Subgroup finding via Bayesian additive regression trees

We provide a Bayesian decision theoretic approach to finding subgroups that have elevated treatment effects. Our approach separates the modeling of the response variable from the task of subgroup finding and allows a flexible modeling of the response variable irrespective of potential subgroups of interest. We use Bayesian additive regression trees to model the response variable and use a utility function defined in terms of a candidate subgroup and the predicted response for that subgroup. Subgroups are identified by maximizing the expected utility where the expectation is taken with respect to the posterior predictive distribution of the response, and the maximization is carried out over an a priori specified set of candidate subgroups. Our approach allows subgroups based on both quantitative and categorical covariates. We illustrate the approach using simulated data set study and a real data set. Copyright © 2017 John Wiley & Sons, Ltd.     

The role of social support and integration for understanding socioeconomic disparities in self-rated health and hypertension

This paper examines socioeconomic (socioeconomic status, SES) disparities in self-rated health and hypertension among 29,816 US adults aged 25 and older using data from the 2001 wave of the National Health Interview Survey. Our purpose is to examine how influential measures of social support and social integration are for each health outcome, and whether support and integration operate by mediating, or buffering, the effects of SES on health. Multivariate regression models show no significant influence of emotional support, but do indicate that many aspects of social integration are directly associated with self-rated health and hypertension, although these measures do not mediate the relationship between SES and health. However, interaction tests show substantial evidence that measures of social integration buffer some of the negative effects of low SES, particularly the negative influence of not working on self-rated health. In addition, findings indicate potential evidence of help-seeking behavior among adults who did not finish high school or who report financial barriers to medical care. Overall, our findings suggest that social programs designed to foster social integration (e.g., free or low-cost bus fare to promote visits with friends and family) may improve health among persons with low levels of education, who are not working, or who have problems accessing medical care because of financial problems.     

Informatics for Neurocritical Care: Challenges and Opportunities

Neurocritical care relies on the continuous, real-time measurement of numerous physiologic parameters. While our capability to obtain such measurements from patients has grown markedly with multimodal monitoring in many neurologic or neurosurgical intensive care units (ICUs), our ability to transform the raw data into actionable information is limited. One reason is that the proprietary nature of medical devices and software often prevents neuro-ICUs from capturing and centrally storing high-density data. Also, ICU alarm systems are often unreliable because the data that are captured are riddled with artifacts. Informatics is the process of acquiring, processing, and interpreting these complex arrays of data. The development of next-generation informatics tools allows for detection of complex physiologic events and brings about the possibility of decision support tools to improve neurocritical care. Although many different approaches to informatics are discussed and considered, here we focus on the Bayesian probabilistic paradigm. It quantifies the uncertainty inherent in neurocritical care instead of ignoring it, and formalizes the natural clinical thought process of updating prior beliefs using incoming patient data. We review this and other opportunities, as well as challenges, for the development and refinement of informatics tools in neurocritical care.     

Statistical assessment of mediational effects for logistic mediational models

The concept of mediation has broad applications in medical health studies. Although the statistical assessment of a mediational effect under the normal assumption has been well established in linear structural equation models (SEM), it has not been extended to the general case where normality is not a usual assumption. In this paper, we propose to extend the definition of mediational effects through causal inference. The new definition is consistent with that in linear SEM and does not rely on the assumption of normality. Here, we focus our attention on the logistic mediation model, where all variables involved are binary. Three approaches to the estimation of mediational effects-Delta method, bootstrap, and Bayesian modelling via Monte Carlo simulation are investigated. Simulation studies are used to examine the behaviour of the three approaches. Measured by 95 per cent confidence interval (CI) coverage rate and root mean square error (RMSE) criteria, it was found that the Bayesian method using a non-informative prior outperformed both bootstrap and the Delta methods, particularly for small sample sizes. Case studies are presented to demonstrate the application of the proposed method to public health research using a nationally representative database. Extending the proposed method to other types of mediational model and to multiple mediators are also discussed.     

Association of magnetic resonance imaging-based knee cartilage T2 measurements and focal knee lesions with knee pain: data from the Osteoarthritis Initiative

Objective

To evaluate the association of magnetic resonance imaging (MRI)–based knee cartilage T2 measurements and focal knee lesions with knee pain in knees without radiographic osteoarthritis (OA) among subjects with OA risk factors.

Methods

We studied the right knees of 126 subjects from the Osteoarthritis Initiative database. We randomly selected 42 subjects ages 45–55 years with OA risk factors, right knee pain (Western Ontario and McMaster Universities Osteoarthritis Index [WOMAC] pain score ≥5), no left knee pain (WOMAC pain score 0), and no radiographic OA (Kellgren/Lawrence [K/L] score ≤1) in the right knee. We also selected 2 comparison groups: 42 subjects without knee pain in either knee and 42 with bilateral knee pain. Both groups were frequency matched to subjects with right knee pain only by sex, age, body mass index, and K/L score. All of the subjects underwent 3T MRI of the right knee. Focal knee lesions were assessed and cartilage T2 measurements were performed.

