Benchmark of filter methods for feature selection in high-dimensional gene expression survival data.

Feature selection is crucial for the analysis of high-dimensional data, but benchmark studies for data with a survival outcome are rare. We compare 14 filter methods for feature selection based on 11 high-dimensional gene expression survival data sets. The aim is to provide guidance on the choice of filter methods for other researchers and practitioners. We analyze the accuracy of predictive models that employ the features selected by the filter methods. Also, we consider the run time, the number of selected features for fitting models with high predictive accuracy as well as the feature selection stability. We conclude that the simple variance filter outperforms all other considered filter methods. This filter selects the features with the largest variance and does not take into account the survival outcome. Also, we identify the correlation-adjusted regression scores filter as a more elaborate alternative that allows fitting models with similar predictive accuracy. Additionally, we investigate the filter methods based on feature rankings, finding groups of similar filters.

Development of a clinical algorithm to predict phenotypic switches between atopic dermatitis and psoriasis (the "Flip-Flop" phenomenon).

BACKGROUND: Atopic dermatitis (AD) and psoriasis vulgaris (PV) are almost mutually exclusive diseases with different immune polarizations, mechanisms and therapeutic targets. Switches to the other disease ("Flip-Flop" [FF] phenomenon) can occur with or without systemic treatment and are often referred to as paradoxical reactions under biological therapy. METHODS: The objective was to develop a diagnostic algorithm by combining clinical criteria of AD and PV to identify FF patients. The algorithm was prospectively validated in patients enrolled in the CK-CARE registry in Bonn, Germany. Afterward, algorithm refinements were implemented based on machine learning. RESULTS: Three hundred adult Caucasian patients were included in the validation study (n = 238 with AD, n = 49 with PV, n = 13 with FF; mean age 41.2 years; n = 161 [53.7%] female). The total FF scores of the PV and AD groups differed significantly from the FF group in the validation data (p < .001). The predictive mean generalized Youden-Index of the initial model was 78.9% [95% confidence interval 72.0%-85.6%] and the accuracy was 89.7%. Disease group-specific sensitivity was 100% (FF), 95.0% (AD), and 61.2% (PV). The specificity was 89.2% (FF), 100% (AD), and 100% (PV), respectively. CONCLUSION: The FF algorithm represents the first validated tool to identify FF patients.

IL-13, periostin and dipeptidyl-peptidase-4 reveal endotype-phenotype associations in atopic dermatitis.

BACKGROUND: The heterogeneous (endo)phenotypes of atopic dermatitis (AD) require precision medicine. Currently, systemic therapy is recommended to patients with an Eczema Area and Severity Index (EASI)≥16. Previous studies have demonstrated an improved treatment response to the anti-interleukin (IL)-13 antibody tralokinumab in AD subgroups with elevated levels of the IL-13-related biomarkers dipeptidyl-peptidase (DPP)-4 and periostin. METHODS: Herein, 373 AD patients aged≥12 years were stratified by IL-13high , periostinhigh and DPP-4high endotypes using cross-sectional data from the ProRaD cohort Bonn. "High" was defined as >80th quantile of 47 non-atopic controls. We analyzed endotype-phenotype associations using machine-learning gradient boosting compared to logistic regression. RESULTS: AD severity and eosinophils correlated with IL-13 and periostin levels. Correlations of IL-13 with EASI were stronger in patients with increased (rs=0.482) than with normal (rs=0.342) periostin levels. We identified eosinophilia>6% and an EASI range of 5.5-17 dependent on the biomarker combination to be associated with increasing probabilities of biomarkerhigh endotypes. Also patients with mild-to-low-moderate severity (EASI<16) featured increased biomarkers (IL-13high : 41%, periostinhigh : 48.4%, DPP-4high : 22.3%). Herthoge sign (adjusted Odds Ratio (aOR)=1.89, 95% Confidence Interval (CI) [1.14-3.14]) and maternal allergic rhinitis (aOR=2.79-4.47) increased the probability of an IL-13high -endotype, "dirty neck" (aOR=2.83 [1.32-6.07]), orbital darkening (aOR=2.43 [1.08-5.50]), keratosis pilaris (aOR=2.21 [1.1-4.42]) and perleche (aOR=3.44 [1.72-6.86]) of a DPP-4high -endotype. CONCLUSIONS: A substantial proportion of patients with EASI<16 featured high biomarker levels suggesting systemic impact of skin inflammation already below the current cut-off for systemic therapy. Our findings facilitate the identification of patients with distinct endotypes potentially linked to response to IL-13-targeted therapy.

