260908 – Clinical Guideline Development Process

Guidelines Are Made by People:

Notes on Dr. Henry Masur’s Grand Rounds on Trustworthy Clinical Practice Guidelines

By Thomas Lee Abshier, ND | 14 November 2026

Watch the original lecture: Creating and Evaluating Trustworthy Clinical Practice Guidelines — Providence Portland Medical Center Grand Rounds, September 8, 2026

Why I am writing about this lecture

I attend the Grand Rounds lectures at Providence Portland Medical Center for my continuing medical education, and I am grateful that Providence makes them available. The September 8, 2026 session was a special one. It was the annual Dr. Richie Wernick Memorial Lectureship, honoring a physician who served Providence Portland for 37 years and who was remembered that morning as a true master of evidence-based medicine and a devoted teacher of residents, students, and faculty. The session also marked the 25th anniversary of the Providence Center for Healthcare Ethics.

The speaker was Dr. Henry Masur, Chief of the Critical Care Medicine Department at the NIH Clinical Center, a position he has held since 1989. He was introduced by Dr. David Gilbert of Providence, a past president of the Infectious Diseases Society of America and an editor of the Sanford Guide, which nearly every clinician who prescribes antibiotics has relied on at some point. Dr. Gilbert reminded the audience that Dr. Masur was the lead author of one of the first papers describing what we now know as AIDS, in the New England Journal of Medicine in 1981, and that he went on to lead the NIH guidelines for HIV treatment and for HIV-associated opportunistic infections.

Few people alive are better qualified to explain how clinical practice guidelines are actually made. What follows is my organized summary of the lecture, with my own commentary in clearly marked sections. I have paraphrased throughout; any errors in the summary are mine, not Dr. Masur’s. I strongly encourage every clinician and student to watch the lecture itself. The warmth, humor, and candor of the speaker do not come through fully on the page.

In brief. Guidelines are advice from panels of experts, not commands. The good ones are clear, referenced, easy to find, honest about how strong the evidence is, and independent of outside pressure. Whether a panel uses the formal GRADE system or structured expert consensus, evidence alone never writes the recommendation; human judgment always enters. The careful clinician reads the boxed recommendation at 3 a.m., and then, later, reads the reasoning behind it to see whether it fits his patient.

How Dr. Masur came to guidelines

Dr. Masur began with his own story, and it is worth retelling because it explains why guidelines exist at all.

As a fellow interested in tropical medicine, he was told by his division chief that the one organism worth studying was Pneumocystis carinii. At the time it was thought to be an intracellular protozoan; it later turned out to be an extracellular fungus. It caused disease only occasionally, in children with congenital immunodeficiencies, in some kidney transplant recipients, and in outbreaks among malnourished infants in post-war Eastern Europe. The entire world literature, he said, was about fifty articles, so he had read all of them.

One thing that was not widely known was the work of Dr. Walter Hughes at St. Jude Children’s Research Hospital. Pneumocystis was killing children with acute lymphocytic leukemia, and Dr. Hughes, reasoning that a drug active against the malaria protozoan might work, tested trimethoprim-sulfamethoxazole (TMP-SMX) first in animals and then in a placebo-controlled study in his patients. The children who received it did not get pneumocystis, and the infection was essentially eliminated from that population.

Then, in 1981, previously healthy young men began appearing with pneumocystis pneumonia. Infectious disease specialists who knew the leukemia work knew what to do, but most of the country did not. There was no internet, journal articles took a year or more to appear, and not everyone read them. Pneumocystis became the presenting illness in about half of patients with AIDS, and most survivors had repeat episodes. When Dr. Margaret Fischl’s simple randomized trial showed that TMP-SMX prevented pneumocystis in patients with AIDS, the question became how to get that message to every clinician quickly. Neither the professional societies nor NIH was writing guidelines at the time. So a group of experts convened and published recommendations through the CDC’s weekly Morbidity and Mortality Weekly Report, which reached clinicians far faster than a journal could. That was the beginning of what are now five sets of federal HIV guidelines.

Dr. Masur also noted that he spoke as a private citizen, not as a federal employee, and that he has no financial conflicts of interest.

Where clinicians get information, and why guidelines still matter

Most of us, Dr. Masur acknowledged, look things up in UpToDate or in an AI tool such as OpenEvidence, which he uses himself. These sources have real strengths: they are current because they are updated constantly, and they are written by reputable people. But each topic reflects the opinion of one or a few authors, without the geographic and practice-setting diversity of a large panel, and the method of selecting and weighing evidence is not laid out in the open. And, tellingly, when these sources want to validate what they say, they quote the guidelines. Each source has a role; the question is which one to go to first, second, and third.

