Several methodologies have been developed to calculate the enrichment, in order to improve the meaning of the list and to improve the predictive capability; some examples are Onto-Express, MAPPFinder, GoMiner, DAVID or EASE .
As a methodology for this analysis, we have worked with GSEA . Gene Set Enrichment Analysis (GSEA) is a computational method that determines whether an a priori defined set of genes shows statistically significant, concordant differences between two biological states. The results of this analysis allow finding common pathways and important differences between data clusters, thus linking the clinicaldata to gene expression. This method was developed to avoid common disadvantages of other methods : (i) Single gene methods tend to ignore effects on pathways, and work better when individual genes have a very large and consistent effects on a specific phenotype. Subtle effects are commonly ignored and after correcting for multiple tests, very few genes may be still significant. (ii) When testing different samples of the same system, the gene overlap tend to be very small, then the real effect of a single gene is unclear if it significant in a sample and it is not in another.
The assessment of the cardiovascular safety profile of any newly developed antihyperglycemic drug is mandatory before registration, as a meta-analysis raised alarm describing a significant increase in myocardial infarction with the thiazolidinedione rosiglitazone. The first results from completed cardiovascular outcome trials are already available: TECOS, SAVOR-TIMI, and EXAMINE investigated dipeptidyl peptidase 4 (DPP-4) inhibitors, ELIXA, LEADER, and SUSTAIN-6 investigated glucagon-like peptide 1 (GLP-1) receptor agonists, and EMPA-REG OUTCOME and CANVAS investigated sodium-dependent glucose transporter 2 (SGLT-2) inhibitors. LEADER, SUSTAIN-6, EMPA-REG OUTCOME, and CANVAS showed potential beneficial results, while the SAVOR-TIMI trial had an increased rate of hospitalization for heart failure. Meanwhile, the same drugs are investigated in preclinical experiments mainly using various animal models, which aim to find interactions and elucidate the underlying downstream mechanisms between the antihyperglycemic drugs and the cardiovascular system. Yet the direct link for observed effects, especially for DPP-4 and SGLT-2 inhibitors, is still unknown. Further inquiry into these mechanisms is crucial for the interpretation of the clinical trials’ outcome and, vice versa, the clinical trials provide hints for an involvement of the cardiovascular system. The synopsis of preclinical and clinicaldata is essential for a detailed understanding of benefits and risks of new antihyperglycemic drugs.
model. The problem however is that large amounts (about 80%) of clinicaldata is unstructured [ Ter13 ]. Even though some data can be entered by clinicians in structured form such as medication, diagnosis, procedures etc., structured report- ing in general is not widely applied in practice. Especially finding descriptions and observations are commonly documented in free text clinical reports or in images. In subsequent decision processes however, clinicians do not have the time to review all potentially valuable images and reports. To efficiently include information from unstructured data in clinical decisions one needs to extract structured representations. In the context of Natural Language Processing (NLP) information extraction from text involves the following five steps [ Cun06 ]: Named Entity Recognition (NER), co-reference resolution, template element construction, template relation construction and template scenario production. The first two steps extract entities from text. Biomedical ontologys are commonly used within NER, which is also referred to as semantic annotation. The ontology then allows the semantic understanding of the extracted entities. For instance from a sentence “lymph nodes in the mediastinum with a size up to 1.6 cm” the entities lymph node, mediastinum, mediastinal lymph node as well as the measurement 1.6 cm is extracted. The problem however is that a set of extracted entities does not represent the asserted content precisely enough to be useful in further applications. Thus, the remaining steps resolve the relations between the extracted entities. For instance that the measurement describes the size of the mediastinal lymph node (see figure 1.1).
