International Journal of Risk & Safety in Medicine - Volume 27, issue s1

Authors: Kamalov, M. | Dobrynin, V. | Balykina, J. | Kolbin, A. | Verbitskaya, E. | Kasimova, M.

Article Type: Abstract

Abstract: BACKGROUND: The actively developing approach in modern medicine is the approach focused on principles of evidence-based medicine. The assessment of quality and reliability of studies is needed. However, in some cases studies corresponding to the first level of evidence may contain errors in randomized control trials (RCTs). Solution of the problem is the Grading of Recommendations Assessment, Development and Evaluation (GRADE) system. Studies both in the fields of medicine and information retrieval are conducted for developing search engines for the MEDLINE database [1 ]; combined techniques for summarization and information retrieval targeted to solving problems of finding the best medication …based on the levels of evidence are being developed [2 ]. OBJECTIVE: Based on the relevance and demand for studies both in the field of medicine and information retrieval, it was decided to start the development of a search engine for the MEDLINE database search on the basis of the Saint-Petersburg State University with the support of Pavlov First Saint-Petersburg State Medical University and Tashkent Institute of Postgraduate Medical Education. Novelty and value of the proposed system are characterized by the use of ranking method of relevant abstracts. It is suggested that the system will be able to perform ranking based on studies level of evidence and to apply GRADE criteria for system evaluation. METHODS: The assigned task falls within the domain of information retrieval and machine learning. Based on the results of implementation from previous work [3 ], in which the main goal was to cluster abstracts from MEDLINE database by subtypes of medical interventions, a set of algorithms for clustering in this study was selected: K-means, K-means ++, EM from the sklearn (http://scikit-learn.org ) and WEKA (http://www.cs.waikato.ac.nz/~ml/weka/ ) libraries, together with the methods of Latent Semantic Analysis (LSA) [4 ] choosing the first 210 facts and the model “bag of words” [5 ] to represent clustered documents. During the process of abstracts classification, few algorithms were tested including: Complement Naive Bayes [6 ], Sequential Minimal Optimization (SMO) [7 ] and non linear SVM from the WEKA library. RESULTS: The first step of this study was to markup abstracts of articles from the MEDLINE by containing and not containing a medical intervention. For this purpose, based on our previous work [8 ] a web-crawler was modified to perform the necessary markuping. The next step was to evaluate the clustering algorithms at the markup abstracts. As a result of clustering abstracts by two groups, when applying the LSA and choosing first 210 facts, the following results were obtained: 1) K-means: Purity = 0,5598, Normalized Entropy = 0.5994; 2) K-means ++: Purity = 0,6743, Normalized Entropy = 0.4996; 3) EM: Purity = 0,5443, Normalized Entropy = 0.6344. When applying the model “bag of words”: 1) K-means: Purity = 0,5134, Normalized Entropy = 0.6254; 2) K-means ++: Purity = 0,5645, Normalized Entropy = 0.5299; 3) EM: Purity = 0,5247, Normalized Entropy = 0.6345. Then, studies which contain medical intervention have been considered and classified by the subtypes of medical interventions. At the process of classification abstracts by subtypes of medical interventions, abstracts were presented as a "bag of words" model with the removal of stop words. The results: 1) Complement Naive Bayes: macro F-measure = 0.6934, micro F-measure = 0.7234; 2) Sequantial Minimal Optimization: macro F-measure = 0.6543, micro F-measure = 0.7042; 3) Non linear SVM: macro F-measure = 0.6835, micro F-measure = 0.7642. CONCLUSIONS: Based on the results of computational experiments, the best results of abstract clustering by containing and not containing medical intervention were obtained using the K-Means ++ algorithm together with LSA, choosing the first 210 facts. The quality of classification abstracts by subtypes of medical interventions value for existing ones [8 ] has been improved using non linear SVM algorithm, with “bag of words” model and the removal of stop words. The results of clustering obtained in this study will help in grouping abstracts by levels of evidence, using the classification by subtypes of medical interventions and it will be possible to extract information from the abstracts on specific types of interventions. Show more

Keywords: Data retrieval, computational experiments

DOI: 10.3233/JRS-150710

Citation: International Journal of Risk & Safety in Medicine, vol. 27, no. s1, pp. S106-S107, 2015

Authors: Isakova, J.

