Saturday, July 13, 2024

Can large language models handle the complexity of family medicine?

Should family physicians be excited or apprehensive about the potential applications of artificial intelligence (AI) and large language models (LLMs) in primary care? An article by Dr. Richard Young and colleagues in the Journal of the American Board of Family Medicine recently made the case for both. Observing that primary care is a “complex adaptive system,” the authors suggested that AI “will likely work when its tasks are limited in scope, have clean data that are mostly linear and deterministic, and fit well into existing workflows.” On the other hand, AI may struggle to incorporate contextual and relational factors, process noisy and inaccurate data, or document vague symptoms that do not indicate a single disease condition.

In an editorial on chatbots and LLMs in the June 2024 issue of American Family Physician, Dr. Aaron Saguil discussed how family medicine practices are turning to LLMs to “help decrease administrative burden and combat burnout.” These tools can already compose visit notes, remotely monitor patients with interactive chats, and draft replies to patient portal messages.

In the future, LLMs may be integrated into electronic health records to provide real-time clinical decision support, suggesting “diagnostic possibilities, recommended ancillary evaluations, and possible treatment strategies.” To minimize the risks of LLMs propagating biased data, generating misinformation, or usurping the family physician’s role on health care teams, Dr. Saguil advised being actively involved in their implementation:

The best defense against AI risks becoming realities is conscientious physicians guiding the development and implementation of LLMs into clinical care settings, pointing out what LLMs can do and what they cannot. In family medicine, no LLM can yet address a complex patient in a unique sociocultural situation with overlapping comorbidities and health states from the vantage point of a longitudinal relationship.

A related FPM article by Dr. Steven Waldren, chief medical informatics officer at the American Academy of Family Physicians, explored other uses of LLMs in primary care, such as rewriting medical or legal forms for patients with lower health literacy or native languages other than English; summarizing information from a medical record, guideline, or research articles; drafting referral letters, prior authorization requests, and insurance appeals; and populating clinical registries. Dr. Waldren recommended three safeguards when using AI in medical practice: using LLMs only “when the physician or other user is able to easily verify the accuracy of the AI output”; not entering protected health or private organizational information in open online LLMs such as ChatGPT; and for now, using LLMs only in low-risk (nonclinical) situations. Echoing Dr. Saguil, Dr. Waldren called on family physicians to “weigh in on the design, development, and deployment of AI in medicine to ensure it is more helpful than harmful to patients, primary care physicians, and practices.”


This post first appeared on the AFP Community Blog.

Monday, July 1, 2024

Clinical documentation and health inequities

Poor treatment by a clinician leads to suspicion of the motives of the health care system, increasing one’s reluctance to seek necessary care and the risk of a having an uncomfortable health care interaction in the future. For many patients—particularly those who identify with one or more racial, sexual, gender, and religious minority groups—this vicious cycle often leads to worse health outcomes.

A Lown Right Care article in the June 2024 issue of American Family Physician addresses the consequences of stigmatizing clinical documentation. In the case scenario, a patient feels stereotyped by his new primary care physician, who incorrectly describes him as African American (he self-identifies as Dominican Hispanic) and noncompliant for not filling a prescription or seeing a subspecialist he could not afford. Drs. Alan Roth and Andy Lazris explain how this language could negatively affect encounters with other clinicians:

Patients who identify as Black are more likely than those who identify as White to have comments using negative words or connotations in their history and physical documentation and may be subject to systemic bias in physicians' perceptions of their credibility.… The stigmatizing language used to describe patients in medical records can influence other clinicians and physicians-in-training in their attitudes toward the patient and their medication-prescribing behavior. This is an important and potentially damaging pathway by which bias can be propagated from one clinician to another.

In a previous AFP editorial, Drs. Megan Healy and Khameer Kidia presented several strategies to reduce bias and avoid stigmatizing language in medical records and other clinical communications. These strategies include not using labels, not blaming patients for their conditions, and not beginning presentations with race, ethnicity, language, socioeconomic status, or other social identifiers that may activate unconscious bias.