In this work, methods for constructing a statistical shape model without the need of manually delineated ground truth data have been proposed. The training data is assumed to be the result of any segmentation algo- rithm or may originated from non-expert annotators. Depending on this data acquisition, the shape examples will contain regions with erroneous boundary segmentations. In order to handle such corrupted parts, each landmark point is assigned a probability of being a boundary. These esti- mated probabilities rely on image-based methods and have an average de- viation to the optimal probabilities of about 22%. During the further model building procedure, two different approaches were introduced, to treat the erroneous data appropriately, before PCA is applied. The statistics inher- ent in the training data is used for reconstruction of the corrupted shapes. The imputation method is a rather brute-force approach, where low prob- ability points are replaced with the mean of corresponding landmarks. Re- cent advances in sparse optimization yield to a robustified version of PCA, where a low-rank matrix is recovered from the corrupted training data ma- trix. Outliers are separated from the data, by solving a convex optimiza- tion problem. By incorporating the boundary probabilities into the RPCA method, the prior knowledge can be exploited, by weighting the selection of the outliers in RPCA. After the training data is reconstructed, using ei- ther the imputation or RPCA approach, the data is assumed to be free of outliers. By applying PCA to the data, the low-dimensional linear subspace is found, to perform dimensionality reduction and model the variability of the training data.
Physiological Differences Between Mice and Humans Most prominently, humans and mice differ considerably in their circulating immune cell composition. In humans, neutrophils constitute the most abundant leukocyte subpopulation (40–70%), whereas mice show up to 84% lymphocytes ( 193 ), which may impact on the relative contributions of innate and adaptive immunity to host responses. As already mentioned, distinct receptor expression regarding TLRs, PARs and Fc receptors determines immune cell capacity of platelets. This makes it often difficult to translate results from animal models to the clinical situation. To overcome this problem, mice expressing human receptors have been generated. While this was successful in some cases (FcγRIIA) ( 194 ), other attempts failed so far to lead to functional receptor expression (PAR-1) ( 195 , 196 ). However, the role of FcγRIIA was never addressed in a murine sepsis model. Furthermore, while LPS challenge yields similar inflammatory responses in mice and men, including cytokine production and lymphopenia, humans are more sensitive to LPS than mice, which necessitate the use of LPS concentrations in mouse models that surpass those required to induce septic shock in humans about 1,000–10,000-fold ( 197 , 198 ).
The clinical diagnosis of herpes zoster corre- sponded well with the proof of VZV in 81.3% of the cases. When the clinical investigator had sent samples from any site of a clinically herpetic lesion, excluding the genital or oral region, specified with the diagnosis herpes infection, a positive result was achieved for HSV-1 in 30.0%, for HSV-2 in 20.6% and for VZV in 14.1%, with a total of 64.7% positive results. The classical diagnosis of labial herpes was supported by an exclusively positive PCR for HSV-1 in 61.9%, the same applied for the scenario where the clinicians stated (aphthous) gingivostomatitis or oral ulcer, cor- responding to positive PCR results for HSV-1 in 40.0% and 15.0%, respectively, but no HSV-2 and VZV could be isolated only on one occasion. Eventually these samples derived from patients with primary HSV-1 infections. Swabs from lesions with the diagnosis her- petic keratoconjunctivitis led to a positive PCR result in 41.2%; in 29.4% HSV-1, in 5.9% HSV-2 and in 5.9% VZV were isolated.
Vitamin D deficiency is considered a worldwide public health problem, in particular because in most countries, large parts of the general population do not meet the dietary vitamin D requirements as recommended by nutritional vitamin D guidelines [ 1 – 10 ]. Vitamin D is important for musculoskeletal health and, historically, is known to be effective for the prevention and treatment of rickets and osteomalacia, and may also reduce fractures and falls in the elderly [ 11 – 14 ]. Several observational studies have shown that a poor vitamin D status is associated with various extra-skeletal diseases such as cardiovascular and metabolic diseases, cancer, autoimmune and neurological diseases [ 14 , 15 ]. By contrast, randomized controlled trials (RCTs) have, in the majority, failed to show clinically relevant effects of vitamin D supplementation on these outcomes [ 16 – 18 ]. Therefore, it has been suggested that vitamin D deficiency may be rather a risk marker for ill health than a causal factor for many diseases [ 19 ]. Consequently, there is still an ongoing scientific controversy on potential extraskeletal effects of vitamin D that is beyond the scope of the present review. It must, however, be underlined that meta-analyses of RCTs support the notion that vitamin D supplementation may reduce the risk of mortality, infections, asthma exacerbations and some pregnancy outcomes, although these data have certain limitations and it might be premature to clearly establish or claim causality [ 20 – 31 ].
