MKMichalina Justyna Kozłowska

Michalina Justyna Kozłowska

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Michalina Justyna Kozłowska has a total of 9 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    ARTIFICIAL INTELLIGENCE IN SUICIDE RISK PREDICTION: A STRUCTURED REVIEW OF MODEL PERFORMANCE, ETHICAL CHALLENGES, AND CLINICAL IMPLEMENTATION

    by Michalina Kozłowska, Maciej Kamil Majchrzak, Faustyna Magdalena Kołodziej, Alicja Kwiatkowska, Jakub Juskowiak, Mateusz Lis, Zuzanna Julia Komin, Maryan Liasota, Eliza Zofia Kabat, Krzysztof Konopka

    Background: Suicide remains a critical global public health crisis, demanding proactive intervention strategies. Recent advancements in artificial intelligence (AI), including machine learning and natural language processing, offer unprecedented capabilities for early suicide risk prediction utilizing electronic health records, clinical notes, and digital phenotyping. Objective: This structured review evaluates current evidence regarding the performance of AI models in predicting suicide risk, analyzes the ethical challenges associated with their deployment, and identifies primary barriers and facilitators to successful clinical implementation. Methodology: A structured literature search was conducted utilizing PubMed/MEDLINE and Google Scholar for peer-reviewed articles published between 2018 and 2026. Following specific inclusion and exclusion criteria, the final corpus of studies was analyzed using a thematic synthesis framework focusing on predictive accuracy, ethical implications, and workflow integration. Results: While AI algorithms frequently demonstrate robust predictive accuracy, their clinical utility is heavily complicated by significant ethical and operational hurdles. Key challenges include algorithmic bias exacerbating health inequities, severe data privacy concerns, and a notable lack of clinician trust stemming from the "black box" nature of complex machine learning models. Furthermore, poor integration into existing psychiatric workflows limits real-world feasibility. Conclusion: AI represents a powerful, yet imperfect, socio-technical tool for suicide prevention. Bridging the translational gap from algorithmic validation to routine clinical use requires rigorous ethical governance, prioritized model explainability, and human-centered implementation strategies tailored to frontline healthcare settings.

  • Open Access

    LARGE LANGUAGE MODELS IN MENTAL HEALTH TRIAGE: A SYSTEMATIC REVIEW OF ACCURACY, BIAS, AND ETHICAL DEPLOYMENT ACROSS DIGITAL PLATFORMS

    by Alicja Kwiatkowska, Faustyna Magdalena Kołodziej, Maciej Kamil Majchrzak, Krzysztof Konopka, Michalina Justyna Kozłowska, Eliza Zofia Kabat, Maryan Liasota, Zuzanna Julia Komin, Mateusz Lis, Jakub Juskowiak

    Background: The intersection of a global psychiatric workforce shortage and the rapid advancement of Generative Artificial Intelligence (GenAI) has catalyzed the integration of Large Language Models (LLMs) into digital healthcare. LLMs are increasingly deployed as preliminary triage agents to assess psychological distress, yet their diagnostic reliability, demographic neutrality, and safety protocols in crisis scenarios remain heavily debated within the psychiatric community. Objective: This review evaluates current research to determine the trustworthiness of large language models (LLMs) used in psychiatric screening. In doing so, it shines a light on the algorithmic biases present in practical use and maps out the necessary ethical standards to ensure digital health services remain safe and fair for all patients. Methodology: We sourced peer-reviewed publications released between 2023 and 2026 by running a systematic literature search across Google Scholar, PubMed/MEDLINE, and IEEE Xplore. The synthesis focused on comparative accuracy studies, observational evaluations, and sociotechnical analyses of leading LLMs (e.g., GPT-4, Claude 3.5 Sonnet, Gemini 1.5 Pro) applied specifically to psychiatric screening and automated clinical routing. Results: LLMs demonstrate high diagnostic concordance with human clinicians in structured assessments for high-prevalence conditions like major depressive disorder, effectively parsing unstructured clinical narratives. However, they exhibit a consistent tendency toward conservative over-triage in psychiatric emergencies. Profound algorithmic biases persist; models frequently recommend divergent, often inferior, treatment pathways for marginalized demographic groups and display gender-based stigmatization. Furthermore, models struggle with complex psychopathologies and display inconsistent safety mechanisms during acute suicidal ideation queries. Conclusion: While LLMs offer substantial utility for scaling preliminary psychiatric screening and reducing administrative triage burdens, their propensity for factual hallucination, demographic bias, and uncalibrated risk escalation currently precludes autonomous clinical deployment. Robust, human-in-the-loop governance and culturally representative training corpora are imperative prior to further integration into global digital health ecosystems.