Dementia risk predictions from German claims data using methods of machine learning.

INTRODUCTION: We examined whether German claims data are suitable for dementia risk prediction, how machine learning (ML) compares to classical regression, and what the important predictors for dementia risk are. METHODS: We analyzed data from the largest German health insurance company, including 117,895 dementia-free people age 65+. Follow-up was 10 years. Predictors were: 23 age-related diseases, 212 medical prescriptions, 87 surgery codes, as well as age and sex. Statistical methods included logistic regression (LR), gradient boosting (GBM), and random forests (RFs). RESULTS: Discriminatory power was moderate for LR (C-statistic = 0.714; 95% confidence interval [CI] = 0.708-0.720) and GBM (C-statistic = 0.707; 95% CI  = 0.700-0.713) and lower for RF (C-statistic = 0.636; 95% CI  = 0.628-0.643). GBM had the best model calibration. We identified antipsychotic medications and cerebrovascular disease but also a less-established specific antibacterial medical prescription as important predictors. DISCUSSION: Our models from German claims data have acceptable accuracy and may provide cost-effective decision support for early dementia screening.

A consistent version of distance covariance for right-censored survival data and its application in hypothesis testing.

Distance covariance is a powerful new dependence measure that was recently introduced by Székely et al. and Székely and Rizzo. In this work, the concept of distance covariance is extended to measuring dependence between a covariate vector and a right-censored survival endpoint by establishing an estimator based on an inverse-probability-of-censoring weighted U-statistic. The consistency of the novel estimator is derived. In a large simulation study, it is shown that induced distance covariance permutation tests show a good performance in detecting various complex associations. Applying the distance covariance permutation tests on a gene expression dataset from breast cancer patients outlines its potential for biostatistical practice.

Subdistribution hazard models for competing risks in discrete time.

A popular modeling approach for competing risks analysis in longitudinal studies is the proportional subdistribution hazards model by Fine and Gray (1999. A proportional hazards model for the subdistribution of a competing risk. Journal of the American Statistical Association94, 496-509). This model is widely used for the analysis of continuous event times in clinical and epidemiological studies. However, it does not apply when event times are measured on a discrete time scale, which is a likely scenario when events occur between pairs of consecutive points in time (e.g., between two follow-up visits of an epidemiological study) and when the exact lengths of the continuous time spans are not known. To adapt the Fine and Gray approach to this situation, we propose a technique for modeling subdistribution hazards in discrete time. Our method, which results in consistent and asymptotically normal estimators of the model parameters, is based on a weighted ML estimation scheme for binary regression. We illustrate the modeling approach by an analysis of nosocomial pneumonia in patients treated in hospitals.

Effects of the dementia care toolbox on personnel's self-reported confidence in patient care: a CRT in general practices.