In machine learning, model validation and tuning of parameters is usually done by a tech- nique called K-fold Cross Validation. It generally works as follows: The available data is divided into K folds. Then, the model is fit on K − 1 of these folds and evaluated on the remaining fold in terms of prediction accuracy. This process is repeated K times, i.e. the observations in each fold are used as test set once and K times as training set. Whereas the procedure is straightforward in normal statistical analysis, it gets more complicated for cox regression and in particular penalized versions. For cox regression, it is not quite obvious how to quantify prediction accuracy on a test set. The estimated coefficients only allow to set the risk of each patient in relation to other observations. A very basic approach to assess the model’s predictive accuracy is to calculate the partial likelihood based on the observations in the test set. Then, the risk for each patient can be set into relation to the risks of other members of the test set. However, this approach is only of limited use if the dataset does not contain many events. To overcome this issue, several versions of cross validation for survival methods have been proposed and implemented, e.g. by Simon et al. (2011) in the software package glmnet or more recently by Dai and Breheny (2019).
Sixty-nine consecutive patients were tested at the nephrology inpatient ward (n = 52) and the oncology ward (n = 17) of the University Hospital of Vienna from 2013 to 2014. Inclusion criteria were the following: age over 18 years, clinical signs of comorbidities constituting a high risk of AKI or evident AKI. Only patients admit- ted to the intensive care unit and the nephrology in- patient clinic were selected for testing. These included kidney transplant recipients for the evaluation of ische- mic reperfusion injury. Preexistent chronic kidney dis- ease was no exclusion criterion in the nephrology group. Inclusion criteria in the cisplatin group were age higher than 18 years, no history of CKD, and the initi- ation of chemotherapy based on cisplatin. Written and oral informed consent to the study and publication of the results was obtained from all patients. The investiga- tion was approved by the ethics committee of the Medical University of Vienna (approval number 1598/2013). The patients’ clinicaldata, demographics, medical history, medication and laboratory data were obtained from their medical files and the databases of the hospital. Representa- tive patients (oncology inpatients, inpatients with more than four data points, and hospitalization for more than 4 days) were selected. Twelve patients were chosen for the report; a further four cases have been described in the supplement (Additional file 1: Table A).
A medical scientist must demonstrate responsibility to patients and their symptoms, and pay particular attention to clinicaldata, as Feinstein stated. 39 In his opinion, in the early 1980s, physicians focused heavily on paraclinical data and on therapies. 39 He complained that quality control was not given the importance that it should be. 39 He indicated that it was more difficult to express observations as data because not a great number of rating scales, such as scores, had been created by 1982. 39 Some rating scales already existed, but they were not suitable because of their poor repeatability. 39 Therefore, a lot of important information was omitted at that time and quality control was not adequate. 39
The question of why the application of CAP, the e ﬀectivness in terms of decontamination, stimulation the healing of chronic wounds or reducing growth of tumor cells, presents mostly no side eﬀects or dis- comfort is calling for further clinicaldata and biological interpretation. From the beginning, safety of plasma medicine was an important issue in research. With the increasing knowledge about details of plasma-cell and plasma-tissue interactions including reports on plasma impact on DNA integrity of cells in vitro, at ﬁrst the risk of potential genotoxic eﬀects of CAP was in the focus, and the basic insight that biological plasma eﬀects are mainly based on the activity of reactive oxygen and nitrogen species (ROS, RNS)  . Because these ROS and RNS are the same as occur in regular physiological and biochemical processes in the body, mammalian cells have mechanisms to save from excess levels of ROS and RNS  . It was demonstrated that such antioxidative pro- tection mechanisms are up-regulated in cells in response to plasma treatment  . Further detailed investigations could demonstrate that detrimental plasma eﬀects on cells result either in cellular repair pro- cesses or in induction of programmed cell death (apoptosis)  . Meanwhile, several studies using well-established and accepted ex- perimental procedures have proven that CAP treatment does not cause increased risk for genotoxicity [16 –18] . This was supported by ﬁrst clinical experience  . However, besides this inevitable exclusion of such fundamental side eﬀects, it is also necessary to monitor and learn about acute and short-term side e ﬀects of CAP treatment.
