Knowledge, Attitudes, and Use of Artificial Intelligence Among Adult and Pediatric Hematologists in Türkiye: A National Cross- Sectional Survey

Technical Summary: Knowledge, Attitudes, and Use of Artificial Intelligence Among Adult and Pediatric Hematologists in Türkiye

Problem Statement
While artificial intelligence (AI), particularly generative AI and large language models (LLMs), is rapidly integrating into clinical and research workflows, its safe adoption depends less on algorithmic performance and more on clinician knowledge, attitudes, governance, and training. In hematology—a data-rich field involving morphology, flow cytometry, and genomics—evidence regarding hematologists' specific readiness, usage patterns, and trust is scarce. No prior national study in Türkiye had simultaneously assessed knowledge, clinical and research application, trust, governance, patient-facing attitudes, and educational needs within this specialty. Furthermore, a gap exists between physician interest in AI and their formal preparation, creating risks regarding data privacy, scientific integrity (e.g., hallucinations), and accountability.

Methodology
The authors conducted a national, cross-sectional, self-administered online survey targeting practicing adult and pediatric hematologists in Türkiye.

• Sampling: The study utilized a census-style invitation sent via verified email to 639 members of the Turkish Society of Hematology (THD). It was a closed survey (members-only) conducted between May 15–30, 2026.
• Instrument: A 41-item questionnaire developed by the THD Artificial Intelligence Scientific Subcommittee (YZ-BAK) covered seven domains: participant profile, awareness/knowledge, clinical use, research/writing use, trust/risk/ethics/governance, attitudes toward patient chatbot use, and educational needs. The survey followed CHERRIES guidelines.
• Data Analysis: Categorical data were summarized as counts and percentages; ordinal data as medians and interquartile ranges (IQR). Group comparisons utilized non-parametric tests (Mann–Whitney U, Kruskal–Wallis) with Benjamini–Hochberg correction for exploratory analyses. Multivariable logistic and ordinal regression models were used to identify independent correlates of AI use and attitudes. Analyses were performed in R.
• Ethics: The study received approval from the Non-Interventional Clinical Research Ethics Committee of Istanbul Medipol University. Participation was voluntary and anonymous.

Key Results
Out of 639 invited members, 146 completed the survey (response rate 22.8%). The cohort consisted of 65.8% adult hematologists and 34.2% pediatric hematologists.

• Adoption vs. Training: Clinical AI use in the prior three months was reported by 87% of respondents. However, only 14.4% had received structured AI training. Familiarity with core AI concepts was modest (median score 3/5).
• Usage Patterns:
  • Clinical: Primary uses included literature summary (81.1%), differential diagnosis support (60.6%), and drug information checking (60.6%). Hematology-specific image interpretation was rare (14.2%).
  • Research: 55.5% used generative AI for research/writing. Among these users, 81.2% had encountered fabricated references or incorrect information ("hallucinations").
• Governance and Risk:
  • Data Privacy: 48.6% of respondents reported entering patient data into non-approved public chatbots; 8.2% admitted to entering identifiable information.
  • Policy: Only 4.8% of respondents reported the existence of a written institutional AI policy.
  • Norms: Despite high usage, responsible-use norms were strong: 89.7% endorsed independent verification of AI output, and 73.3% endorsed declaring AI use.
• Attitudes:
  • Physicians were largely neutral regarding patient-initiated chatbot use but expressed significant concern about potential harm (anxiety/misdirection).
  • 93.2% expressed openness to an official, supervised patient-facing AI tool integrated into their clinic.
  • Few felt professionally threatened (8.2%).
• Determinants of Use: Higher self-rated digital literacy was independently associated with increased research/writing AI use (Adjusted OR 1.87 per point). Experience >10 years was inversely associated with research use but positively associated with endorsing mandatory AI-use declaration.
• Subspecialty Differences: No significant differences were found between adult and pediatric hematologists regarding primary outcomes.

Key Contributions

1. First National Baseline: This is the first national study in Türkiye to jointly assess hematologists' knowledge, clinical/research use, trust, governance, and educational needs.
2. Quantifying the Norm-Practice Gap: The study quantifies a critical divergence between stated norms and actual behavior: while 90% of hematologists endorse verification, nearly half (48.6%) enter patient data into non-approved public chatbots, and 81% of research users encounter hallucinations, all within a context where institutional policies are nearly absent (4.8%).
3. Specialty-Wide Uniformity: The findings indicate that AI adoption patterns and governance needs are consistent across adult and pediatric hematology, suggesting a unified approach to policy and training is required.
4. Operationalizing Integrity Risks: The study moves beyond theoretical concerns to measured exposure, documenting the high prevalence of fabricated references in research workflows.

Significance and Claims
The paper claims that Turkish hematologists have rapidly adopted AI as an efficiency and information tool while maintaining cautious, verification-oriented attitudes and feeling little professional threat. However, this adoption is outpacing formal training and institutional governance.

The authors conclude that the primary barrier to responsible integration is not attitudinal but institutional. They argue that closing the gap between current practice and safe integration requires:

• Hematology-specific governance: Clear policies on permissible tools and patient-data handling.
• Targeted training: Structured education focused on safe generative-AI use and academic applications, particularly for those with lower digital literacy.
• Supervised Tooling: The development and deployment of institution-approved, supervised AI tools, which aligns with the high physician openness (93%) to such solutions.

As the inaugural output of the THD AI Scientific Subcommittee, these findings provide an actionable baseline for policy, education, and the responsible integration of AI into hematology practice.