BACKGROUND: In rapidly aging populations, general practitioners (GPs) are challenged in dementia care of patients with and without migration background. Uncertainties in treating dementia in migrant patients due to language barriers or information deficits are reported. To address these deficits, we developed the Dementia Care Toolbox which was judged helpful by GP practice personnel. This two-armed cluster-randomised trial (CRT) investigated the effects of this toolbox on German GPs' and practice assistants' (PrAs) attitudes and confidence in dementia care, especially in patients with migration background. METHODS: A total of 32 GP practices were recruited and randomised into intervention (toolbox use for 3 months) and waiting-list control (toolbox after follow-up). After 3 months all participating GPs and PrAs received a standardised questionnaire addressing their levels of self-reported confidence in dementia care for patients with and without migration background. A generalized estimating equation model that took practice cluster effects into account was applied to assess GPs and PrAs self-reported confidence in dementia care in patients with and without migration background. RESULTS: Overall, the intervention had no significant effect on self-reported confidence in dementia care. However, the use of the dementia care toolbox showed a tendency for a learning effect on knowledge about local support structures for migrant patients with dementia (odds ratio 1.43; 95% CI 0.68-3.03, p = 0.35) and for less communication difficulties with dementia patients in general (odds ratio 0.72; 95% CI 0.33-1.56; p = 0.40). Moreover, intervention practices showed a tendency towards more awareness of own limitations: less self-confidence regarding answering questions (odds ratio 0.82; 95% CI 0.36-1.86, p = 0.64) and providing information for patients with migration background (odds ratio 0.60; 95% CI 0.25-1.45, p = 0.26). CONCLUSION: The Toolbox Dementia Care increased awareness on the respective topic. Given a small sample size, further studies on its effectiveness in primary care are needed. TRIAL REGISTRATION: German Clinical Trials Register, DRKS00014632. Registered 02/08/2018.

Analysis of cesarean section rates in two German hospitals applying the 10-Group Classification System.

OBJECTIVES: In Germany, cesarean section (CS) rates more than doubled within the past two decades. For analysis, auditing and inter-hospital comparison, the 10-Group Classification System (TGCS) is recommended. We used the TGCS to analyze CS rates in two German hospitals of different levels of care. METHODS: From October 2017 to September 2018, data were prospectively collected. Unit A is a level three university hospital, unit B a level one district hospital. The German birth registry was used for comparison with national data. We performed two-sample Z tests and bootstrapping to compare aggregated (unit A + B) with national data and unit A with unit B. RESULTS: In both datasets (national data and aggregated data unit A + B), Robson group (RG) 5 was the largest contributor to the overall CS rate. Compared to national data, group sizes in RG 1 and 3 were significantly smaller in the units under investigation, RG 8 and 10 significantly larger. Total CS rates between the two units differed (40.7 vs. 28.4%, p<0.001). The CS rate in RG 5 and RG 10 was different (p<0.01 for both). The most relative frequent RG in both units consisted of group 5, followed by group 10 and 2a. CONCLUSIONS: The analysis allowed us to explain different CS rates with differences in the study population and with differences in the clinical practice. These results serve as a starting point for audits, inter-hospital comparisons and for interventions aiming to reduce CS rates.

Multiple aneurysms in subarachnoid hemorrhage - identification of the ruptured aneurysm, when the bleeding pattern is not self-explanatory - development of a novel prediction score.

BACKGROUND: In aneurysmal subarachnoid hemorrhage (SAH) and multiple intracranial aneurysms (MIAs) identification of the bleeding source cannot always be assessed according to the hemorrhage pattern. Therefore, we developed a statistical model for the prediction of the ruptured aneurysm in patients with SAH and multiple potential bleeding sources at the time of ictus. METHODS: Between 2012 and 2015, 252 patients harboring 619 aneurysms were admitted to the authors' institution. Patients were followed prospectively. Aneurysm and patient characteristics, as well as radiological findings were entered into a computerized database. Gradient boosting techniques were used to derive the statistical model for the prediction of the ruptured aneurysm. Based on the statistical prediction model, a scoring system was produced for the use in the clinical setting. The aneurysm with the highest score poses the highest possibility of being the bleeding source. The prediction score was then prospectively applied to 34 patients suffering from SAH and harboring MIAs. RESULTS: According to the statistical prediction model the main factors affecting the rupture in patients harboring multiple aneurysms were: 1) aneurysm size, 2) aneurysm location and 3) aneurysm shape. The prediction score identified correctly the ruptured aneurysm in all the patients that were used in the prospective validation. Even in the five most debatable and challenging cases assessed in the period of prospective validation, for which the score was designed for, the ruptured aneurysm was predicted correctly. CONCLUSIONS: This new and simple prediction score might provide additional support for neurovascular teams for treatment decision in SAH patients harboring multiple aneurysms. In a small prospective sample, the prediction score performed with high accuracy but larger cohorts for external validation are warranted.

