Tech & Innovation

Cohort study: Higher Telehealth Use Linked to Lower Rates of Select Low-Value Services in Medicare

3 Jan, 2025 | 09:30h | UTC

Background: Telehealth has rapidly expanded in recent years, potentially transforming how primary care is delivered. However, questions remain regarding its impact on low-value services—tests or procedures that confer minimal benefit and might be wasteful. Previous research raised concerns that virtual encounters could either reduce or increase unnecessary care, but rigorous data on this matter have been limited.

Objective: To assess whether a primary care practice’s adoption of telehealth is associated with changes in the rate of eight established low-value services, comprising office-based procedures, laboratory tests, imaging studies, and mixed-modality interventions.

Methods: This retrospective cohort study used Medicare fee-for-service claims from 2019 through 2022 for 577,928 beneficiaries attributed to 2,552 primary care practices in Michigan. Practices were grouped into low, medium, or high tertiles of telehealth volume in 2022. A difference-in-differences approach was performed, comparing annualized low-value service rate changes between the prepandemic (2019) and postpandemic (2022) periods.

Results: Overall, high-telehealth practices demonstrated reduced rates of certain office-based low-value services, specifically cervical cancer screening (−2.9 services per 1000 beneficiaries, 95% CI −5.3 to −0.4) among older women. Additionally, high-telehealth practices showed lower rates of select low-value thyroid tests (−40 per 1000 beneficiaries, 95% CI −70 to −9). For five other measures—including imaging for low back pain, imaging for uncomplicated headache, and PSA tests in older men—no significant association was observed between greater telehealth use and low-value service rates. Notably, telehealth volume increased markedly from 2019 to 2022, while in-person visits generally decreased.

Conclusions: These findings suggest that widespread telehealth adoption in Michigan primary care was not associated with elevated low-value service use. In fact, certain office-based low-value tests appeared to decline, possibly owing to fewer face-to-face opportunities to perform unnecessary interventions. Nonetheless, caution is warranted in generalizing these findings, as telehealth’s effects may vary across different clinical contexts.

Implications for Practice: Health care systems should consider structured telehealth protocols that encourage judicious testing and minimize overuse. While telehealth can broaden access, clinicians must remain vigilant to avoid missing necessary care. Clear guidelines, effective triage, and patient education might help balance convenience with quality.

Study Strengths and Limitations: Strengths include a large Medicare population and established low-value service metrics, enhancing the study’s validity. Limitations include a single-state focus (Michigan) and reliance on claims data without detailed clinical information, restricting the scope of outcomes assessed.

Future Research: Further investigation is needed to verify whether these trends extend to other states, different insurance models, and additional low-value services (including medications). Evaluations of telehealth’s role in both low-value and high-value care could offer deeper insights into its broader effects on cost and quality.

Reference: Liu T, Zhu Z, Thompson MP, et al. Primary Care Practice Telehealth Use and Low-Value Care Services. JAMA Network Open. 2024;7(11):e2445436. DOI: http://doi.org/10.1001/jamanetworkopen.2024.45436

Phase 2 RCT: CRISPR-Based Therapy Reduces Attacks in Hereditary Angioedema

2 Jan, 2025 | 10:00h | UTC

Background: Hereditary angioedema (HAE) is a rare autosomal dominant disorder characterized by unpredictable attacks of angioedema involving cutaneous tissues, the gastrointestinal tract, and, potentially, the larynx, posing a risk of asphyxiation. Current prophylactic treatments require frequent administration, often leading to suboptimal adherence and ongoing disease burden. NTLA-2002 is an in vivo CRISPR-Cas9–based therapy designed to permanently inactivate the KLKB1 gene in hepatocytes, thereby reducing plasma kallikrein levels and, hypothetically, lowering attack frequency in patients with HAE.

Objective: To evaluate whether a single intravenous infusion of NTLA-2002 (25 mg or 50 mg) would safely and effectively decrease HAE attack rates and reduce plasma kallikrein protein levels over a 16-week primary observation period, as compared with placebo.

Methods: This phase 2, randomized, double-blind, placebo-controlled trial included 27 adults with confirmed type 1 or type 2 HAE. Participants were assigned in a 2:2:1 ratio to receive a one-time dose of 25 mg or 50 mg of NTLA-2002 or placebo. The primary endpoint was the investigator-confirmed number of angioedema attacks per month from Week 1 through Week 16. Secondary endpoints included the number of moderate-to-severe attacks, use of on-demand therapy, adverse events, and changes in total plasma kallikrein protein levels (analyzed by immunoassays). Exploratory measures encompassed patient-reported outcomes using the Angioedema Quality of Life (AE-QoL) questionnaire.

Results: During the 16-week period, the mean monthly attack rate decreased by 75% in the 25 mg group and 77% in the 50 mg group relative to placebo (estimated rates of 0.70 vs. 0.65 vs. 2.82 attacks per month, respectively). Notably, 4 of 10 patients (40%) in the 25 mg group and 8 of 11 (73%) in the 50 mg group reported no attacks or further prophylaxis use after dosing. Placebo recipients showed only a 16% reduction from baseline. Adverse events were predominantly mild to moderate; headache, fatigue, and nasopharyngitis were most common. Infusion-related reactions occurred in a few patients but resolved without sequelae. A single transient grade 2 elevation in alanine aminotransferase was recorded in one participant given 25 mg of NTLA-2002. By Week 16, total plasma kallikrein levels decreased by 55% in the 25 mg group and 86% in the 50 mg group, with no meaningful changes in placebo.