There are various further options to integrate the heterogeneous group structure of multi-omics data into the model building process. One could simply build separate models for each modality and then combine these into an overall model [Zhao et al., 2014]. Another approach includes some pre-knowledge about the prediction capabilities of certain modalities into the estimation process. Especially clinicaldata is often already known to impact an outcome like cancer prognosis. This part of the data is usually very small compared to molecular measurements like gene expression data and therefore, important clinical covariates might get lost in a model which treats all modalities the same way. In a penalized model, the so called ”favoring” approach would only penalize the molecular part of the covariates and leave the clinical covariates un-penalized. The favoring approach and some alternatives are discussed in a paper by Boulesteix and Sauerbrei .
It is very difficult to try and apply standards and data definitions after a trial database has been designed and the data collected, or to try and change data structures unless a trial has been designed from the beginning with those data structures in mind (eg, it is much easier to map data to ClinicalData Interchange Standards Consortium - Study Data Tabulation Mod- el (CDISC SDTM), the tabular data format used by the Food and Drug Administration (US) (FDA), if it has been collected using ClinicalData Interchange Standards Consortium - ClinicalData Acquisition Standards Harmonisation (CDISC CDASH) data items). Legacy data conversion can be done when there is value in combining data from prior trials, but it is resource-intensive and may compromise data in- tegrity. The time and costs required for retrospective ‘standardisation’ would put such an exercise beyond the resources of many non-commercial units. Instead, it is important that standards are designed from the start, with decisions made about the coding and oth- er systems to be used made as part of the trial design process.
Finally, Ensembl is a genomics centric resource that integrates the information for a comprehensive set of mainly vertebrate genomes and provides automatic annotations derived from genome sequences. Ensembl produces protein sequence sets for each organism directly derived from the gene predictions (11). Ensembl’s major strength is the easy connection between proteomics and biological knowledge as it directly links proteins to genes and transcripts. In PRIDE, these four are by far the most popular searched databases, together with other databases specific to certain model organisms like The Arabidopsis Information Resource (TAIR) (14). When protein identifications have been generated from different databases, making results comparable can be troublesome (15). This is due to the existence of heterogeneous and changing identifiers referring to the same protein in different resources. In order to overcome this common problem in proteomics, the Protein Identifier Cross Referencing (PICR) system was launched in 2007 at the EBI (16). PICR uses the archive database of UniProt (UniParc) (10) as a data warehouse to offer protein cross-references based on 100% sequence identity from over 70 distinct source databases loaded into UniParc.
