Model-based Recursive Partitioning Research Articles

Are species photosynthetic characteristics good predictors of seedling post-hurricane demographic patterns and species spatiotemporal distribution in a hurricane impacted wet montane forest?

In situ measurements of leaf level photosynthetic response to light were collected from seedlings of ten tree species from a tropical montane wet forest, the John Crow Mountains, Jamaica. A model-based recursive partitioning ('mob') algorithm was then used to identify species associations based on their fitted photosynthetic response curves. Leaf area dark respiration (RD) and light saturated maximum photosynthetic (Amax) rates were also used as 'mob' partitioning variables, to identify species associations based on seedling demographic patterns (from June 2007 to May 2010) following a hurricane (Aug. 2007) and the spatiotemporal distribution patterns of stems in 2006 and 2012. RD and Amax rates ranged from 1.14 to 2.02 μmol (CO2) m−2s−1 and 2.97–5.87 μmol (CO2) m−2s−1, respectively, placing the ten species in the range of intermediate shade tolerance. Several parsimonious species 'mob' groups were formed based on 1) interspecific differences among species response curves, 2) variations in post-hurricane seedling demographic trends and 3) RD rates and species spatiotemporal distribution patterns at aspects that are more or less exposed to hurricanes. The composition of parsimonious groupings based on photosynthetic curves was not concordant with the groups based on demographic trends but was partially concordant with the RD - species spatiotemporal distribution groups. Our results indicated that the influence of photosynthetic characteristics on demographic traits and species distributions was not straightforward. Rather, there was a complex pattern of interaction between ecophysiological and demographic traits, which determined species successional status, post-hurricane response and ultimately, species distribution at our study site.

Model-based recursive partitioning to identify risk clusters for metabolic syndrome and its components: findings from the International Mobility in Aging Study

ObjectiveConceptual models underpinning much epidemiological research on ageing acknowledge that environmental, social and biological systems interact to influence health outcomes. Recursive partitioning is a data-driven approach that allows for concurrent...

Subgroup identification in dose-finding trials via model-based recursive partitioning.

An important task in early-phase drug development is to identify patients, which respond better or worse to an experimental treatment. While a variety of different subgroup identification methods have been developed for the situation of randomized clinical trials that study an experimental treatment and control, much less work has been done in the situation when patients are randomized to different dose groups. In this article, we propose new strategies to perform subgroup analyses in dose-finding trials and discuss the challenges, which arise in this new setting. We consider model-based recursive partitioning, which has recently been applied to subgroup identification in 2-arm trials, as a promising method to tackle these challenges and assess its viability using a real trial example and simulations. Our results show that model-based recursive partitioning can be used to identify subgroups of patients with different dose-response curves and improves estimation of treatment effects and minimum effective doses compared to models ignoring possible subgroups, when heterogeneity among patients is present.

A Machine Learning Approach to Identifying Placebo Responders in Late-Life Depression Trials

Despite efforts to identify characteristics associated with medication-placebo differences in antidepressant trials, few consistent findings have emerged to guide participant selection in drug development settings and differential therapeutics in clinical practice. Limitations in the methodologies used, particularly searching for a single moderator while treating all other variables as noise, may partially explain the failure to generate consistent results. The present study tested whether interactions between pretreatment patient characteristics, rather than a single-variable solution, may better predict who is most likely to benefit from placebo versus medication. Data were analyzed from 174 patients aged 75 years and older with unipolar depression who were randomly assigned to citalopram or placebo. Model-based recursive partitioning analysis was conducted to identify the most robust significant moderators of placebo versus citalopram response. The greatest signal detection between medication and placebo in favor of medication was among patients with fewer years of education (≤12) who suffered from a longer duration of depression since their first episode (>3.47 years) (B = 2.53, t(32) = 3.01, p = 0.004). Compared with medication, placebo had the greatest response for those who were more educated (>12 years), to the point where placebo almost outperformed medication (B = -0.57, t(96) = -1.90, p = 0.06). Machine learning approaches capable of evaluating the contributions of multiple predictor variables may be a promising methodology for identifying placebo versus medication responders. Duration of depression and education should be considered in the efforts to modulate placebo magnitude in drug development settings and in clinical practice.