Prevalence, incidence and future projection of diabetic eye disease in Europe: a systematic review and meta-analysis.

To examine the prevalence and incidence of diabetic eye disease (DED) among individuals with diabetes in Europe, a systematic review to identify all published European prevalence and incidence studies of DED in individuals with diabetes managed in primary health care was performed according to the MOOSE and PRISMA guidelines. The databases Medline, Embase and Web of Science were searched to 2 September 2017. Meta-analyses and meta-regressions were performed. The pooled prevalence estimates were applied to diabetes prevalence rates provided by the International Diabetes Foundation atlas and Eurostat population data, and extrapolated to the year 2050. Data of 35 prevalence and four incidence studies were meta-analyzed. Any diabetic retinopathy (DR) and diabetic macular edema (DME) were prevalent in 25.7% (95% CI 22.8-28.8%) and 3.7% (95% CI 2.2-6.2%), respectively. In meta-regression, the prevalence of DR in persons with type 1 diabetes was significantly higher compared to persons with type 2 diabetes (54.4% vs. 25.0%). The pooled mean annual incidence of any DR and DME in in persons with type 2 diabetes was 4.6% (95% CI 2.3-8.8%) and 0.4% (95% CI 0.5-1.4%), respectively. We estimated that persons with diabetes affected by any DED in Europe will increase from 6.4 million today to 8.6 million in 2050, of whom 30% require close monitoring and/or treatment. DED is estimated to be present in more than a quarter of persons with type 2 diabetes and half of persons with type 1 diabetes underlining the importance of regular monitoring. Future health services need to be planned accordingly.

Human spermatozoa of male patients with subfertility express the interleukin-6 receptor.

Male subfertility has been associated with bacterial infections and chronic inflammation. In this context, several studies investigated cytokine levels in seminal plasma, whereas interleukin-6 (IL-6) appears to be crucial. However, little is known about its receptor, the IL-6R expression on human spermatozoa. Thus, the aim of the present study was to screen spermatozoa for IL-6R expression and to identify its localisation. Semen samples of 137 patients (median age 37.69, SD ± 7.82) with subfertility were analysed. Sperm analysis including determination of IL-6 was performed following the World Health Organization criteria. Also, flow cytometry was performed for sperm IL-6R expression. IL-6R+ cells were used for immunofluorescence staining to identify receptor localisation. The results showed positive staining for IL-6R in the midpiece of spermatozoa. Furthermore, a significant correlation between sperm IL-6R expression, seminal plasma IL-6 and total sperm count could be demonstrated, whereas a negative correlation was observed in sperm IL-6R expression and motility. However, no statistical significance could be observed between IL-6R expression, vitality and morphology. Moreover, incubation of spermatozoa with IL-6 led to a slight but significant decrease in motility after 24 hr. These data suggest that IL-6R expression may play a role in impaired sperm function during inflammation.

Correlation-adjusted regression survival scores for high-dimensional variable selection.

BACKGROUND: The development of classification methods for personalized medicine is highly dependent on the identification of predictive genetic markers. In survival analysis, it is often necessary to discriminate between influential and noninfluential markers. It is common to perform univariate screening using Cox scores, which quantify the associations between survival and each of the markers to provide a ranking. Since Cox scores do not account for dependencies between the markers, their use is suboptimal in the presence of highly correlated markers. METHODS: As an alternative to the Cox score, we propose the correlation-adjusted regression survival (CARS) score for right-censored survival outcomes. By removing the correlations between the markers, the CARS score quantifies the associations between the outcome and the set of "decorrelated" marker values. Estimation of the scores is based on inverse probability weighting, which is applied to log-transformed event times. For high-dimensional data, estimation is based on shrinkage techniques. RESULTS: The consistency of the CARS score is proven under mild regularity conditions. In simulations with high correlations, survival models based on CARS score rankings achieved higher areas under the precision-recall curve than competing methods. Two example applications on prostate and breast cancer confirmed these results. CARS scores are implemented in the R package carSurv. CONCLUSIONS: In research applications involving high-dimensional genetic data, the use of CARS scores for marker selection is a favorable alternative to Cox scores even when correlations between covariates are low. Having a straightforward interpretation and low computational requirements, CARS scores are an easy-to-use screening tool in personalized medicine research.

