Prediction-based variable selection for component-wise gradient boosting.

Model-based component-wise gradient boosting is a popular tool for data-driven variable selection. In order to improve its prediction and selection qualities even further, several modifications of the original algorithm have been developed, that mainly focus on different stopping criteria, leaving the actual variable selection mechanism untouched. We investigate different prediction-based mechanisms for the variable selection step in model-based component-wise gradient boosting. These approaches include Akaikes Information Criterion (AIC) as well as a selection rule relying on the component-wise test error computed via cross-validation. We implemented the AIC and cross-validation routines for Generalized Linear Models and evaluated them regarding their variable selection properties and predictive performance. An extensive simulation study revealed improved selection properties whereas the prediction error could be lowered in a real world application with age-standardized COVID-19 incidence rates.

MELD-score for risk stratification in cardiac surgery.

The outcome of the patients undergoing cardiac surgery with cardiopulmonary bypass (CPB) is also influenced by the renal and hepatic organ functions. Risk stratification, using scores such as EURO Score II or STS Short-Term Risk Calculator for patients undergoing cardiac surgery with cardiopulmonary bypass, ignores the quantitative renal and hepatic function; therefore, MELD-Score was applied in these cases. We retrospectively examined patient data using the MELD score as a predictor of mortality. To perform a univariate analysis of the data, patients were classified into three groups based on the MELD Score: MELD < 10 (Group 1), MELD 10 to 19 (Group 2), and MELD ≥ 20 (Group 3). A total of 11,477 participants were included in the study, though several patients with either missing MELD scores or lack of creatinine, bilirubin, or INR levels were dropped from the original cohort. Eventually, 10,882 patients were included in the analysis. The primary outcome was defined as postoperative, in-hospital mortality. Secondary outcomes such as postoperative bleeding, including the requirement for repeat thoracotomy, postoperative neurological complications, and assessment of catecholamines on weaning from cardiopulmonary bypass/ requirement of mechanical circulatory support were examined. A higher MELD score was associated with increased postoperative mortality. Patients with MELD > 20 experienced a 31.2% postoperative mortality, compared to Group 1 (4.6%) and Group 2 (17.5%). The highest rates of postoperative bleeding (13.8%) and, repeat thoracotomy (13.2%) & postoperative pneumonia (17.4%) were seen in Group 3 (threefold increase when compared to Group 1, renal failure requiring dialysis (N = 235, 2.7% in Group 1 v/s N = 78, 22.9% in Group 3) or requiring high dose catecholamines post-operatively or mechanical circulatory support (IABP/ECLS). Incidentally, an increased MELD Score was not associated with a significant increase in the postoperative incidence of stroke/TIA or the presence of sternal wound infections/complications. A higher mortality was observed in patients with reduced liver and renal function, with a significant increase in patients with a MELD score > 20. As the current risk stratification scores do not consider this, we recommend applying the MELD score before considering patients for cardiac surgery.

Bayesian learners in gradient boosting for linear mixed models.

Selection of relevant fixed and random effects without prior choices made from possibly insufficient theory is important in mixed models. Inference with current boosting techniques suffers from biased estimates of random effects and the inflexibility of random effects selection. This paper proposes a new inference method "BayesBoost" that integrates a Bayesian learner into gradient boosting with simultaneous estimation and selection of fixed and random effects in linear mixed models. The method introduces a novel selection strategy for random effects, which allows for computationally fast selection of random slopes even in high-dimensional data structures. Additionally, the new method not only overcomes the shortcomings of Bayesian inference in giving precise and unambiguous guidelines for the selection of covariates by benefiting from boosting techniques, but also provides Bayesian ways to construct estimators for the precision of parameters such as variance components or credible intervals, which are not available in conventional boosting frameworks. The effectiveness of the new approach can be observed via simulation and in a real-world application.

Joint Modelling Approaches to Survival Analysis via Likelihood-Based Boosting Techniques.

