Binary partitioning for continuous longitudinal data: categorizing a prognostic variable

Subgroup detection in linear growth curve models with generalized linear mixed model (GLMM) trees

Growth curve models are popular tools for studying the development of a response variable within subjects over time. Heterogeneity between subjects is common in such models, and researchers are typically interested in explaining or predicting this heterogeneity. We show how generalized linear mixed-effects model (GLMM) trees can be used to identify subgroups with different trajectories in linear growth curve models. Originally developed for clustered cross-sectional data, GLMM trees are extended here to longitudinal data. The resulting extended GLMM trees are directly applicable to growth curve models as an important special case. In simulated and real-world data, we assess performance of the extensions and compare against other partitioning methods for growth curve models. Extended GLMM trees perform more accurately than the original algorithm and LongCART, and similarly accurate compared to structural equation model (SEM) trees. In addition, GLMM trees allow for modeling both discrete and continuous time series, are less sensitive to (mis-)specification of the random-effects structure and are much faster to compute.

Detecting Cohort Effects in Accelerated Longitudinal Designs Using Multilevel Models

Accelerated longitudinal designs allow researchers to efficiently collect longitudinal data covering a time span much longer than the study duration. One important assumption of these designs is that each cohort (a group defined by their age of entry into the study) shares the same longitudinal trajectory. Although previous research has examined the impact of violating this assumption when each cohort is defined by a single age of entry, it is possible that each cohort is instead defined by a range of ages, such as groups that experience a particular historical event. In this paper we examined how including cohort membership in linear and quadratic multilevel models performed in detecting and controlling for cohort effects in this scenario. Using a Monte Carlo simulation study, we assessed the performance of this approach under conditions related to the number of cohorts, the overlap between cohorts, the strength of the cohort effect, the number of affected parameters, and the sample size. Our results indicate that models including a proxy variable for cohort membership based on age at study entry performed comparably to using true cohort membership in detecting cohort effects accurately and returning unbiased parameter estimates. This indicates that researchers can control for cohort effects even when true cohort membership is unknown.

A review on longitudinal data analysis with random forest

In longitudinal studies variables are measured repeatedly over time, leading to clustered and correlated observations. If the goal of the study is to develop prediction models, machine learning approaches such as the powerful random forest (RF) are often promising alternatives to standard statistical methods, especially in the context of high-dimensional data. In this paper, we review extensions of the standard RF method for the purpose of longitudinal data analysis. Extension methods are categorized according to the data structures for which they are designed. We consider both univariate and multivariate response longitudinal data and further categorize the repeated measurements according to whether the time effect is relevant. Even though most extensions are proposed for low-dimensional data, some can be applied to high-dimensional data. Information of available software implementations of the reviewed extensions is also given. We conclude with discussions on the limitations of our review and some future research directions.

Effect of CD34+ Cell Dose on the Outcomes of Allogeneic Stem Cell Transplantation with Post-Transplantation Cyclophosphamide

The impact of infused CD34⁺ cell dose on outcomes after allogeneic hematopoietic stem cell transplantation (alloHSCT) using standard graft-versus-host disease (GVHD) prophylaxis remains controversial. Information on this subject is scarce for alloHSCT using high-dose post-transplantation cyclophosphamide (PTCy). We aimed to assess the effect of CD34⁺ cell dose in peripheral blood stem cell (PBSC) grafts on the outcome of alloHSCT using PTCy-based GVHD prophylaxis. To do so, we conducted a single-center retrospective analysis of 221 consecutive adult patients who underwent PTCy alloHSCT from HLA-matched sibling donors (MSDs; n = 22), HLA-matched unrelated donors (MUDs; n = 83), mismatched unrelated donors (MMUDs; n = 73), and haploidentical donors (n = 43). Based on the binary partitioning method, 5 × 10⁶/kg was used as the optimal cutoff for CD34⁺ cell dose. According to our institutional protocol, the maximum CD34⁺ cell dose was capped at 8 × 10⁶/kg. The study cohort was divided into 2 groups based on CD34⁺ cell dose: high dose (>5 to 8 × 10⁶/kg) and low dose (≤5 × 10⁶/kg). Patients receiving high-dose CD34⁺-containing grafts had significantly shorter median times to neutrophil engraftment and platelet engraftment compared to those who received low-dose CD34+ (19 days versus 21 days [P = .002] and 16 days versus 22 days [P = .04], respectively). There were no differences between the high-dose and low-dose groups in the cumulative incidence of day +100 acute GVHD (grade II-IV: 25% versus 23% [P = .7]; grade III-IV: 5% versus 4% [P = .4], respectively) or 2-year chronic GVHD (moderate/severe GVHD: 9% versus 6%; P = .5). There was no impact of CD34⁺ cell dose on survival outcomes with the use of MSDs, MUDs, or MMUDs. Recipients of haploidentical alloHSCT using low-dose CD34⁺ cells had significantly worse overall survival (hazard ratio [HR], 6.01; P = .004) and relapse-free survival (HR, 4.57; P = .004). In recipients of PBSC PTCy alloHSCT, infused CD34⁺ cell doses >5 to 8 × 10⁶/kg were associated with faster neutrophil and platelet engraftment, independent of donor type. Our study suggests an impact of CD34⁺ cell dose on survival outcomes only with haploidentical donors, for whom the administration of a CD34⁺ cell dose ≤5 × 10⁶/kg significantly decreased survival outcomes.