  • Open Access

    ALGORITHMIC BIAS IN AI-ASSISTED PSYCHIATRIC DIAGNOSIS: A STRUCTURED REVIEW OF DEMOGRAPHIC DISPARITIES AND GOVERNANCE FAILURES

    by Mateusz Lis, Maryan Liasota, Eliza Kabat, Michalina Kozłowska, Krzysztof Konopka, Maciej Majchrzak, Faustyna Kołodziej, Alicja Kwiatkowska, Jakub Juskowiak, Zuzanna Komin

    Background: The use of artificial intelligence (AI) and machine learning (ML) for psychiatric diagnostic aid during clinical triage and mental health assessment has a potential value that is hard to underestimate. The use of these tools will raise many socio-technical issues concerning the biases of the algorithms. Such biases may be especially harmful for those patient populations that are already most vulnerable. Objective: The purpose of this study is to identify the degree of biases of algorithms for AI in mental health tools, understand how the existing demographic imbalances impact the diagnostic gap and identify the existing regulatory crisis that allows these systemic clinical hazards. Method: A structured review of the literature was performed and research on the study was limited to articles published between the years 2018 and 2026 on the databases of PubMed/MEDLINE and Google Scholar. A thematic synthesis framework was used to study evidence through three domains: diagnostic disparities, demographic imbalances, and regulatory inadequacy. Results: Observed literature indicates that AI tools for diagnosis have great variability and a lack of significant accuracy toward different racial, gender, and socioeconomic groups. The review found that the lack of existing governance frameworks allows systemic clinical inequities to persist. Conclusions: Uncontrolled algorithmic bias toward psychiatric AI tools will create direct and significant clinical harm. These will lead to exacerbated health inequities and will require severe policy shifts. Standardized algorithm auditing and human-in-the-loop clinical pathways will also be needed.

  • Open Access

    ARTIFICIAL INTELLIGENCE IN DERMATOLOGICAL IMAGE CLASSIFICATION: DIAGNOSTIC ACCURACY, SKIN TONE DISPARITIES, AND CLINICAL GOVERNANCE

    by Faustyna Magdalena Kołodziej, Mateusz Lis, Zuzanna Julia Komin, Maryan Liasota, Eliza Zofia Kabat, Michalina Justyna Kozłowska, Krzysztof Konopka, Maciej Kamil Majchrzak, Jakub Juskowiak, Alicja Kwiatkowska

    Background: Artificial intelligence (AI) is increasingly used to classify skin lesions from dermoscopic and clinical images, yet high performance in curated datasets may not translate into safe or equitable clinical use. Objective: This structured narrative review evaluates diagnostic accuracy, skin-tone-related performance disparities, dataset representation, and governance requirements for AI-based skin-lesion classification. Methods: PubMed/MEDLINE was used as the primary database, supplemented by PubMed Central, Google Scholar, Semantic Scholar, journal websites, and reference-list screening. English-language, peer-reviewed literature published mainly between 2021 and 2026 was synthesised thematically across diagnostic performance, external validity, skin-tone representation, clinical implementation, and governance. Results: Meta-analytic evidence indicates high sensitivity for AI-based skin-lesion classification, including pooled sensitivity of 0.91 and an HSROC-AUC of 0.88, but lower specificity and substantial heterogeneity limit direct clinical translation. Performance declined in darker skin-tone groups in several external evaluations; in selected models, sensitivity fell between Fitzpatrick types I–II and V–VI. Limited reporting of skin tone, reliance on internally tested or curated datasets, and sparse prospective evidence further constrain generalisability. Conclusion: AI should be implemented as clinician-supervised decision support rather than as an autonomous diagnostic system. Equitable deployment requires representative datasets, external and subgroup-stratified validation, transparent reporting, defined clinical accountability, regulatory oversight, and continuous post-deployment monitoring.