Conclusions: A single intravenous infusion of NTLA-2002 significantly lowered attack frequency and reduced total plasma kallikrein levels in HAE. Most patients treated at 50 mg experienced no attacks, suggesting that long-term prophylaxis might be unnecessary for many. Longer observation supports durability, yet cost and potential long-term effects of gene editing warrant cautious interpretation.

Implications for Practice: If confirmed by larger phase 3 trials, this gene-editing approach could alter the management of HAE, reducing or eliminating the need for continuous prophylaxis. However, clinicians must weigh the high upfront cost, possible unpredictable immune responses, and the novelty of CRISPR-based therapies before integrating them into standard care.

Study Strengths and Limitations: Strengths include a placebo-controlled design, meaningful improvement in patient-reported outcomes, and robust plasma kallikrein protein reduction. Limitations are the small sample size, short primary observation period, and uncertain long-term safety in diverse populations.

Future Research: Ongoing phase 3 studies with larger cohorts and extended follow-up are essential to confirm safety, long-term efficacy, and cost-effectiveness.

Reference: Cohn DM, Gurugama P, Magerl M, et al. CRISPR-Based Therapy for Hereditary Angioedema. New England Journal of Medicine. 2024; DOI: http://doi.org/10.1056/NEJMoa2405734

Editorial: Musunuru K. A Milestone for Gene-Editing Therapies. New England Journal of Medicine. 2024; DOI: http://doi.org/10.1056/NEJMe2412176

Scoping Review of RCTs: AI Interventions Show Positive but Varied Impact in Clinical Practice

28 Dec, 2024 | 00:04h | UTC

Background: The rapid expansion of artificial intelligence (AI) in health care has stimulated a growing number of randomized controlled trials (RCTs) intended to validate AI’s clinical utility. However, many AI models previously tested in retrospective or simulated settings lack real-world evidence. Investigating the breadth and depth of these RCTs is key to understanding the current status of AI in clinical practice.

Objective: This scoping review aimed to identify, classify, and evaluate RCTs that integrate modern AI (non-linear computational models, including deep learning) into patient management. The primary goal was to assess geographic distribution, trial design, outcomes measured (diagnostic performance, care management, patient behavior, clinical decision making), and the overall success rate of AI-based interventions.

Methods: The authors systematically searched PubMed, SCOPUS, CENTRAL, and the International Clinical Trials Registry Platform from January 2018 to November 2023. Included studies featured an AI intervention integrated into clinical workflows, with patient outcomes influenced by clinician–AI interactions or standalone AI systems. Exclusions encompassed linear-risk models and non-English publications. After screening 10,484 records, 86 RCTs were ultimately included. Simple descriptive statistics summarized trial characteristics, endpoints, and results.

Results: Most RCTs (63%) were single-center studies with a median sample size of 359 patients. Gastroenterology (43%) and Radiology (13%) were leading specialties, often focusing on deep learning algorithms for endoscopic or imaging tasks. The USA led in overall trial volume, followed by China, with 81% of all trials reporting positive primary outcomes (improvements or non-inferiority). Diagnostic yield or performance metrics predominated (54%), though some studies evaluated patient-centered endpoints such as adherence or symptom reduction. Despite these promising findings, 60% of trials measuring operational time showed mixed effects—some reported reduced procedural times (p<0.05), while others noted significant increases (p<0.05).

Conclusions: AI-driven interventions generally improved diagnostic measures and care processes, demonstrating potential for augmenting clinical decision making. Nevertheless, the prevalence of single-center designs limits the generalizability of outcomes. Publication bias remains a concern, given that negative or null results may be underreported. More extensive multicenter RCTs, greater demographic transparency, and standardized reporting are critical to fully determine AI’s clinical relevance.

Implications for Practice: AI tools might enhance screening, detection rates, and therapeutic monitoring in areas like gastroenterology, radiology, and cardiology. Clinicians should remain mindful of possible workflow inefficiencies and biases. Thorough validation and robust implementation strategies are essential before widespread adoption can be justified.

Study Strengths and Limitations: Strengths include a timely review capturing diverse RCTs up to late 2023 and strict inclusion criteria requiring true AI integration into patient care. Limitations include the English-only search and reliance on published results, potentially omitting unpublished or negative trials.

Future Research: Further investigations should prioritize multicenter, large-scale RCTs with meaningful clinical endpoints—quality of life, survival, and long-term safety. Enhanced adherence to reporting standards (CONSORT-AI) and recruitment of ethnically diverse populations are necessary steps to advance the field.

Reference: Han R, Acosta JN, Shakeri Z, Ioannidis JPA, Topol EJ, Rajpurkar P, et al. “Randomised controlled trials evaluating artificial intelligence in clinical practice: a scoping review.” The Lancet Digital Health. 2024;6(5). DOI: http://doi.org/10.1016/S2589-7500(24)00047-5

Cohort Study: Higher Telehealth Intensity May Reduce Certain Office-Based Low-Value Services in Medicare Primary Care

2 Jan, 2025 | 08:00h | UTC

Background: The rapid expansion of telehealth has raised concerns about its potential to foster wasteful services, especially in primary care. While telehealth can eliminate certain in-person interventions, it might also increase unnecessary laboratory or imaging requests, given the more limited physical exam. Evaluating how telehealth intensity affects the provision of low-value care is crucial for guiding future policy and clinical practice.