Latent profiles of biological dysregulation and risk of mortality: time-to-event analysis using the Midlife in the US longitudinal study

BACKGROUND

There is a well-established relationship between high allostatic load (AL) and increased risk of mortality. This study expands on the literature by combined latent profile analysis (LPA) with survival data analysis techniques to assess the degree to which AL status is associated with time to death.

METHODS

LPA was employed to identify underlying classes of biological dysregulation among a sample of 815 participants from the Midlife in the US study. Sex-stratified Cox proportional hazards regression models were used to estimate the association between class of biological dysregulation and time to death while controlling for sociodemographic covariates.

RESULTS

The LPA resulted in three classes: low dysregulation, immunometabolic dysregulation and parasympathetic reactivity. Women in the immunometabolic dysregulation group had more than three times the risk of death as compared with women in the low dysregulation group (HR=3.25, 95% CI: 1.47 to 7.07), but that there was not a statistically significant difference between the parasympathetic reactivity group and the low dysregulation group (HR=1.80, 95% CI: 0.62 to 5.23). For men, the risk of death for those in the immunometabolic dysregulation (HR=1.79, 95% CI: 0.88 to 3.65) and parasympathetic reactivity (HR=0.90, 95% CI: 0.34 to 3.65) groups did not differ from the low dysregulation group.

CONCLUSION

The findings are consistent with the previous research that demonstrates increased AL as a risk factor for mortality. Specifically, in women, that increased risk may be associated with immunometabolic dysregulation and not simply a generalised measure of cumulative risk as is typically employed in AL research.

Mixed-effect models with trees

Tree-based regression models are a class of statistical models for predicting continuous response variables when the shape of the regression function is unknown. They naturally take into account both non-linearities and interactions. However, they struggle with linear and quasi-linear effects and assume iid data. This article proposes two new algorithms for jointly estimating an interpretable predictive mixed-effect model with two components: a linear part, capturing the main effects, and a non-parametric component consisting of three trees for capturing non-linearities and interactions among individual-level predictors, among cluster-level predictors or cross-level. The first proposed algorithm focuses on prediction. The second one is an extension which implements a post-selection inference strategy to provide valid inference. The performance of the two algorithms is validated via Monte Carlo studies. An application on INVALSI data illustrates the potentiality of the proposed approach.

Explorations of Individual Change Processes and Their Determinants: A Novel Approach and Remaining Challenges

Over the past 40 years there have been great advances in the analysis of individual change and the analyses of between-person differences in change. While conditional growth models are the dominant approach, exploratory models, such as growth mixture models and structural equation modeling trees, allow for greater flexibility in the modeling of between-person differences in change. We continue to push for greater flexibility in the modeling of individual change and its determinants by combining growth mixture modeling with structural equation modeling trees to evaluate how measured covariates predict class membership using a recursive partitioning algorithm. This approach, referred to as growth mixture modeling with membership trees, is illustrated with longitudinal reading data from the Early Childhood Longitudinal Study with the MplusTrees package in R.

Functional random forests for curve response

The rapid advancement of functional data in various application fields has increased the demand for advanced statistical approaches that can incorporate complex structures and nonlinear associations. In this article, we propose a novel functional random forests (FunFor) approach to model the functional data response that is densely and regularly measured, as an extension of the landmark work of Breiman, who introduced traditional random forests for a univariate response. The FunFor approach is able to predict curve responses for new observations and selects important variables from a large set of scalar predictors. The FunFor approach inherits the efficiency of the traditional random forest approach in detecting complex relationships, including nonlinear and high-order interactions. Additionally, it is a non-parametric approach without the imposition of parametric and distributional assumptions. Eight simulation settings and one real-data analysis consistently demonstrate the excellent performance of the FunFor approach in various scenarios. In particular, FunFor successfully ranks the true predictors as the most important variables, while achieving the most robust variable sections and the smallest prediction errors when comparing it with three other relevant approaches. Although motivated by a biological leaf shape data analysis, the proposed FunFor approach has great potential to be widely applied in various fields due to its minimal requirement on tuning parameters and its distribution-free and model-free nature. An R package named 'FunFor', implementing the FunFor approach, is available at GitHub.

A random forest method with feature selection for developing medical prediction models with clustered and longitudinal data

BACKGROUND

Machine learning methodologies are gaining popularity for developing medical prediction models for datasets with a large number of predictors, particularly in the setting of clustered and longitudinal data. Binary Mixed Model (BiMM) forest is a promising machine learning algorithm which may be applied to develop prediction models for clustered and longitudinal binary outcomes. Although machine learning methods for clustered and longitudinal methods such as BiMM forest exist, feature selection has not been analyzed via data simulations. Feature selection improves the practicality and ease of use of prediction models for clinicians by reducing the burden of data collection. Thus, feature selection procedures are not only beneficial, but are often necessary for development of medical prediction models. In this study, we aim to assess feature selection within the BiMM forest setting for modeling clustered and longitudinal binary outcomes.