Article Type: Abstract

Abstract: BACKGROUND: Rapid development of medicine requires regular update of clinical data evidence. This task accomplishment requires participation of numerous specialists in evidence-based medicine, who are proficient in various statistical methods and can work with big data analysis tools in biomedical sciences. This, in turn, requires significant time and other resources. Today, at the peak of IT development, cognitive systems in the field of medicine with special technologies of data collection and analysis, is the start of a new trend. OBJECTIVE: The development of cognitive IT system for drug prescription with the potential to analyze automatically the information about …drugs effectiveness and safety on the basis of clinical practice experience and scientific data according to evidence levels and patients’ personal characteristics. METHODS: The cognitive system was developed with the use of United Medical Knowledge Base (UMKB). UMKB is a semantic network of medical knowledge, which is structured according to the medical ontologies and the theory of fuzzy logic. UMKB is being filled simultaneously in all the areas of medicine. From one side it is filled by means of the linguistic module analyzing medical texts, from the second side - by academic institutions, from the third side – by the cognitive IT systems with the data from electronic health records (EHRs). Native language of UMKB is Russian. It is designed primarily for use in the Russian clinical practice. However the platform for filling knowledge is multilingual and supports any other languages. This means that the practice of world schools may also be integrated and used in UMKB. The peculiarity lies in the fact that UMKB is presented as a semantic network where biomedical knowledge are structured according to certain medical ontologies (special rules of information storage that 𠄼carries𠄽 data: phenomena, processes, simple and complex concepts in medicine, - in the form of interrelated objects). The keystone underlying UMKB is the model of medical knowledge representation, which is able to describe any area of medicine. With the help of this model one can accurately simulate risk factors, etiology, and pathogenesis of a disease (probability, time of development and the sequence of pathological signs at each stage of a disease). While describing pathological and compensatory mechanisms the database provides an opportunity to clarify a lot of conditions that affect this mechanism. It is also simple to simulate structural and functional features of the concept and its relationships (for example, compensatory mechanisms, reflexes, complex anatomical structures, all the features of variant anatomy and other characteristics), which form reactivity and resistance of the organism. All this is very important for cognitive IT systems concerning personalized and evidence-based medicine. When describing medical knowledge there are often situations of uncertainty, lack of sufficiently complete and accurate data on the subject area, poorly understood phenomena, conflicting theories or imprecise concepts. Semantic network of UMKB presents complex relationships among medical concepts characterized by the following features: type and direction of relationship, its weight and value, accuracy and personalization of the weight or value of relationship, date of actualization. Multifactorial influence on the weight or value of relationship, a lot of elementary and intermediate traits that influence weight, the moment of actualization are supported to formalize. United Medical Knowledge Base is a large-scale project, its main goal is to increase the quality and duration of life through personalized care based on evidence that can only be achieved by combining medical big data from various fields of biomedical sciences. RESULTS: On the basis of UMKB a prototype of the cognitive IT system PharmExpert with analytical potential was developed. PharmExpert is a clinical decision support system for drug prescribing, which is integrated into medical information system at health institutions and analyzes electronic health records (EHRs) in any format of the background mode, correcting drug therapy according to personal patient's profile and data about compatibility of the drugs. The system has a very important function - self-learning that will help it to absorb a huge mountain of medical data from routine clinical practice in the nearest future. Now it works on the basis of data from UMKB, handbooks in pharmacology, summaries of medical products characteristics (SmPCs), available reviews of scientific literature and clinical guidelines on drugs interactions and compatibility. In the short term, at the stage of clinical testing, PharmExpert memorizing all the cases of clinical experience and the reaction of the physicians (accepting or ignoring the recommendations of the system), will be able to realize self–learning function by rebuilding ties and remodeling knowledge of the semantic network according to clinical data and generating the best standards of drug therapy taking into account personal characteristics of the patient and levels of data evidence. Working in the background mode is one of the most important advantages of the system. The physician is not asked to enter any additional data beyond that the specialist enters into the EHR on an everyday basis. Now PharmExpert is installed in the medical information systems of the range of clinical centers in the Russian Federation. CONCLUSIONS: We developed a prototype of cognitive IT system for drug prescription with the potential to analyze automatically the information about drugs effectiveness and safety on the basis of clinical practice experience and scientific data according to evidence levels and patients’ personal characteristics. The system is based on the structured semantic network of medical knowledge from UMKB. Show more

Keywords: Cognitive IT-systems, big data, medicine, medical knowledge, semantic, network

DOI: 10.3233/JRS-150711

Citation: International Journal of Risk & Safety in Medicine, vol. 27, no. s1, pp. S108-S109, 2015

Authors: Khakimov, N. | Khasanova, G. | Ershova, K. | Gibadullina, L. | Vetkina, T. | Lobisheva, G. | Chumakova, A.