Incidence of Rhegmatogenous Retinal Detachment in Europe - A Systematic Review and Meta-Analysis.

PURPOSE: To examine the prevalence and incidence of rhegmatogenous retinal detachment (RRD) in Europe. METHODS: A systematic review of the prevalence and incidence of RRD in Europe according to the meta-analyses of observational studies in epidemiology (MOOSE) guidelines was performed using the databases PubMed, Embase, and Web of Science followed by a meta-analysis of incidence data. RESULTS: 658 studies were identified, 76 full-text articles were assessed, and 5 studies were eligible for meta-analysis. Random-effects pooled mean annual incidence of RRD in Europe was 13.3 cases per 100,000 inhabitants (95% CI: 11.3-15.6) based on the examination of 95,765 individuals from the 5 incidence studies. No European studies reporting prevalence of retinal detachment were identified. CONCLUSION: Published articles on the prevalence and incidence of RRD in Europe are limited. Further epidemiological research is needed for adequate healthcare planning.

Prevalence of Retinal Vein Occlusion in Europe: A Systematic Review and Meta-Analysis.

PURPOSE: The aim of this study was to examine the prevalence and incidence of retinal vein occlusions (RVO) in Europe. METHODS: A systematic review and meta-analysis of the prevalence and incidence of RVO in Europe according to the Meta-Analyses of Observational Studies in Epidemiology (MOOSE) guidelines was performed in the databases PubMed, Embase and Web of Science. Based on Eurostat data, the total number of affected individuals in the EU was calculated and projected to the year 2050. RESULTS: The random-effects pooled prevalence of RVO in Europe was 0.7% (95% CI: 0.5-0.9%) in persons aged 55 years and older based on the examination of 25,002 individuals from 4 prevalence studies. No European incidence studies were identified. The total number of affected individuals in the EU was estimated to rise by almost 20% from 900,000 today to 1.1 million in 2050. CONCLUSION: Published articles on the prevalence and incidence of RVO in Europe are limited. With the expected increase in affected persons, further epidemiological research is warranted for adequate healthcare planning.

Effectiveness of the online-eLearning program KeepCoool at improving the vaccine cold chain in general practices.

BACKGROUND: Protecting vaccines from freeze damage is a poorly addressed problem. We describe the effectiveness of the eLearning KeepCoool on cold chain maintenance in general practices. METHODS: For this intervention study, temperatures of vaccine refrigerators were logged at one-minute intervals. Personnel from practices with cold chain breaches was offered the eLearning. The primary outcome was the intervention's effectiveness to achieve temperatures in the target range (2 to 8°C) in the sixth week (follow-up) compared to the first (baseline). Using continuous temperature data, a generalized additive model for location, scale and shape was estimated. RESULTS: The practice response rate was 38% (64 of 168). At baseline, 73% of the practices and 68% of the refrigerators (51 of 75) showed cold chain breaches. 47% of the practices (n = 22 with 24 refrigerators) participated in the eLearning (55 physicians and practice assistants). At follow-up, 17% of those refrigerators were in the target range continuously, 38% reached targets 95% of the time while always >0°C, and temperatures ≤0°C decreased by 63%. Based on 2 million temperature data, the average Euclidian distance based on regression showed a significant improvement (p<0.05). CONCLUSION: The eLearning KeepCoool improved the practices´ vaccine cold chain. It is freely available at https://keepcoool.ukbonn.de.