What a clinician should expect from a guideline

Dr. Masur described the qualities that make a guideline useful to the person reading it.

Authoritative authors, with conflicts handled sensibly. He told of serving on an NIH panel on knee surgery from which anyone who had ever performed the operation was excluded, to avoid financial conflict. That is not the panel he would want to rely on. Conflict of interest is easier to define in theory than in practice. Does having run a trial create an intellectual conflict? Does money paid to one’s institution count? He believes one needs well-informed experts and must accept some nuance in managing conflicts.

Easy access, ratings, and depth. A reader should be able to find one item without scrolling through everything, see how strong each recommendation is and what kind of evidence it rests on, and then find the supporting detail: the patient populations studied, adverse effects, who did not respond, and the references.

Brevity. Guidelines are not textbook chapters. A chapter covers epidemiology, diagnosis, therapy, and prevention. A guideline should tell you how to practice; otherwise it becomes unbearably long to write and to read.

Clarity. “We recommend TMP-SMX as the drug of choice” is clear. Phrases such as “you should consider” or “you may use” are not; one can consider almost anything. He would rather a panel make a definite recommendation and say plainly whether it rests on evidence or on expert opinion. He also finds it unhelpful when a guideline reports that “many providers” do something; he would rather know what the majority of the panel, or the whole panel, does, even in the absence of evidence.

Ratings a busy reader can understand. The NIH panels he has led rate each recommendation as strong, moderate, or weak, and rate the evidence as coming from randomized trials, well-designed non-randomized studies, or expert opinion. He finds that intuitive. He contrasted it, gently and with self-deprecating humor, with the GRADE-based ratings in the Surviving Sepsis Campaign guidelines, which he co-authored. Categories such as “low” versus “very low” quality, or “not rated, but good practice,” required him to reread the definitions several times.

Something usable at 3 a.m. At night, a clinician mostly needs a table of preferred and alternative therapy, or a color-coded chart: green means do it, yellow means do it in certain circumstances, red means never. But good medicine, he said, means going back later to read the text and judge whether the data and reasoning apply to the patient in front of you, especially in a training program.

What makes a guideline trustworthy, and for whom it is written

When a guideline comes from the CDC, FDA, NIH, or the Infectious Diseases Society of America, Dr. Masur suggested asking four questions. Was it sponsored by a credible organization? Were the experts reasonably free of conflict of interest? Are the methods of analysis transparent? Can a quick reader understand the ratings?

He added a fifth consideration: who is the intended audience? Guidelines get into trouble when they try to serve too many masters. The federal HIV guidelines are written for clinicians only. They are advice from a panel of experts on how best to care for a patient. They are not mandates or regulations, and they do not require any hospital to adopt them as quality-assurance standards. Others may choose to use them, but that is their choice. When a guideline is also aimed at patients, insurers, quality programs, and regulators, it drifts from giving advice toward telling people what to do. He observed that during COVID, medicine sometimes got into trouble when it moved from science and clinical practice into public policy.

What it takes to build a guideline

Dr. Masur asked the audience to remember, when criticizing a guideline (which he said they have every right to do), how hard guidelines are to make.

  • Leadership. Someone has to care deeply about sepsis, endocarditis, or urinary tract infection and want to produce a great guideline.
  • Volunteers. Panel leaders and members are, in almost every case, unpaid. He spoke with evident pride about a profession whose members donate so much time, while acknowledging how hard it is to carve out that time alongside clinical duties and RVU expectations.
  • Administrative support and money. Someone must arrange meetings, keep minutes and a record of decisions, distribute literature searches, and publish the result on the web.
  • Independence from endorsing organizations. Sponsors may review and suggest, and panels should listen and compromise, but sponsors cannot dictate recommendations. He described a period, before the current administration, when CDC leadership required that all CDC-endorsed recommendations on topics such as syphilis in people with HIV be identical to other CDC guidance. Because there is no point in an independent panel that must copy another group’s conclusions, the panel asked the CDC not to endorse, and it did not.

To give a sense of scale, the adult HIV opportunistic infection guidelines cover about thirty organisms with thirty working groups, about 150 members, three full-time administrative staff, roughly 800 pages of text, and about 3,000 references, at a cost of roughly a million dollars a year. A guideline on a narrower topic might need only ten people and little support, but none of this is easy.

Are guidelines used? The opportunistic infection and antiretroviral guidelines receive roughly 500,000 to 600,000 page views a year, which his analytics team believes reflect real people, not bots. Hospitals adopt guideline recommendations as quality metrics, such as the preferred therapy for cryptococcal meningitis or TMP-SMX prophylaxis for every patient with a CD4 count below 200. Proving that guidelines improve outcomes is harder, but the evidence suggests they are having an impact.

