The rapid increase in telemedicine coupled with recent advances in diagnostic artificial intelligence (AI) create the imperative to consider the opportunities and risks of inserting AI-based support into new paradigms of care. Here we build on recent achievements in the accuracy of image-based AI for skin cancer diagnosis to address the effects of varied representations of AI-based support across different levels of clinical expertise and multiple clinical workflows. We find that good quality AI-based support of clinical decision-making improves diagnostic accuracy over that of either AI or physicians alone, and that the least experienced clinicians gain the most from AI-based support. We further find that AI-based multiclass probabilities outperformed content-based image retrieval (CBIR) representations of AI in the mobile technology environment, and AI-based support had utility in simulations of second opinions and of telemedicine triage. In addition to demonstrating the potential benefits associated with good quality AI in the hands of non-expert clinicians, we find that faulty AI can mislead the entire spectrum of clinicians, including experts. Lastly, we show that insights derived from AI class-activation maps can inform improvements in human diagnosis. Together, our approach and findings offer a framework for future studies across the spectrum of image-based diagnostics to improve human–computer collaboration in clinical practice.

We investigated whether human preferences hold the potential to improve diagnostic artificial intelligence (AI)-based decision support using skin cancer diagnosis as a use case. We utilized nonuniform rewards and penalties based on expert-generated tables, balancing the benefits and harms of various diagnostic errors, which were applied using reinforcement learning. Compared with supervised learning, the reinforcement learning model improved the sensitivity for melanoma from 61.4% to 79.5% (95% confidence interval (CI): 73.5–85.6%) and for basal cell carcinoma from 79.4% to 87.1% (95% CI: 80.3–93.9%). AI overconfidence was also reduced while simultaneously maintaining accuracy. Reinforcement learning increased the rate of correct diagnoses made by dermatologists by 12.0% (95% CI: 8.8–15.1%) and improved the rate of optimal management decisions from 57.4% to 65.3% (95% CI: 61.7–68.9%). We further demonstrated that the reward-adjusted reinforcement learning model and a threshold-based model outperformed naïve supervised learning in various clinical scenarios. Our findings suggest the potential for incorporating human preferences into image-based diagnostic algorithms.

While contributing to an improved continuity of care, Shared Electronic Health Record (EHR) systems may also lead to information overload of healthcare providers. Document-oriented architectures, such as the commonly employed IHE XDS profile, which only support information retrieval at the level of documents, are particularly susceptible for this problem. The objective of the EHR-ARCHE project was to develop a methodology and a prototype to efficiently satisfy healthcare providers' information needs when accessing a patient's Shared EHR during a treatment situation. We especially aimed to investigate whether this objective can be reached by integrating EHR Archetypes into an IHE XDS environment. Using methodical triangulation, we first analysed the information needs of healthcare providers, focusing on the treatment of diabetes patients as an exemplary application domain. We then designed ISO/EN 13606 Archetypes covering the identified information needs. To support a content-based search for fine-grained information items within EHR documents, we extended the IHE XDS environment with two additional actors. Finally, we conducted a formative and summative evaluation of our approach within a controlled study. We identified 446 frequently needed diabetes-specific information items, representing typical information needs of healthcare providers. We then created 128 Archetypes and 120 EHR documents for two fictive patients. All seven diabetes experts, who evaluated our approach, preferred the content-based search to a conventional XDS search. Success rates of finding relevant information was higher for the content-based search (100% versus 80%) and the latter was also more time-efficient (8-14min versus 20min or more). Our results show that for an efficient satisfaction of health care providers' information needs, a content-based search that rests upon the integration of Archetypes into an IHE XDS-based Shared EHR system is superior to a conventional metadata-based XDS search.

Shared Electronic Health Record (EHR) systems, which provide a health information exchange (HIE) within a community of care, were found to be a key enabler of informational continuity of diabetes mellitus (DM) care. Quantitative analyses of the actual contribution of Shared EHR systems to informational continuity of care are rare. The goal of this study was to quantitatively analyze (i) the degree of fragmentation of DM care in Austria as an indicator for the need for HIE, and (ii) the quantity of information (i.e. number of documents) from Austrian DM patients that would be made available by a nationwide Shared EHR system for HIE. Our analyses are based on social security claims data of 7.9 million Austrians from 2006 and 2007. DM patients were identified through medication data and inpatient diagnoses. The degree of fragmentation was determined by the number of different healthcare providers per patient. The amount of information that would be made available by a nationwide Shared EHR system was estimated by the number of documents that would have been available to a healthcare provider if he had access to information on the patient's visits to any of the other healthcare providers. As a reference value we determined the number of locally available documents that would have originated from the patient's visits to the healthcare provider himself. We performed our analysis for two types of systems: (i) a "comprehensive" Shared EHR system (SEHRS), where each visit of a patient results in a single document (progress note), and (ii) the Austrian ELGA system, which allows four specific document types to be shared. 391,630 DM patients were identified, corresponding to 4.7% of the Austrian population. More than 90% of the patients received health services from more than one healthcare provider in one year. Both, the SEHRS as well as ELGA would have multiplied the available information during a patient visit in comparison to an isolated local EHR system; the median ratio of external to local medical documents was between 1:1 for a typical visit at a primary care provider (SEHRS as well as ELGA) and 39:1 (SEHRS) respectively 28:1 (ELGA) for a typical visit at a hospital. Due to the high degree of care fragmentation, there is an obvious need for HIE for Austrian DM patients. Both, the SEHRS as well as ELGA could provide a substantial contribution to informational continuity of care in Austrian DM treatment. Hospitals and specialists would have gained the most amount of external information, primary care providers and pharmacies would have at least doubled their available information. Despite being the most important potential feeders of a national Shared EHR system according to our analysis, primary care providers will not tap their full corresponding potential under the current implementation scenario of ELGA.