Antonija Mijatović
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Antonija Mijatović has a total of 53 co-authors. Here are the top co-authors by number of shared publications:
- Ana Marušić
Department of Research, University Hospital of Split, Spinčićeva 1, 21000 Split, Croatia
12 shared publications - Ivan Buljan
Department of Psychology, Faculty of Humanities and Social Sciences, University of Split, 21000 Split, Croatia
4 shared publications - Andrijana Perković Paloš
Faculty of Humanities and Social Sciences, University of Split
3 shared publications - Daniel Garcia-Costa
Computer Science Department, Universitat de València, Valencia, Spain
3 shared publications - Francisco Grimaldo
Computer Science Department, Universitat de València, Valencia, Spain
3 shared publications - Marija Franka Žuljević
Department of Medical Humanities, University of Split School of Medicine, Split, Croatia
3 shared publications
Latest Publications
- Open Access
- 5 November 2026
- Published by MIT Press in Quantitative Science Studies
- p. 1-17
Characteristics of Croatian clinical medicine: a 22-year bibliometric study
Bibliometric analyses have been increasingly used for assessing science, as they provide valuable insights into the dynamics of scientific research. Using the affiliations of Croatian hospitals, we retrieved publications records from Scopus covering the 2003-2024 period and conducted a bibliometric performance analysis. The total number of publications did not statistically significantly increase over time (incidence rate ratio [IRR] per year=1.05, p=0.18). In contrast, the proportion of open-access publications increased significantly (odds ratio [OR] per year=1.21, p<0.001). Median citations per publication exhibited a strong increasing monotonic trend over time (Spearman’s ρ=0.78, p<0.001). The 18 journals with the highest number of publications were all published in Croatia and collectively accounted for 38.5 % of all retrieved publications. Among them, a publication from a single Croatian journal, Biochemia Medica, was in the 100 most-cited publications. The 10 most prolific authors by number of publications were predominantly affiliated with institutions located in the capital of Croatia – Zagreb, which accounted for 64.9 % of all affiliations among these authors. This analysis demonstrated the evolution of scientific research of Croatian clinicians, characterized by growth in open access and citations, as well as a strong geographical centralization.
- 5 November 2026
- Published by MIT Press in Quantitative Science Studies
- p. 1-17
Open AccessCharacteristics of Croatian clinical medicine: a 22-year bibliometric study
Bibliometric analyses have been increasingly used for assessing science, as they provide valuable insights into the dynamics of scientific research. Using the affiliations of Croatian hospitals, we retrieved publications records from Scopus covering the 2003-2024 period and conducted a bibliometric performance analysis. The total number of publications did not statistically significantly increase over time (incidence rate ratio [IRR] per year=1.05, p=0.18). In contrast, the proportion of open-access publications increased significantly (odds ratio [OR] per year=1.21, p<0.001). Median citations per publication exhibited a strong increasing monotonic trend over time (Spearman’s ρ=0.78, p<0.001). The 18 journals with the highest number of publications were all published in Croatia and collectively accounted for 38.5 % of all retrieved publications. Among them, a publication from a single Croatian journal, Biochemia Medica, was in the 100 most-cited publications. The 10 most prolific authors by number of publications were predominantly affiliated with institutions located in the capital of Croatia – Zagreb, which accounted for 64.9 % of all affiliations among these authors. This analysis demonstrated the evolution of scientific research of Croatian clinicians, characterized by growth in open access and citations, as well as a strong geographical centralization.
- Open Access
- 8 October 2026
- Published by Taylor & Francis in Accountability in Research
- Vol. 33 (8) , 2744222
Anticipating the unpredictable: A qualitative study on ethics experts’ perspectives on review challenges related to AI
1. Artificial intelligence (AI) has rapidly proliferated in research and healthcare over the past decade, with machine learning (ML) algorithms becoming an increasingly important tool in medical di...
- 8 October 2026
- Published by Taylor & Francis in Accountability in Research
- Vol. 33 (8) , 2744222
Open AccessAnticipating the unpredictable: A qualitative study on ethics experts’ perspectives on review challenges related to AI
1. Artificial intelligence (AI) has rapidly proliferated in research and healthcare over the past decade, with machine learning (ML) algorithms becoming an increasingly important tool in medical di...
