A psychological approach to learning causal networks

We examine the role of a common cognitive heuristic in unsupervised learning of Bayesian probability networks from data. Human beings perceive a larger association between causal than diagnostic relationships. This psychological principal can be used to orient the arcs within Bayesian networks by prohibiting the direction that is less predictive. The heuristic increased predictive accuracy by an average of 0.51 % percent, a small amount. It also increased total agreement between different network learning algorithms (Max Spanning Tree, Taboo, EQ, SopLeq, and Taboo Order) by 25 %. Prior to use of the heuristic, the multiple raters Kappa between the algorithms was 0.60 (95 % confidence interval, CI, from 0.53 to 0.67) indicating moderate agreement among the networks learned through different algorithms. After the use of the heuristic, the multiple raters Kappa was 0.85 (95 % CI from 0.78 to 0.92). There was a statistically significant increase in agreement between the five algorithms (alpha <?0.05). These data suggest that the heuristic increased agreement between networks learned through use of different algorithms, without loss of predictive accuracy. Additional research is needed to see if findings persist in other data sets and to explain why a heuristic used by humans could improve construct validity of mathematical algorithms.     

A weight of evidence approach to causal inference

ObjectiveThe Bradford Hill criteria are the best available criteria for causal inference. However, there is no information on how the criteria should be weighed and they cannot be combined into one probability estimate for causality. Our objective is to provide an empirical basis for weighing the Bradford Hill criteria and to develop a transparent method to estimate the probability for causality.Study Design and SettingAll 159 agents classified by International Agency for Research of Cancer as category 1 or 2A carcinogens were evaluated by applying the nine Bradford Hill criteria. Discriminant analysis was used to estimate the weights for each of the nine Bradford Hill criteria.ResultsThe discriminant analysis yielded weights for the nine causality criteria. These weights were used to combine the nine criteria into one overall assessment of the probability that an association is causal. The criteria strength, consistency of the association and experimental evidence were the three criteria with the largest impact. The model correctly predicted 130 of the 159 (81.8%) agents.ConclusionThe proposed approach enables using the Bradford Hill criteria in a quantitative manner resulting in a probability estimate of the probability that an association is causal.     

A Bayesian semiparametric latent variable approach to causal mediation

In assessing causal mediation effects in randomized studies, a challenge is that the direct and indirect effects can vary across participants due to different measured and unmeasured characteristics. In that case, the population effect estimated from standard approaches implicitly averages over and does not estimate the heterogeneous direct and indirect effects. We propose a Bayesian semiparametric method to estimate heterogeneous direct and indirect effects via clusters, where the clusters are formed by both individual covariate profiles and individual effects due to unmeasured characteristics. These cluster‐specific direct and indirect effects can be estimated through a set of regression models where specific coefficients are clustered by a stick‐breaking prior. To let clustering be appropriately informed by individual direct and indirect effects, we specify a data‐dependent prior. We conduct simulation studies to assess performance of the proposed method compared to other methods. We use this approach to estimate heterogeneous causal direct and indirect effects of an expressive writing intervention for patients with renal cell carcinoma.     

A unified approach to age-dependent metabolic modeling

A generalized approach for estimating the effect of age on deposition fractions and retention half-times is presented. The theory employs measured physiological and anatomical parameters and relates these to metabolic parameters within a general framework intended for use in setting secondary limits for exposure of the general public. Calculated results indicate that younger ages will be characterized by increased deposition fractions and decreased half-times for retention following radionuclide uptakes. Representative examples are provided for radionuclides of current interest in health physics practice.     

Tutorial: causal modeling and patient satisfaction

Causal modeling is used in a variety of sciences because it allows exploration of complex relationships among several variables simultaneously. Although not used extensively in health care as yet, causal modeling could be helpful, given the complexity of the current health care system. The purpose of this article is to provide a general introduction to causal modeling and the syntax used in developing and testing a model. To illustrate the method, a sample model is tested using a combination of hypothetical and actual patient satisfaction data.     

Pharmacological and residual effects in randomized placebo-controlled trials. A structural causal modelling approach

BackgroundDistinguishing between pharmacological and residual effects, this paper considers the problem of causal assessment in the case of a particular model, namely a Sure Outcome of Random Events (SORE) model developed for the analysis of data from a randomized placebo-controlled double-blind trial of a drug.MethodThis model takes into account two kinds of observable effects, a therapeutic effect and a side-effect. For each observable effect, two latent factors are considered, i.e. a pharmacological (or explained) factor and a residual (or unexplained) one.ResultsThe model presents a plausible mechanism generating the observed and latent outcomes, recursively decomposed into an ordered sequence of sub-mechanisms.ConclusionsThe characteristics of this model leads to a novel assessment of causality that evaluates the effect of latent variables and of the bias resulting from ignoring the structural features of the data generating process. This approach is illustrated by a numerical example, along with a case study based on a secondary analysis of real data.     

