Diagnostic reasoning prompts reveal the potential for large language model interpretability in medicine.

One of the major barriers to using large language models (LLMs) in medicine is the perception they use uninterpretable methods to make clinical decisions that are inherently different from the cognitive processes of clinicians. In this manuscript we develop diagnostic reasoning prompts to study whether LLMs can imitate clinical reasoning while accurately forming a diagnosis. We find that GPT-4 can be prompted to mimic the common clinical reasoning processes of clinicians without sacrificing diagnostic accuracy. This is significant because an LLM that can imitate clinical reasoning to provide an interpretable rationale offers physicians a means to evaluate whether an LLMs response is likely correct and can be trusted for patient care. Prompting methods that use diagnostic reasoning have the potential to mitigate the "black box" limitations of LLMs, bringing them one step closer to safe and effective use in medicine.

Use of Voice-Based Conversational Artificial Intelligence for Basal Insulin Prescription Management Among Patients With Type 2 Diabetes: A Randomized Clinical Trial.

Importance: Optimizing insulin therapy for patients with type 2 diabetes can be challenging given the need for frequent dose adjustments. Most patients receive suboptimal doses and do not achieve glycemic control. Objective: To examine whether a voice-based conversational artificial intelligence (AI) application can help patients with type 2 diabetes titrate basal insulin at home to achieve rapid glycemic control. Design, Setting, and Participants: In this randomized clinical trial conducted at 4 primary care clinics at an academic medical center from March 1, 2021, to December 31, 2022, 32 adults with type 2 diabetes requiring initiation or adjustment of once-daily basal insulin were followed up for 8 weeks. Statistical analysis was performed from January to February 2023. Interventions: Participants were randomized in a 1:1 ratio to receive basal insulin management with a voice-based conversational AI application or standard of care. Main Outcomes and Measures: Primary outcomes were time to optimal insulin dose (number of days needed to achieve glycemic control), insulin adherence, and change in composite survey scores measuring diabetes-related emotional distress and attitudes toward health technology and medication adherence. Secondary outcomes were glycemic control and glycemic improvement. Analysis was performed on an intent-to-treat basis. Results: The study population included 32 patients (mean [SD] age, 55.1 [12.7] years; 19 women [59.4%]). Participants in the voice-based conversational AI group more quickly achieved optimal insulin dosing compared with the standard of care group (median, 15 days [IQR, 6-27 days] vs >56 days [IQR, >29.5 to >56 days]; a significant difference in time-to-event curves; P = .006) and had better insulin adherence (mean [SD], 82.9% [20.6%] vs 50.2% [43.0%]; difference, 32.7% [95% CI, 8.0%-57.4%]; P = .01). Participants in the voice-based conversational AI group were also more likely than those in the standard of care group to achieve glycemic control (13 of 16 [81.3%; 95% CI, 53.7%-95.0%] vs 4 of 16 [25.0%; 95% CI, 8.3%-52.6%]; difference, 56.3% [95% CI, 21.4%-91.1%]; P = .005) and glycemic improvement, as measured by change in mean (SD) fasting blood glucose level (-45.9 [45.9] mg/dL [95% CI, -70.4 to -21.5 mg/dL] vs 23.0 [54.7] mg/dL [95% CI, -8.6 to 54.6 mg/dL]; difference, -68.9 mg/dL [95% CI, -107.1 to -30.7 mg/dL]; P = .001). There was a significant difference between the voice-based conversational AI group and the standard of care group in change in composite survey scores measuring diabetes-related emotional distress (-1.9 points vs 1.7 points; difference, -3.6 points [95% CI, -6.8 to -0.4 points]; P = .03). Conclusions and Relevance: In this randomized clinical trial of a voice-based conversational AI application that provided autonomous basal insulin management for adults with type 2 diabetes, participants in the AI group had significantly improved time to optimal insulin dose, insulin adherence, glycemic control, and diabetes-related emotional distress compared with those in the standard of care group. These findings suggest that voice-based digital health solutions can be useful for medication titration. Trial Registration: ClinicalTrials.gov Identifier: NCT05081011.

Comparison of History of Present Illness Summaries Generated by a Chatbot and Senior Internal Medicine Residents.

This prognostic study assesses the ability of a chatbot to write a history of present illness compared with senior internal medicine residents.

Reactivation of Chagas Disease in a Patient With an Autoimmune Rheumatic Disease: Case Report and Review of the Literature.