Cost Sharing of Disease-Modifying Treatments (DMTs) as Policy Lever to Improve DMTs’ Access in Multiple Sclerosis

ObjectivesTo understand the nonlinear relationship between out-of-pocket (OOP) payments and disease-modifying treatment (DMT) use and adherence, primarily to pinpoint the threshold at which the use of DMTs becomes price sensitive. MethodsIndividuals with more than two multiple sclerosis (MS) diagnoses (International Classification of Diseases, Ninth Revision code 340) were identified from the MarketScan database (2006–2009). Heterogeneity in treatment was normalized by calculating an annual OOP payment as the average OOP payment for purchasing a fixed basket of DMTs at the insurance plan level. A local linear regression with a model-based recursive partitioning algorithm was applied to explore the relationship between OOP and consequently lower DMT use and adherence as measured by days covered by DMT. ResultsWe identified the inflection points in annual OOP payments as $442 for DMT use and $890 for DMT adherence. For patients with annual OOP payments of more than $442, a $100-increase in OOP payment was associated with a decline of 0.6% in DMT use; for annual OOP payments of more than $890, a $100-increase in OOP payment was associated with two fewer days of DMT treatments. ConclusionsAlthough the use of DMTs and DMT adherence appeared unassociated with OOP payment below $442 and $890, respectively, an excessive OOP payment was a barrier to DMT access. This information can inform maximum monthly and yearly payment caps when designing valued-based insurance plans.

Detecting treatment-subgroup interactions in clustered data with generalized linear mixed-effects model trees

Identification of subgroups of patients for whom treatment A is more effective than treatment B, and vice versa, is of key importance to the development of personalized medicine. Tree-based algorithms are helpful tools for the detection of such interactions, but none of the available algorithms allow for taking into account clustered or nested dataset structures, which are particularly common in psychological research. Therefore, we propose the generalized linear mixed-effects model tree (GLMM tree) algorithm, which allows for the detection of treatment-subgroup interactions, while accounting for the clustered structure of a dataset. The algorithm uses model-based recursive partitioning to detect treatment-subgroup interactions, and a GLMM to estimate the random-effects parameters. In a simulation study, GLMM trees show higher accuracy in recovering treatment-subgroup interactions, higher predictive accuracy, and lower type II error rates than linear-model-based recursive partitioning and mixed-effects regression trees. Also, GLMM trees show somewhat higher predictive accuracy than linear mixed-effects models with pre-specified interaction effects, on average. We illustrate the application of GLMM trees on an individual patient-level data meta-analysis on treatments for depression. We conclude that GLMM trees are a promising exploratory tool for the detection of treatment-subgroup interactions in clustered datasets.

Varying sensitivity of mountainous streamwater base-flow {bf{N}}{{bf{O}}}_{{bf{3}}}^{-}concentrations to N deposition in the northern suburbs of Tokyo

Ecosystems of suburban landscapes (i.e., forest, inland water ecosystem) are threatened by high nitrogen (N) loadings derived from urban air pollutants. Forest ecosystems under high chronic N loadings tend to leach more N via streams. In the northern suburbs of Tokyo, N deposition loading on terrestrial ecosystems has increased over the past 30 years. In this region, we investigated nitrate concentrations in 608 independent small forested catchment water samples from northeastern suburbs of Tokyo. The nitrate concentrations varied from 0.07 to 3.31 mg-N L−1 in this region. We evaluated the effects of N deposition and catchment properties (e.g., meteorological and topographic factors, vegetation and soil types) on nitrate concentrations. In the random forest model, simulated N deposition rates from an atmospheric chemistry transportation model explained most of the variance of nitrate concentration. To evaluate the effects of afforestation management in the catchment, we followed a model-based recursive partitioning method (MOB). MOB succeeded in data-driven identification of subgroups with varying sensitivities to N deposition rate by vegetation composition in the catchment. According to MOB, the catchment with dominant coniferous coverage that mostly consisted of plantation with old tree age tended to have strong sensitivity of nitrate concentrations to N deposition loading.