Bayesian graphical modeling for heterogeneous causal effects

There is a growing interest in current medical research to develop personalized treatments using a molecular-based approach. The broad goal is to implement a more precise and targeted decision-making process, relative to traditional treatments based primarily on clinical diagnoses. Specifically, we consider patients affected by Acute Myeloid Leukemia (AML), an hematological cancer characterized by uncontrolled proliferation of hematopoietic stem cells in the bone marrow. Because AML responds poorly to chemotherapeutic treatments, the development of targeted therapies is essential to improve patients' prospects. In particular, the dataset we analyze contains the levels of proteins involved in cell cycle regulation and linked to the progression of the disease. We evaluate treatment effects within a causal framework represented by a Directed Acyclic Graph (DAG) model, whose vertices are the protein levels in the network. A major obstacle in implementing the above program is represented by individual heterogeneity. We address this issue through a Dirichlet Process (DP) mixture of Gaussian DAG-models where both the graphical structure as well as the allied model parameters are regarded as uncertain. Our procedure determines a clustering structure of the units reflecting the underlying heterogeneity, and produces subject-specific estimates of causal effects based on Bayesian Model Averaging (BMA). With reference to the AML dataset, we identify different effects of protein regulation among individuals; moreover, our method clusters patients into groups that exhibit only mild similarities with traditional categories based on morphological features.     

A robustified modeling approach to analyze pediatric length of stay

PURPOSE: Length of stay (LOS) is an important measure of the cost of pediatric hospitalizations, but the guidelines developed so far are not rigorously evidence-based. This study demonstrates a robust gamma mixed regression approach to analyze the positively skewed LOS variable, which has implications for future studies of pediatric health care management. METHODS: The robustified approach is applied to analyze hospital discharge data on childhood gastroenteritis in Western Australia (n=514). The model accounts for demographic characteristics and co-morbidities of the patients, as well as the dependency of LOS outcomes nested within the 58 hospitals in the State. The method is compared with the standard linear mixed regression with trimming of extreme observations. RESULTS: For the empirical application, the linear mixed regression results are sensitive to the magnitude of trimming. The identified significant factors from the robust regression model, namely infection, failure to thrive, and iron deficiency anemia are resistant to high-LOS outliers. CONCLUSIONS: Robust gamma mixed regression appears to be a suitable alternative to analyze the clustered and positively skewed pediatric LOS, without transforming and trimming the data arbitrarily.     

A Bayesian approach to net health benefits: an illustration and application to modeling HIV prevention.

PURPOSE: To conduct a cost-effectiveness analysis of HIV prevention when costs and effects cannot be measured directly. To quantify the total estimation of uncertainty due to sampling variability as well as inexact knowledge of HIV transmission parameters. METHODS: The authors focus on estimating the incremental net health benefit (INHB) in a randomized trial of HIV prevention with intervention and control conditions. Using a Bernoulli model of HIV transmission, changes in the participants' risk behaviors are converted into the number of HIV infections averted. A sampling model is used to account for variation in the behavior measurements. Bayes's theorem and Monte Carlo methods are used to attain the stated objectives. RESULTS: The authors obtained a positive mean INHB of 0.0008, indicating that advocacy training is just slightly favored over the control condition for men, assuming a $50,000 per quality-adjusted life year (QALY) threshold. To be confident of a positive INHB, the decision maker would need to spend more than $100,000 per QALY.     

A causal modelling approach to the development of theory-based behaviour change programmes for trial evaluation

Theory-based intervention programmes to support health-related behaviour change aim to increase health impact and improve understanding of mechanisms of behaviour change. However, the science of intervention development remains at an early stage. We present a causal modelling approach to developing complex interventions for evaluation in randomized trials. In this approach a generic model links behavioural determinants, causally through behaviour, to physiological and biochemical variables, and health outcomes. It is tailored to context, target population, behaviours and health outcomes. The development of a specific causal model based on theory and evidence is illustrated by the ProActive programme, supporting increased physical activity among individuals at risk of Type 2 diabetes. The model provides a rational guide to appropriate measures, intervention points and intervention techniques, and can be tested quantitatively. Causal modelling is critically compared to other approaches to intervention development and evaluation, and research directions are indicated.     

A Bayesian approach to estimating causal vaccine effects on binary post‐infection outcomes

To estimate causal effects of vaccine on post‐infection outcomes, Hudgens and Halloran (2006) defined a post‐infection causal vaccine efficacy estimand VE_I based on the principal stratification framework. They also derived closed forms for the maximum likelihood estimators of the causal estimand under some assumptions. Extending their research, we propose a Bayesian approach to estimating the causal vaccine effects on binary post‐infection outcomes. The identifiability of the causal vaccine effect VE_I is discussed under different assumptions on selection bias. The performance of the proposed Bayesian method is compared with the maximum likelihood method through simulation studies and two case studies — a clinical trial of a rotavirus vaccine candidate and a field study of pertussis vaccination. For both case studies, the Bayesian approach provided similar inference as the frequentist analysis. However, simulation studies with small sample sizes suggest that the Bayesian approach provides smaller bias and shorter confidence interval length. Copyright © 2015 John Wiley & Sons, Ltd.     