METHODS

We conducted a simulation study to compare BiMM forest with feature selection (backward elimination or stepwise selection) to standard generalized linear mixed model feature selection methods (shrinkage and backward elimination). We also evaluated feature selection methods to develop models predicting mobility disability in older adults using the Health, Aging and Body Composition Study dataset as an example utilization of the proposed methodology.

RESULTS

BiMM forest with backward elimination generally offered higher computational efficiency, similar or higher predictive performance (accuracy and area under the receiver operating curve), and similar or higher ability to identify correct features compared to linear methods for the different simulated scenarios. For predicting mobility disability in older adults, methods generally performed similarly in terms of accuracy, area under the receiver operating curve, and specificity; however, BiMM forest with backward elimination had the highest sensitivity.

CONCLUSIONS

This study is novel because it is the first investigation of feature selection for developing random forest prediction models for clustered and longitudinal binary outcomes. Results from the simulation study reveal that BiMM forest with backward elimination has the highest accuracy (performance and identification of correct features) and lowest computation time compared to other feature selection methods in some scenarios and similar performance in other scenarios. Many informatics datasets have clustered and longitudinal outcomes and results from this study suggest that BiMM forest with backward elimination may be beneficial for developing medical prediction models.

Score-based tests for detecting heterogeneity in linear mixed models

Cross-level interactions among fixed effects in linear mixed models (also known as multilevel models) can be complicated by heterogeneity stemming from random effects and residuals. When heterogeneity is present, tests of fixed effects (including cross-level interaction terms) are subject to inflated type I or type II error. While the impact of variance change/heterogeneity has been noticed in the literature, few methods have been proposed to detect this heterogeneity in a simple, systematic way. In addition, when heterogeneity among clusters is detected, researchers often wish to know which clusters' variances differed from the others. In this study, we utilize a recently proposed family of score-based tests to distinguish between cross-level interactions and heterogeneity in variance components, also providing information about specific clusters that exhibit heterogeneity. These score-based tests only require estimation of the null model (when variance homogeneity is assumed to hold), and they have been previously applied to psychometric models to detect measurement invariance. In this paper, we extend the tests to linear mixed models, allowing us to use the tests to differentiate between interaction and heterogeneity. We detail the tests' implementation and performance via simulation, provide an empirical example of the tests' use in practice, and provide code illustrating the tests' general application.

Refining Severe Tricuspid Regurgitation Definition by Echocardiography with a New Outcomes-Based "Massive" Grade

BACKGROUND

Current echocardiographic guidelines recommend that tricuspid regurgitation (TR) severity be graded in three categories, following assessment of specific parameters. Findings from recent trials have shown that the severity of TR frequently far exceeds the current definition of severe. We postulated that a grading approach that emphasizes outcomes could be useful to identify patients with severe TR at increased risk of mortality.

METHODS

We identified 284 patients with echocardiograms demonstrating severe functional TR, defined as vena contracta (VC) ≥ 0.7 cm. Demographics and mortality data were obtained from the medical records. Patients were divided into study (n = 122 patients with three-dimensional images) and validation (n = 162) cohorts. The VC was measured in both the right ventricular (RV) inflow and apical four-chamber views and averaged. For the study cohort, tricuspid annular, RV end-diastolic (basal, mid, long axis) dimensions, tricuspid leaflet tenting height and area, RV free-wall longitudinal strain, and RV volumes were measured from two- and three-dimensional data sets. A K-partition algorithm was used in the study cohort to derive a mortality-related cutoff VC value, above which TR was termed "massive." The ability of this VC cutoff to identify patients at greater mortality risk was then tested in the validation cohort using Kaplan-Meier survival analysis.

RESULTS

In the study cohort, VC > 0.92 cm (massive TR) was optimally associated with worse survival. Tricuspid annular and RV size were larger in the massive group (P < .05), while there were no significant differences in demographics between the TR groups. Importantly, in the independent validation cohort, the above VC cutoff also correlated with increased mortality in the massive group (log-rank P < .05).

CONCLUSIONS

Among patients traditionally defined as having severe TR, a subset exists with massive TR, resulting in greater adverse RV remodeling and increased mortality. These patients may derive the greatest benefit from emerging percutaneous therapies.

The estimated prevalence of exposure to carcinogens, asthmagens, and ototoxic agents among healthcare workers in Australia

BACKGROUND

Healthcare workers are occupationally exposed to various hazardous chemicals and agents that can potentially result in long-term adverse health effects. These exposures have not been comprehensively examined at a population level. The aim of this study was to examine occupational exposures to a wide range of asthmagens, carcinogens, and ototoxic agents among healthcare workers in Australia.

METHODS

Data were collected as part of the Australian Work Exposures Studies, which were computer-assisted telephone surveys conducted in 2011, 2014, and 2016 to assess the prevalence of occupational exposures to carcinogens, asthmagens, and ototoxic agents, respectively, among Australian workers. Using data on healthcare workers, the prevalence of exposures to these agents was calculated and associations of demographic variables and occupation groups with exposure status were examined.

