Jana Winterstein

Mrs. Jana Winterstein

Johannes Gutenberg University Mainz
orcid

Research Focus & Metrics

Academic Career Length

7Years (2020-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
17 January 2025 in Nature Communications
Verified
Journal Article
Open access
PDF file
SHSarah Haggenmüller
Sarah Haggenmüller
CWChristoph Wies
Christoph Wies
JAJulia Abels
Julia Abels
JWJana Winterstein
Jana Winterstein
LHLukas Heinlein
Lukas Heinlein
TBTitus Josef Brinker
Titus Josef Brinker
Accurate melanoma diagnosis is crucial for patient outcomes and reliability of AI diagnostic tools. We assess interrater variability among eight expert pathologists reviewing histopathological images and clinical metadata of 792 melanoma-suspicious lesions prospectively collected at eight German hospitals. Moreover, we provide access to the largest panel-validated dataset featuring dermoscopic and histopathological images with metadata. Complete agreement is achieved in 53.5% of cases (424/792), and a majority vote ( ≥ five pathologists) in 90.9% (720/792). Considerable discordance is observed for non-invasive melanomas (complete agreement in only 10/73 cases). The expert panel disagrees with the local pathologists’ and dermatologists’ diagnoses in 14.9% and 33.5% of cases, respectively. This variability highlights the diagnostic challenges of early-stage melanomas and the need to reconsider how ground truth is established in routine care and AI research. Including at least two pathologists or virtual panels may contribute to more consistent diagnostic results.

18 January 2025 in European Journal of Cancer
Verified
Journal Article
Open access
JWJana Winterstein
Jana Winterstein
JAJulia Abels
Julia Abels
AKArlene Kuehn
Arlene Kuehn
NCNicolas Carl
Nicolas Carl
CWChristoph Wies
Christoph Wies
TBTitus Josef Brinker
Titus Josef Brinker
This study explores the potential of Artificial Intelligence (AI)-generated social media influencers to disseminate cancer prevention messages. Utilizing a Generative AI (GenAI) application, we created a virtual persona, "Wanda", to promote cancer awareness on Instagram. We created five posts, addressing the five most modifiable risk factors for cancer: tobacco consumption, unhealthy diet, sun exposure, alcohol consumption, and Human Papillomavirus (HPV) infection. To amplify the campaign's reach, posts were boosted using a custom-targeted as well as an automated advertisement algorithm. An overall budget of €100 was equally distributed between the two algorithms. Campaign performance was assessed based on the number of users reached and the age distribution of the audience. The campaign achieved a total of 9902 recognitions, with a cost-efficiency analysis revealing an average expenditure of €0.013 per reach. The most economical intervention achieved a cost of only €0.006 per reach. In comparing the two advertisement strategies, we observed similar overall reach but noted differences in the age demographics of the audience. Our findings underscore the potential of combining generative AI with strategically targeted advertisement to promote cancer prevention messages effectively, with minimal time and financial investment. We discuss chances presented by GenAI applications in health communication, their implication, and the impact of parasocial relationships on content perception. This study highlights the potential of AI-driven influencers as scalable tools for digital health communication.

18 November 2025 in European Journal of Cancer
Verified
Journal Article
Open access
NMNicolas B. Merl
Nicolas B. Merl
FSFranziska Schramm
Franziska Schramm
CWChristoph Wies
Christoph Wies
JWJana Winterstein
Jana Winterstein
TBTitus Josef Brinker
Titus Josef Brinker
Artificial intelligence (AI) is increasingly used to generate digital health communication, but its impact on user engagement and message perception in oncology prevention remains unclear. Understanding how AI-generated content influences attitudes and behaviors is critical for designing effective and trustworthy prevention strategies. We conducted a systematic review and meta-analysis following PRISMA 2020 guidelines (PROSPERO CRD420251021036). PubMed, Scopus, Web of Science, and Google Scholar were searched for studies published between 2020 and 2025 that compared AI-generated and human-generated social media content regarding user attitudes, interaction, or health-related outcomes. Risk of bias was assessed using RoB 2, ROBINS-I, AXIS, MMAT, and CASP tools. Random-effects meta-analyses pooled standardized effect sizes for user interaction and perceived quality. Thirty-three studies (28 quantitative, five qualitative) met inclusion criteria across health, marketing, political, and social media domains. AI-generated content significantly increased user interaction compared with human-generated content (pooled ratio = 1.12; 95 % CI 1.04-1.20), while perceived quality showed a positive but nonsignificant trend. Credibility and emotional resonance consistently mediated user engagement across modalities. AI-generated communication can expand the reach and personalization of cancer prevention messages but carries risks when transparency and factual accuracy are lacking. Ethical frameworks emphasizing disclosure, credibility cues, and expert verification are essential to ensure safe use. Integrating AI tools into oncology prevention strategies may strengthen engagement, trust, and adherence to evidence-based health communication.