  • Open Access

    ELECTRONIC HEALTH RECORD-BASED PREDICTIVE APPROACHES FOR HOSPITAL READMISSION: A STRUCTURED REVIEW OF SOCIOECONOMIC VARIABLES AND ALGORITHMIC EQUITY

    by Maciej Kamil Majchrzak, Faustyna Magdalena Kołodziej, Krzysztof Konopka, Mateusz Lis, Zuzanna Julia Komin, Maryan Liasota, Eliza Zofia Kabat, Jakub Juskowiak, Alicja Kwiatkowska, Michalina Justyna Kozłowska

    Background: Hospital readmission rates are often used to judge the quality of care and how well healthcare systems are working. Predictive models based on electronic health records (EHRs) can help estimate which patients are at higher risk, but when they don’t properly reflect socioeconomic realities, they can unintentionally reinforce existing inequalities (Obermeyer et al., 2019). This review looks at how well these EHR-based models predict 30-day readmissions, how Social Determinants of Health (SDOH) are included, and what this means for fairness and health equity in real-world use. Methodology: A structured literature review was conducted using PubMed/MEDLINE and Google Scholar. It included peer-reviewed studies published between 2018 and 2026, focusing on model validation studies, retrospective cohort analyses, and research on bias and fairness in predictive healthcare models. Results: Synthesis reveals that incorporating unstructured and structured SDOH variables yields mixed impacts on discriminatory performance (Area Under the Receiver Operating Characteristic curve changes ranging from -0.01 to +0.04), yet significantly rectifies systemic algorithmic underestimation of risk for marginalized demographics. Unadjusted models consistently show racial, ethnic, and socioeconomic bias, often because they rely on proxy variables embedded in historical healthcare data (Gianfrancesco et al., 2018). As a result, preventive transition-of-care resources can be systematically directed away from the patients who need them most (Pierson et al., 2021). Conclusion: To make predictive healthcare tools truly useful in practice, it is not enough to focus only on accuracy. They also need to account for fairness, real clinical context, and the actual consequences of their use in everyday healthcare settings (Wiens et al., 2019). Policy-level interventions must mandate standardized SDOH extraction alongside equity-focused calibration algorithms before widespread clinical deployment (Ghassemi et al., 2021).

  • Open Access

    WEARABLE BIOSENSORS IN REMOTE CHRONIC DISEASE MONITORING: A STRUCTURED REVIEW OF IMPLEMENTATION BARRIERS AND HEALTH EQUITY IMPLICATIONS

    by Jakub Juskowiak, Zuzanna Julia Komin, Maryan Liasota, Eliza Zofia Kabat, Michalina Justyna Kozłowska, Krzysztof Konopka, Maciej Kamil Majchrzak, Faustyna Magdalena Kołodziej, Alicja Kwiatkowska, Mateusz Lis

    Background Clinical-grade wearable biosensors offer significant potential for tracking chronic diseases outside the hospital. However, while the hardware itself is highly accurate, the medical field still struggles with the logistical reality of deploying these tools. The structural barriers preventing widespread clinical use, and the ways these hurdles impact overall health equity, lack rigorous synthesis in the current literature. Objective Moving beyond pure hardware accuracy, this analysis maps the practical realities of continuous patient surveillance. It specifically investigates the institutional workflow bottlenecks, hardware usability hurdles, and underlying economic disparities that dictate how monitoring programs actually function for populations managing heart failure, diabetes, or COPD. Methodology Peer-reviewed literature spanning 2018 to 2026 was extracted from PubMed/MEDLINE and Google Scholar. The inclusion criteria deliberately filtered out purely technical validation studies. Instead, the analysis targeted data documenting the logistical friction, long-term adherence failures, and unequal access surrounding continuous glucose monitors, ambulatory cardiac patches, and multi-parameter tracking devices. Results The synthesized data show that severe workflow friction and poor electronic health record interoperability actively block hospital networks from scaling these programs. On the patient side, a distinct "compliance paradox" emerges. High initial motivation frequently collapses into device abandonment, usually triggered by physical skin irritation, convoluted software interfaces, and low technical literacy. The literature also exposes a severe digital divide. A patient's income bracket, geographic location, and advanced age strictly gatekeep their ability to access continuous medical monitoring. Conclusion Healthcare networks cannot rely solely on precise hardware to improve chronic disease management. Deploying continuous monitors successfully requires a fundamental shift toward repairing broken clinical workflows and addressing systemic exclusion. Without structural interventions targeting financial and technical accessibility, digital health initiatives will simply reinforce existing medical inequalities.