RESULTS

The prevalence of exposure to at least one asthmagen, carcinogen, and ototoxic agent was 92.3%, 50.7%, and 44.6%, respectively. The most common exposures were to (a) cleaning and sterilizing agents in the asthmagen group; (b) shift work in the carcinogen group; and (c) toluene and p-xylene among ototoxic agents. Exposure varied by occupation, with exposure to carcinogens and ototoxic agents highest among personal carers and exposure to carcinogens most likely among nursing professionals and health and welfare support workers.

CONCLUSION

The results demonstrate that a substantial proportion of Australian healthcare workers are occupationally exposed to asthmagens, carcinogens, and ototoxic agents. These exposures are more common among certain occupational groups. The information provided by this study will be useful in prioritizing and implementing control strategies.

Generalized linear mixed-model (GLMM) trees: A flexible decision-tree method for multilevel and longitudinal data

Objective: Decision-tree methods are machine-learning methods which provide results that are relatively easy to interpret and apply by human decision makers. The resulting decision trees show how baseline patient characteristics can be combined to predict treatment outcomes for individual patients, for example. This paper introduces GLMM trees, a decision-tree method for multilevel and longitudinal data. Method: To illustrate, we apply GLMM trees to a dataset of 3,256 young people (mean age 11.33, 48% girls) receiving treatment at one of several mental-health service providers in the UK. Two treatment outcomes (mental-health difficulties scores corrected for baseline) were regressed on 18 demographic, case and severity characteristics at baseline. We compared the performance of GLMM trees with that of traditional GLMMs and random forests. Results: GLMM trees yielded modest predictive accuracy, with cross-validated multiple R values of .18 and .25. Predictive accuracy did not differ significantly from that of traditional GLMMs and random forests, while GLMM trees required evaluation of a lower number of variables. Conclusion: GLMM trees provide a useful data-analytic tool for clinical prediction problems. The supplemental material provides a tutorial for replicating the GLMM tree analyses in R.

Identification of tumor size as the only factor associated with nondiagnostic biopsies in patients with small renal masses

INTRODUCTION

As greater numbers of small renal masses (SRMs) are discovered incidentally, renal tumor biopsy (RTB) is an increasingly recognized step for the management of these lesions, ideally for the prevention of surgical overtreatment for benign disease. While the diagnosis can often be obtained preoperatively by RTB, indeterminate results create greater difficulty for patients and clinicians. This study examines a series of RTBs, identifying the portion of these that were able to yield a diagnosis, and correlates patient factors, including RENAL and PADUA scoring, with the outcome of a non-diagnostic result.

METHODS

Patients were identified as having undergone RTB at the Princess Margaret Cancer Centre in Ontario, Canada, between January 2000 and December 2009. Data was compiled from these 423 patients and analyzed using CART methodology to determine the level of association between various patient and tumor factors and the outcome of a non-diagnostic biopsy. Tumor size was further used to develop a classification tree to describe the prediction of a non-diagnostic biopsy.

RESULTS

Of these 423 patients undergoing RTB, 66 (16%) resulted in a non-diagnostic biopsy. The only patient or tumor factor that was found to be associated with a non-diagnostic outcome was mass size, where small masses (<1.28 cm diameter) were found to have a 38% chance of being non-diagnostic, compared with a 13% chance in those tumors >1.28 cm diameter (86% accuracy, 95% confidence interval [CI] 0.82-0.89).

CONCLUSIONS

When evaluating SRMs for diagnostic workup, mass size is the only tumor or patient characteristic associated with a non-diagnostic RTB.

Precision medicine: Subgroup identification in longitudinal trajectories

In clinical studies, the treatment effect may be heterogeneous among patients. It is of interest to identify subpopulations which benefit most from the treatment, regardless of the treatment's overall performance. In this study, we are interested in subgroup identification in longitudinal studies when nonlinear trajectory patterns are present. Under such a situation, evaluation of the treatment effect entails comparing longitudinal trajectories while subgroup identification requires a further evaluation of differential treatment effects among subgroups induced by moderators. To this end, we propose a tree-structured subgroup identification method, termed "interaction tree for longitudinal trajectories", which combines mixed effects models with regression splines to model the nonlinear progression patterns among repeated measures. Extensive simulation studies are conducted to evaluate its performance and an application to an alcohol addiction pharmacogenetic trial is presented.

A New Multilevel CART Algorithm for Multilevel Data with Binary Outcomes

The multilevel logistic regression model (M-logit) is the standard model for modeling multilevel data with binary outcomes. However, many assumptions and restrictions should be considered when applying this model for unbiased estimation. To overcome these limitations, we proposed a multilevel CART (M-CART) algorithm which combines the M-logit and single level CART (S-CART) within the framework of the expectation-maximization. Simulation results showed that the proposed M-CART provided substantial improvements on classification accuracy, sensitivity, and specific over the M-logit, S-CART, and single level logistic regression model when modeling multilevel data with binary outcomes. This benefit of using M-CART was consistently found across different conditions of sample size, intra-class correlation, and when relationships between predictors and outcomes were nonlinear and nonadditive.