- Open Access
- 8 July 2026
- Published by Springer Nature in Research Integrity and Peer Review
- Vol. 11 (1)
Retraction and correction of scientific articles due to conflicts of interest: a cross‑sectional analysis of the retraction watch database
Background: Conflicts of interest (CoI) can undermine trust in scientific publishing, yet empirical evidence on how often CoI lead to editorial action and how journals respond remains limited. This study aimed to characterise the frequency, patterns, and editorial handling of CoI‑related notices in the scholarly literature using an openly available, multidisciplinary dataset. Methods: We conducted a descriptive cross‑sectional analysis of publicly available records from the Retraction Watch (RW) database, accessed via Crossref, from inception to 23 April 2026. All editorial notices in which CoI was explicitly mentioned as the sole or one of several stated reasons were included, encompassing retractions, corrections, expressions of concern, and reinstatements. For each record, we extracted article and notice dates, article type, scientific field, country of the corresponding author, number of authors and institutions, publisher, and all stated reasons for the notice. Data were analysed using descriptive statistics and bootstrap‑based confidence intervals. Results: We identified 886 CoI‑related records (748 retractions, 39 corrections, 97 expressions of concern, and 2 reinstatements). CoI was the sole stated reason in 95 cases (10.7%), including 32 retractions, 11 corrections, and 52 expressions of concern; in the remaining notices, CoI most commonly co‑occurred with concerns such as compromised peer review, authorship or affiliation problems, journal or institutional investigations, and broader ethical violations. The median time from publication to editorial action was longest for expressions of concern and shortest for corrections, with wide variability across records and notice types. CoI‑related notices were concentrated in basic life, biomedical, and health sciences, and were unevenly distributed across high‑output countries and a small number of large publishers, particularly open‑access outlets. Most affected articles were original research papers with multi‑author, multi‑institution teams. Among articles with determinable author gender, both first and last authors were more often male than female, although these patterns were exploratory and lacked a baseline comparator. Conclusions: Editorial actions explicitly linked to CoI remain relatively uncommon compared with the broader volume of retractions and corrections, and CoI are rarely cited in isolation. The frequent co‑occurrence of CoI with other forms of misconduct or procedural problems suggests that undisclosed or poorly managed CoI often signal deeper weaknesses in disclosure practices, editorial oversight, and research governance rather than isolated administrative lapses. Strengthening CoI policies, transparency, verification mechanisms, and linked metadata systems may help detect problems earlier and support more consistent, graduated editorial responses.
- 8 July 2026
- Published by Springer Nature in Research Integrity and Peer Review
- Vol. 11 (1)
Open AccessRetraction and correction of scientific articles due to conflicts of interest: a cross‑sectional analysis of the retraction watch database
Background: Conflicts of interest (CoI) can undermine trust in scientific publishing, yet empirical evidence on how often CoI lead to editorial action and how journals respond remains limited. This study aimed to characterise the frequency, patterns, and editorial handling of CoI‑related notices in the scholarly literature using an openly available, multidisciplinary dataset. Methods: We conducted a descriptive cross‑sectional analysis of publicly available records from the Retraction Watch (RW) database, accessed via Crossref, from inception to 23 April 2026. All editorial notices in which CoI was explicitly mentioned as the sole or one of several stated reasons were included, encompassing retractions, corrections, expressions of concern, and reinstatements. For each record, we extracted article and notice dates, article type, scientific field, country of the corresponding author, number of authors and institutions, publisher, and all stated reasons for the notice. Data were analysed using descriptive statistics and bootstrap‑based confidence intervals. Results: We identified 886 CoI‑related records (748 retractions, 39 corrections, 97 expressions of concern, and 2 reinstatements). CoI was the sole stated reason in 95 cases (10.7%), including 32 retractions, 11 corrections, and 52 expressions of concern; in the remaining notices, CoI most commonly co‑occurred with concerns such as compromised peer review, authorship or affiliation problems, journal or institutional investigations, and broader ethical violations. The median time from publication to editorial action was longest for expressions of concern and shortest for corrections, with wide variability across records and notice types. CoI‑related notices were concentrated in basic life, biomedical, and health sciences, and were unevenly distributed across high‑output countries and a small number of large publishers, particularly open‑access outlets. Most affected articles were original research papers with multi‑author, multi‑institution teams. Among articles with determinable author gender, both first and last authors were more often male than female, although these patterns were exploratory and lacked a baseline comparator. Conclusions: Editorial actions explicitly linked to CoI remain relatively uncommon compared with the broader volume of retractions and corrections, and CoI are rarely cited in isolation. The frequent co‑occurrence of CoI with other forms of misconduct or procedural problems suggests that undisclosed or poorly managed CoI often signal deeper weaknesses in disclosure practices, editorial oversight, and research governance rather than isolated administrative lapses. Strengthening CoI policies, transparency, verification mechanisms, and linked metadata systems may help detect problems earlier and support more consistent, graduated editorial responses.