  • Open Access

    DIGITAL THERAPEUTICS FOR ANXIETY AND DEPRESSION: A STRUCTURED REVIEW OF CLINICAL EVIDENCE, REGULATORY FRAMEWORKS, AND ACCESS DISPARITIES

    by Maryan Liasota, Michalina Justyna Kozłowska, Krzysztof Konopka, Maciej Kamil Majchrzak, Faustyna Magdalena Kołodziej, Alicja Kwiatkowska, Jakub Juskowiak, Mateusz Lis, Zuzanna Julia Komin, Eliza Zofia Kabat

    Background: The integration of Digital Therapeutics (DTx) into psychiatric care represents a paradigm shift in the management of anxiety and depression. With the continuous technological advancement of Software as a Medical Device (SaMD), its potential to alleviate shortages in psychiatric care is facing rigorous evaluation. As a result, ongoing studies aim to verify whether these technologies deliver genuine medical improvements, while simultaneously exploring the structural hurdles of deploying them within traditional clinical settings. Objective: This structured review evaluates the intersection of clinical evidence for DTx in treating Generalized Anxiety Disorder (GAD) and Major Depressive Disorder (MDD) with existing regulatory frameworks and socioeconomic barriers to equitable access. Methodology: An investigative sweep of the PubMed/MEDLINE and Google Scholar databases was executed to identify literature published from 2018 to 2026. Inclusion criteria focused on peer-reviewed RCTs and policy analyses of evidence-based DTx platforms. The assembled data underwent systematic organization via a thematic approach, with material being classified into three distinct categories: clinical efficacy, regulatory limitations, and accessibility equity. Results: Findings indicate that while AI-driven CBT and chatbot-delivered interventions demonstrate significant symptom reduction with moderate-to-high effect sizes, their long-term efficacy is moderated by user engagement patterns. Regulatory analysis reveals a fragmentation between the US FDA and European DiGA pathways, complicating global reimbursement. Furthermore, structural disparities—driven by the "digital divide" and low technological literacy—persistently limit the reach of these interventions among underserved populations. Conclusion: Digital interventions present an easily expandable mechanism for combating global psychiatric shortages; however, inconsistent legal guidelines and socioeconomic barriers heavily restrict their medical utility. Subsequent academic efforts need to emphasize multidimensional adoption models to prevent these technological advancements from worsening present-day medical inequalities.

  • Open Access

    AI-DRIVEN EARLY WARNING SYSTEMS FOR SEPSIS DETECTION IN ICU SETTINGS: A STRUCTURED REVIEW OF PREDICTIVE PERFORMANCE, CLINICAL INTEGRATION AND EQUITY OUTCOMES

    by Zuzanna Julia Komin, Eliza Zofia Kabat, Michalina Justyna Kozłowska, Krzysztof Konopka, Maciej Kamil Majchrzak, Faustyna Magdalena Kołodziej, Alicja Kwiatkowska, Jakub Juskowiak, Mateusz Lis, Maryan Liasota

    Background: Sepsis remains a primary cause of mortality and organ dysfunction in Intensive Care Units (ICUs) globally. Although AI models show considerable promise for the early identification of sepsis, moving these tools from retrospective performance to bedside utility is hindered by socio-technical barriers, workflow friction, and algorithmic equity issues. Objective: This structured review evaluates the state of AI-driven sepsis Early Warning Systems (EWS), identifying systemic barriers to clinical integration and assessing algorithmic fairness across demographic cohorts. Methodology: A structured review of peer-reviewed literature published in PubMed/MEDLINE and Google Scholar between January 2018 and June 2026 was conducted, focusing on predictive performance, clinical implementation, and algorithmic equity. Results: Retrospective assessments of deep sequential architectures demonstrate high diagnostic accuracy (AUROC>0.95) (Boussina et al., 2024; Siddiqui, 2026). Prospective trials of landmark systems such as TREWS (Adams et al., 2022), COMPOSER (Shashikumar et al., 2025), and CONCERN EWS (Rossetti et al., 2025), confirm clinical utility, yielding mortality reductions of 17% to 35.6%. However, real-world effectiveness is frequently hindered by severe alert fatigue (Wong et al., 2021). Furthermore, because models rely on historical health data, they often inherit systemic biases, risking the exacerbation of existing disparities related to race, gender and socioeconomic status (Chen et al., 2023; Obermeyer et al., 2019; Seyyed-Kalantari et al., 2021). Conclusion: Maximizing the clinical impact of AI-driven EWS requires a paradigm shift from purely performance-centric model development to rigorous, equity-oriented clinical governance guided by international reporting frameworks like TRIPOD-AI (Collins et al., 2024) and DECIDE-AI (Vasey et al., 2022). To operationalize this, we propose the integration of the Clinical AI Safety Officer (CAISO) to maintain real-time algorithmic safety, audit demographic equity, and manage the clinical accountability loop.