BiMM forest: A random forest method for modeling clustered and longitudinal binary outcomes

Clustered binary outcomes and datasets with many predictor variables are frequently encountered in clinical research (e.g. longitudinal studies). Generalized linear mixed models (GLMMs) typically employed for clustered endpoints have challenges for some scenarios, particularly for complex datasets which contain many interactions among predictors and nonlinear predictors of outcome. We propose a new method called Binary Mixed Model (BiMM) forest, which combines random forest and GLMM methodology. BiMM forest offers a flexible and stable method which naturally models interactions among predictors and can be employed in the setting of clustered data. Simulation studies show that BiMM forest achieves similar or superior prediction accuracy compared to standard random forest, GLMMs and its tree counterpart (BiMM tree) for clustered binary outcomes. The method is applied to a real dataset from the Acute Liver Failure Study Group. BiMM forest offers an alternative method for modeling clustered binary outcomes which may be applied in myriad research settings.

Regression Trees for Longitudinal Data with Baseline Covariates

Longitudinal changes in a population of interest are often heterogeneous and may be influenced by a combination of baseline factors. In such cases, traditional linear mixed effects models (Laird and Ware, 1982) assuming common parametric form for the mean structure may not be applicable. We show that the regression tree methodology for longitudinal data can identify and characterize longitudinally homogeneous subgroups. Most of the currently available regression tree construction methods are either limited to a repeated measures scenario or combine the heterogeneity among subgroups with the random inter-subject variability. We propose a longitudinal classification and regression tree (LongCART) algorithm under conditional inference framework (Hothorn, Hornik and Zeileis, 2006) that overcomes these limitations utilizing a two-step approach. The LongCART algorithm first selects the partitioning variable via a parameter instability test and then finds the optimal split for the selected partitioning variable. Thus, at each node, the decision of further splitting is type-I error controlled and thus it guards against variable selection bias, over-fitting and spurious splitting. We have obtained the asymptotic results for the proposed instability test and examined its finite sample behavior through simulation studies. Comparative performance of LongCART algorithm were evaluated empirically via simulation studies. Finally, we applied LongCART to study the longitudinal changes in choline levels among HIV-positive patients.

Classification tree for the prediction of malignant disease and the prediction of non-diagnostic biopsies in patients with small renal masses

INTRODUCTION

Preoperative prediction of benign vs. malignant small renal masses (SRMs) remains a challenge. This study: 1) validates our previously published classification tree (CT) with an external cohort; 2) creates a new CT with the combined cohort; and 3) evaluates the RENAL and PADUA scoring systems for prediction of malignancy.

METHODS

This study includes a total of 818 patients with renal masses; 395 underwent surgical resection and 423 underwent biopsy. A CT to predict benign disease was developed using patient and tumour characteristics from the 709 eligible participants. Our CT is based on four parameters: tumour volume, symptoms, gender, and symptomatology. CART modelling was also used to determine if RENAL and PADUA scoring could predict malignancy.

RESULTS

When externally validated with the surgical cohort, the predictive accuracy of the old CT dropped. However, by combining the cohorts and creating a new CT, the predictive accuracy increased from 74% to 87% (95% confidence interval 0.84-0.89). RENAL and PADUA score alone were not predictive of malignancy. One limitation was the lack of available histological data from the biopsy series.

CONCLUSIONS

The validated old CT and new combined-cohort CT have a predictive value greater than currently published nomograms and single-biopsy cohorts. Overall, RENAL and PADUA scores were not able to predict malignancy.

BiMM tree: A decision tree method for modeling clustered and longitudinal binary outcomes

Clustered binary outcomes are frequently encountered in clinical research (e.g. longitudinal studies). Generalized linear mixed models (GLMMs) for clustered endpoints have challenges for some scenarios (e.g. data with multi-way interactions and nonlinear predictors unknown a priori). We develop an alternative, data-driven method called Binary Mixed Model (BiMM) tree, which combines decision tree and GLMM within a unified framework. Simulation studies show that BiMM tree achieves slightly higher or similar accuracy compared to standard methods. The method is applied to a real dataset from the Acute Liver Failure Study Group.

Recursive Partitioning with Nonlinear Models of Change

In this article, we introduce nonlinear longitudinal recursive partitioning (nLRP) and the R package longRpart2 to carry out the analysis. This method implements recursive partitioning (also known as decision trees) in order to split data based on individual- (i.e., cluster) level covariates with the goal of predicting differences in nonlinear longitudinal trajectories. At each node, a user-specified linear or nonlinear mixed-effects model is estimated. This method is an extension of Abdolell et al.'s (2002) longitudinal recursive partitioning while permitting a nonlinear mixed-effects model in addition to a linear mixed-effects model in each node. We give an overview of recursive partitioning, nonlinear mixed-effects models for longitudinal data, describe nLRP, and illustrate its use with empirical data from the Early Childhood Longitudinal Study-Kindergarten Cohort.

A recursive partitioning approach for subgroup identification in individual patient data meta-analysis

Background

Motivated by the setting of clinical trials in low back pain, this work investigated statistical methods to identify patient subgroups for which there is a large treatment effect (treatment by subgroup interaction). Statistical tests for interaction are often underpowered. Individual patient data (IPD) meta‐analyses provide a framework with improved statistical power to investigate subgroups. However, conventional approaches to subgroup analyses applied in both a single trial setting and an IPD setting have a number of issues, one of them being that factors used to define subgroups are investigated one at a time. As individuals have multiple characteristics that may be related to response to treatment, alternative exploratory statistical methods are required.