Reactivation of Chagas disease has been described in immunosuppressed patients, but there is a paucity of literature describing reactivation in patients on immunosuppressive therapies for the treatment of autoimmune rheumatic diseases. We describe a case of Chagas disease reactivation in a woman taking azathioprine and prednisone for limited cutaneous systemic sclerosis (lcSSc). Reactivation manifested as indurated and erythematous cutaneous nodules. Sequencing of a skin biopsy specimen confirmed the diagnosis of Chagas disease. She was treated with benznidazole with clinical improvement in the cutaneous lesions. However, her clinical course was complicated and included disseminated CMV disease and subsequent septic shock due to bacteremia. Our case and review of the literature highlight that screening for Chagas disease should be strongly considered for patients who will undergo immunosuppression for treatment of autoimmune disease if epidemiologically indicated.

Patient-centered design in developing a mobile application for oral anticancer medications.

OBJECTIVES: To develop and test the usability and feasibility of a customizable mobile application (app) designed to help educate patients about their oral anticancer medications (OAMs) and regimens. SETTING: Outpatient cancer center and oncology pharmacy for urban, Midwestern academic health system. PRACTICE DESCRIPTION: Clinically-supervised educational intervention to support patients learning about OAMs. PRACTICE INNOVATION: With input from patient partners, our interdisciplinary team designed the first known tablet-based educational app that can interface with a patient's electronic medical record. The app is based on learning style and adherence theories and is customizable for individually prescribed OAMs. The app can accommodate multiple learning styles through text at 6th-grade reading level, pictures, animations, and audio voiceovers. Functionalities include interactive educational modules on 11 OAMs and case-based patient stories on common barriers to OAM adherence. EVALUATION: Early phase testing provided the opportunity to observe the user interface with the app and app functionality. Data were summarized descriptively from observations and comments of patient subjects. RESULTS: Thirty patient subjects provided input-19 in phase 1 usability testing and 11 in phase 2 feasibility testing. Comments provided by patient subjects during usability testing were largely positive. Responses included self-identification with patient stories, usefulness of drug information, preferences for text messages, and app limitations (e.g., perceived generational digital divide in technology use and potential patient inability to receive text messages). Using their feedback, modifications were made to the prototype app. Responses in feasibility testing demonstrated the app's usefulness across a wide range of ages. Highest opinion ratings on app usefulness were stated by patients who were newer to OAM therapy. CONCLUSION: User feedback suggests the potential benefit of the app as a tool to help patients with cancer, particularly after the first months for those starting new OAM regimens. Processes and lessons learned are transferable to other settings.

Visual Recognition Software for Binary Classification and Its Application to Spruce Pollen Identification.

Discriminating between black and white spruce (Picea mariana and Picea glauca) is a difficult palynological classification problem that, if solved, would provide valuable data for paleoclimate reconstructions. We developed an open-source visual recognition software (ARLO, Automated Recognition with Layered Optimization) capable of differentiating between these two species at an accuracy on par with human experts. The system applies pattern recognition and machine learning to the analysis of pollen images and discovers general-purpose image features, defined by simple features of lines and grids of pixels taken at different dimensions, size, spacing, and resolution. It adapts to a given problem by searching for the most effective combination of both feature representation and learning strategy. This results in a powerful and flexible framework for image classification. We worked with images acquired using an automated slide scanner. We first applied a hash-based "pollen spotting" model to segment pollen grains from the slide background. We next tested ARLO's ability to reconstruct black to white spruce pollen ratios using artificially constructed slides of known ratios. We then developed a more scalable hash-based method of image analysis that was able to distinguish between the pollen of black and white spruce with an estimated accuracy of 83.61%, comparable to human expert performance. Our results demonstrate the capability of machine learning systems to automate challenging taxonomic classifications in pollen analysis, and our success with simple image representations suggests that our approach is generalizable to many other object recognition problems.

Interacting boson problems can be QMA hard.

Computing the ground-state energy of interacting electron problems has recently been shown to be hard for quantum Merlin Arthur (QMA), a quantum analogue of the complexity class NP. Fermionic problems are usually hard, a phenomenon widely attributed to the so-called sign problem. The corresponding bosonic problems are, according to conventional wisdom, tractable. Here, we demonstrate that the complexity of interacting boson problems is also QMA hard. Moreover, the bosonic version of N-representability problem is QMA complete. Consequently, these problems are unlikely to have efficient quantum algorithms.