Are guidelines evidence-based?

This was the question Dr. Gilbert had asked him to address, and Dr. Masur joked that he had managed to talk for thirty minutes before reaching it.

He began with David Sackett of McMaster University, one of the founders of evidence-based medicine, who described it as the conscientious, explicit, and judicious use of current best evidence in making decisions about the care of individual patients. Dr. Masur corrected a common misconception: Sackett did regard the randomized trial as the gold standard, but he never claimed it was the only acceptable evidence. Systematic reviews, controlled observational studies, case series, and sometimes expert opinion all have a place, because trials do not exist for everything. Dr. Masur believes even a strong recommendation can sometimes rest on evidence other than a randomized trial.

Two ways to weigh the evidence

Option A: structured expert consensus. This is the approach his NIH panels have used. A multidisciplinary panel is assembled, including not only infectious disease physicians but methodologists, biostatisticians, pulmonologists, maternal-fetal medicine specialists, and others as needed, all screened for financial conflicts. A librarian sends members quarterly PubMed searches. Rather than fixing a list of questions in advance, the panel is asked to write, for example, the section on managing pneumocystis; members identify the key issues and decide what to recommend. Some panels reach decisions by consensus, others by vote, but the process is transparent about how each decision was reached. Although these guidelines were removed from a national guideline compendium for not using GRADE, he does not believe that harmed their credibility.

Option B: GRADE (Grading of Recommendations, Assessment, Development and Evaluation). GRADE is the most widely adopted system in the world for rating the certainty of evidence and the strength of recommendations, used by the World Health Organization, Cochrane, the American College of Physicians, and most professional societies. Dr. Masur was generous in his praise. He called it a wonderful, systematic, objective way to learn to evaluate data, and said that being trained in it is extraordinarily valuable. GRADE frames questions in the PICO format (population, intervention, comparison, outcome) and evaluates both the quality of evidence and the strength of recommendations.

He also described its limitations as he has experienced them. It can overemphasize methodology, downgrading evidence for methodological issues that may matter less than clinical considerations. It concentrates on questions defined before the panel meets rather than those that arise along the way. It does not eliminate the need for expert interpretation (which, he noted, its developers never claimed it would). Above all, it is slow and expensive. Methodologists usually must be hired, and every study goes through a laborious process. A guideline may take five years to produce, during which the field moves on, and volunteers burn out.

Currency matters. A well-funded organization such as the National Comprehensive Cancer Network updates its guidelines annually. By his count, 27 of the Infectious Diseases Society’s 53 guidelines are at least five years old, and the Surviving Sepsis guidelines are updated every four to five years, in a field where much is changing.

His summary: whichever option a panel chooses, clear recommendations require evidence, and both require judgment.

Six cases: where evidence ends and judgment begins

The heart of the lecture was a series of examples showing that even recommendations that look clear-cut are often the product of long deliberation.

1. Some recommendations are easy. In Dr. Fischl’s trial, no patients on TMP-SMX developed pneumocystis, while about half of those without it did. In an early zidovudine (AZT) trial, the separation between the treated and placebo groups in life-threatening events was obvious. One still has to read the study, check the population, and confirm the endpoints were valid, but no elaborate apparatus is needed to reach the conclusion.

2. Different methods often reach the same answer. Dr. Masur has long wished someone would give one panel GRADE and another panel structured consensus, keep them apart, and compare the results. He suspects they would usually agree. As an example, the NIH antiretroviral guideline (which does not use GRADE) and the British HIV Association and IAS-USA guidelines (which use GRADE or a modified version) all reach the same recommendation on which antiretroviral regimen to start. GRADE has its place, he concluded, but many situations do not require it.

3. Data are rarely binary. The rule to start pneumocystis prophylaxis at a CD4 count below 200 seems precise. But the data show cases occurring mostly at low counts, with some at 300, 400, and even 500. A threshold of 600 would treat many people to prevent a few cases; a threshold of 50 would miss many. With disarming candor, he explained that after long debate among 150, 200, 250, 300, and 400, the panel chose 200 in part because clinicians needed a number they could remember. It is a sound threshold, but the data could have supported 150 or 300.