Joint models are a powerful class of statistical models which apply to any data where event times are recorded alongside a longitudinal outcome by connecting longitudinal and time-to-event data within a joint likelihood allowing for quantification of the association between the two outcomes without possible bias. In order to make joint models feasible for regularization and variable selection, a statistical boosting algorithm has been proposed, which fits joint models using component-wise gradient boosting techniques. However, these methods have well-known limitations, i.e., they provide no balanced updating procedure for random effects in longitudinal analysis and tend to return biased effect estimation for time-dependent covariates in survival analysis. In this manuscript, we adapt likelihood-based boosting techniques to the framework of joint models and propose a novel algorithm in order to improve inference where gradient boosting has said limitations. The algorithm represents a novel boosting approach allowing for time-dependent covariates in survival analysis and in addition offers variable selection for joint models, which is evaluated via simulations and real world application modelling CD4 cell counts of patients infected with human immunodeficiency virus (HIV). Overall, the method stands out with respect to variable selection properties and represents an accessible way to boosting for time-dependent covariates in survival analysis, which lays a foundation for all kinds of possible extensions.

Gradient boosting for linear mixed models.

Gradient boosting from the field of statistical learning is widely known as a powerful framework for estimation and selection of predictor effects in various regression models by adapting concepts from classification theory. Current boosting approaches also offer methods accounting for random effects and thus enable prediction of mixed models for longitudinal and clustered data. However, these approaches include several flaws resulting in unbalanced effect selection with falsely induced shrinkage and a low convergence rate on the one hand and biased estimates of the random effects on the other hand. We therefore propose a new boosting algorithm which explicitly accounts for the random structure by excluding it from the selection procedure, properly correcting the random effects estimates and in addition providing likelihood-based estimation of the random effects variance structure. The new algorithm offers an organic and unbiased fitting approach, which is shown via simulations and data examples.

Addressing cluster-constant covariates in mixed effects models via likelihood-based boosting techniques.

Boosting techniques from the field of statistical learning have grown to be a popular tool for estimating and selecting predictor effects in various regression models and can roughly be separated in two general approaches, namely gradient boosting and likelihood-based boosting. An extensive framework has been proposed in order to fit generalized mixed models based on boosting, however for the case of cluster-constant covariates likelihood-based boosting approaches tend to mischoose variables in the selection step leading to wrong estimates. We propose an improved boosting algorithm for linear mixed models, where the random effects are properly weighted, disentangled from the fixed effects updating scheme and corrected for correlations with cluster-constant covariates in order to improve quality of estimates and in addition reduce the computational effort. The method outperforms current state-of-the-art approaches from boosting and maximum likelihood inference which is shown via simulations and various data examples.

Development of a new sternal dehiscence prediction scale for decision making in sternal closure techniques after cardiac surgery.

BACKGROUND: After sternotomy, the spectrum for sternal osteosynthesis comprises standard wiring and more complex techniques, like titanium plating. The aim of this study is to develop a predictive risk score that evaluates the risk of sternum instability individually. The surgeon may then choose an appropriate sternal osteosynthesis technique that is risk- adjusted as well as cost-effective. METHODS: Data from 7.173 patients operated via sternotomy for all cardiovascular indications from 2008 until 2017 were retrospectively analyzed. Sternal dehiscence occurred in 2.5% of patients (n = 176). A multivariable analysis model examined pre- and intraoperative factors. A multivariable logistic regression model and a backward elimination based on the Akaike Information Criterion (AIC) a logistic model were selected. RESULTS: The model showed good sensitivity and specificity (area under the receiver-operating characteristic curve, AUC: 0.76) and several predictors of sternal instability could be evaluated. Multivariable logistic regression showed the highest Odds Ratios (OR) for reexploration (OR 6.6, confidence interval, CI [4.5-9.5], p < 0.001), obesity (body mass index, BMI > 35 kg/m2) (OR 4.23, [CI 2.4-7.3], p < 0.001), insulin-dependent diabetes mellitus (IDDM) (OR 2.2, CI [1.5-3.2], p = 0.01), smoking (OR 2.03, [CI 1.3-3.08], p = 0.001). After weighting the probability of sternum dehiscence with each factor, a risk score model was proposed scaling from - 1 to 5 points. This resulted in a risk score ranging up to 18 points, with an estimated risk for sternum complication up to 74%. CONCLUSIONS: A weighted scoring system based on individual risk factors was specifically created to predict sternal dehiscence. High-scoring patients should receive additive closure techniques.