Objective: To determine whether higher telehealth utilization at the practice level is associated with changes in the rates of common low-value services among Medicare fee-for-service beneficiaries in Michigan.

Methods: Using Medicare claims data from January 1, 2019, to December 31, 2022, this retrospective cohort employed a difference-in-differences design. A total of 577,928 beneficiaries attributed to 2,552 primary care practices were included. Practices were stratified into low, medium, or high telehealth tertiles based on the volume of virtual visits per 1,000 beneficiaries in 2022. Eight low-value services relevant to primary care were grouped into four main categories: office-based (e.g., cervical cancer screening in women older than 65), laboratory-based, imaging-based, and mixed-modality services.

Results: Among the 577,928 beneficiaries (332,100 women; mean age, 76 years), practices with high telehealth utilization had a greater reduction in office-based cervical cancer screening (−2.9 [95% CI, −5.3 to −0.4] services per 1,000 beneficiaries) and low-value thyroid testing (−40 [95% CI, −70 to −9] tests per 1,000 beneficiaries), compared with low-utilization practices. No significant association emerged for other laboratory- or imaging-based low-value services, including PSA testing for men over 75 or imaging for uncomplicated low back pain. These findings suggest that while telehealth can lower certain office-based low-value services, it does not appear to substantially increase other types of wasteful care.

Conclusions: High telehealth intensity was linked to reductions in specific low-value procedures delivered in-office, without raising the overall rates of other potentially unnecessary interventions. These data may alleviate some policy concerns that telehealth drives excessive or wasteful care due to its convenience. Instead, substituting certain in-person visits with virtual encounters might curtail opportunities for procedures with minimal clinical benefit.

Implications for Practice: For clinicians and policymakers, these results underscore the possibility that carefully implemented telehealth may reduce some low-value services. Nonetheless, sustained monitoring is needed to confirm whether telehealth encourages or discourages appropriate clinical decision-making across a broader range of interventions.

Study Strengths and Limitations: Strengths include a sizable cohort, a pre- versus post-pandemic time frame, and comprehensive analysis of multiple low-value outcomes. Limitations involve the exclusive focus on beneficiaries in Michigan, the inability to capture prescription-related low-value practices (e.g., antibiotic overuse), and the reliance on claims-based measures, which lack clinical details.

Future Research: Subsequent studies should expand to different geographic areas, assess additional low-value endpoints such as overtreatment with medications, and explore whether demographic or socioeconomic factors modify telehealth’s impact on care quality.

Reference: Liu T, Zhu Z, Thompson MP, et al. Primary Care Practice Telehealth Use and Low-Value Care Services. JAMA Netw Open. 2024;7(11):e2445436. DOI: http://doi.org/10.1001/jamanetworkopen.2024.45436

VisionFM: A Generalist AI Surpasses Single-Modality Models in Ophthalmic Diagnostics

25 Dec, 2024 | 13:41h | UTC

Background: Ophthalmic AI models typically address single diseases or modalities. Their limited generalizability restricts broad clinical application. This study introduces VisionFM, a novel foundation model trained on 3.4 million images from over 500,000 individuals. It covers eight distinct ophthalmic imaging modalities (e.g., fundus photography, OCT, slit-lamp, ultrasound, MRI) and encompasses multiple diseases. Compared with prior single-task or single-modality approaches, VisionFM’s architecture and large-scale pretraining enable diverse tasks such as disease screening, lesion segmentation, prognosis, and prediction of systemic markers.

Objective: To develop and validate a generalist ophthalmic AI framework that can handle multiple imaging modalities, recognize multiple diseases, and adapt to new clinical tasks through efficient fine-tuning, potentially easing the global burden of vision impairment.

Methods: VisionFM employs individual Vision Transformer–based encoders for each of the eight imaging modalities, pretrained with self-supervised learning (iBOT) focused on masked image modeling. After pretraining, various task-specific decoders were fine-tuned for classification, segmentation, and prediction tasks. The model was evaluated on 53 public and 12 private datasets, covering eight disease categories (e.g., diabetic retinopathy, glaucoma, cataract), five imaging modalities (fundus photographs, OCT, etc.), plus additional tasks (e.g., MRI-based orbital tumor segmentation). Performance metrics included AUROCs, Dice similarity coefficients, F1 scores, and comparisons with ophthalmologists of varying clinical experience.

Results: VisionFM achieved an average AUROC of 0.950 (95% CI, 0.941–0.959) across eight disease categories in internal validation. External validation showed AUROCs of 0.945 (95% CI, 0.934–0.956) for diabetic retinopathy and 0.974 (95% CI, 0.966–0.983) for AMD, surpassing baseline deep learning approaches. In a 12-disease classification test involving 38 ophthalmologists, VisionFM’s accuracy matched intermediate-level specialists. It successfully handled modality shifts (e.g., grading diabetic retinopathy on previously unseen OCTA), with an AUROC of 0.935 (95% CI, 0.902–0.964). VisionFM also predicted glaucoma progression (F1, 72.3%; 95% CI, 55.0–86.3) and flagged possible intracranial tumors (AUROC, 0.986; 95% CI, 0.960–1.00) from fundus images.