Methods

Tree‐based methods are a promising alternative that systematically searches the covariate space to identify subgroups defined by multiple characteristics. A tree method in particular, SIDES, is described and extended for application in an IPD meta‐analyses setting by incorporating fixed‐effects and random‐effects models to account for between‐trial variation. The performance of the proposed extension was assessed using simulation studies. The proposed method was then applied to an IPD low back pain dataset.

Results

The simulation studies found that the extended IPD‐SIDES method performed well in detecting subgroups especially in the presence of large between‐trial variation. The IPD‐SIDES method identified subgroups with enhanced treatment effect when applied to the low back pain data.

Conclusions

This work proposes an exploratory statistical approach for subgroup analyses applicable in any research discipline where subgroup analyses in an IPD meta‐analysis setting are of interest.

The estimated prevalence of exposure to asthmagens in the Australian workforce, 2014

Background

There is very little information available on a national level as to the number of people exposed to specific asthmagens in workplaces.

Methods

We conducted a national telephone survey in Australia to investigate the prevalence of current occupational exposure to 277 asthmagens, assembled into 27 groups. Demographic and current job information were obtained. A web-based tool, OccIDEAS, was used to collect job task information and assign exposure to each asthmagen group.

Results

In the Australian Workplace Exposure Study – Asthma (AWES- Asthma) we interviewed 4878 participants (2441 male and 2437 female). Exposure to at least one asthmagen was more common among men (47 %) than women (40 %). Extrapolated to the Australian population, approximately 2.8 million men and 1.7 million women were estimated to be exposed. Among men, the most common exposures were bioaerosols (29 %) and metals (27 %), whilst the most common exposures among women were latex (25 %) and industrial cleaning and sterilising agents (20 %).

Conclusions

This study provides information about the prevalence of exposure to asthmagens in Australian workplaces which will be useful in setting priorities for control and prevention of occupational asthma.

Electronic supplementary material

The online version of this article (doi:10.1186/s12890-016-0212-6) contains supplementary material, which is available to authorized users.

Older Age and Leg Pain Are Good Predictors of Pain and Disability Outcomes in 2710 Patients Who Receive Lumbar Fusion

BACKGROUND

Identifying appropriate candidates for lumbar spine fusion is a challenging and controversial topic. The purpose of this study was to identify baseline characteristics related to poor/favorable outcomes at 1 year for a patient who received lumbar spine fusion.

QUESTIONS/PURPOSES

The aims of this study were to describe baseline characteristics of those who received lumbar surgery and to identify baseline characteristics from a spine repository that were related to poor and favorable pain and disability outcomes for patient who received lumbar fusion (with or without decompression), who were followed up for 1 full year and discriminate predictor variables that were either or in contrast to prognostic variables reported in the literature.

METHODS

This study analyzed data from 2710 patients who underwent lumbar spine fusion. All patient data was part of a multicenter, multi-national spine repository. Ten relatively commonly captured data variables were used as predictors for the study. Univariate/multivariate logistic regression analyses were run against outcome variables of pain/disability.

RESULTS

Multiple univariate findings were associated with pain/disability outcomes at 1 year including age, previous surgical history, baseline disability, baseline pain, baseline quality of life scores, and leg pain greater than back pain. Notably significant multivariate findings for both pain and disability include older age, previous surgical history, and baseline mental summary scores, disability, and pain.

CONCLUSION

Leg pain greater than back pain and older age may yield promising value when predicting positive outcomes. Other significant findings may yield less value since these findings are similar to those that are considered to be prognostic regardless of intervention type.

N-terminal pro-B-type Natriuretic Peptides’ Prognostic Utility Is Overestimated in Meta-analyses Using Study-specific Optimal Diagnostic Thresholds

Background:

N-terminal fragment B-type natriuretic peptide (NT-proBNP) prognostic utility is commonly determined post hoc by identifying a single optimal discrimination threshold tailored to the individual study population. The authors aimed to determine how using these study-specific post hoc thresholds impacts meta-analysis results.

Methods:

The authors conducted a systematic review of studies reporting the ability of preoperative NT-proBNP measurements to predict the composite outcome of all-cause mortality and nonfatal myocardial infarction at 30 days after noncardiac surgery. Individual patient-level data NT-proBNP thresholds were determined using two different methodologies. First, a single combined NT-proBNP threshold was determined for the entire cohort of patients, and a meta-analysis conducted using this single threshold. Second, study-specific thresholds were determined for each individual study, with meta-analysis being conducted using these study-specific thresholds.

Results:

The authors obtained individual patient data from 14 studies (n = 2,196). Using a single NT-proBNP cohort threshold, the odds ratio (OR) associated with an increased NT-proBNP measurement was 3.43 (95% CI, 2.08 to 5.64). Using individual study-specific thresholds, the OR associated with an increased NT-proBNP measurement was 6.45 (95% CI, 3.98 to 10.46). In smaller studies (&lt;100 patients) a single cohort threshold was associated with an OR of 5.4 (95% CI, 2.27 to 12.84) as compared with an OR of 14.38 (95% CI, 6.08 to 34.01) for study-specific thresholds.

Conclusions:

Post hoc identification of study-specific prognostic biomarker thresholds artificially maximizes biomarker predictive power, resulting in an amplification or overestimation during meta-analysis of these results. This effect is accentuated in small studies.