Safety of medical compression stockings in patients with diabetes mellitus or peripheral arterial disease.

INTRODUCTION: Compression therapy is highly effective in the treatment of many venous diseases, including leg edema. However, its relevance in patients with peripheral arterial disease (PAD) or diabetes mellitus is critically discussed. The aim of the present study was to assess the influence of compression therapy on microperfusion and its safety in patients with PAD or diabetes mellitus. RESEARCH DESIGN AND METHODS: A prospective analysis of 94 consecutive patients (44 patients with diabetes, 45 patients with PAD and 5 healthy controls) undergoing medical compression therapy was performed. Microperfusion was assessed by a combined method of white light tissue spectrometry and laser Doppler flowmetry under medical compression therapy (classes I and II), in different body positions (supine, sitting, standing and elevated position of the leg) and at different locations (great toe, lateral ankle and calf). RESULTS: During the entire study, no compression-related adverse events occurred. Evaluation of microcirculation parameters (oxygen saturation of hemoglobin and flow) at the different locations and in sitting and standing positions (patients with diabetes and PAD) under compression therapy classes I and II revealed no tendency for reduced microperfusion in both groups. In contrast, in the elevated leg position, all mean perfusion values decreased in the PAD and diabetes groups. However, the same effect was seen in the healthy subgroup. CONCLUSIONS: In consideration of the present inclusion criteria, use of medical compression stockings is safe and feasible in patients with diabetes or PAD. This study did not find relevant impairment of microperfusion parameters under compression therapy in these patient subgroups in physiologic body positions. TRIAL REGISTRATION NUMBER: NCT03384758.

Influence of end-stage renal disease on long-term survival after major amputation.

Background: To analyze long-term outcomes and possible influencing factors in patients with endstage renal disease (ESRD) and critical limb ischemia (CLI) after major amputation compared to patients with normal renal function and non-dialysis-dependent chronic kidney disease. Patients and methods: Abstraction of single-center medical records of patients undergoing above knee (AKA) and below knee (BKA) amputation over a 10 years period (n = 436; 2009-2018). Excluded were amputations due to trauma or tumor. Patients were subdivided according to renal function in three categories: ESRD patients (n = 98), non-dialysis dependent chronic kidney disease (CKD, n = 98) and normal renal function (NF, n = 240). Predefined endpoints were survival and postoperative complications. Cox-regression models were built to analyze independent risk factors for outcome parameters. Results: In total, 298 AKA, 133 BKA and 5 knee joint exarticulations were performed. ESRD patients showed inferior in-hospital results as to death (ESRD 36.7 % vs. CKD 19.4 % and NF 20.0 %, P = .002). Similarly, long-term survival rates (6 months: ESRD 55.0 % vs. CKD 69.4 %, NF 67.9 % 1 year: ESRD 48.6 %, CKD 60.2 %, NF 60.8 % 5 years: ESRD 9.9 %, CKD 31.8 %, NF 37.1 %, P < .001) were significantly decreased for ESRD patients. Median postoperative survival was 10 months in ERSD, and 22 months in CKD and NF, respectively. Analysis of postoperative surgical complications revealed no differences between groups (ESRD 19.4 %, CKD 17.3 %, NF 17.0 %; P = 0.433). Cox regression analysis indicated that dialysis (HR 1.63; 95 % CI 1.22-2.16; P = .001), hypertension (HR 1.59; 95 % CI 0.99-2.54) and smoking (HR 1.22; 95 % CI 1.03-1.44; P = .022) was associated with increased risk of death during follow-up. Conclusions: Mortality after limb amputation in ERSD patients remains high. Survival of ERSD patients is lower in relation to chronic kidney disease and patients with normal renal function. Due to poor in hospital outcomes and absent long-term survival, benefit of primary amputation in ERSD seems scarce.