In 2021, a controversial JAMA social media post questioned the existence of structural racism and made the incredible claim that “no physician is racist.” This unfortunate episode contrasted with AFP’s active approach to advancing health equity, including sharing and promoting antiracist practices. Nonetheless, the medical profession has a long way to go to align its behaviors with its ideals. In a first-person narrative published in the New England Journal of Medicine, a Black health equity researcher visiting the emergency department receives callous treatment from two nurses and is mocked by a physician for suggesting that her area of expertise had something to do with the indifferent care she received. Making excuses for the health professionals involved (e.g., the emergency department was overcrowded, the nurses were tired and harried, the doctor was just having a bad day) amounts to being part of the problem. Instead, family physicians and other primary care clinicians can choose to be part of the solution to eliminating health inequities in kidney transplant, infertility care, skin conditions, asthma and lung diseases, cardiovascular disease, end-of-life care, and preventive care, among others.


This post first appeared on the AFP Community Blog.

Tuesday, June 18, 2024

Contemplating colorectal cancer screening choices

Invitations to patients eligible for colorectal cancer screening need not be limited to office visits. A 2018 systematic review and meta-analysis of 73 randomized clinical trials of U.S.-based interventions found that mailing fecal tests more than doubled the likelihood that targeted patients received colorectal cancer screening. A recent trial in Clinical Gastroenterology and Hepatology compared screening completion in community health center patients randomly offered one of three options in an outreach mailing: colonoscopy referral only, fecal immunochemical test (FIT) only, or an active choice of colonoscopy or FIT. At 6 months, 12.8% of patients in the active choice arm had completed screening compared with 11.3% in the FIT-only arm and 5.6% in the colonoscopy-only arm.

As I discussed in a previous post, one problem with screening colonoscopy is that it is frequently repeated at shorter intervals than the recommended 10 years without a good reason. In fact, 10 years may not be long enough. A cohort study in JAMA Oncology suggested that a 15-year rescreening interval may be appropriate for average risk patients without a family history of colorectal cancer and with negative findings on their first screening colonoscopy. Using Swedish register-based data sources, researchers showed that individuals meeting these two criteria between 1990 and 2016 had 15-year standardized colorectal cancer incidence and mortality ratios that were lower than the 10-year cumulative risks in a matched control group.

A U.S. cross-sectional study that relied on data from the national Gastrointestinal Quality Improvement Consortium registry found that most patients with an episode of acute diverticulitis were not more likely to have colorectal cancer diagnosed on a follow-up colonoscopy than asymptomatic patients undergoing a screening colonoscopy. Only those with complicated diverticulitis (i.e., diverticulitis with perforation or abscess) were significantly more likely to have colorectal cancer (adjusted odds ratio = 3.57; 95% CI, 1.59 to 8.01).

Regarding the multitarget stool DNA (MT-sDNA) test for colorectal cancer screening, a study of 500 randomly selected patients in a Midwest health system found that about 1 in 5 had the test ordered inappropriately. The most common reasons for inappropriate ordering were having had a colonoscopy within the previous 10 years, having a family history of colorectal cancer, reporting symptoms suggestive of possible colorectal cancer, being younger than 45 years old, and having a previous diagnosis of adenomatous polyps.

A multitarget stool RNA (MT-sRNA) test with performance characteristics similar to those of the MT-sDNA test was approved in May 2024 by the U.S. Food and Drug Administration. Both tests are more sensitive for colorectal cancer and advanced adenomas than FIT but have lower specificity, resulting in higher false positive rates and more diagnostic colonoscopies. Of note, a research letter demonstrated that lowering the threshold for a positive FIT produced similar sensitivities and specificity as the MT-sRNA test, even without the RNA component of the test.


This post first appeared on the AFP Community Blog.

