Oncology Trainee Perceptions of the Prior Authorization Process: A National Survey

Large Language Models to Help Appeal Denied Radiotherapy Services

PURPOSE

Large language model (LLM) artificial intelligences may help physicians appeal insurer denials of prescribed medical services, a task that delays patient care and contributes to burnout. We evaluated LLM performance at this task for denials of radiotherapy services.

METHODS

We evaluated generative pretrained transformer 3.5 (GPT-3.5; OpenAI, San Francisco, CA), GPT-4, GPT-4 with internet search functionality (GPT-4web), and GPT-3.5ft. The latter was developed by fine-tuning GPT-3.5 via an OpenAI application programming interface with 53 examples of appeal letters written by radiation oncologists. Twenty test prompts with simulated patient histories were programmatically presented to the LLMs, and output appeal letters were scored by three blinded radiation oncologists for language representation, clinical detail inclusion, clinical reasoning validity, literature citations, and overall readiness for insurer submission.

RESULTS

Interobserver agreement between radiation oncologists' scores was moderate or better for all domains (Cohen's kappa coefficients: 0.41-0.91). GPT-3.5, GPT-4, and GPT-4web wrote letters that were on average linguistically clear, summarized provided clinical histories without confabulation, reasoned appropriately, and were scored useful to expedite the insurance appeal process. GPT-4 and GPT-4web letters demonstrated superior clinical reasoning and were readier for submission than GPT-3.5 letters (P < .001). Fine-tuning increased GPT-3.5ft confabulation and compromised performance compared with other LLMs across all domains (P < .001). All LLMs, including GPT-4web, were poor at supporting clinical assertions with existing, relevant, and appropriately cited primary literature.

CONCLUSION

When prompted appropriately, three commercially available LLMs drafted letters that physicians deemed would expedite appealing insurer denials of radiotherapy services. LLMs may decrease this task's clerical workload on providers. However, LLM performance worsened when fine-tuned with a task-specific, small training data set.

Perceptions of prior authorization burden and solutions

The prior authorization (PA) process consumes time and money on the part of patients, providers, and payers. While some research shows substantial possible savings in the PA process, identifying what different groups can do is not as well known. Thus, organizations have struggled to capture this opportunity. To understand different perspectives on PA burden and receptivity to possible changes in the PA process, we surveyed 1005 patients, 1010 provider employees, and 115 private payer employees. Patients reported the longest perceived wait times but indicated the highest perceived approval rates and lowest perceived burden. The relatively low burden for patients is because most do not have to engage in PA directly. Provider respondents reported spending time equivalent of more than 100 000 full-time registered nurses per year on prior authorization. Artificial intelligence (AI) represents a possible solution: 65% of private payer respondents reported that their organizations planned to incorporate AI into the process in the next 3 to 5 years. Intended adoption by provider respondents is much smaller (11%). Private payer respondents cited cybersecurity concerns and a lack of technical infrastructure as barriers; provider respondents cited lack of budget and limited trust in the technology.

Nearly All Peer-to-Peer Reviews for CT and MRI Prior Authorization Denials for Orthopedic Specialists Are Approved

In the event of prior authorization denial, physicians may request peer-to-peer review, which may delay treatment and increase administrative burden. The purpose of this study was to quantify the approval rate of peer-to-peer review and evaluate its efficiency in the context of advanced imaging use in an orthopedic practice. Patients at a single outpatient orthopedic clinic initially receiving an insurance denial for computed tomography or magnetic resonance imaging requiring peer-to-peer review from March to December 2022 were prospectively enrolled. Characteristics of the request, peer-to-peer review, and the reviewer and dates in the process were collected. If the study was approved after peer-to-peer review, the date of the imaging study and brief results were recorded. A total of 62 denials were included. One denial was approved prior to peer-to-peer review. Fifty-eight (of 61, 95.1%) reviews were approved, of which 51 (of 58, 87.9%) studies were completed by patients. Reviewers were always physicians (61 of 61, 100%), but of those whose specialty was known, none were orthopedic surgeons. Forty-four of 61 (72.1%) reviewers reported reviewing clinical notes in advance. The median number of days from visit to peer-to-peer review was 9.0 (interquartile range, 7.0-13.25). The median number of days from visit to imaging center appointment was 13.5 (interquartile range, 9.0-20.75) for approved studies. Of the 51 approved studies completed by patients, the results of 38 (74.5%) confirmed the suspected diagnosis. In an orthopedic specialty practice, almost all peer-to-peer reviews were approved, with the majority of the completed studies confirming the suspected diagnosis. Thus, patient care was delayed. Reform is crucial to improve the efficiency of the review process, especially in light of additional administrative and financial burden. [Orthopedics. 2024;47(3):141-146.].