Conclusions: VisionFM offers a versatile, scalable platform for comprehensive ophthalmic tasks. Through self-supervised learning and efficient fine-tuning, it extends specialist-level performance to multiple clinical scenarios and imaging modalities. The study demonstrates that large-scale, multimodal pretraining can enable robust generalization to unseen data, potentially reducing data annotation burdens and accelerating AI adoption worldwide.

Implications for Practice: VisionFM may help address global shortages of qualified ophthalmologists and expand care in low-resource settings, though clinical decision-making still requires appropriate human oversight. Further multicenter studies are needed before widespread implementation, especially for higher-risk use cases such as tumor detection.

Study Strengths and Limitations: Strengths include its unique multimodal design, large-scale pretraining, and extensive external validation. Limitations involve demographic bias toward Chinese datasets, the need for larger cohorts in certain applications (e.g., intracranial tumor detection), and the challenges of matching real-world clinical complexity when only image-based data are used.

Future Research: Further validation in diverse populations, integration of new imaging modalities (e.g., widefield imaging, ultrasound variants), and expansion to additional diseases are planned. Hybridization with large language models could facilitate automatic generation of clinical reports.

Reference: Qiu J, Wu J, Wei H, et al. Development and Validation of a Multimodal Multitask Vision Foundation Model for Generalist Ophthalmic Artificial Intelligence. NEJM AI 2024;1(12). DOI: http://doi.org/10.1056/AIoa2300221

Large Language Models for Structured Reporting in Radiology: Past, Present, and Future

21 Dec, 2024 | 14:12h | UTC

Introduction/Context: Structured reporting (SR) has been a topic of discussion for decades as a way to standardize and improve the quality of radiology reports. Although there is evidence that SR can reduce errors and enhance guideline adherence, widespread adoption remains limited. Large language models (LLMs), transformer-based and trained on large volumes of text, have emerged as a promising solution to automate and facilitate structured radiology reporting. This narrative review provides an overview of LLM-based SR, while also discussing limitations, regulatory challenges, and future applications.

Key Points/Findings/Recommendations:

Application of LLMs in Structured Reporting: Studies focus on models such as GPT-3.5 and GPT-4, showing promising results in converting free-text radiology reports into structured formats, including multilingual capabilities.
Main Advantages:
- Consistency and improved standardization of reports.
- Potential to reduce errors and enhance comprehensiveness.
- Streamlining the use of predefined templates and formats.
Multilingual and Translational Capabilities: Research shows that several LLMs can process reports in different languages, promoting broader global collaboration.
Technical Limitations: Hallucination (fabricated content), terminology inconsistencies, and misinterpretations remain hurdles to large-scale adoption.
Regulatory and Privacy Challenges: Proprietary model opacity and a lack of regulated pathways pose difficulties for safely integrating these systems into clinical practice.

Implications for Practice:

Automation and Efficiency: LLMs can streamline the reporting process, reduce typing effort, and facilitate standardized descriptions, offering efficiency gains.
System Integration: Incorporation of LLMs into radiology systems (PACS, RIS) could help with documentation and report formatting, improving communication across teams.
Broader Clinical Perspective: Standardized reports may enhance information sharing and potentially patient safety, especially in multidisciplinary settings.

Limitations/Considerations:

Hallucination: Even advanced models may generate inaccurate or fictitious content unrelated to the evidence.
Lack of Transparency: Proprietary models often do not disclose their training data or algorithms, hindering external validation.
Evolving Regulations: The European Union and other countries are developing AI-specific legal frameworks. Meeting safety, reliability, and transparency requirements is critical for clinical use.
Availability of Open-Source Models: While open-source models offer more regulatory flexibility, they still require significant validation and refinement for clinical application.

Conclusion: LLMs have significant potential to transform structured radiology reporting, offering increased efficiency and accuracy. However, regulatory issues, model opacity, and current technical limitations must be addressed before these tools can be safely and effectively integrated into clinical practice. Future research should explore clinical acceptance of LLM-generated reports, compare them to radiologist-produced reports, and investigate how these models can be best integrated into existing systems.

Sources and Links:

Busch, F., Hoffmann, L., dos Santos, D.P. et al. Large language models for structured reporting in radiology: past, present, and future. Eur Radiol (2024). https://doi.org/10.1007/s00330-024-11107-6

Review: Enhancing Interpretability and Accuracy of AI Models in Healthcare

16 Dec, 2024 | 11:06h | UTC

Introduction: Artificial intelligence (AI) has shown remarkable potential in healthcare for improving diagnostics, predictive modeling, and treatment planning. However, the “black-box” nature of many high-performing AI models limits their trustworthiness and clinical utility. Challenges such as limited generalizability, variable performance across populations, and difficulty in explaining model decisions remain critical barriers to widespread adoption. This review synthesizes current evidence on AI models in healthcare, focusing on the interplay between model accuracy and interpretability. By highlighting these issues, we aim to guide healthcare professionals and researchers toward strategies that balance performance with transparency and reliability.