Results

Prevalences of meniscal, bone marrow, and ligamentous lesions and joint effusion were not significantly different between the groups (P > 0.05), while cartilage lesions were more frequent in subjects with right knee pain only compared to subjects without knee pain (P < 0.05). T2 values averaged over all of the compartments were similar in subjects with right knee pain only (mean ± SD 34.4 ± 1.8 msec) and in subjects with bilateral knee pain (mean ± SD 34.7 ± 4.7 msec), but were significantly higher compared to subjects without knee pain (mean ± SD 32.4 ± 1.8 msec; P < 0.05).

Conclusion

These results suggest that elevated cartilage T2 values are associated with findings of pain in the early phase of OA, whereas among morphologic knee abnormalities only knee cartilage lesions are significantly associated with knee pain status.     

Correlation of magnetic resonance imaging–based knee cartilage T2 measurements and focal knee lesions with body mass index: Thirty‐six–month followup data from a longitudinal,observational multicenter study

Objective

To compare magnetic resonance imaging (MRI)–based knee cartilage T2 measurements and focal knee lesions and 36‐month changes in these parameters among knees of normal controls and knees of normal weight, overweight, and obese subjects with risk factors for knee osteoarthritis (OA).

Methods

A total of 267 subjects ages 45–55 years from the Osteoarthritis Initiative database were analyzed in this study. Two hundred thirty‐one subjects had risk factors for knee OA, but no radiographic OA (Kellgren/Lawrence score ≤1) at baseline. Thirty‐six subjects were normal controls. Subjects with OA risk factors were stratified in 3 groups: normal weight (n = 78), overweight (n = 84), and obese (n = 69). All subjects underwent 3T MRI of the right knee at baseline and after 36 months. Focal knee lesions were assessed and cartilage T2 measurements (mean T2 and T2 texture analysis) were performed.

Results

The baseline prevalence and severity of meniscal and cartilage lesions were highest in obese subjects and lowest in normal controls (P < 0.05). Obese subjects had the highest mean T2 values and the most heterogeneous cartilage (as assessed by T2 texture analysis), while normal controls had the lowest mean T2 values and the most homogeneous cartilage at baseline (P < 0.05). Increased body mass index (BMI) was significantly (P < 0.05) associated with greater progression of cartilage lesions and constantly elevated cartilage T2 entropy over 36 months.

Conclusion

In preclinical OA, increased BMI is associated with more severe cartilage degeneration as assessed by both morphologic and quantitative MRI measurements.     

An objective Bayesian analysis of a crossover design via model selection and model averaging

Inference about the treatment effect in a crossover design has received much attention over time owing to the uncertainty in the existence of the carryover effect and its impact on the estimation of the treatment effect. Adding to this uncertainty is that the existence of the carryover effect and its size may depend on the presence of the treatment effect and its size. We consider estimation and testing hypothesis about the treatment effect in a two‐period crossover design, assuming normally distributed response variable, and use an objective Bayesian approach to test the hypothesis about the treatment effect and to estimate its size when it exists while accounting for the uncertainty about the presence of the carryover effect as well as the treatment and period effects. We evaluate and compare the performance of the proposed approach with a standard frequentist approach using simulated data, and real data. Copyright © 2016 John Wiley & Sons, Ltd.     

Liver detection algorithm: its efficacy for CT noise reduction in the liver

Objectives

This study was performed to evaluate the efficacy of a novel computed tomography (CT) liver detection algorithm (LDA), which allows for targeted increase of radiation dose to the upper abdomen, on image quality of the liver.

Methods

We retrospectively evaluated the LDA by comparing 40 consecutive patients who had portal venous CT abdomen performed without use of the algorithm, to 40 patients in whom the algorithm was used. Image quality was assessed objectively by comparing the standard deviation (SD) of attenuation values in Hounsfield units (HU) of the abdominal organs. Qualitative analysis was performed by two blinded radiologists who independently graded the image quality of abdominal organs

Results

There was significant noise reduction in the liver (P < 0.001) and spleen (P < 0.001) in the LDA group compared to the conventional group. There was also a significant improvement in image quality of the liver (P < 0.001), kidney (P < 0.001), spleen (P < 0.001), pancreas (P < 0.001), and psoas (P = 0.005) in the LDA group compared to the conventional group. Overall dose between the two groups was similar.

Conclusions

This liver detection algorithm improves the subjective image quality of upper abdominal organs, in particular the liver, without increasing overall radiation dose.