- Open Access
- 14 April 2026
- Published by JMIR Publications Inc. in JMIR Medical Informatics
- Vol. 14, e72657
Classification of Cochrane Plain Language Summaries by Conclusiveness Using Transformer-Based Models and ChatGPT: Retrospective Observational Study
Background: Cochrane plain language summaries (PLSs) aim to make systematic review findings more accessible to the general public. However, inconsistencies in how conclusions are presented may impact comprehension and decision-making. Classifying PLSs based on conclusiveness can improve clarity and facilitate informed health decisions. Objective: This study aimed to develop and evaluate deep learning language models for the classification of PLSs according to 3 levels of conclusiveness (conclusive, inconclusive, and unclear) and to compare their performance with a general-purpose large language model (GPT-4o). Methods: We used a publicly available dataset containing 4405 Cochrane PLSs of systematic reviews published until 2019, already classified by humans according to 9 categories of conclusiveness regarding the intervention’s effectiveness or safety. We merged these categories into 3 classes based on the strength of conclusiveness: conclusive, inconclusive, and unclear. For the fine-tuning, we used Scientific Bidirectional Encoder Representations from Transformers (SciBERT), a pretrained language model trained on 1.14 million papers primarily from the health sciences, and Longformer, a transformer model designed specifically to process long documents. The script was developed using the Python programming language and the PyTorch framework. We computed evaluation metrics using the scikit-learn machine learning library and determined the area under the curve of the receiver operating characteristic (AUCROC) to measure the model performance in balancing sensitivity and specificity. We also analyzed a separate set of 213 PLSs and compared the predictions of our pretrained models with both manual verification and outputs generated by ChatGPT. Results: The model based on SciBERT achieved a balanced accuracy of 56.6%. The AUCROC was 0.91 for “conclusive,” 0.67 for “inconclusive,” and 0.75 for “unclear” conclusiveness classes. The Longformer-based model had a balanced accuracy of 60.9%, with AUCROCs of 0.86 for “conclusive,” 0.67 for “inconclusive,” and 0.72 for “unclear” conclusiveness classes. Both models underperformed compared with ChatGPT, which demonstrated higher accuracy (74.2%), better precision and recall, and a higher Cohen κ (0.57). Conclusions: Fine-tuning 2 transformer-based language models showed mixed results in classifying Cochrane PLSs by conclusiveness, likely due to semantic overlap and subtle linguistic differences. Despite satisfactory internal test metrics, the fine-tuned models failed to generalize to newly published PLSs, where performance dropped to near-chance levels. These findings suggest that general-purpose large language models like GPT-4o may currently offer more reliable results for practical classification tasks in biomedical applications.
- 14 April 2026
- Published by JMIR Publications Inc. in JMIR Medical Informatics
- Vol. 14, e72657
Open AccessClassification of Cochrane Plain Language Summaries by Conclusiveness Using Transformer-Based Models and ChatGPT: Retrospective Observational Study
Background: Cochrane plain language summaries (PLSs) aim to make systematic review findings more accessible to the general public. However, inconsistencies in how conclusions are presented may impact comprehension and decision-making. Classifying PLSs based on conclusiveness can improve clarity and facilitate informed health decisions. Objective: This study aimed to develop and evaluate deep learning language models for the classification of PLSs according to 3 levels of conclusiveness (conclusive, inconclusive, and unclear) and to compare their performance with a general-purpose large language model (GPT-4o). Methods: We used a publicly available dataset containing 4405 Cochrane PLSs of systematic reviews published until 2019, already classified by humans according to 9 categories of conclusiveness regarding the intervention’s effectiveness or safety. We merged these categories into 3 classes based on the strength of conclusiveness: conclusive, inconclusive, and unclear. For the fine-tuning, we used Scientific Bidirectional Encoder Representations from Transformers (SciBERT), a pretrained language model trained on 1.14 million papers primarily from the health sciences, and Longformer, a transformer model designed specifically to process long documents. The script was developed using the Python programming language and the PyTorch framework. We computed evaluation metrics using the scikit-learn machine learning library and determined the area under the curve of the receiver operating characteristic (AUCROC) to measure the model performance in balancing sensitivity and specificity. We also analyzed a separate set of 213 PLSs and compared the predictions of our pretrained models with both manual verification and outputs generated by ChatGPT. Results: The model based on SciBERT achieved a balanced accuracy of 56.6%. The AUCROC was 0.91 for “conclusive,” 0.67 for “inconclusive,” and 0.75 for “unclear” conclusiveness classes. The Longformer-based model had a balanced accuracy of 60.9%, with AUCROCs of 0.86 for “conclusive,” 0.67 for “inconclusive,” and 0.72 for “unclear” conclusiveness classes. Both models underperformed compared with ChatGPT, which demonstrated higher accuracy (74.2%), better precision and recall, and a higher Cohen κ (0.57). Conclusions: Fine-tuning 2 transformer-based language models showed mixed results in classifying Cochrane PLSs by conclusiveness, likely due to semantic overlap and subtle linguistic differences. Despite satisfactory internal test metrics, the fine-tuned models failed to generalize to newly published PLSs, where performance dropped to near-chance levels. These findings suggest that general-purpose large language models like GPT-4o may currently offer more reliable results for practical classification tasks in biomedical applications.