Article Type: Abstract

Abstract: BACKGROUND: The relevance of the problem of colorectal cancer (CRC) is evident because of extremely high morbidity and mortality rates, associated with this disease. CRC is mostly diagnosed only at very advanced stages. The reduction of mortality can be achieved by the popularization of screening-methods for early identification of CRC and adenomatous polyps of the colon, which are proved to be precancerous condition. Fecal occult blood test is a well-known method of screening for CRC. The advantages of this method when compared, for example, with colonoscopy are its simplicity and cost-effectiveness. Two techniques are usually used for detection of occult …blood in the stool: Hemoccult (Guaiac) test and immunochemical test for hemoglobin. There is no consensus among researchers regarding the validity of these tests for the diagnosis of colorectal cancer. For example, J.S. Mandel (1996) notes 60% sensitivity of Guaiac-test for the detection of the early forms of colorectal cancer, while O.I. Kit (2014) suggets that it is not higher than 30%. There are also various opinions about specificity of these two tests. OBJECTIVE: To review the literature on the validity of the fecal occult blood tests for the diagnosis of CRC. METHODS: We looked for articles (electronic versions) available for free in the full-text versions, published from June 1, 1990 to December 31, 2014 in Russian or English. The following databases were used for search: E-LIBRARY; Cochrane; MEDLINE; EMBASE; Google search. Only original research papers were analyzed. Literature reviews or systematic reviews were not taken for analyses. Selection criteria: 1) use of Guaiac and/or immunochemical fecal occult blood test as screening-tests for the detection of colorectal cancer and/or colon polyps (1 cm or more in diameter) in people older than 45 years; 2) comparing of results with the results of colonoscopy (colonoscopy is counted by majority of the authors as a “gold standard” for the diagnosis of CRC and adenomatous polyps). RESULTS: Initial keyword search returned 803 000 results, of which 449 sources were selected. After reading the abstracts, 29 articles that met inclusion criteria were kept. 10 other articles were excluded after that because they did not contain enough data for extraction or did not contain a control group. At the final step 19 articles were used for meta-analysis. Forest plot and Rock curve, which were developed with inclusion of the data from all studies, showed heterogeneity of the data. Additional analyzes were performed in subgroups with different diagnoses and various tests. The sensitivity of the Guaiac test for the diagnosis of colorectal cancer varied from 0.13 to 1.00, and specificity - from 0.69 to 0.99. The sensitivity of the immunochemical test for the diagnosis of CRC ranged from 0.42 to 0.94 with specificity ranging from 0.40 to 1.00. The sensitivity of the Guaiac test for the diagnosis of the colon polyps was between 0.05 and 0.69, and its specificity - from 0.67 to 0.98. The sensitivity of the immunochemical test for the diagnosis of polyps was from 0.24 to 0.75, and its specificity - from 0.40 to 0.97. Bivariate analysis of the validity of Guaiac test and immunochemical method for the diagnosis of colorectal cancer showed better results for the immunochemical test compared to Guaiac test. The tests showed very similar results when used for the diagnosis of polyposis. Bivariate analysis, comparing the validity of tests for the diagnosis of colorectal cancer versus polyposis demonstrated better results for CRC. Multivariate analysis of the validity of the Guaiac and immunochemical tests for the diagnosis of colorectal cancer and polyps also showed better results for detection of colorectal cancer compared with the polyps for both tests. At the same time the highest validity for the diagnosis of CRC was demonstrated for immunochemical analysis. CONCLUSIONS: 1. The sensitivity of the Guaiac test for occult blood in stool is lower than its specificity. 2. Broad dispersion of the validity characteristics of the fecal occult blood tests was observed. 3. The validity of tests for occult blood was higher when they were used for detection of colorectal cancer than of colon polyposis. 4. The highest validity rate has been demonstrated for the immunochemical test when it was used for colon cancer screening. Show more

Keywords: Screening, colon cancer, occult blood

DOI: 10.3233/JRS-150712

Citation: International Journal of Risk & Safety in Medicine, vol. 27, no. s1, pp. S110-S111, 2015

Authors: Yudina, E.V. | Ziganshina, L.E.