Predictors of pain and disability outcomes in one thousand, one hundred and eight patients who underwent lumbar discectomy surgery

BACKGROUND

A key component toward improving surgical outcomes is proper patient selection. Improved selection can occur through exploration of prognostic studies that identify variables which are associated with good or poorer outcomes with a specific intervention, such as lumbar discectomy. To date there are no guidelines identifying key prognostic variables that assist surgeons in proper patient selection for lumbar discectomy. The purpose of this study was to identify baseline characteristics that were related to poor or favourable outcomes for patients who undergo lumbar discectomy. In particular, we were interested in prognostic factors that were unique to those commonly reported in the musculoskeletal literature, regardless of intervention type.

METHODS

This retrospective study analysed data from 1,108 patients who underwent lumbar discectomy and had one year outcomes for pain and disability. All patient data was part of a multicentre, multi-national spine repository. Ten relatively commonly captured data variables were used as predictors for the study: (1) age, (2) body mass index, (3) gender, (4) previous back surgery history, (5) baseline disability, unique baseline scores for pain for both (6) low back and (7) leg pain, (8) baseline SF-12 Physical Component Summary (PCS) scores, (9) baseline SF-12 Mental Component Summary (MCS) scores, and (10) leg pain greater than back pain. Univariate and multivariate logistic regression analyses were run against one year outcome variables of pain and disability.

RESULTS

For the multivariate analyses associated with the outcome of pain, older patients, those with higher baseline back pain, those with lesser reported disability and higher SF-12 MCS quality of life scores were associated with improved outcomes. For the multivariate analyses associated with the outcome of disability, presence of leg pain greater than back pain and no previous surgery suggested a better outcome.

CONCLUSIONS

For this study, several predictive variables were either unique or conflicted with those advocated in general prognostic literature, suggesting they may have value for clinical decision making for lumbar discectomy surgery. In particular, leg pain greater than back pain and older age may yield promising value. Other significant findings such as quality of life scores and prior surgery may yield less value since these findings are similar to those that are considered to be prognostic regardless of intervention type.

Identifying cut points for biomarker defined subset effects in clinical trials with survival endpoints

The advancements in biotechnology and genetics lead to an increasing research interest in personalized medicine, where a patient's genetic profile or biological traits contribute to choosing the most effective treatment for the patient. The process starts with finding a specific biomarker among all possible candidates that can best predict the treatment effect. After a biomarker is chosen, identifying a cut point of the biomarker value that splits the patients into treatment effective and non-effective subgroups becomes an important scientific problem. Numerous methods have been proposed to validate the predictive marker and select the appropriate cut points either prospectively or retrospectively using clinical trial data. In trials with survival outcomes, the current practice applies an interaction testing procedure and chooses the cut point that minimizes the p-values for the tests. Such method assumes independence between the baseline hazard and biomarker value. In reality, however, this assumption is often violated, as the chosen biomarker might also be prognostic in addition to its predictive nature for treatment effect. In this paper we propose a block-wise estimation and a sequential testing approach to identify the cut point in biomarkers that can group the patients into subsets based on their distinct treatment outcomes without assuming independence between the biomarker and baseline hazard. Numerical results based on simulated survival data show that the proposed method could pinpoint accurately the cut points in biomarker values that separate the patient subpopulations into subgroups with distinctive treatment outcomes.

Growth kinetics of small renal masses: A prospective analysis from the Renal Cell Carcinoma Consortium of Canada

INTRODUCTION

Most small renal masses (SRMs) are diagnosed incidentally and have a low malignant potential. As more elderly patients and infirm patients are diagnosed with SRMs, there is an increased interest in active surveillance (AS) with delayed intervention. Patient and tumour characteristics relating to aggressive disease have not been well-studied. The objective was to determine predictors of growth of SRMs treated with AS.

METHODS

A multicentre prospective phase 2 clinical trial was conducted on 207 SRMs in 169 patients in 8 institutions in Canada from 2004 to 2009; in these patients treatment was delayed until disease progression. Patient and tumour characteristics were evaluated to determine predictors of growth of SRMs by measuring rates of change in growth (on imaging) over time. All patients underwent AS for presumed renal cell carcinoma (RCC) based on diagnostic imaging. We used the following factors to develop a predictive model of tumour growth with binary recursive partitioning analysis: patient characteristics (age, symptoms at diagnosis) and tumour characteristics (consistency [solid vs. cystic] and maximum diameter at diagnosis.

RESULTS

With a median follow-up of 603 days, 169 patients (with 207 SRMs) were followed prospectively. Age, symptoms at diagnosis, tumour consistency and maximum diameter of the renal mass were not predictors of growth. This cohort was limited by lack of availability of patient and tumour characteristics, such as sex, degree of endophytic component and tumour location.

CONCLUSION

Slow growth rates and the low malignant potential of SRMs have led to AS as a treatment option in the elderly and infirm population. In a large prospective cohort, we have shown that age, symptoms, tumour consistency and maximum diameter of the mass at diagnosis are not predictors of growth of T1a lesions. More knowledge on predictors of growth of SRMs is needed.