Teaming in Graduate Medical Education: Ward Rounds and Beyond

Teamwork in graduate medical education (GME) is often hindered in clinical learning environments where discontinuity among residents, supervisors, and other health care professionals is typical. Teaming is a conceptual approach to teamwork in dynamic environments with constantly changing team members and goals. Teaming is built on principles of project management and team leadership, which together provide an attractive strategy for addressing teamwork challenges in GME. Indeed, teaming is now a requirement of the Accreditation Council for Graduate Medical Education Clinical Learning Environment Review program. However, many clinician-educators and leaders may be unfamiliar with teaming and how to integrate it into their GME programs. In this article, the teaming framework is described with a specific example of how it can be applied to improve hospital ward rounds, a common setting of teamwork breakdown. The goal of this article is to educate and encourage GME leaders as they learn new ways to implement teaming to improve patient care and education in their programs.

Active steps to reduce administrative spending associated with financial transactions in US health care

US health care administrative spending is approximately $1 trillion annually. A major operational area is the financial transactions ecosystem, which has approximately $200 billion in spending annually. Efficient financial transactions ecosystems from other industries and countries exhibit 2 features: immediate payment assurance and high use of automation throughout the process. The current system has an average transaction cost of $12 to $19 per claim across private payers and providers for more than 9 billion claims per year; each claim on average takes 4 to 6 weeks to process and pay. For simple claims, the transaction cost is $7 to $10 across private payers and providers; for complex claims, $35 to $40. Prior authorization on approximately 5000 codes has an average cost of $40 to $50 per submission for private payers and $20 to $30 for providers. Interventions aligned with a more efficient financial transactions ecosystem could reduce spending by $40 billion to $60 billion; approximately half is at the organizational level (scaling interventions being implemented by leading private payers and providers) and half at the industry level (adopting a centralized automated claims clearinghouse, standardizing medical policies for a subset of prior authorizations, and standardizing physician licensure for a national provider directory).

Impact of Prior Authorization on Patient Access to Cancer Care

Prior authorization (PA) is a type of utilization review that health insurers apply to control service delivery, payments, and reimbursements of health interventions. The original stated intent of PA was to ensure high-quality standards in treatment delivery while encouraging evidence-based and cost-effective therapeutic choices. However, as currently implemented in clinical practice, PA has been shown to affect the health workforce, adding administrative burden to authorize needed health interventions for patients and often requiring time-consuming peer-to-peer reviews to challenge initial denials. PA is presently required for a wide range of interventions, including supportive care medicines and other essential cancer care interventions. Patients who are denied coverage are commonly forced to receive second-choice options, including less effective or less tolerable options, or are exposed to financial toxicity because of substantial out-of-pocket expenditures, affecting patient-centric outcomes. The development of tools informed by national clinical guidelines to identify standard-of-care interventions for patients with specific cancer diagnoses and the implementation of evidence-based clinical pathways as part of quality improvement efforts of cancer centers have improved patient outcomes and may serve to establish new payment models for health insurers, thereby also reducing administrative burden and delays. The definition of a set of essential interventions and guidelines- or pathways-driven decisions could facilitate reimbursement decisions and thus reduce the need for PAs. Structural changes in how PA is applied and implemented, including a redefinition of its real need, are needed to optimize patient-centric outcomes and support high-quality care of patients with cancer.