4. Sometimes the only data come from another era. The guidelines advise continuing prophylaxis until the CD4 count has been above 200 for three months. That rule comes from studies in the 1980s and 1990s. Today’s antiretroviral drugs suppress the virus quickly, and we now know that CD4 cells function much better when the viral load is undetectable, so a count of 100 today may be safer than it was then. Yet there is little new data beyond the COHERE cohort analysis, which suggested very low risk with counts of 100 to 200 and a suppressed viral load. The panel kept the older rule out of caution, which he considers reasonable because the downside of a few more months of prophylaxis is small, and added that clinicians “can consider” stopping when the count is 100 to 200 and the virus is undetectable, a phrasing he admitted he is not enthusiastic about.

5. Sometimes practice gets ahead of the data. Pneumocystis is classically diagnosed by seeing the organism in sputum or bronchoalveolar lavage fluid, which is well validated against histopathology. But many laboratories now offer only PCR. PCR is very sensitive, so a negative result effectively rules out pneumocystis; a positive result, however, proves the organism is present in the airway without proving it is causing the pneumonia rather than merely colonizing. The guideline notes that PCR has replaced staining, but does not say how far one can rely on a positive result. In the Q&A he said plainly that he thinks guidelines should not shirk this kind of responsibility: even without data, clinicians want to know what the experts actually do.

6. External validity: does the trial apply to my patient? External validity is the extent to which a study’s results can reasonably be applied to patients and settings beyond those actually studied. His example was cryptococcal meningitis. U.S. guidance calls for about two weeks of liposomal amphotericin B plus flucytosine, with close in-hospital monitoring for raised intracranial pressure, a regimen supported by solid data. The AMBITION trial, published in the New England Journal of Medicine in 2022 and conducted in five African countries, found that a single high dose of liposomal amphotericin with oral flucytosine and fluconazole was non-inferior to the control regimen. That is simpler and potentially transformative. But he pointed out that the trial was open-label and designed for non-inferiority, that the control arm used conventional amphotericin B deoxycholate rather than the liposomal regimen recommended in the United States, and that mortality in both arms was roughly twice what is typically seen in the U.S. Whether the single-dose approach should be adopted here, with different patients and different antiretroviral therapy, remains a matter of legitimate debate.

And a seventh factor: people. Guideline panels are groups of human beings. Group dynamics, including aggressive consensus-building, the persistence of a trial’s strongest advocates, and plain fatigue at the end of a long debate, sometimes shape outcomes. Some decisions, he said frankly, are influenced by personality rather than evidence alone.

Dr. Masur’s conclusions

He closed with four points:

  1. There are many reasonable processes for analyzing data, and he would welcome the views of evidence-based medicine experts on guidelines that do not use GRADE.
  2. Evidence-based decisions are essential, but every system requires judgment.
  3. Practical, human factors, including personalities, affect what guidelines say.
  4. Physicians should be proud to belong to a profession whose members donate their time so that those who know the data best can advise those who know it less well.

He thanked Providence for its tradition of developing academic leaders and practitioners, and noted with a smile that the NIH is always looking for good fellowship applicants in critical care and critical care infectious disease.

From the question period

When should a topic get a formal national guideline rather than a local recommendation? There is no objective formula, he said. It is a qualitative judgment about the clinical and economic importance of a problem, such as the overuse of antibiotics for asymptomatic bacteriuria, or how often a condition such as Staphylococcus aureus bacteremia is mismanaged. Societies have limited volunteers and money, and in practice a guideline usually happens because a committed leader persuades the organization it is needed.

Will AI replace some of the labor of guideline-making? He was candid that he did not know, and that the best programs always ask questions a speaker cannot answer. For now, he said, he would rather know what an experienced expert such as Dr. Gilbert actually uses for Acinetobacter. He uses OpenEvidence himself, but has seen AI tools cite an older edition of a guideline instead of the current one, or lose the nuance of a recommendation by taking it out of context, so he still reads the articles and seeks out people with practical experience. His advice: buyer beware. He also described a safeguard his panels use, the “sniff test.” Before finalizing a recommendation, members are asked whether anyone actually does this. If no one raises a hand, the recommendation is reconsidered, because a panel should not recommend something nobody does. The resident who asked the question replied that, while she shares some skepticism, the models are improving quickly, and clinicians will need to think about how they remain useful. Dr. Masur added that every patient is to some extent unique, in anatomy, drug penetration, and much else, and that he hopes there will always be a role for the clinician who knows the patient and the data better than any machine.

What about expert “guidance documents,” such as the IDSA guidance on antimicrobial-resistant infections? A questioner praised such documents for giving focused expert opinion, drawing on molecular, in vitro, case-report, and other evidence, on rare and complex problems. Dr. Masur agreed that both GRADE and consensus approaches have room for this. If he faced coccidioidal meningitis, he would want to know what the handful of leading mycology experts in the world actually do, which is better than inventing something himself. Expert opinion based on weak evidence or clinical experience is valuable, he said, as long as it is labeled as such.

