12 weeks). Another idea for future research would be to include other outcomes such as functional impairment or dysfunctional behaviours and cognitions as well as the classical outcomes such as symptom severity, depression, or anxiety.

Pharmacological interventions for somatoform disorders in adults.

Statistical procedures for missing data have vastly improved, yet misconception and unsound practice still abound. The authors frame the missing-data problem, review methods, offer advice, and raise issues that remain unresolved. They clear up common misunderstandings regarding the missing at random (MAR) concept. They summarize the evidence against older procedures and, with few exceptions, discourage their use. They present, in both technical and practical language, 2 general approaches that come highly recommended: maximum likelihood (ML) and Bayesian multiple imputation (MI). Newer developments are discussed, including some for dealing with missing data that are not MAR. Although not yet in the mainstream, these procedures may eventually extend the ML and MI methods that currently represent the state of the art.

/pdf/missing-data-our-view-of-the-state-of-the-art-3xnwc88wmd.pdf

Missing data: Our view of the state of the art.

XI. STRATEGIES FOR IMPROVING DIABETES CARE D iabetes is a chronic illness that requires continuing medical care and patient self-management education to prevent acute complications and to reduce the risk of long-term complications. Diabetes care is complex and requires that many issues, beyond glycemic control, be addressed. A large body of evidence exists that supports a range of interventions to improve diabetes outcomes. These standards of care are intended to provide clinicians, patients, researchers, payors, and other interested individuals with the components of diabetes care, treatment goals, and tools to evaluate the quality of care. While individual preferences, comorbidities, and other patient factors may require modification of goals, targets that are desirable for most patients with diabetes are provided. These standards are not intended to preclude more extensive evaluation and management of the patient by other specialists as needed. For more detailed information, refer to Bode (Ed.): Medical Management of Type 1 Diabetes (1), Burant (Ed): Medical Management of Type 2 Diabetes (2), and Klingensmith (Ed): Intensive Diabetes Management (3). The recommendations included are diagnostic and therapeutic actions that are known or believed to favorably affect health outcomes of patients with diabetes. A grading system (Table 1), developed by the American Diabetes Association (ADA) and modeled after existing methods, was utilized to clarify and codify the evidence that forms the basis for the recommendations. The level of evidence that supports each recommendation is listed after each recommendation using the letters A, B, C, or E.

Standards of Medical Care in Diabetes

PART I 1. A framework for investigating change over time 2. Exploring Longitudinal Data on Change 3. Introducing the multilevel model for change 4. Doing data analysis with the multilevel mode for change 5. Treating TIME more flexibly 6. Modelling discontinuous and nonlinear change 7. Examining the multilevel model's error covariance structure 8. Modelling change using covariance structure analysis PART II 9. A Framework for Investigating Event Occurrence 10. Describing discrete-time event occurrence data 11. Fitting basic Discrete-Time Hazard Models 12. Extending the Discrete-Time Hazard Model 13. Describing Continuous-Time Event Occurrence Data 14. Fitting Cox Regression Models 15. Extending the Cox Regression Model

Applied Longitudinal Data Analysis

In medical science, studies are often designed to investigate changes in a specific parameter which is measured repeatedly over time in the participating subjects. This allows one to model the process of change within individuals. Although this process occurs in every individual, the inter subject variability can be high. For example, using data of 955 men, Brant et al showed that the average rates of increase of systolic blood pressure (SBP) are smallest in the younger age groups, and greatest in the older age groups, that obese individuals tend to have a higher SBP than non-obese individuals, and that individuals in more recent birth cohorts have lower SBP’s than those born before 1910. However, these factors are not sufficient to explain all the heterogeneity between individuals since, after correction for age, obesity and birth cohort, individuals with SBP’s above (below) average at initial examination, still have slower (faster) rates of longitudinal change in SBP.

Linear Mixed Models for Longitudinal Data

Introduction.- Motivating Studies.- Generalized Linear Models.- Linear Mixed Models for Gaussian Longitudinal Data.- Model Families.- The Strength of Marginal Models.- Likelihood-based Models.- Generalized Estimating Equations.- Pseudo-likelihood.- Fitting Marginal Models with SAS.- Conditional Models.- Pseudo-likehood.- From Subject-Specific to Random-Effects Models.- Generalized Linear Mixed Models (GLMM).- Fitting Generalized Linear Mixed Models with SAS.- Marginal Versus Random-Effects Models.- Ordinal Data.- The Epilepsy Data.- Non-linear Models.- Psuedo-likelihood for a Hierarchical Model.- Random-effects Models with Serial Correlation.- Non-Gaussian Random Effects.- Joint Continuous and Discrete Responses.- High-dimensional Multivariate Repeated Measurements.- Missing Data Concepts.- Simple Methods, Direct Likelikhood and WGEE.- Multiple Imputation and the Expectation-Maximization Algorithm.- Selection Models.- Pattern-mixture Models.- Sensitivity Analysis.- Incomplete Data and SAS.

Models for Discrete Longitudinal Data

Chromosome instability is a hallmark of tumorigenesis. This study establishes that chromosome instability is also common during early human embryogenesis. A new array-based method allowed screening of genome-wide copy number and loss of heterozygosity in single cells. This revealed not only mosaicism for whole-chromosome aneuploidies and uniparental disomies in most cleavage-stage embryos but also frequent segmental deletions, duplications and amplifications that were reciprocal in sister blastomeres, implying the occurrence of breakage-fusion-bridge cycles. This explains the low human fecundity and identifies post-zygotic chromosome instability as a leading cause of constitutional chromosomal disorders.

/pdf/chromosome-instability-is-common-in-human-cleavage-stage-1ivaf0i5qr.pdf

Chromosome instability is common in human cleavage-stage embryos

Mixed models have become very popular for the analysis of longitudinal data, partly because they are flexible and widely applicable, partly also because many commercially available software packages offer procedures to fit them. They assume that measurements from a single subject share a set of latent, unobserved, random effects which are used to generate an association structure between the repeated measurements. In this chapter, we give an overview of frequently used mixed models for continuous as well as discrete longitudinal data, with emphasis on model formulation and parameter interpretation. The fact that the latent structures generate associations implies that mixed models are also extremely convenient for the joint analysis of longitudinal data with other outcomes such as dropout time or some time-to-event outcome, or for the analysis of multiple longitudinally measured outcomes. All models will be extensively illustrated with the analysis of real data.

/pdf/random-effects-models-for-longitudinal-data-1jcr7vk0zp.pdf

Random Effects Models for Longitudinal Data

This article investigates the impact of the normality assumption for random effects on their estimates in the linear mixed-effects model. It shows that if the distribution of random effects is a finite mixture of normal distributions, then the random effects may be badly estimated if normality is assumed, and the current methods for inspecting the appropriateness of the model assumptions are not sound. Further, it is argued that a better way to detect the components of the mixture is to build this assumption in the model and then “compare” the fitted model with the Gaussian model. All of this is illustrated on two practical examples.

https://www.jstor.org/stable/pdfplus/2291398.pdf

Geert Verbeke

Papers

Linear Mixed Models for Longitudinal Data

Models for Discrete Longitudinal Data

Chromosome instability is common in human cleavage-stage embryos

Random Effects Models for Longitudinal Data

A Linear Mixed-Effects Model with Heterogeneity in the Random-Effects Population