11 December 2025 in European Journal of Cancer
Verified
Journal Article
Open access
NCNicolas Carl
Nicolas Carl
MHMartin Joachim Hetz
Martin Joachim Hetz
CWChristoph Wies
Christoph Wies
SHSarah Haggenmüller
Sarah Haggenmüller
JWJana Winterstein
Jana Winterstein
TBTitus Josef Brinker
Titus Josef Brinker
Large language models (LLMs) are utilized to answer queries in urology and oncology, yet the performance is limited due to outdated data and missing source transparency, which undermines clinical reliability and therefore adoption. We developed UroBot, a urology-specific chatbot integrating retrieval-augmented generation (RAG) to provide in-line references and source text previews for each response. In a randomized controlled reader study, UroBot and ChatGPT were compared across ten uro-oncological case rounds. Thirty urologists assessed recommendation correctness, source verifiability and trust with preference ratings collected after each round. UroBot performed significantly better than ChatGPT in recommendation correctness (73 % vs. 50 %; p < 0.001), source attribution (74 % vs. 30 %; p < 0.001) and verifiability of sources (84 % vs. 35 %; p < 0.001). Furthermore, clinicians consistently preferred UroBot for accuracy, source verifiability and trust. Qualitative analysis showed that ChatGPT often produced vague or incorrect citations, with 28 % being non-existent or outdated and 83 % lacking specific sections, whereas UroBot achieved complete alignment on guideline sub-section and page level. These gains in citation precision were mirrored by higher clinician ratings for verifiability and trust. Limitations include the small sample size of ten cases due to feasibility, which may not cover the full uro-oncological spectrum. Our findings show that combining LLMs with RAG with in-line references and source text previews markedly enhances perceived source attribution and verifiability compared to state-of-the-art conventional LLMs. Importantly, this approach is readily transferable across medical subspecialties, enabling reliable and up-to-date clinical decision support.

Verified
Journal Article
Open access
PDF file
KKKlytaimnistra Kiouptsi
Klytaimnistra Kiouptsi
SJSven Jäckel
Sven Jäckel
EWEivor Wilms
Eivor Wilms
GPGIULIA PONTAROLLO
GIULIA PONTAROLLO
JWJana Winterstein
Jana Winterstein
CRChristoph Reinhardt
Christoph Reinhardt
The commensal microbiota is a recognized enhancer of arterial thrombus growth. While several studies have demonstrated the prothrombotic role of the gut microbiota, the molecular mechanisms promoting arterial thrombus growth are still under debate. Here, we demonstrate that germ-free (GF) mice, which from birth lack colonization with a gut microbiota, show diminished static deposition of washed platelets to type I collagen compared with their conventionally raised (CONV-R) counterparts. Flow cytometry experiments revealed that platelets from GF mice show diminished activation of the integrin αIIbβ3 (glycoprotein IIbIIIa) when activated by the platelet agonist adenosine diphosphate (ADP). Furthermore, washed platelets from Toll-like receptor-2 (Tlr2)-deficient mice likewise showed impaired static deposition to the subendothelial matrix component type I collagen compared with wild-type (WT) controls, a process that was unaffected by GPIbα-blockade but influenced by von Willebrand factor (VWF) plasma levels. Collectively, our results indicate that microbiota-triggered steady-state activation of innate immune pathways via TLR2 enhances platelet deposition to subendothelial matrix molecules. Our results link host colonization status with the ADP-triggered activation of integrin αIIbβ3, a pathway promoting platelet deposition to the growing thrombus.

19 October 2025 in Current Oncology
Verified
Journal Article
Open access
PDF file
LMLasse Maywald
Lasse Maywald
LNLisa Thuy Nguyen
Lisa Thuy Nguyen
JWJana Winterstein
Jana Winterstein
MHMartin Joachim Hetz
Martin Joachim Hetz
MMMaurin Helen Mangold
Maurin Helen Mangold
NCNicolas Carl
Nicolas Carl
Large language models (LLMs) are increasingly explored as chatbots for patient education, including applications in urooncology. Since only 12% of adults have proficient health literacy and most patient information materials exceed recommended reading levels, improving readability is crucial. Although LLMs could potentially increase the readability of medical information, evidence is mixed, underscoring the need to assess chatbot outputs in clinical settings. Therefore, this study evaluates the measured and perceived readability of chatbot responses in speech-based interactions with urological patients. Urological patients engaged in unscripted conversations with a GPT-4-based chatbot. Transcripts were analyzed using three readability indices: Flesch–Reading-Ease (FRE), Lesbarkeitsindex (LIX) and Wiener-Sachtextformel (WSF). Perceived readability was assessed using a survey covering technical language, clarity and explainability. Associations between measured and perceived readability were analyzed. Knowledge retention was not assessed in this study. A total of 231 conversations were evaluated. The most frequently addressed topics were prostate cancer (22.5%), robotic-assisted prostatectomy (19.9%) and follow-up (18.6%). Objectively, responses were classified as difficult to read (FRE 43.1 ± 9.1; LIX 52.8 ± 6.2; WSF 11.2 ± 1.6). In contrast, perceived readability was rated highly for technical language, clarity and explainability (83–90%). Correlation analyses revealed no association between objective and perceived readability. Chatbot responses were objectively written at a difficult reading level, exceeding recommendations for optimized health literacy. Nevertheless, most patients perceived the information as clear and understandable. This discrepancy suggests that perceived comprehensibility is influenced by factors beyond measurable linguistic complexity.