Prevalence and Incidence of Registered Severe Visual Impairment and Blindness Due to Uveitis in Germany.

PURPOSE: To assess prevalence and incidence of severe visual impairment (VI) and blindness due to uveitis in Northrhine, Germany. METHODS:  A random sample of 5088 out of 20 390 cases of the blindness registry in Northrhine was evaluated. The number of cases of severe VI and blindness were projected up to the year 2030. RESULTS: The prevalence of registered severe VI and blindness due to uveitis was 1123.9 (95% CI: 1059.5 to 1191.2) per 100 000 in our study cohort and 550.31 (95% CI: 365.98-794.38) per 100 000 in Northrhine, respectively. The incidence of registered severe VI and blindness due to uveitis was 311.8 per 100 000 PYs (95% CI: 149.6-572.7) in our study cohort and 11.1 per 100 000 (95% CI: 5.6-19.8) in Northrhine, respectively. CONCLUSIONS: We report prevalence and incidence of severe VI and blindness due to uveitis in a large German cohort.

Modifications in ocular microperfusion after transcatheter aortic valve implantation.

Cerebral embolization is a known complication of transcatheter aortic valve implantation (TAVI) but the effect of the procedure on the ocular perfusion is currently unclear. Thus, we investigated post-procedural morphologic and perfusion changes of the retina and choroid, using optical coherence tomography angiography (OCTA) and color fundus photography (CFP) in a prospective cohort study. Ophthalmic examinations were conducted pre- and post-TAVI. OCTA images were analyzed quantitatively based on vessel density and skeleton density of the superficial and deep retinal plexus as well as the signal intensity and flow deficits in the choriocapillaris. CFP images were assessed for presence of acute retinal ischemia, optic nerve swelling, vessel emboli, hemorrhages and cotton wool spots. Data was analyzed using linear mixed models. Twenty patients (9 women; 11 men) at a mean age of 81 ± 6 years were included. Pre- and post-interventional ocular imaging data were available for 32 eyes. The analysis revealed a significant impairment of the choriocapillaris perfusion after TAVI with an increased proportion of flow deficits (p = 0.044). When controlling for blood pressure, the average size of choriocapillaris flow voids was significantly increased (systolic and diastolic, p = 0.039 and 0.029). Qualitatively, focal areas of retinal ischemia were detected on OCTA in 33% of participants. Silent emboli or cotton wool spots were identified on CFP in 21%. Our findings indicate a reduced choroidal perfusion as well as areas of retinal ischemia and embolization in a considerable proportion of patients following TAVI. Pending confirmation in a larger sample, these complications merit monitoring as well as inclusion in consent procedures for TAVI.

Altered Serum Phospholipids in Atopic Dermatitis and Association with Clinical Status.

Circulating phospholipids have been considered as biomarkers and therapeutic targets in multiple disorders. Atopic dermatitis (AD) is the most common inflammatory skin disease. Although there are numerous studies having addressed stratum corneum lipids in the context of epidermal barrier, little is known about the circulating lipids in patients with AD. In this study, we explored the changes of serum phospholipids in AD using liquid chromatography coupled to tandem mass spectrometry and sought serum lipids' contribution to clinical status. Several serum levels of phospholipids were altered in the AD group (n = 179) compared with that in healthy controls (n = 47) and patients without AD with atopic comorbidities (n = 22); lipids exhibiting the apparent changes included increased sphingosine, multiple variants of phosphatidylcholine, and decreased ceramide (16:0) in patients with AD. Moreover, serum levels of sphingosine correlated with the severity of AD, and sphingosine and ceramide(16:0) were also detected as the risk-increasing effect and risk-reduction effect of AD, respectively. In summary, alterations in the serum concentration of phospholipids are seen in patients with AD. Although more detailed investigations will be needed to evaluate the significance of the changes in circulating lipids in AD, these findings can provide, to our knowledge, previously unreported insight into AD's pathogenesis and therapeutic strategies.