My reflections

This section is my own commentary, not part of the lecture.

What I valued most in this lecture was its honesty. Dr. Masur has spent more than four decades at the center of guideline-making in HIV, critical care, and infectious disease, and he chose to show the audience the seams: the threshold chosen partly for memorability, the rule carried forward from an earlier era of therapy, the diagnostic test that outran its validation, the trial whose results may not travel well, the influence of strong personalities. None of this diminishes guidelines. It does the opposite. A guideline whose makers are open about where judgment entered is easier to trust, because the reader knows what kind of claim is being made.

Three lessons from the lecture seem to me worth carrying into any field of medicine, including my own practice as a naturopathic physician.

First, a guideline is advice, and the clinician still owns the decision. Dr. Masur’s insistence that the NIH guidelines are addressed to clinicians, not written as mandates, honors the relationship between a doctor and his patient. Sackett’s definition of evidence-based medicine ends with the words “individual patients” for a reason. The best summary of the evidence is still a summary of other people’s patients.

Second, read beneath the box. The boxed recommendation is the right tool at 3 a.m. The supporting text is the right tool the next week. That is where one learns whether a recommendation rests on a randomized trial, a cohort from 1990, or a panel’s best judgment, and whether the people studied resemble the person in the exam room.

Third, external validity applies to every study we read. The AMBITION example is a model of how to read any trial: who was enrolled, what the comparison really was, how sick the patients were, and how different the setting is from one’s own. That habit of mind protects patients whether the question is antifungal therapy, a new drug, or a dietary or lifestyle intervention.

I also appreciated his closing note about volunteerism. Thousands of hours of unpaid expert work stand behind the guidelines most of us consult every week, and that generosity deserves recognition.

I am grateful to Dr. Masur for a lecture that was learned, practical, and humane, to Dr. David Gilbert for bringing him to Portland, and to Providence Portland Medical Center for continuing a Grand Rounds tradition that Dr. Wernick helped to build. Please watch the original: Creating and Evaluating Trustworthy Clinical Practice Guidelines.

A reader’s checklist, drawn from the lecture

When you open a clinical practice guideline, ask:

  • Who sponsored it, and is the panel independent of its sponsors?
  • Were the experts screened for conflicts of interest, and how were conflicts handled?
  • Is the method of finding and weighing evidence described openly?
  • Is each recommendation rated for strength and for the kind of evidence behind it, in terms you can understand quickly?
  • Is the language definite (“we recommend”) or vague (“consider”)?
  • How old is it, and has the field changed since it was written?
  • Do the studies behind the recommendation involve patients and settings like yours?
  • Where the evidence is thin, does the panel say so and still tell you what experts actually do?

Editorial notes

These notes clarify points from the lecture for readers who want to follow up. They are mine, not the speaker’s.

  • The Fischl trial. In the recording, the trial numbers are stated slightly differently at different points. The published study (Fischl, Dickinson, and La Voie, JAMA 1988) randomized 60 patients with Kaposi sarcoma: none of the 30 receiving TMP-SMX developed pneumocystis pneumonia, compared with 16 of the 30 who received no prophylaxis. Adverse reactions occurred in half of the treated patients.
  • Naming. The organism that infects humans is now called Pneumocystis jirovecii; Pneumocystis carinii is the historical name and is now reserved for the rat organism. Clinicians still commonly say “PCP” for pneumocystis pneumonia.
  • NIH rating scheme. The NIH HIV guidelines pair a letter for the strength of a recommendation (A, B, or C) with a Roman numeral for the quality of evidence (I for randomized trials, II for well-designed non-randomized or observational studies, III for expert opinion).
  • AMBITION-cm. Jarvis and colleagues, New England Journal of Medicine 2022, enrolled 844 adults at sites in Botswana, Malawi, South Africa, Uganda, and Zimbabwe. A single 10 mg/kg dose of liposomal amphotericin B given with 14 days of flucytosine and fluconazole was non-inferior to one week of amphotericin B deoxycholate plus flucytosine, with fewer serious adverse events. The WHO subsequently adopted the single-dose regimen. The trial investigators have argued in print that the regimen is also applicable in high-income settings; Dr. Masur’s caution reflects the other side of that legitimate debate.
  • The 1989 guideline. The early pneumocystis prophylaxis recommendations convened through NIH and published by the U.S. Public Health Service appeared in MMWR on June 16, 1989.

Sources

Leave a Reply

Your email address will not be published. Required fields are marked *