Article Type: Abstract

Abstract: BACKGROUND: Cochrane collaboration has made a huge contribution to the development of evidence-based medicine; Cochrane work is the international gold standard of independent, credible and reliable high-quality information in medicine. Over the past 20 years the Cochrane Collaboration helped transforming decision-making in health and reforming it significantly, saving lives and contributing to longevity [1 ]. Until recently, Cochrane evidence were available only in English, which represents a significant barrier to their wider use in non-English speaking countries. To provide access to evidence, obtained from Cochrane Reviews, for health professionals and general public (from non-English-speaking countries), bypassing language barriers, Cochrane collaboration …in 2014 initiated an international project of translating Plain language summaries of Cochrane Reviews into other languages [2 , 3 ]. Russian translations of Plain language summaries were started in May 2014 by the team from Kazan Federal University (Department of Basic and Clinical Pharmacology; 2014–2015 as an Affiliated Centre in Tatarstan of the Nordic Cochrane Centre, since August 2015 as Cochrane Russia, a Russian branch of Cochrane Nordic, Head - Liliya Eugenevna Ziganshina) on a voluntary basis. OBJECTIVE: To assess the quality of Russian translations of Cochrane Plain Language Summaries (PLS) and their potential impact on the Russian speaking community through user feedback with the overarching aim of furthering the translations project. METHODS: We conducted the continuous online survey via Google Docs. We invited respondents through the electronic Russian language discussion forum on Essential Medicines (E-lek), links to survey on the Russian Cochrane.org website, invitations to Cochrane contributors registered in Archie from potential Russian-speaking countries. We set up the survey in Russian and English. The respondents were asked to respond to the questionnaire regarding the relevance and potential impact of the Cochrane Russian translations project, topics of interest in the field of health and health care, the quality and clarity of translated content, the preferred style of presentation and suggestions to improve the quality of translations of Plain language summaries of Cochrane Reviews. RESULTS: Currently the team of translators includes volunteers from the staff, Masters and PhD students of the Department of Basic and Clinical Pharmacology of the Kazan Federal University, and Kazan Medical University, our colleagues from Kazan and other cities of Russia, from the Republic of Armenia and the USA. By September 20th 2015, 446 Plain language summaries of Cochrane Reviews were translated into Russian and published on the web-site http://www.cochrane.org/ru/evidence . Our project “Russian translations for Cochrane” has already covered a wide range of health priority areas with translations of Plain language summaries and abstracts of the most topical and priority Cochrane reviews. During the period from 03.03.2015 to 20.09.2015 we received 113 answers from our respondents (103 answers in Russian and 10 answers in English). These were representatives of the medical and pharmaceutical professions (60%), representatives of non-medical professions (17%), students/graduate students (16%), retirees (4%) and others categories of citizens among the respondents. Half of the respondents (50%) belonged to the age group of 36–60 years, followed by the group of 18–35 years (41%). According to the survey the vast majority of respondents consider that the Cochrane Russian translations project is needed for Russia and Russian speaking countries (94%; n = 106), it is needed for their work, studies, and life in general (91%; n = 103). Nobody answered “No” to the question: “Do you think that this project is needed for Russia and Russian-speaking countries?” Information from the Cochrane evidence can affect (change) individual practice and/or attitude to drugs or diagnostic procedures of 87% (n = 98) of respondents. Only two people answered negatively to this question. However, only one third of respondents would like to become volunteer members of the translations project. The Russian texts of translations of Cochrane summaries and their main message were completely understandable or mostly clear to the vast majority of respondents (92%; n = 104). Respondents, proficient in English (n = 61), answered that the Russian-language translations fully complied (43%; n = 26) or in general corresponded to (57%; n = 35) the original English text. The majority of respondents (85%, n = 96) rated the quality of the translated texts as excellent and good. “More than half of respondents (61%; n = 69) would prefer the translations to be adapted to the usual style of presentation in Russian. The respondents agreed that mistakes, or typos or both very few. Our respondents provided valuable suggestions for further improvement of the Russian translations project. We would like to present here some of these: “More translations needed”, “The ultimate goal... is to try to adapt the summaries to Russian language style as much as possible. This is a very challenging task, however and at present format the summaries are already great”, “Go great as you do!” “Move forward and be efficient!” “Distribute information about the project through social networks and different means of social media”, “Studying Cochrane Database should be included in the Russian medical school's curriculum at a much larger extent than it is included (if at all) now. It would be beneficial for high school students as well.” CONCLUSIONS: The survey provided positive feedback on the Russian translations project concerning the clarity and quality of Russian texts and overall satisfaction of the readers. It confirmed the importance and relevance of the Russian translations project for Russian speaking audience, representing various professions and age groups. The survey results with detailed feedback contribute to further improvement of the Russian translations project. Limitations: Selective and subjective evaluation of translations by the respondents, difficulties with clear criteria for the objective evaluation. Further quality improvement of original PLS texts would contribute to higher translation quality. Acknowledgments: We would like to thank Juliane Reed, Coordinator of the Cochrane Translations Project, Professor Peter C Gøtzsche, Director of the Cochrane Nordic, co-founder of the Cochrane Collaboration, Cochrane leadership and the global Cochrane network together with the leadership of the Kazan Federal University for continuous encouragement, spirit and support. Show more

Keywords: Cochrane, russian translations, plain language summaries

DOI: 10.3233/JRS-150713

Citation: International Journal of Risk & Safety in Medicine, vol. 27, no. s1, pp. S112-S113, 2015