ELISA is the main approach for the sensitive quantification of protein biomarkers in body flu- ids and is currently employed in clinical laboratories for the measurement of clinical mark- ers. As such, it also constitutes the main methodological approach for biomarker validation and further qualification. For the latter, specific assay performance requirements have to be met, as described in respective guidelines of regulatory agencies. Even though many clini- cal ELISA assays in serum are regularly used, ELISA clinical applications in urine are signif- icantly less. The scope of our study was to evaluate ELISA assay analytical performance in urine for a series of potential biomarkers for bladder cancer, as a first step towards their large scale clinical validation. Seven biomarkers (Secreted protein acidic and rich in cyste- ine, Survivin, Slit homolog 2 protein, NRC-Interacting Factor 1, Histone 2B, Proteinase-3 and Profilin-1) previously described in the literature as having differential expression in blad- der cancer were included in the study. A total of 11 commercially available ELISA tests for these markers were tested by standard curve analysis, assay reproducibility, linearity and spiking experiments. The results show disappointing performance with coefficients of varia- tion >20% for the vast majority of the tests performed. Only 3 assays (for Secreted protein acidic and rich in cysteine, Survivin and Slit homolog 2 protein) passed the accuracy thresh- olds and were found suitable for further application in marker quantification. These results collectively reflect the difficulties in developing urine-based ELISA assays of sufficient ana- lytical performance for clinical application, presumably attributed to the urine matrix itself and/or presence of markers in various isoforms.
with hemodialysis (HD) is unknown.
Methods e-Ultimaster is a prospective, single-arm, multi-center registry with clinical follow-up at 3 months and 1 year.
Results A total of 19,475 patients were enrolled, including 1466 patients with CKD, with 167 undergoing HD. Patients with CKD had a higher prevalence of overall comorbidities, multiple/small vessel disease (≤ 2.75 mm), bifurcation lesions, and more often left main artery treatments (all p < 0.0001) when compared with patients with normal renal function (reference). CKD patients had a higher risk of target lesion failure (unadjusted OR, 2.51 [95% CI 2.04–3.08]), target vessel failure (OR, 2.44 [95% CI 2.01–2.96]), patient-oriented composite end point (OR, 2.19 [95% CI 1.87–2.56]), and major adverse cardio- vascular events (OR, 2.34 [95% CI 1.93–2.83, p for all < 0.0001]) as reference. The rates of target lesion revascularization (OR, 1.17 [95% CI 0.79–1.73], p = 0.44) were not different. Bleeding complications were more frequently observed in CKD than in the reference (all p < 0.0001).
Methods A cross-sectional survey was performed in the province of la Ngou- nie, Gabon between December 2015 and Februrary 2016 and 947 participants of all ages were recruited. Clinical parameters and behavioural exposure fac- tors were ascertained by questionnaire-based interviews. Parasitological anal- ysis of blood samples was performed for L. loa detection. Diagnostic scores consisting of clinical and behavioural factors were modelled to predict loiasis in sub-groups residing in endemic regions.
In Bulgaria, neither clinical guidelines nor clinical algo- rithms were successfully integrated into most of the CPs. This occurred primarily because only a few clinical guidelines had been introduced in Bulgaria when CPs were implemented. Secondly, over 80% of the Bulgarian CPs are a combination of similar diagnoses and conditions. Like the diagnosis related groups (DRGs) for patients differing in their medical and biological characteristics, they are rather broad in scope and therefore cannot contain a clinical algorithm as an option for solving a particular problem, but rather offer a range of possible diagnostic and therapeutic strategies. Incorporating multiple diagnoses, conditions and critical procedures into one CP, parallel to the DRGs, impedes the implementation of clinical guidelines and clinical algorithms.
Students in their final year of undergraduate medial education showed significantly higher scores for factor 3 (“Securing information”) and for its two items “Summar- izing” and “Checking with the patient”, respectively, than students from semester 10. This is an interesting finding which could reflect that medical students up to semester 10 are still taught history taking in the traditional way by obtaining the history in separate sequential categories (e.g. “history of the present illness”, “past medical his- tory”, “review of systems” etc.) [ 24 ]. It has been shown over 40 years ago, that his method to teach history taking is deficient when used as the only teaching method [ 25 ]. The combination of content and process of history taking leads to better learning outcomes with respect to clinical reasoning [ 26 , 27 ]. Students in their final year have more learning opportunities with real history taking while work- ing full time on hospital wards. Therefore, their learning is less static and rather resembles problem-based learning tutorials [ 28 ]. This might be a reason why they achieved higher scores for factor 3 in our study.