Using recursive partitioning to account for parameter heterogeneity in multinomial processing tree models.

In multinomial processing tree (MPT) models, individual differences between the participants in a study can lead to heterogeneity of the model parameters. While subject covariates may explain these differences, it is often unknown in advance how the parameters depend on the available covariates, that is, which variables play a role at all, interact, or have a nonlinear influence, etc. Therefore, a new approach for capturing parameter heterogeneity in MPT models is proposed based on the machine learning method MOB for model-based recursive partitioning. This procedure recursively partitions the covariate space, leading to an MPT tree with subgroups that are directly interpretable in terms of effects and interactions of the covariates. The pros and cons of MPT trees as a means of analyzing the effects of covariates in MPT model parameters are discussed based on simulation experiments as well as on two empirical applications from memory research. Software that implements MPT trees is provided via the mpttree function in the psychotree package in R.

Big Data is not only about data: The two cultures of modelling

The contribution of Big Data to social science is not limited to data availability but includes the introduction of analytical approaches that have been developed in computer science, and in particular in machine learning. This brings about a new ‘culture’ of statistical modelling that bears considerable potential for the social scientist. This argument is illustrated with a brief discussion of model-based recursive partitioning which can bridge the theory and data-driven approach. Such a method is an example of how this new approach can help revise models that work for the full dataset: it can be used for evaluating different models, a traditional weakness of the ‘traditional’ statistical approach used in social science.

Comparing Four Methods for Estimating Tree-Based Treatment Regimes.

When multiple treatment alternatives are available for a certain psychological or medical problem, an important challenge is to find an optimal treatment regime, which specifies for each patient the most effective treatment alternative given his or her pattern of pretreatment characteristics. The focus of this paper is on tree-based treatment regimes, which link an optimal treatment alternative to each leaf of a tree; as such they provide an insightful representation of the decision structure underlying the regime. This paper compares the absolute and relative performance of four methods for estimating regimes of that sort (viz., Interaction Trees, Model-based Recursive Partitioning, an approach developed by Zhang et al. and Qualitative Interaction Trees) in an extensive simulation study. The evaluation criteria were, on the one hand, the expected outcome if the entire population would be subjected to the treatment regime resulting from each method under study and the proportion of clients assigned to the truly best treatment alternative, and, on the other hand, the Type I and Type II error probabilities of each method. The method of Zhang et al. was superior regarding the first two outcome measures and the Type II error probabilities, but performed worst in some conditions of the simulation study regarding Type I error probabilities.

Moderators of change in an Internet-based intervention for eating disorders with different levels of therapist support: What works for whom?

This study investigated moderators of intervention response in a fully automated Internet-based monitoring and feedback intervention (‘Featback’) with different levels of therapist support for individuals with eating disorder (ED) symptoms. This study was part of a randomized controlled trial comparing four conditions: 1) Featback, 2) Featback with low-intensity (weekly) therapist support, 3) Featback with high-intensity (three times a week) therapist support, and 4) a waiting list. A total of 273 participants completed baseline and post-intervention assessments. The outcome measure was ED psychopathology. Model-based recursive partitioning was applied. Baseline levels of ED psychopathology were found to moderate intervention response. Specifically, in terms of improvement in symptoms of anorexia nervosa, participants with higher baseline levels of anorectic psychopathology showed better outcomes in the waiting list condition and the Featback conditions with low- and high-intensity therapist support in comparison with Featback without therapist support. In terms of improvement in symptoms of bulimia nervosa, participants with mild to moderate bulimic symptoms had better outcomes in the Featback conditions compared with the waiting list. Thus, the fully automated Internet-based intervention with and without therapist support may be particularly suitable in improving mild to moderate bulimic psychopathology, whereas the intervention without therapist support may be less effective in improving severe anorectic psychopathology. Further investigating differential intervention responses is important, as this could help to optimize the delivery and dissemination of E-health interventions and therapist support, which in turn could help moving toward personalized (E-)care.