- 17 March 2026
- Published by Taylor & Francis in Accountability in Research
- Vol. 33 (6) , 2642202
Ethics tools for strengthening ethics review: A scoping Review
1. Ethics review processes in research and innovation activities have become more complex due to rapid scientific and technological advances, the rise of interdisciplinary research, and the growing...
- 17 March 2026
- Published by Taylor & Francis in Accountability in Research
- Vol. 33 (6) , 2642202
Ethics tools for strengthening ethics review: A scoping Review
1. Ethics review processes in research and innovation activities have become more complex due to rapid scientific and technological advances, the rise of interdisciplinary research, and the growing...
- Top Cited Papers
- 23 January 2026
- Published by Elsevier in Journal of Clinical Epidemiology
- Vol. 192, 112170
Consistency of protocol and safety data reporting in clinical trial registrations and corresponding publications of interventions involving MDMA and psilocybin
- 23 January 2026
- Published by Elsevier in Journal of Clinical Epidemiology
- Vol. 192, 112170
Top Cited PapersConsistency of protocol and safety data reporting in clinical trial registrations and corresponding publications of interventions involving MDMA and psilocybin
- Open Access
- 12 January 2026
- Published by International Society of Global Health in Journal of Global Health
- Vol. 16, 04017
Impact of the COVID-19 pandemic on university performance: a retrospective follow-up study of the University of Split, Croatia
The COVID-19 outbreak, proclaimed a pandemic by the World Health Organization in March 2020 [1], caused extensive disruptions to higher education institutions (HEIs) throughout the world. The ensuing transition to distance learning and remote work created major difficulties for students and staff members, as well as institutional management [2,3]. Members of university communities, including students and academic staff, experienced higher stress levels of anxiety and depression, together with decreased […]
- 12 January 2026
- Published by International Society of Global Health in Journal of Global Health
- Vol. 16, 04017
Open AccessImpact of the COVID-19 pandemic on university performance: a retrospective follow-up study of the University of Split, Croatia
The COVID-19 outbreak, proclaimed a pandemic by the World Health Organization in March 2020 [1], caused extensive disruptions to higher education institutions (HEIs) throughout the world. The ensuing transition to distance learning and remote work created major difficulties for students and staff members, as well as institutional management [2,3]. Members of university communities, including students and academic staff, experienced higher stress levels of anxiety and depression, together with decreased […]
- Open Access
- 8 August 2025
- Published by Springer Nature in Research Integrity and Peer Review
- Vol. 10 (1) , 14
How good are medical students and researchers in detecting duplications in digital images from research articles: a cross-sectional survey
Inappropriate manipulations of digital images pose significant risks to research integrity. Here we assessed the capability of students and researchers to detect image duplications in biomedical images. We conducted a pen-and-paper survey involving medical students who had been exposed to research paper images during their studies, as well as active researchers. We asked them to identify duplications in images of Western blots, cell cultures, and histological sections and evaluated their performance based on the number of correctly and incorrectly detected duplications. A total of 831 students and 26 researchers completed the survey during 2023/2024 academic year. Out of 34 duplications of 21 unique image parts, the students correctly identified a median of 10 duplications (interquartile range [IQR] = 8-13), and made 2 mistakes (IQR = 1-4), whereas the researchers identified a median of 11 duplications (IQR = 8-14) and made 1 mistake (IQR = 1-3). There were no significant differences between the two groups in either the number of correctly detected duplications (p = .271, Cliff's δ = 0.126) or the number of mistakes (p = .731, Cliff's δ = 0.039). Both students and researchers identified higer percentage of duplications in the Western blot images than cell or tissue images (p < .005 and Cohen's d = 0.72; p < .005 and Cohen's d = 1.01, respectively). For students, gender was a weak predictor of performance, with female participants finding slightly more duplications (p < .005, Cliff's δ = 0.158), but making more mistakes (p < .005, Cliff's δ = 0.239). The study year had no significant impact on student performance (p = .209; Cliff's δ = 0.085). Despite differences in expertise, both students and researchers demonstrated limited proficiency in detecting duplications in digital images. Digital image manipulation may be better detected by automated screening tools, and researchers should have clear guidance on how to prepare digital images in scientific publications.
