New sepsis guidelines: An improvement, but not a solution
Volume 8, Issue 12 | August 30, 2023
In This Issue
Does cancer screening really lengthen lives?
Rethinking disease surveillance
A potential biomarker for early onset pre-eclampsia
The AI revolution will depend on integration among modalities
New and Noteworthy
New guidelines for managing sepsis are great. Early diagnosis would be better.
Last week, the CDC released new sepsis guidelines for hospitals. The guidelines focus on how hospitals should manage and report these cases, but they do not address the bigger underlying problem: Once the patient gets to hospital, it is often too late. Clinicians need to be able to accurately diagnose infections before they reach the point of sepsis.
Up to 20% of all deaths worldwide are attributed to “sepsis,” a vicious cascade of ever-more-life-threatening events that resists clear description as a unit. Official clinical definitions have been frequently updated - the most recent is “Sepsis-3,” published in February 2016 - but are still inadequate.
Part of the trouble is that sepsis often starts with a more-or-less routine local infection (e.g., a normal cold, a finger cut; usually bacterial, but can be fungal). The situation only becomes dangerous when the infection goes rogue, overcoming local constraints. If not rapidly treated, these infections can become resistant to frontline antibiotics, enter the bloodstream, and trigger sepsis: A deadly overreaction of the immune system that can lead to multiple organ failure, coma, and death (more than 50% of the time).
The case of Jim Henson (Muppets founder) is still all too typical. After three days of cold or flu symptoms, streptococcal pneumonia became established. Identified and treated earlier, penicillin would have been effective, but by the time he was rushed to hospital ICU, no antibiotics worked. He entered a coma after two hours, and never recovered consciousness. He was 53 years old.
Commentary: Guidelines for identifying and treating sepsis patients in the hospital are essential, but we need to use newer diagnostics early enough to catch the risk before the patient gets there. We can start by eliminating a viral cause (see our July 26th newsletter), and then adopting newer, rapidly evolving diagnostic tools to identify the specific pathogen and its resistance profile (traditional culture-based microbiology is frequently just too slow).
Cancer screening lengthens lives, right? (Maybe not.)
Cancer screening tests are routinely credited with saving lives by catching malignancies at stages early enough to allow treatment to succeed. Based on that success, we generally assume that cancer screening helps people live longer lives. A new meta-analysis in JAMA Internal Medicine calls that assumption into question.
The review looked at 18 randomized clinical trials, together involving 2.1 million people, and covered six common cancer screening tests: mammography, colonoscopy, sigmoidoscopy, fecal occult blood testing, CT (for lung cancer in smokers or former smokers), and prostate-specific antigen testing. All the trials included more than nine years of follow-up and reported all-cause mortality and estimated lifetime gained.
And you know what? The only one that did anything for longevity was sigmoidoscopy, which added . . . wait for it . . . a little more than three months of life.
Commentary: One of science’s great superpowers is its ability to tease out the truth, even when it contradicts something “everyone knows.” This study looks like a great example of that ability - on a population level, at least. Your personal mileage may vary.
DNA methylation may predict early onset pre-eclampsia
If you want to prevent early onset pre-eclampsia (a potentially fatal high-blood-pressure disorder that can occur during pregnancy), you’ve got to catch it really early. A study published this week in Nature Medicine looked at whether DNA methylation level in blood might be a good biomarker for the condition. It was: The test correctly identified 72% of patients who went on to develop early onset pre-eclampsia (with 80% specificity), enabling both risk stratification and pre-symptomatic diagnosis.
Commentary: While the result is exciting, the sample size was small, at only 498 patients. Here’s hoping it’s confirmed by future research.
Food for Thought
Rethinking the hierarchy of disease surveillance
STAT News ran a great article about Christian Happi at Nigeria’s Redeemer’s University and Pardis Sabeti at the Broad Institute in Cambridge, Mass. Together, the two created Sentinel, a disease-surveillance system in West Africa that “hopes to shift surveillance toward a more bottom-up approach, bringing low-cost diagnostics to community settings and rural hospitals, as well as empowering frontline workers to track the spread of disease in real time.” One of their success stories: an easy-to-use CRISPR-based diagnostic test called SHINE that costs only a few dollars and is, according to Happi, essentially equivalent to doing a PCR test on a sheet of paper.
Commentary: Some of the challenges that Sentinel faces are specific to developing nations. But many are the same here in the US. We’d love to see these techniques applied here, too.
If AI is to revolutionize diagnostics, integration will be key
Eric Topol and collaborators have been on the forefront of evaluating the clinical implications of AI. He recently published two highly relevant items: a Lancet Perspective on Digital Medicine, illustrating the need for, and promise of, multimodal data integration across images, genetics, liquid biopsies, and text EHRs in earlier cancer detection; and a Ground Truths blog discussion with John Halamka, a leading voice on IT in hospital medicine. Halamka tells an anecdote of one way AI can help: “It's Sunday at three in the morning. Mrs. Smith has just come in. She has a 3,000 page chart, 75 hospitalizations, and four OR visits. Her complaint tonight is: “I feel weak.” AI can extract key past data and answer questions from the chart that no human would be able to even find, never mind interpret.
Commentary: The critical point here is integration from multiple diagnostic and therapeutic modalities. For many patients and diseases, the answer to the question, “What is wrong?” will not be as simple as a single test with a single answer. AI promises to extract clinically relevant conclusions from more complex, more subtle, less invasive, and heterogeneous multi-modal data than is possible by unaided human evaluation alone.
However, the big questions for AI are
Will it work reliably in the highly consequential world of clinical diagnosis?
Will it provide a substantial enough improvement over steadily improving simpler techniques to justify its higher cost, novel error profile, and privacy challenges?
We will discuss specific impacts, their opportunities, and risks in future issues.
Texas joins the cool kids (when it comes to biomarker testing)
Tomorrow Texas will join 10 other states (AZ, AR, GA, IL, KY, LA, MD, MN, NM, OK) that have mandated health insurance coverage for biomarker testing, thanks to a law that was passed back in May. The details are important - the mandate covers tests that are “evidence-backed,” which was defined as:
FDA-approved
Covered by CMS, and
Included in national guidelines.
It is expected that most of this testing will be related to cancer diagnosis, most often aimed at discovering whether a patient has a cancer sub-type that responds to a specific drug. Several other states have similar bills pending.
Quick Hits
A July 31 commentary in the Lancet noted that while recent efforts to improve tuberculosis treatment worldwide (i.e., the 1/4/6x24 Campaign to cure tuberculosis quickly, launched in 2022) are laudable, diagnosis remains “the weakest link in the cascade of care.” The article estimates that 40% of people who develop active TB are never diagnosed (that’s more than 4.2 million people).
During a clinical safety analysis of their study on the use of AI assistance in reading mammograms, researchers at Sweden’s Lund University discovered that using AI resulted in the detection of 20% more cancers compared with standard screening, without affecting false positives. At the same time, the screen-reading workload for radiologists was reduced by 44%.