Growth kinetics of renal masses: analysis of a prospective cohort of patients undergoing active surveillance

BACKGROUND

Active surveillance (AS) represents a treatment option for renal masses in patients who are not surgical candidates either because of existing comorbidities or patient choice. Among renal masses undergoing AS, some grow rapidly and require treatment or progress to metastatic disease. Patient and tumour characteristics related to this more aggressive behaviour have been poorly studied.

OBJECTIVE

To report the analysis of a multi-institutional cohort of patients undergoing AS for small renal masses.

DESIGN, SETTING, AND PARTICIPANTS

This prospective study included 82 patients with 84 renal masses who underwent AS in three Canadian institutions between July 2001 and June 2009.

INTERVENTION

All patients underwent AS for renal masses presumed to be renal cell carcinoma (RCC) as based on diagnostic imaging.

MEASUREMENTS

Age, sex, symptoms at presentation, maximum diameter at diagnosis (cm), tumour location (central/peripheral), degree of endophytic component (1-100%), and tumour consistency (solid/cystic) were used to develop a predictive model of the tumour growth rate using binary recursive partitioning analysis with a repeated measures outcome.

RESULTS AND LIMITATIONS

With a median follow-up of 36 mo (range: 6-96), the mean annual renal mass growth rate for the entire cohort was 0.25 cm/yr (standard deviation [SD]: 0.49 cm/yr). Only one patient (1.2%) developed metastatic RCC. Amongst all variables, maximum diameter at diagnosis was the only predictor of tumour growth rate, and two distinct growth rates were identified. Masses that are ≥2.45 cm in largest diameter at diagnosis grow faster than smaller masses. This series was limited by its moderate sample size, although it is the largest published prospective series to date.

CONCLUSIONS

We confirm that most renal masses grow slowly and carry a low metastatic potential. Tumour size is a predictor of tumour growth rate, with renal masses <2.45 cm growing more slowly than masses >2.45 cm.

Biostatistics: a toolkit for exploration, validation, and interpretation of clinical data

Biostatistics plays a key role in all phases of clinical research starting from the design to the monitoring, data collection, data analysis, and interpretation of the results. A clear understanding of the statistical framework as it relates to the study hypothesis, reported results, and interpretation is vital for the scientific integrity of the study and its acceptance in the general medical community. In this brief report, we will put in perspective the general analytical framework for exploring and validating prognostic factors using data from large databases.

Overweight, obesity and breast cancer prognosis: optimal body size indicator cut-points

BACKGROUND

Evidence from the data provided in numerous published articles indicates that obesity and overweight can have a negative prognosis role in breast cancer. However, different Body Size Indicators (BSI) and cut-points have been employed and may partly explain discrepancies between the findings of various studies.

MATERIAL AND METHODS

14,709 women were recruited, treated and followed for a first unilateral breast cancer. After randomly splitting the patients' data into two groups, a maximum statistical outcome approach was used to select optimal BSI cut-points from a "training sample", when prognosis events were investigated. External validation was then carried out using a "validation sample", and agreement between the selected optimal BSI cut-points was assessed. Body Mass Index (BMI), weight (W), Ideal Weight Ratio (IWR) and Body Surface Area (BSA) were used, and were assessed at the time of diagnosis.

RESULTS

The selected optimal BSI cut-points were reliable when overall survival, metastasis recurrence and disease free interval events were investigated. The chosen BMI cut-point values matched the overweight cut-point value given by the World Health Organization. Agreement between defined binary BSI was acceptable; however, it varied from "fair" to "very good". Analysis of second primary cancer occurrence and contralateral recurrence events was not conclusive. When local and node recurrence events were taken into account, the results were inconsistent and were linked to an unconfirmed relationship between stoutness and these prognosis events.

CONCLUSIONS

Efficient, optimal BSI cut-points indicate a poorer prognosis, illustrated by a shortened overall survival and an increase of metastasis recurrences, from a BMI value of 25 kg/m(2), a W value of 60 kg, an IWR value of 20% and a BSA value of 1.7 m(2). Further BSI cut-point investigations are needed, taking into account contralateral recurrence and second primary cancer events.

Is there a critical period for cochlear implantation in congenitally deaf children? Analyses of hearing and speech perception performance after implantation

A range of basic and applied studies have demonstrated that during the development of the auditory system, early experimental manipulations or clinical interventions are generally more effective than those made later. We present a short review of these studies. We investigated this age-related plasticity in relation to the timing of cochlear implantation in deaf-from-birth children. Cochlear implantation is a standard intervention for providing hearing in children with severe to profound deafness. An important practical question is whether there is a critical period or cutoff age of implantation after which hearing outcomes are significantly reduced. In this article, we present data from prelingually deaf children (mostly congenitally deaf) implanted at ages ranging from 1 to 15 years. Each child was tested with auditory and speech understanding tests before implantation, and at regular intervals up to 8 years postimplantation. We measured the improvement in performance of speech understanding tests in younger implanted children and compared it with the results of those implanted at a later age. We also used a binary partitioning algorithm to divide the data systematically at all ages at implant to determine the optimum split, i.e., to determine the age at implant which best separates performance of early implanted versus later implanted children. We observed distinct age-of-implant cutoffs, and will discuss whether these really represent critical periods during development.