Tree-Based Global Model Tests for Polytomous Rasch Models.

Psychometric measurement models are only valid if measurement invariance holds between test takers of different groups. Global model tests, such as the well-established likelihood ratio (LR) test, are sensitive to violations of measurement invariance, such as differential item functioning and differential step functioning. However, these traditional approaches are only applicable when comparing previously specified reference and focal groups, such as males and females. Here, we propose a new framework for global model tests for polytomous Rasch models based on a model-based recursive partitioning algorithm. With this approach, a priori specification of reference and focal groups is no longer necessary, because they are automatically detected in a data-driven way. The statistical background of the new framework is introduced along with an instructive example. A series of simulation studies illustrates and compares its statistical properties to the well-established LR test. While both the LR test and the new framework are sensitive to differential item functioning and differential step functioning and respect a given significance level regardless of true differences in the ability distributions, the new data-driven approach is more powerful when the group structure is not known a priori-as will usually be the case in practical applications. The usage and interpretation of the new method are illustrated in an empirical application example. A software implementation is freely available in the R system for statistical computing.

The tree analysis model as a guide for the selection of the optimum ovulation induction method to improve the quality of oocytes

The tree analysis model as a guide for the selection of the optimum ovulation induction method to improve the quality of oocytes

Integrated paediatric fever management and antibiotic over-treatment in Malawi health facilities: data mining a national facility census.

BackgroundThere are growing concerns about irrational antibiotic prescription practices in the era of test-based malaria case management. This study assessed integrated paediatric fever management using malaria rapid diagnostic tests (RDT) and Integrated Management of Childhood Illness (IMCI) guidelines, including the relationship between RDT-negative results and antibiotic over-treatment in Malawi health facilities in 2013–2014.MethodsA Malawi national facility census included 1981 observed sick children aged 2–59 months with fever complaints. Weighted frequencies were tabulated for other complaints, assessments and prescriptions for RDT-confirmed malaria, IMCI-classified non-severe pneumonia, and clinical diarrhoea. Classification trees using model-based recursive partitioning estimated the association between RDT results and antibiotic over-treatment and learned the influence of 38 other input variables at patient-, provider- and facility-levels.ResultsAmong 1981 clients, 72 % were tested or referred for malaria diagnosis and 85 % with RDT-confirmed malaria were prescribed first-line anti-malarials. Twenty-eight percent with IMCI-pneumonia were not prescribed antibiotics (under-treatment) and 59 % ‘without antibiotic need’ were prescribed antibiotics (over-treatment). Few clients had respiratory rates counted to identify antibiotic need for IMCI-pneumonia (18 %). RDT-negative children had 16.8 (95 % CI 8.6–32.7) times higher antibiotic over-treatment odds compared to RDT-positive cases conditioned by cough or difficult breathing complaints.ConclusionsIntegrated paediatric fever management was sub-optimal for completed assessments and antibiotic targeting despite common compliance to malaria treatment guidelines. RDT-negative results were strongly associated with antibiotic over-treatment conditioned by cough or difficult breathing complaints. A shift from malaria-focused ‘test and treat’ strategies toward ‘IMCI with testing’ is needed to improve quality fever care and rational use of both anti-malarials and antibiotics in line with recent global commitments to combat resistance.Electronic supplementary materialThe online version of this article (doi:10.1186/s12936-016-1439-7) contains supplementary material, which is available to authorized users.