A sensitivity approach to modeling longitudinal bivariate ordered data subject to informative dropouts

Incomplete data abound in epidemiological and clinical studies. When the missing data process is not properly investigated, inferences may be misleading. An increasing number of models that incorporate nonrandom incomplete data have become available. At the same time, however, serious doubts have arisen about the validity of these models, known to rely on strong and unverifiable assumptions. A common conclusion emerging from the current literature is the clear need for a sensitivity analysis. We propose in this paper a detailed sensitivity analysis using graphical and analytical techniques to understand the impact of missing-data assumptions on inferences. Specifically, we explore the influence of perturbing a missing at random model locally in the direction of non-random dropout models. Data from a psychiatric trial are used to illustrate the methodology.     

Vertical modeling: A pattern mixture approach for competing risks modeling

We study an alternative approach for estimation in the competing risks framework, called vertical modeling. It is motivated by a decomposition of the joint distribution of time and cause of failure. The two elements of this decomposition are (1) the time of failure and (2) the cause of failure condition on time of failure. Both elements of the model are based on observable quantities, namely the total hazard and the relative cause‐specific hazards. The model can be implemented using the standard software. The relative cause‐specific hazards are flexibly estimated using multinomial logistic regression and smoothing splines. We show estimates of cumulative incidences from vertical modeling to be more efficient statistically than those obtained from the standard nonparametric model. We illustrate our methods using data of 8966 leukemia patients from the European Group for Blood and Marrow Transplantation. Copyright © 2010 John Wiley & Sons, Ltd.     

The future of measles in highly immunized populations. A modeling approach

Little is known about how an intensive measles elimination program changes the overall immune status of the population. A computer model was created to study the effect of the measles elimination program in the United States on the number of susceptibles in the population. The simulation reveals that in the prevaccine era, approximately 10.6% of the population was susceptible to measles, most of whom were children less than 10 years of age. With the institution of the measles immunization program, the proportion of susceptibles in the population fell to 3.1% from 1978 through 1981, but then began to rise by approximately 0.1% per year to reach about 10.9% in the year 2050. The susceptibles at this time were distributed evenly throughout all age groups. The model did not consider the potential effect of waning immunity. The results of this study suggest that measles elimination in the United States has been achieved by an effective immunization program aimed at young susceptibles combined with a highly, naturally immunized adult population. However, despite short-term success in eliminating the disease, long-range projections demonstrate that the proportion of susceptibles in the year 2050 may be greater than in the prevaccine era. Present vaccine technology and public health policy must be altered to deal with this eventuality.     

A Bayesian approach to modeling the interaction between air pollution and temperature

PurposeInvestigating the interaction between particulate matter air pollution (PM) and temperature is important for quantifying the effects of PM on mortality. One approach is stratification—estimating the effect of PM within different temperature strata—but this treats the cutpoints that define the strata as fixed, when in fact they are unknown. The purpose of this paper is to propose a new approach that appropriately accounts for uncertainty regarding the cutpoints, and to apply this approach to data from two Australian cities.MethodsWe propose a Bayesian model which allows the effects of PM to differ within different temperature strata. The cutpoints that define the strata are parameters that are jointly estimated along with the other model parameters. This is in contrast with the standard stratification approach, where cutpoints are specified a priori and treated as fixed constants. Also, the Bayesian model is formulated in a way that ensures continuity in the effects of PM at the stratum boundaries. Markov chain Monte Carlo methods are used to perform the inferences.ResultsAnalysis of daily data over several years provides evidence for an interactive effect between PM and temperature in Sydney and no support for such an effect in Melbourne.ConclusionsThe proposed Bayesian model provides a means for investigating interactions between PM and temperature which appropriately incorporates uncertainty.     

Challenges to institutionalization: a women-in-development perspective

Institutionalization, or sustainable program development, isa prime concern of donor agencies, governments and practitioners.How can host country institutions apply a project methodologyin an ongoing way as part of normal routine once expatriatetechnical assistance and funding is withdrawn? The questionis particularly pertinent within the health sector, and increasinglyrepresents a quid pro quo on the part of large bilateral donors.The issue becomes even more complex when viewed from a frequentlyoverlooked women-in-development perspective. This paper reviewsthe scant literature, articulates a set of issues regardinginstitutionalization and discusses challenges to sustained behaviorchange from a women-in-development vantage point.     

Bulimia as a manifestation of the stress process: A LISREL causal modeling analysis

Two studies conducted to develop and cross validate a causal model of bulimia are reported. It was hypothesized that a stress process, comprised of three components, the sources of stress, the mediators of stress, and the manifestations of stress would provide the basis of a causal model of bulimia, using linear structural relations analysis (LISREL). In Study 1, 144 female introductory psychology students were assessed for bulimia as well as 10 other variables representative of the stress process. The proposed model was comprised of two sources of stress: Environmental stressors (life events and daily hassles), and Depression (depression and risk for depression). The model indicated that coping skills are an important mediator of stress, and that having a high frequency of environmental stressors and/or the presence of depression or risk for depression, may lead an individual to resort to ineffective coping mechanisms, which, in turn, results in the expression of bulimic behavior. The model was cross-validated in Study 2, in which 150 female introductory psychology students were assessed on the same variables. Treatment implications are discussed.