Thursday, June 13, 2024

Canadian mammography kerfuffle echoes U.S. screening debate, 15 years later

In November 2009, I was a medical officer with the U.S. Preventive Services Task Force (USPSTF) at the Agency for Healthcare Research and Quality. The health reform legislation that would eventually become law as the Affordable Care Act (or "Obamacare" after then-president Obama) was being fiercely debated in Congress. Politicians who opposed expanding health insurance to the poor, self-employed, and employees of small businesses launched all sorts of spurious charges, the worst being that the law would establish "death panels" that would determine whether elderly patients with chronic medical conditions would be allowed to live or die. Into this political maelstrom stepped the USPSTF, releasing an ill-timed update that recommended against routinely screening women aged 40-49 for breast cancer. Contrary to popular belief, this language didn't mean they were advising all women in this age group NOT to be screened; instead, they were empowering patients to make this decision individually in consultation with their physicians, based on their preferences and values.

The current Task Force looked at essentially the same evidence and come to a different conclusion: start screening everybody at age 40, never mind the potential harms. In a Medscape commentary, I explained why I don't believe this change is justified. However, it put a great deal of pressure on our neighbors to the north and the Canadian Task Force on Preventive Health Care to reassess its recommendations and come to a similar conclusion. If Americans are getting screened for breast cancer in their 40s, why shouldn't Canadians too?

The Canadian Task Force released its draft recommendations two weeks ago. In short, they echo the USPSTF's recommendations in 2009 and 2016 and maintain that breast cancer screening should be a "personal choice," particularly for females younger than age 50. "For women aged 40 to 49, based on the current evidence (trials, observational studies, modelling and a review on values and preferences), we suggest not to systematically screen with mammography [emphasis mine]. Because individual values and preferences may differ, those who want to be screened after being informed of the benefits and harms should be offered screening every 2 to 3 years." The supporting data, much of it derived from systematic reviews, is extensive and compelling, including discussion tools for women in various age groups.

Unlike the U.S. 15 years ago, Canada isn't in the midst of a major health reform debate - they've already had an equitable universal health care system for decades, thank you very much. But that hasn't stopped one health official from trying to score cheap political points. The Canadian Minister of Health, Mark Holland, a lifelong politician without any health professions training, has forcefully objected to the CTFPHC's draft recommendations and ordered an unprecedented "external review" that will, no doubt, include conflicted experts such as radiologists who have obvious financial incentives to perform as many mammograms on asymptomatic women as possible.

Here is Holland being interviewed on an Ottawa news channel. When the anchor asks him what he would do if the external review confirms the Task Force's recommendations, he dodges and weaves and avoids answering.

As a former USPSTF member once said, "we can follow the evidence wherever it leads," or we can start with a preordained conclusion and cherry-pick data that supports what we already know to be true. Canada, Mark Holland, and the CTFPHC would do well not to mimic the most expensive, inequitable health care system in the world and associate higher percentages of women receiving mammography in their 40s with better preventive care "quality."

Tuesday, June 4, 2024

Why medications for AUD should be as popular as GLP-1 agonists for obesity

A recent commentary in the Journal of General Internal Medicine compared anti-obesity medications with medications for alcohol use disorder (AUD). Both chronic conditions are “characterized by behavioral patterns that pose risks of adverse health consequences” and “subject to societal stigma including … the idea that they reflect a lack of personal willpower.” While prescriptions for costly glucagon-like-peptide-1 (GLP-1) agonists for obesity such as semaglutide have skyrocketed, use of less expensive drugs for AUD remains low. The authors suggested that public perceptions that the latter are ineffective or unnecessary, implicit biases of clinicians, and delayed health benefits of alcohol cessation compared to weight loss contribute to the differences in use.

Given the magnitude of the problem, which worsened during the pandemic, an American College of Physicians policy brief advocated “training, payment, and delivery system policies to enable physicians and other qualified health professionals to screen, diagnose, and treat excessive alcohol use and AUD.” Recognizing patients with excessive alcohol use remains a challenge despite a U.S. Preventive Task Force recommendation to routinely screen adults, including pregnant patients, for unhealthy alcohol use and provide brief behavioral counseling interventions to persons engaged in risky or hazardous drinking.

A systematic review in JAMA’s Rational Clinical Examination series concluded that the Alcohol Use Disorders Identification Test (AUDIT) is most the useful tool for identifying AUD in adults and postpartum individuals, while the abbreviated AUDIT-Consumption (AUDIT-C) best identifies excessive alcohol use in children aged 9 to 18 years and older adults. Other studies have found that a single question alcohol screen (“How many times in the past year have you had five (men)/four (women) or more drinks in a day?”) is comparable to the AUDIT-C in detecting unhealthy alcohol use and current AUD in primary care.