  • Open Access

    CHATBOT-DELIVERED COGNITIVE BEHAVIOURAL THERAPY: A STRUCTURED REVIEW OF EFFICACY, USER ENGAGEMENT, AND EQUITY OF ACCESS

    by Krzysztof Konopka, Michalina Justyna Kozłowska, Eliza Zofia Kabat, Maryan Liasota, Zuzanna Julia Komin, Mateusz Lis, Jakub Juskowiak, Alicja Kwiatkowska, Faustyna Magdalena Kołodziej, Maciej Kamil Majchrzak

    Background: Chatbots are increasingly used to deliver cognitive behavioural therapy (CBT) via mobile and web applications. Many mental health systems cannot meet demand with face-to-face care alone. CBT chatbots have therefore been introduced in app form, but data on outcomes, dropout, and unequal uptake remain limited. Objective: This review evaluates whether CBT chatbots reduce anxiety and depression symptoms, which user- and design-related factors are linked to engagement and dropout, and which structural barriers limit who can practically use these tools. Methodology: Peer-reviewed studies from 2018 to 2026 were identified through searches in PubMed/MEDLINE and Google Scholar. Included studies examined automated CBT chatbots (e.g., Woebot, Wysa, Youper, Limbic Care) and reported clinical outcomes, engagement or dropout, and/or access-related findings. Twenty-four publications were synthesised thematically. Results: Short-term trials and observational studies reported reductions in anxiety and depression symptoms, with small-to-moderate effects1. Generative conversational designs were linked to higher engagement than static digital programmes in at least one RCT, which reported roughly 2.4 times the use frequency of a generative CBT app relative to workbook controls3. Attrition remained notable; one meta-analysis estimated intervention-arm dropout at 21.84%5. Unequal access persisted because of connectivity, device availability, and digital literacy, despite the privacy and round-the-clock availability of app-based care23. Conclusion: CBT chatbots may serve as useful low-intensity interventions, but their public health value depends on improved retention and fairer access. Implementation should address digital literacy, culturally appropriate design, and governance that does not restrict validated tools to already advantaged users.

  • Open Access

    PASSIVE SMARTPHONE SENSING FOR MENTAL HEALTH MONITORING: A STRUCTURED REVIEW OF BEHAVIORAL BIOMARKERS, PREDICTIVE VALIDITY, AND ETHICAL RISKS

    by Eliza Zofia Kabat, Krzysztof Konopka, Maciej Kamil Majchrzak, Faustyna Magdalena Kołodziej, Alicja Kwiatkowska, Jakub Juskowiak, Mateusz Lis, Zuzanna Julia Komin, Maryan Liasota, Michalina Justyna Kozłowska

    Methods: This structured review adhered to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA 2020) guidelines. Systematic searches were carried out in PubMed/MEDLINE and Google Scholar to identify peer-reviewed articles published between January 2018 and June 2026. Studies met inclusion criteria if they examined passive sensing frameworks (e.g., AWARE, Beiwe, RADAR-Base) collecting GPS, accelerometry, or communication metadata for psychiatric monitoring, resulting in a final sample of 15 core empirical studies. Findings: The evidence shows that changes in spatial mobility, sleep architecture, and communication latency display strong longitudinal correlations with symptom severity scores for major depression (PHQ-9) and generalized anxiety (GAD-7). Machine learning models incorporating multi-modal data streams reached high predictive accuracy (AUC 0.72–0.88) when forecasting depressive shifts and spotting prodromal relapse signatures in bipolar disorder. Even with these advancements, notable translation barriers persist, especially concerning algorithmic generalizability across diverse socioeconomic groups and the ethical threat of "surveillance creep." Conclusion: Passive smartphone sensing offers incredible potential for proactive, continuous psychiatric care. However, actual clinical translation will require strict cross-cultural validation of digital biomarkers alongside the creation of solid data governance frameworks to protect patient autonomy and privacy.