A Comparative Study for Modelling the Survival of Breast Cancer Patients in the West of Iran

<p><strong>BACKGROUND</strong><strong>: </strong>Breast cancer is the main cause of women cancer mortality. Therefore, precise prediction of patients’ risk level is the major concern in therapeutic strategies. Although statistical learning algorithms are high quality risk prediction methods, but usually their better prediction quality leads to more loss of interpretability. Therefore, the aim of this study is to compare ‘Model-Based Recursive Partitioning’ and ‘Random Survival Forest’; whether the partitioning, as the more interpretable learning technique, could be a suitable successor for forests.</p><p><strong>PATIENTS & METHODS</strong><strong>: </strong>The applied dataset for this retrospective cohort study includes the information of 539 Iranian females with breast cancer. To model the patients’ survival, various learning algorithms were fitted and their accuracy measures were statistically compared by means of several precision criteria.</p><p><strong>RESULTS</strong><strong>:</strong> This study verified the stability of ‘Model-based Recursive Partitioning’, further to ‘Random Survival Forest’ deficiency to present a unique pervasive model. Moreover, except ‘Log-Logistic-Based Recursive Partitioning’, none of the methods significantly outperformed ‘Exponential- Based Recursive Partitioning’.</p><p><strong>CONCLUSIONS:</strong> Briefly, it was concluded that the loss of interpretability due to the use of over complex models, may not always counterbalanced by the amount of prediction improvements.</p>

Model-Based Recursive Partitioning for Subgroup Analyses.

The identification of patient subgroups with differential treatment effects is the first step towards individualised treatments. A current draft guideline by the EMA discusses potentials and problems in subgroup analyses and formulated challenges to the development of appropriate statistical procedures for the data-driven identification of patient subgroups. We introduce model-based recursive partitioning as a procedure for the automated detection of patient subgroups that are identifiable by predictive factors. The method starts with a model for the overall treatment effect as defined for the primary analysis in the study protocol and uses measures for detecting parameter instabilities in this treatment effect. The procedure produces a segmented model with differential treatment parameters corresponding to each patient subgroup. The subgroups are linked to predictive factors by means of a decision tree. The method is applied to the search for subgroups of patients suffering from amyotrophic lateral sclerosis that differ with respect to their Riluzole treatment effect, the only currently approved drug for this disease.

Heterogeneous Effects of Housing Vouchers on the Mental Health of US Adolescents.

To assess the mental health effects on adolescents of low-income families residing in high-poverty public housing who received housing vouchers to assist relocation. We defined treatment effects to compare 2829 adolescents aged 12 to 19 years in families offered housing vouchers versus those living in public housing in the Moving to Opportunity experiment (1994-1997; Boston, MA; Baltimore, MD; Chicago, IL; Los Angeles, CA; New York, NY). We employed model-based recursive partitioning to identify subgroups with heterogeneous treatment effects on psychological distress and behavior problems measured in 2002. We tested 35 potential baseline treatment modifiers. For psychological distress, Chicago participants experienced null treatment effects. Outside Chicago, boys experienced detrimental effects, whereas girls experienced beneficial effects. Behavior problems effects were null for adolescents who were aged 10 years or younger at baseline. For adolescents who were older than 10 years at baseline, violent crime victimization, unmarried parents, and unsafe neighborhoods increased adverse treatment effects. Adolescents who were older than 10 years at baseline without learning problems or violent crime victimization, and whose parents moved for better schools, experienced beneficial effects. Health effects of housing vouchers varied across subgroups. Supplemental services may be necessary for vulnerable subgroups for whom housing vouchers alone may not be beneficial.

Identifying Moderators of Response to the Penn Resiliency Program: A Synthesis Study.