Key Recommendations:

Adopt Interpretable Models: Incorporate interpretable frameworks (e.g., LIME, SHAP, Grad-CAM) that allow clinicians to understand model outputs. Such techniques facilitate better acceptance of AI-driven diagnostics, reducing the risk of misinterpretations and erroneous clinical decisions.
Balance Accuracy and Transparency: Consider hybrid models that achieve high accuracy while maintaining a degree of explainability. Techniques that blend deep learning with simpler statistical or machine learning approaches can offer clinical insights into AI-derived predictions without substantially sacrificing performance.
Integrate Uncertainty Quantification: Employ methods that estimate uncertainty in AI predictions, providing clinicians with confidence intervals or probability distributions. This approach helps manage expectations, enabling safer decision-making in critical clinical scenarios.
Use Multimodal Data and Diverse Populations: Encourage the integration of multiple data sources—imaging, clinical notes, genomics—to improve model generalizability. Validate models on diverse patient cohorts to ensure robust performance across various healthcare environments and patient demographics.
Collaborative, User-Centered Development: Engage healthcare professionals in the model development process to ensure clinical relevance and usability. User-centered design can guide AI solutions that seamlessly fit into existing workflows and foster greater trust among end-users.

Conclusion: The path to fully leveraging AI in healthcare involves overcoming interpretability challenges while maintaining accuracy. Implementing interpretable, uncertainty-aware models validated on diverse datasets will enhance trust, foster responsible adoption, and ultimately improve patient outcomes. By focusing on both technological innovation and user-centered development, we can drive AI models toward safer, more transparent, and clinically beneficial applications.

Reference: Ennab M, Mcheick H. Enhancing interpretability and accuracy of AI models in healthcare: a comprehensive review on challenges and future directions. Frontiers in Robotics and AI. 2024;11:Article 1444763. DOI: https://doi.org/10.3389/frobt.2024.1444763

RCT: Tele-ICU Intervention Did Not Significantly Reduce ICU Length of Stay in Critically Ill Patients

10 Oct, 2024 | 17:40h | UTC

Background: Telemedicine in critical care, particularly through tele-ICU interventions, has gained traction as a potential solution to the global shortage of intensivists. These systems, which include remote intensivist-led care, have shown promise in improving outcomes, but robust evidence from randomized clinical trials is lacking. The TELESCOPE trial was conducted to assess whether daily remote multidisciplinary rounds combined with monthly audit and feedback meetings could reduce ICU length of stay (LOS) compared with standard care.

Objective: The primary objective of the TELESCOPE trial was to determine if a tele-ICU intervention, involving remote daily multidisciplinary rounds and monthly performance audits led by a board-certified intensivist, could reduce ICU LOS compared to usual care.

Methods: This was a cluster randomized clinical trial involving 30 general ICUs in Brazil, enrolling all consecutive adult patients admitted between June 2019 and April 2021. A total of 17,024 patients were included, with 15 ICUs receiving the tele-ICU intervention and 15 receiving standard care. The intervention consisted of daily remote rounds led by an intensivist, monthly audit meetings, and the provision of evidence-based protocols. The primary outcome was ICU LOS, and secondary outcomes included hospital mortality, ICU efficiency, and various infection rates.

Results: There was no statistically significant difference in ICU LOS between the intervention and control groups (mean LOS: 8.1 days in the tele-ICU group vs. 7.1 days in the usual care group; percentage change, 8.2%; 95% CI, −5.4% to 23.8%; P = .24). Hospital mortality was also similar (41.6% vs. 40.2%; odds ratio, 0.93; 95% CI, 0.78-1.12). No significant differences were found in secondary outcomes, including rates of central line-associated bloodstream infections, ventilator-associated events, or ventilator-free days at 28 days.

Conclusions: The tele-ICU intervention did not reduce ICU LOS in critically ill patients. The lack of observed benefit may be due to suboptimal implementation, variable adherence by local teams, and the high severity of illness in the patient population.

Implications for Practice: While tele-ICU models hold potential, this study suggests that remote intensivist-led care, as implemented in the TELESCOPE trial, may not be sufficient to improve outcomes in high-resource ICU settings with critically ill patients.

Study Strengths and Limitations: The study’s strengths include its pragmatic design, the large number of patients enrolled, and its reflection of real-world ICU settings. However, limitations include the unblinded nature of the trial, suboptimal adherence to the tele-ICU protocol in some centers, and the strain on ICU resources during the COVID-19 pandemic, which may have affected the trial’s outcomes.

Future Research: Further studies should explore how tele-ICU interventions can be optimized, with a focus on identifying the ICU environments and patient populations most likely to benefit. Trials should also address potential barriers to effective implementation, such as staff engagement and local resource constraints.

Reference: Pereira AJ, et al. (2024) Effect of Tele-ICU on Clinical Outcomes of Critically Ill Patients: The TELESCOPE Randomized Clinical Trial. JAMA. DOI: http://doi.org/10.1001/jama.2024.20651

RCT: Telehealth-Delivered Early Palliative Care Equivalent to In-Person Care in Advanced Lung Cancer

26 Sep, 2024 | 15:06h | UTC

Background: Patients with advanced lung cancer often face a high symptom burden and decreased quality of life (QOL), but access to early palliative care, which can improve these outcomes, remains limited. While telehealth has become increasingly utilized due to the COVID-19 pandemic, it is unclear whether virtual palliative care is as effective as in-person care.

Objective: To compare the effect of early palliative care delivered via secure video vs in-person visits on the quality of life of patients with advanced non–small cell lung cancer (NSCLC).

