Health Information Professionals – Respondent Characteristics
Of the 101 respondents to the health informatician survey, the majority were over 40 years old (54.5%), male (62.0%), and white (70.1%) (Table 1). Respondents self-identified in multiple roles, with most identifying as clinicians (44.6%), followed by data scientists (36.6%). The majority of respondents reported having 10 or more years of experience in informatics (54.5%).
Most respondents feel they are very familiar (41.6%) or fairly familiar (48.5%) with AI/ML, and the majority have taken courses on the subject (61.4%) or are currently researching the topic (68.3%). Additionally, the majority of respondents say they have worked on implementing AI/ML (42.6%).
Clinician Survey – Respondent Characteristics
Of the 607 U.S.-based clinician survey respondents, the majority were under 40 years of age (69.9%), female (55.7%), and white (68.2%) (Table 2). 70.7% of respondents were physicians, of which 45.4% were residents. The most common specialty was hospital medicine (52.1%), followed by hematology (20.8%). The majority of respondents (65.8%) reported that they determine daily whether a patient requires VTE prophylaxis. Only 20.5% of respondents reported that they have used AI/ML in their clinical practice, with the majority either not (57.9%) or not sure (21.6%).
Healthcare Informatics Specialist – AI/ML Experience
The majority of informaticians (62.6%) said their organization is using or developing AI/ML for healthcare. Of the 62 respondents, the majority described their organization's AI/ML landscape as having at least one model implemented and in use (82.3%). Additionally, they said their organization primarily develops models in-house (81.4%). Less than half (45.8%) said they use a third-party vendor or partner with a local university (28.8%).
Respondents who developed AI/ML systems used Python (76.6%), R (45.3%), and toolkits (42.2%). Among those who listed their preferred toolkits, Scikit-learn and TensorFlow were most commonly cited.
Healthcare Informatics – Attitudes Towards AI/ML
The majority of informaticians agreed that AI/ML could have a positive impact on patient care (95.0%), help healthcare organizations meet regulatory requirements (95.0%), and have a positive economic impact on healthcare organizations (81.0%) (Fig. 1, Supplementary Table 1). Informatics mostly agreed that AI/ML could perform better than humans (76.3%) and could replace human employees in some job functions (60.4%). Respondents felt that AI/ML was trustworthy overall (58.5%) and would trust their care to an AI/ML system (49.5%); however, less than half would trust a closed, proprietary system (39.7%). Most informaticians agreed that AI/ML should be independently validated and standardized (96.0%), regulated (95.6%), and evaluated in randomized controlled trials (81.2%) before use in clinical practice.
Informatics scholars' attitudes towards AI/ML (n = 101).
The three most common reasons cited by respondents as barriers to the success of AI/ML in healthcare were data quality (67.3%), lack of standardization (39.8%), and difficulty in adoption by healthcare providers (35.7%).
Healthcare informaticists – attitudes towards AI/ML in thrombus management
The majority of informaticians agreed that AI/ML can be used in the clinical management of thrombus (56.0% 95% CI 46–66%). Of these 56 respondents, most agreed that AI/ML can be used for risk stratification (94.6%), radiological accuracy (87.5%), surveillance (80.4%), diagnosis (73.2%), and treatment (73.2%) (Figure 2). Four respondents suggested monitoring the thrombolysis process, warfarin dosing, shared decision-making, and treatment in acute and chronic convalescent phases as potential additional uses of AI/ML. When all respondents were asked about perceived barriers, the most commonly cited barriers were lack of transparency of AI/ML systems (48.5%), concerns that clinicians will not use AI/ML systems (34.7%), and concerns about liability (24.8%) (Supplementary Table 2). Informatics scientists who self-identified as clinicians were more likely to believe that AI could help with VTE than informaticians who did not identify as clinicians, but the difference was not statistically significant (66.7% vs. 47.3%, respectively, p = 0.052). Respondents from organizations that had implemented AI were not significantly more likely to believe that AI could help with VTE than respondents from organizations that had not implemented AI (59.0% vs. 48.7%, respectively, p = 0.32).
Potential application of AI/ML in thrombus management (n = 56).
All respondents were asked the open-ended question: “What other considerations should be made when using AI/ML to assist in the clinical management of thrombus?” Thirty-seven responses were received, of which six did not answer the question and were therefore excluded from the analysis. Of the remaining 31 responses, almost all were related to system validation and discussed factors related to testing, bias, and transparency. Several responses discussed implementation and several mentioned the importance of clinician oversight (themes in Table 3, coding tree in Supplementary Table 3).