Articles in the January 2024 and May 2024 issues of American Family Physician discussed FDA-approved and off-label pharmacotherapies for adults with AUD. According to an Agency for Healthcare Research and Quality review, oral naltrexone, acamprosate, and topiramate have the strongest evidence for reducing alcohol consumption, while injectable naltrexone, baclofen, and gabapentin have weaker supporting evidence. Of the two first-line treatments approved by the FDA for AUD, acamprosate is contraindicated in patients with a creatinine clearance of ≤ 30 mL/min, while naltrexone should be avoided in patients who use opioids or have advanced liver disease. Disulfiram is not more effective than placebo in reducing alcohol consumption.


This post first appeared on the AFP Community Blog.

Friday, May 31, 2024

Measles outbreaks and the future of public health

In early March, a one year-old boy living in a shelter in Chicago was diagnosed with measles. Since measles is a highly contagious virus, more than 2000 shelter residents, many without confirmed histories of measles vaccination, were considered to be exposed. Over the next 3 days, the Chicago Department of Public Health (CDPH) swung into action, vaccinating 882 residents and confirming previous vaccination status in several hundred more, resulting in an estimated measles, mumps, and rubella (MMR) vaccine coverage of 93%. Nonetheless, in the next two months, 57 more cases of measles had been diagnosed among people residing in or having contact with residents of the shelter. A dynamic disease simulation model from the Centers for Disease Control and Prevention (CDC) estimated that in the absence of mass vaccination and case finding by CDPH, there was a 69% probability that this single infection would have led to more than 100 new measles cases.

The outlook for measles in the next decade in the U.S. is mixed, at best. Driven by the COVID-19 pandemic, vaccine skepticism and anti-vaccine sentiment has led to increases in religious and philosophical exemptions, causing the national kindergarten vaccination rate to fall to 93% in 2021-2022. Each percentage point decline increases the risk of a measles outbreak, as illustrated by the CDC's National Center for Immunization and Respiratory Diseases, with 93% MMR vaccination coverage associated with a 36% risk that a single infectious child could create an outbreak at a school with 100 children.

Unfortunately, decreasing childhood vaccination rates were not the only negative effect of the pandemic. The compelling PBS mini-series "The Invisible Shield" juxtaposes the historical successes of public health officials in controlling diseases and saving lives with the angry and occasionally violent pushback they received for ordering business closures and masking mandates to slow the spread of COVID-19. A recent Health Affairs article documented 112 judicial decisions from 2020-2023 that constrained public health's legal powers. As a result, the health commissioner for Columbus, Ohio was unable to close a daycare center where a measles case was reported in 2021, and a county in Michigan came close to shutting down the entire county health department over a dispute about masks.

An even bigger threat to public health is awaiting a decision from the Supreme Court:

On January 17, 2024, the US Supreme Court heard oral arguments in 2 combined cases, Loper Bright Enterprises v Raimondo and Relentless, Inc v Department of Commerce, that will determine the fate of Chevron deference, a bedrock principle of administrative law that obligates courts to defer to an agency’s reasonable interpretations of ambiguous statutes. ... Agencies like the FDA and CMS have, for decades, regulated medicine and public health in reliance of these principles.

If the Supreme Court decides that this deferential statutory framework is unconstitutional, it would effectively open every federal health regulation that does not derive directly from specific legislative language (the vast majority of thousands and thousands of regulations) to a lawsuit, and it would put the burden on a dysfunctional Congress to account for "all possible regulatory scenarios and unforeseen circumstances" in writing unambiguous legislation. This would be a disaster.

I wonder how federal and state governments would respond if public health actions became necessary to protect people from H5N1 ("bird flu"), which so far has only been confirmed in 3 dairy workers this year but has likely infected many more? So far H5N1 has only caused mild symptoms in humans, but the fact that this virus has spilled over from cows to people and little surveillance is being done in to track its spread in either population is worrisome.