Methods: This multisite, randomized comparative effectiveness trial enrolled 1250 adults with advanced NSCLC from 22 cancer centers in the US between June 2018 and May 2023. Participants were randomized to receive either early palliative care via video visits or in person every four weeks. The primary outcome was QOL measured by the Functional Assessment of Cancer Therapy-Lung (FACT-L) questionnaire at 24 weeks. Secondary outcomes included caregiver participation in palliative care visits and patient and caregiver satisfaction with care, mood symptoms, coping, and prognostic understanding.

Results: By week 24, patients in both groups reported equivalent QOL scores, with the video visit group scoring a mean of 99.7 compared to 97.7 in the in-person group (difference of 2.0 points, 90% CI, 0.1-3.9; P = .04 for equivalence). Both groups experienced similar improvements in QOL from baseline (mean increase of 8.4 points for video visits and 6.9 points for in-person care). Caregiver participation in palliative care visits was lower in the video visit group (36.6% vs 49.7%; P < .001). No significant differences were found between the groups in caregiver QOL, patient or caregiver satisfaction with care, mood symptoms, or coping strategies.

Conclusions: Early palliative care delivered via telehealth was equivalent to in-person visits in improving QOL for patients with advanced NSCLC. This underscores the potential of telehealth to increase access to essential palliative care services for this population without compromising care quality.

Implications for Practice: Telehealth can provide a feasible alternative to in-person palliative care, especially for patients with advanced lung cancer who face barriers to in-person visits, such as transportation challenges. However, strategies to enhance caregiver involvement in virtual visits may need to be developed.

Study Strengths and Limitations: Strengths include the large, multisite randomized design and the use of validated outcome measures. Limitations involve the COVID-19 pandemic’s impact, which caused some intervention contamination due to unavoidable video visits in the in-person group. Additionally, caregiver participation was lower than expected, potentially limiting the generalizability of results regarding caregiver outcomes.

Future Research: Further studies should explore the long-term impact of telehealth on palliative care outcomes and investigate ways to enhance caregiver involvement in virtual care.

Reference: Greer, J. A., et al. (2024). Telehealth vs In-Person Early Palliative Care for Patients With Advanced Lung Cancer: A Multisite Randomized Clinical Trial. JAMA. DOI: http://doi.org/10.1001/jama.2024.13964

Retrospective Study: AI Tool Accurately Excludes Pathology in Up to 52.7% of Unremarkable Chest Radiographs with Low Critical Misses – Radiology

24 Aug, 2024 | 16:14h | UTC

Study Design and Population: This retrospective study assessed the effectiveness of a commercial AI tool in correctly identifying unremarkable chest radiographs, thus potentially reducing the workload in radiology departments. The study analyzed 1,961 chest radiographs from adult patients (median age: 72 years) across four Danish hospitals. The radiographs were labeled as remarkable or unremarkable by thoracic radiologists, and the AI tool’s performance was evaluated at varying sensitivity thresholds.

Main Findings: The AI tool demonstrated a specificity of 24.5% to 52.7% at sensitivity thresholds of 99.9% to 98.0%, respectively, effectively excluding pathology in unremarkable chest radiographs. At sensitivities of 95.4% or higher, the AI had equal or lower rates of critical misses compared to radiology reports, with the AI missing only 2.2% of critical findings compared to 1.1% by radiologists at similar sensitivity levels.

Implications for Practice: The results suggest that AI tools could autonomously report up to 52.7% of unremarkable chest radiographs, potentially reducing radiologist workload without compromising patient safety. However, prospective studies are necessary to confirm these findings and optimize AI deployment in clinical practice.

Reference: Plesner LL, Müller FC, Brejnebøl MW, et al. (2024). Using AI to Identify Unremarkable Chest Radiographs for Automatic Reporting. Radiology, 312(2), e240272. DOI: https://doi.org/10.1148/radiol.240272

Innovative Antimicrobial Susceptibility Testing Bypasses Blood Culture, Promising Faster Sepsis Diagnosis – Nature

18 Aug, 2024 | 14:09h | UTC

Study Design and Population: This study introduces a novel ultra-rapid antimicrobial susceptibility testing (AST) method that bypasses the traditional blood culture process, potentially reducing diagnostic time by 40-60 hours. The method was evaluated using a cohort of 190 hospitalized patients in Korea with suspected sepsis, including those with blood cancers.

Main Findings: The new AST method identified bacterial species in all patients with positive blood infections, achieving a 100% match in species identification. For antimicrobial susceptibility, the method demonstrated a 94.9% categorical agreement with conventional AST methods, with a theoretical turnaround time of 13 ± 2.53 hours, significantly faster than current workflows.

Implications for Practice: This method could improve sepsis treatment by providing same-day results, potentially reducing sepsis-related mortality and the use of broad-spectrum antibiotics. However, further validation in a more diverse patient population is necessary to confirm its clinical efficacy and value.

Reference: Kim, T. H., Kang, J., Jang, H., Joo, H., Lee, G. Y., Kim, H., et al. (2024). Blood culture-free ultra-rapid antimicrobial susceptibility testing. Nature, (2024).

Cluster RCT: AI-ECG Shows Potential to Reduce Door-to-Balloon Time and Cardiac Deaths in STEMI Patients – NEJM AI

10 Aug, 2024 | 21:57h | UTC

Study Design and Population: This open-label, cluster randomized controlled trial assessed the impact of AI-powered electrocardiogram (AI-ECG) on reducing treatment delays for ST-elevation myocardial infarction (STEMI). The study involved 43,234 patients, with an average age of 60 years, at Tri-Service General Hospital in Taiwan. Patients were randomized 1:1 into an intervention group (AI-ECG-assisted STEMI detection) or a control group receiving standard care.

