The AI Scribe Dilemma: Efficiency Expectations Versus Financial Practicality
Healthcare systems worldwide are adopting AI ambient scribes that promise to eliminate the busywork of documentation. Is this documentation relief translating into time efficiency? A recent report from the Peterson Health Technology Institute (PHTI) critically assesses the technology’s impact and raises troubling issues regarding its financial and operational benefits.
AI Scribes: Provider Burnout Resolved, But at What Expense?
The PHTI report, an evaluation of health systems from across the nation, cites the aim of alleviating provider burnout as one of the most crucial reasons driving adoption of AI scribes. In addressing the burnout problem, the focus has been on careful editing, and the attempt to capture other ROI has been minimal. The report suggests that this imbalance will lead health systems to regard AI scribes as common tools that need to justify their real costs.
Financial Implications: Two Sides of a Coin
As with any new innovation, AI scheduling assistants have potentiated both cost saving and associated expenses within current health systems. The authors express concern that AI-enabled scribes may lead to over-documentation, also known as “enhanced documentation quality,” which may increase reimbursements. While this could mitigate some costs, better AI interfaces could exacerbate problems for patients and the broader market. Alternatively, the software cost might capture the value of efficiency, decreased administrative workload, lower attrition of clinicians, and net savings through improved clinical systems management.
A Proposal for Evaluation Frameworks and Longitudinal Assessment
The PHTI has called out a need for comprehensive research and uniform approaches to benchmark the effectiveness of AI scribes with respect to various metrics. The report points out the need to capture the implications of efficiency, expense, and patient care in the long term. The report is accompanied by tables that capture AI scribes, their corresponding companies, and the available conducted studies—a highlight for time-stressed readers is found in the “Looking Ahead” section on page 28.
Integration of Telehealth Services to Pharmaceutical Firms Analyzed
Pharma companies like Pfizer and Eli Lilly have also come under scrutiny for their use of telehealth platforms to administer services to patients. There are allegations from senators that the firms might be infringing upon the federal anti-kickback statute. Professors Ateev Mehrotra, Olivier Wouters, and Erin Fuse Brown published in the New England Journal of Medicine, discussing how these mergers may increase access but also consider care, which leads to unnecessary prescribing. The authors discuss the difficulties that come with evaluating the impact of these “black boxy” arrangements that lack adequate open disclosure.
Devices that Aid and AI Features in Breakthrough Diagnostics
STAT has also expanded its list of tracked experimental medical devices with ”breakthrough” accomplishments from the Food and Drug Administration (FDA). These devices are considered for the designation when they provide superior treatment or diagnosis compared to the current standard, which also serves to draw interest from investors. Recently added to the list is a generative AI-powered chatbot, PathChat DX, developed by Modella AI for use by pathologists for case diagnosing. Others included are a breast cancer imaging device, a breast cancer detection device and two tests for Alzheimer’s blood tests.
AI in Drug Development: Apheris and Google DeepMind
With Apheris recently putting together its consortium of drugmakers, the Life Sciences Data Company announced they will be sponsoring OpenFold3, an open-sourced protein folding model hosted by Columbia University that is in many ways like AlphaFold. This partnership looks to further develop AI-based drug discovery algorithms. In another geburt, AstraZeneca presented a study during a conference this week employing the use of an AI model created by Altis Labs to evaluate the CT images and estimate the survival chances of lung cancer patients. The model is also being validated on breast cancer and colorectal cancer scans, which can assist drug makers in designing clinical trials. Google too advanced Ggemma, a collection of large language models for drug development pretrained by Google DeepMind.
Navigating the Complexities of AI in Healthcare
The AI in Medicine section of the report also underlines one of the main challenges the industry is facing, which is the sophistication associated with the innovation of AI technologies. Although the potential for improving efficiency, accessibility, and treatment outcomes remains plentiful, it’s accompanied by concerns around cost, ethical dimensions, and policy on mitigation. The delicate balance between applying AI in healthcare technologies and upholding the safety and well-being of patients is a deep concern embracing the industry.