To identify moderators of a cognitive-behavioral depression prevention program's effect on depressive symptoms among youth in early adolescence, data from three randomized controlled trials of the Penn Resiliency Program (PRP) were aggregated to maximize statistical power and sample diversity (N = 1145). Depressive symptoms, measured with the Children's Depression Inventory (CDI; Kovacs 1992), were assessed at six common time points over two-years of follow-up. Latent growth curve models evaluated whether PRP and control conditions differed in the rate of change in CDI and whether youth- and family-level characteristics moderated intervention effects. Model-based recursive partitioning was used as a supplementary analysis for identifying moderators. There was a three-way interaction of PRP, initial symptom severity, and intervention site on growth in depressive symptoms. There was considerable variability in PRP's effects, with the nature of the interaction between PRP and initial symptom levels differing considerably across sites. PRP reduced depressive symptoms among youth with unmarried parents, but not among those with married parents. Finally, PRP's effects differed across school grade levels. Although initial symptom severity moderated PRP's effect on depressive symptoms, it was not a reliable indicator of how well the intervention performed, limiting its utility as a prescriptive variable. Our primary analyses suggest that PRP's effects are limited to youth whose parents are unmarried. The small number of fifth grade students (n = 25; 2%) showed a delayed and sustained intervention response. Our findings underscore the importance of evaluating site, family, and contextual characteristics as moderators in future studies.

A Comparison between Cure Model and Recursive Partitioning: A Retrospective Cohort Study of Iranian Female with Breast Cancer.

Background. Breast cancer which is the most common cause of women cancer death has an increasing incidence and mortality rates in Iran. A proper modeling would correctly detect the factors' effect on breast cancer, which may be the basis of health care planning. Therefore, this study aimed to practically develop two recently introduced statistical models in order to compare them as the survival prediction tools for breast cancer patients. Materials and Methods. For this retrospective cohort study, the 18-year follow-up information of 539 breast cancer patients was analyzed by “Parametric Mixture Cure Model” and “Model-Based Recursive Partitioning.” Furthermore, a simulation study was carried out to compare the performance of mentioned models for different situations. Results. “Model-Based Recursive Partitioning” was able to present a better description of dataset and provided a fine separation of individuals with different risk levels. Additionally the results of simulation study confirmed the superiority of this recursive partitioning for nonlinear model structures. Conclusion. “Model-Based Recursive Partitioning” seems to be a potential instrument for processing complex mixture cure models. Therefore, applying this model is recommended for long-term survival patients.

Crop yield forecasting on the Canadian Prairies by remotely sensed vegetation indices and machine learning methods

Crop yield forecast models for barley, canola and spring wheat grown on the Canadian Prairies were developed using vegetation indices derived from satellite data and machine learning methods. Hierarchical clustering was used to group the crop yield data from 40 Census Agricultural Regions (CARs) into several larger regions for building the forecast models. The Normalized Difference Vegetation Index (NDVI) and Enhanced Vegetation Index (EVI) derived from the Moderate-resolution Imaging Spectroradiometer (MODIS), and NDVI derived from the Advanced Very High Resolution Radiometer (AVHRR) were considered as predictors for crop yields. Multiple linear regression (MLR) and two nonlinear machine learning models – Bayesian neural networks (BNN) and model-based recursive partitioning (MOB) – were used to forecast crop yields, with various combinations of MODIS-NDVI, MODIS-EVI and NOAA-NDVI as predictors. Crop yield forecasts made using predictors from July and earlier were evaluated by the cross-validated mean absolute error skill score (in reference to climatological forecasts) during 2000–2011. While MODIS-NDVI was found to be the most effective predictor for all three crops, having MODIS-EVI as an additional predictor enhanced the forecast skills. While MLR, BNN and MOB all showed significantly higher skills than climatological forecasts for all three crops, barley was the only case where the nonlinear BNN and MOB models showed slightly higher skills than MLR. The lack of skill improvement by nonlinear models over MLR is likely due to the short (12 years) record available for MODIS data, which limits our study to 2000–2011, with very low yields coming from a single severe drought year (2002).