Main Findings: AI-ECG significantly reduced the median door-to-balloon time for emergency department patients (82.0 vs. 96.0 minutes, P=0.002) and the ECG-to-balloon time across all settings (78.0 vs. 83.6 minutes, P=0.011). While the AI-ECG intervention did not significantly affect all-cause mortality or new-onset heart failure, it led to a notable reduction in cardiac death rates (85 vs. 116 cases; odds ratio, 0.73; P=0.029).

Implications for Practice: AI-ECG can expedite the critical time to treatment for STEMI patients, potentially reducing cardiac death. Although overall mortality remained unchanged, the reduction in cardiac deaths suggests that AI-ECG could be a valuable tool in emergency and inpatient settings to improve outcomes for STEMI patients.

Reference: Lin C. et al. (2024). Artificial Intelligence–Powered Rapid Identification of ST-Elevation Myocardial Infarction via Electrocardiogram (ARISE) — A Pragmatic Randomized Controlled Trial. NEJM AI, 1(7). DOI: 10.1056/AIoa2400190.

Retrospective Study: Automated Multiorgan CT Markers Predict Diabetes and Cardiometabolic Comorbidities – Radiology

10 Aug, 2024 | 21:36h | UTC

Study Design and Population: This retrospective study analyzed data from 32,166 Korean adults (mean age, 45 years) who underwent health screenings, including fluorodeoxyglucose PET/CT scans, between 2012 and 2015. The study aimed to evaluate the predictive ability of automated CT-derived markers, such as visceral and subcutaneous fat, muscle area, bone density, liver fat, and aortic calcification, for diabetes and associated cardiometabolic conditions.

Main Findings: Visceral fat index showed the highest predictive performance for both prevalent and incident diabetes, with an AUC of 0.70 for men and 0.82 for women in cross-sectional analyses. Combining visceral fat, muscle area, liver fat, and aortic calcification improved prediction, yielding a C-index of 0.69 for men and 0.83 for women. Additionally, the study found that these CT markers were effective in identifying metabolic syndrome, fatty liver, coronary artery calcium scores >100, sarcopenia, and osteoporosis, with AUCs ranging from 0.80 to 0.95.

Implications for Practice: Automated CT-derived markers can effectively predict diabetes and multiple cardiometabolic comorbidities, surpassing traditional anthropometric measures. These findings suggest that integrating such automated assessments into routine clinical practice could enhance risk stratification and preventive care, particularly through opportunistic screening during routine CT scans.

Reference: Chang Y, Yoon SH, Kwon R, et al. (2024). Automated Comprehensive CT Assessment of the Risk of Diabetes and Associated Cardiometabolic Conditions. Radiology, 312(2), e233410. DOI: https://doi.org/10.1148/radiol.233410.

Deep Learning Model Noninferior to Radiologists in Detecting Clinically Significant Prostate Cancer at MRI – Radiology

10 Aug, 2024 | 21:31h | UTC

Study Design and Population: This retrospective study evaluated the performance of a deep learning (DL) model for detecting clinically significant prostate cancer (csPCa) using multiparametric MRI (mpMRI) images from 5215 patients (5735 examinations) with a mean age of 66 years. The study included patients who underwent prostate MRI between January 2017 and December 2019 at a single academic institution. The DL model was trained on T2-weighted, diffusion-weighted, and contrast-enhanced MRI sequences, with pathologic diagnosis as the reference standard.

Main Findings: The DL model achieved an area under the receiver operating characteristic curve (AUC) of 0.89 on the internal test set and 0.86 on an external test set, demonstrating noninferiority to radiologists, who had AUCs of 0.89 and 0.84, respectively. Additionally, the combination of the DL model and radiologists improved diagnostic performance (AUC of 0.89). Gradient-weighted class activation maps (Grad-CAMs) effectively localized csPCa lesions, overlapping with true-positive cases in 92% of internal test set and 97% of external test set cases.

Implications for Practice: The DL model showed comparable performance to experienced radiologists in detecting csPCa at MRI, suggesting its potential to assist radiologists in improving diagnostic accuracy and reducing interobserver variability. Future research should focus on integrating the model into clinical workflows and assessing its impact on biopsy targeting.

Reference: Cai JC, Nakai H, Kuanar S, et al. (2024). Fully Automated Deep Learning Model to Detect Clinically Significant Prostate Cancer at MRI. Radiology, 312(2): e232635. DOI: https://doi.org/10.1148/radiol.232635.

Cross-Sectional Study: AI Model Accurately Detects Myopia, Strabismus, and Ptosis in Children Using Smartphone Photos – JAMA Netw Open

10 Aug, 2024 | 21:21h | UTC

Study Design and Population: This cross-sectional study was conducted at Shanghai Ninth People’s Hospital from October 2022 to September 2023, including 476 children diagnosed with myopia, strabismus, or ptosis. A total of 1,419 images were used to develop an AI model to detect these conditions based on mobile phone photographs.

Main Findings: The AI model demonstrated strong performance with a sensitivity of 0.84 for myopia, 0.73 for strabismus, and 0.85 for ptosis. The model achieved overall accuracies exceeding 0.80 for myopia and strabismus and 0.92 for ptosis, indicating its effectiveness in early detection of these pediatric eye conditions using only smartphone images.