- 8 August 2025
- Published by Springer Nature in Research Integrity and Peer Review
- Vol. 10 (1) , 14
Open AccessHow good are medical students and researchers in detecting duplications in digital images from research articles: a cross-sectional survey
Inappropriate manipulations of digital images pose significant risks to research integrity. Here we assessed the capability of students and researchers to detect image duplications in biomedical images. We conducted a pen-and-paper survey involving medical students who had been exposed to research paper images during their studies, as well as active researchers. We asked them to identify duplications in images of Western blots, cell cultures, and histological sections and evaluated their performance based on the number of correctly and incorrectly detected duplications. A total of 831 students and 26 researchers completed the survey during 2023/2024 academic year. Out of 34 duplications of 21 unique image parts, the students correctly identified a median of 10 duplications (interquartile range [IQR] = 8-13), and made 2 mistakes (IQR = 1-4), whereas the researchers identified a median of 11 duplications (IQR = 8-14) and made 1 mistake (IQR = 1-3). There were no significant differences between the two groups in either the number of correctly detected duplications (p = .271, Cliff's δ = 0.126) or the number of mistakes (p = .731, Cliff's δ = 0.039). Both students and researchers identified higer percentage of duplications in the Western blot images than cell or tissue images (p < .005 and Cohen's d = 0.72; p < .005 and Cohen's d = 1.01, respectively). For students, gender was a weak predictor of performance, with female participants finding slightly more duplications (p < .005, Cliff's δ = 0.158), but making more mistakes (p < .005, Cliff's δ = 0.239). The study year had no significant impact on student performance (p = .209; Cliff's δ = 0.085). Despite differences in expertise, both students and researchers demonstrated limited proficiency in detecting duplications in digital images. Digital image manipulation may be better detected by automated screening tools, and researchers should have clear guidance on how to prepare digital images in scientific publications.
- Open Access
- 28 March 2025
- Published by University of Zadar in PUBMET
- p. 42-44
Data retrieval, data cleaning and data merging
Introduction: Bibliometric analyses have become fairly frequent in today's science (Donthu et al., 2021; Klarin, 2024; Öztürk et al., 2024). They help map out research areas and evaluate the quality of scientific research. Most bibliometric analyses today gather their data from Web of Science or Scopus. Yet, the methodology sections on obtaining that data often leave us wanting, especially in the few cases where datasets from both databases were merged (Echchakoui, 2020). The aim of our research was to create a comprehensive database of all publications by Croatian clinicians, regardless of the research area, so we could map their areas of interest during the period 2005-2022, as well as study potential effects of Croatia’s European Union (EU) membership. Our focus on publications by clinicians rather than publications in clinical medicine led us to create a more complex data retrieval strategy, using affiliations instead of research areas.
- 28 March 2025
- Published by University of Zadar in PUBMET
- p. 42-44
Open AccessData retrieval, data cleaning and data merging
Introduction: Bibliometric analyses have become fairly frequent in today's science (Donthu et al., 2021; Klarin, 2024; Öztürk et al., 2024). They help map out research areas and evaluate the quality of scientific research. Most bibliometric analyses today gather their data from Web of Science or Scopus. Yet, the methodology sections on obtaining that data often leave us wanting, especially in the few cases where datasets from both databases were merged (Echchakoui, 2020). The aim of our research was to create a comprehensive database of all publications by Croatian clinicians, regardless of the research area, so we could map their areas of interest during the period 2005-2022, as well as study potential effects of Croatia’s European Union (EU) membership. Our focus on publications by clinicians rather than publications in clinical medicine led us to create a more complex data retrieval strategy, using affiliations instead of research areas.
- Open Access
- 3 March 2025
- Published by Taylor & Francis in Journal of Responsible Innovation
- Vol. 12 (1)
Knowledge needs of research ethics committees for the integration of ethics in research and technology development
- 3 March 2025
- Published by Taylor & Francis in Journal of Responsible Innovation
- Vol. 12 (1)
Open AccessKnowledge needs of research ethics committees for the integration of ethics in research and technology development