Implications for Practice: The findings suggest that AI-based screening tools could enable early detection of common pediatric eye diseases at home, reducing the reliance on hospital-based screenings. This approach could facilitate timely intervention, improving visual outcomes and reducing the burden on healthcare systems.

Reference: Shu, Q. et al. (2024). Artificial Intelligence for Early Detection of Pediatric Eye Diseases Using Mobile Photos. JAMA Network Open, 7(8), e2425124. DOI: 10.1001/jamanetworkopen.2024.25124.

Meta-Analysis: Effectiveness of therapist-guided remote vs. in-person cognitive behavioral therapy

20 Mar, 2024 | 19:32h | UTC

Study Design and Population: This systematic review and meta-analysis investigated the efficacy of therapist-guided remote cognitive behavioral therapy (CBT) compared to traditional in-person CBT. The authors conducted a comprehensive search across several databases, including MEDLINE, Embase, PsycINFO, CINAHL, and the Cochrane Central Register of Controlled Trials, up to July 4, 2023. A total of 54 randomized controlled trials (RCTs) were included, encompassing 5463 adult patients presenting with various clinical conditions. The study meticulously assessed the risk of bias and extracted data using a standardized approach, and outcomes were analyzed using a random-effects model.

Main Findings: The primary analysis focused on patient-important outcomes, comparing the effectiveness of remote and in-person CBT across diverse conditions such as anxiety and related disorders, depressive symptoms, insomnia, chronic pain or fatigue syndromes, body image or eating disorders, tinnitus, alcohol use disorder, and mood and anxiety disorders. The meta-analysis, based on moderate-certainty evidence, found little to no difference in effectiveness between remote and in-person CBT (standardized mean difference [SMD] −0.02, 95% confidence interval −0.12 to 0.07), suggesting that both delivery methods are comparably effective across a range of disorders.

Implications for Practice: The findings indicate that therapist-guided remote CBT is nearly as effective as in-person CBT for treating a variety of mental health and somatic disorders. This supports the potential for remote CBT to significantly increase access to evidence-based care, especially in settings where in-person therapy is not feasible or is limited by geographic, economic, or social barriers. Future research should explore optimizing remote CBT delivery methods to further enhance accessibility and efficacy.

Reference: Zandieh, S. et al (2024). Therapist-guided remote versus in-person cognitive behavioural therapy: a systematic review and meta-analysis of randomized controlled trials. CMAJ, 196(10), E327-E340. [Link]

AI-Powered GPTs for Doctors: Evidence-Based Medicine & Clinical Decision Support Prompts

17 Dec, 2023 | 18:39h | UTC

Welcome to IntelliDoctor’s innovative collection of AI-Powered GPT Prompts, specifically designed for doctors seeking to enhance their clinical practice with evidence-based information. Utilizing the latest Natural Language Processing (NLP) technology, our prompts provide decisive support in various aspects of medicine, from identifying drug interactions to differential diagnoses, treatments, and more. Each prompt has been meticulously developed to provide accurate and up-to-date information, assisting healthcare professionals in quickly accessing crucial data for patient care decision-making. Important: access to these specialized prompts requires a GPT-4 subscription. With our tool, doctors can easily obtain relevant clinical insights, optimizing time and improving the quality of medical care. Explore our 6 specialized prompts and discover how artificial intelligence can transform your medical practice.

Disclaimer: These tools are intended for use by doctors and healthcare professionals only, and are not recommended for use by other individuals. GPT models can make errors, so please use them with extreme caution and always verify the information before applying it to patient care.

All Purpose: Comprehensive solutions for general medical inquiries.

Medications: Detailed information on various medications.

Interactions: Analysis of potential drug interactions.

Diseases: Information on a wide range of diseases.

Signs and Symptoms: Assistance in interpreting clinical signs and symptoms.

Differential Diagnosis: A tool to aid in differential diagnosis.

Perspective | AI predicted to play major role in cardiac CT and CV care in the coming decade

11 Aug, 2023 | 15:30h | UTC

AI Predicted to Play Major Role in Cardiac CT and CV Care in the Coming Decade – TCTMD

Perspective | Creation and adoption of large language models in medicine

9 Aug, 2023 | 15:38h | UTC

Creation and Adoption of Large Language Models in Medicine – JAMA (free for a limited period)

Commentary: Rethinking large language models in medicine – Stanford Medicine

Commentary on Twitter

21% of medical journal articles cited by GPT-4 were found to be fake; GPT-3.5 cited an estimated 98% fake articles. Narrower topics had more fake articles than broader topics. Despite its promise, ChatGPT is currently not a reliable source of medical data. https://t.co/DCTIkT1OkZ

— JAMA Network Open (@JAMANetworkOpen) August 9, 2023

Consensus Paper | Surgical video data use, structure, and exploration (for research in AI, quality improvement, and education)

9 Aug, 2023 | 15:20h | UTC

SAGES consensus recommendations on surgical video data use, structure, and exploration (for research in artificial intelligence, clinical quality improvement, and surgical education) – Surgical Endoscopy

Review | Artificial intelligence, machine learning and deep learning: Potential resources for the infection clinician

9 Aug, 2023 | 15:18h | UTC

Artificial intelligence, machine learning and deep learning: Potential resources for the infection clinician – Journal of Infection