ICU and You · Journal Club · Number 101
163 young adults with severe sepsis in a PICU, and the trap waiting in their mortality rate
Zakutansky SK, McCaffery H, Viglianti EM, Carlton EF. Characteristics and outcomes of young adult patients with severe sepsis admitted to pediatric intensive care units versus medical/surgical intensive care units. Journal of Intensive Care Medicine 2023;38(3):290–298.
Read the paperNot open access — unlike most of what gets reviewed here, this one is behind a publisher paywall, so you will need your hospital library. Nine pages, three tables and two figures. Table 1 is the paper: read the two columns of comorbidities side by side and you will have worked out the problem before the authors raise it.
Because they are the news. Not the formal, ceremonial version of reading — the daily kind. The journals are our news pages, our opinion columns, our form guide and our sports results. People who would not dream of going a week without knowing what happened in the world will go a year without knowing what happened in their own specialty, and then wonder why the registrar knows something they do not.
It does not have to be a session. You do not need an hour, a highlighter or a plan. Read the abstracts. Skim the contents page of one journal while the kettle boils. Open the one paper in ten that looks interesting and abandon it at paragraph three if it is not. Nobody is marking this.
What matters is staying curious rather than staying current — nobody is current, and the people who claim to be are reading less than they say. What you are actually maintaining is the habit of noticing: that a question is open, that a practice has shifted, that somebody has published the thing you were arguing about last month.
And then do the part that makes it stick. Mention it on a ward round. Argue about it with somebody who disagrees. Ask what the study could not have shown. A paper read alone is half a paper.
Publish or perish is the line we all know. The quieter one is this: absorb it, or lose your edge.
A nineteen-year-old with relapsed leukaemia arrives in septic shock. She has been looked after by the same children’s hospital since she was six. Does it matter whether tonight she goes to the paediatric intensive care unit that knows her, or to an adult one that does not?
Journal Club 100 asked where the sixteen-year-old goes. This asks the harder half: once they get there, does it change anything.
Truven MarketScan is a very large American claims database — employer-based insurance plus Medicaid from a selection of states. The authors took every hospitalisation of a patient aged 0 to 26 between 2010 and 2015, pulled out the young adults aged 18 to 26 with severe sepsis by diagnosis code, and then split them by the type of provider who billed for their critical care: paediatric critical care medicine, or adult medical and surgical critical care.
18,900
young adult admissions with severe sepsis1,491
of them treated by an ICU physician — fewer than one in twelve163 / 952
final cohort, paediatric unit against adult unitThat middle number deserves a pause. Of nearly nineteen thousand young adults hospitalised with severe sepsis, fewer than eight per cent were billed as having been treated by an intensivist at all. Some of that is real — plenty of severe sepsis is managed on a ward. A good deal of it is the method: identifying intensive care by who submitted the bill undercounts it, and the authors say so. Either way, the paper is describing a narrow and unusual slice.
The two groups are not the same patients, and the gap is not subtle.
Paediatric unit · 163
Median age 19. A complex chronic condition in 95.7%. Malignancy in 70.6%, transplant 10.4%, a genetic condition 9.2%. More often male, more often on Medicaid.
Adult unit · 952
Median age 22. A complex chronic condition in 85.3%. Malignancy in 42.5%, transplant 3.9%, a genetic condition 2.8%.
The same in both
Technology dependence — 34.4% against 33.1%. Gastrostomies, tracheostomies and shunts are distributed evenly. It is the diagnoses that differ, not the hardware.
Read the first column again. A nineteen-year-old in a paediatric intensive care unit with severe sepsis is, seven times in ten, a cancer patient. These are not young adults who happened to end up in the wrong building. They are the children of that hospital, grown up and still coming back.
Unadjusted mortality was 13.5% in the paediatric units and 8.2% in the adult ones. Adjusted for age, sex, insurance status and comorbidity, admission to an adult unit carried roughly half the odds of dying: odds ratio 0.50, 95% confidence interval 0.29 to 0.89. The second model, using a different comorbidity index, gives the same answer. Length of stay was identical — a median of nine days in both. Paediatric admissions cost about forty per cent more — a median of $48,484 against $35,050.
You can already see how that will be quoted.
It rests on twenty-two deaths. That is the entire paediatric numerator — 22 of 163.
And there is no adjustment for severity of illness on admission. No SOFA, no PRISM, no APACHE, nothing. The models know the patient’s age, sex, insurance and a count of chronic conditions. They do not know how sick she was when she arrived, which is the single largest determinant of whether she lives.
Nor is there any information on advance directives — and in a population where seven in ten have cancer, limitations of treatment are not a footnote.
There is a deeper problem, and it is the one I would raise first at a journal club. Severe sepsis was identified by diagnosis code — but paediatric and adult medicine do not define sepsis the same way. Adult practice uses Sepsis-3 and the SOFA score; paediatric practice at the time used systemic inflammatory response criteria. The authors put it plainly: given that difference, comparisons between the two groups “may be unequal”. Two cohorts, two definitions, one odds ratio.
No validated comorbidity index exists for young adults, so the authors used two — a paediatric one and an adult one — and ran the models separately. Then they looked at what the two indices actually caught, and found the overlap was almost nothing: of all the comorbidities that distinguished the paediatric group, only malignancy appeared in both.
Their conclusion is a sentence most papers would have buried: neither index is likely adequate to capture the comorbidity profile of this population. Which is to say the thing they adjusted for is the thing they could not measure.
Their stated verdict on the mortality gap is correspondingly careful: the variation is “likely due to significant differences in the patient populations”. Not the unit. The patients.
The first is in the paper, in a sensitivity analysis almost nobody will quote. The authors widened the lens beyond intensive care and compared every young adult looked after by paediatric medical and surgical providers against every one looked after by adult providers — 138 against 9,049, ICU and ward together. Mortality was 6.5% against 5.8%, with a p-value of 0.71. No difference at all. Widen the denominator and the signal evaporates, which is what case-mix confounding looks like.
The second arrived four years later and is the study this question actually needed. Using a national database of nearly 20,000 patients aged 14 to 22 across 856 hospitals, Badesch and colleagues exploited the fact that the age-18 boundary allocates near-identical patients to different units almost arbitrarily — a regression discontinuity design, which is about as close to randomisation as observational data gets. They found the same practice differences you would expect, including adrenaline as the first-line vasopressor in paediatric units and noradrenaline in adult ones, and no significant effect of paediatric unit admission on mortality at any bandwidth around the cutoff, adjusted or not. Their own conclusion was that there is now enough equipoise to randomise.
One caveat, and it matters: that second study is so far a conference abstract only — presented to the American Thoracic Society in 2026, and not yet a peer-reviewed paper. I include it because it is the right design and it points the same way as the sensitivity analysis above, not because an abstract settles anything. Watch for the full publication.
This paper excluded pregnancy-related hospitalisations, on the reasonable-sounding grounds that maternal sepsis is usually treated as a separate entity. The podcast running alongside this one is about a specialty that does the same thing in the other direction.
So in a single fortnight we have the largest study of young-adult sepsis removing the pregnant patient from the denominator, and an intensivist arguing that the great mistake we make with her is treating her as a separate entity. Sepsis is among the leading causes of maternal death in both the United Kingdom and Australia. She is a young adult with sepsis. She is simply never in the cohort.
None of which makes it a bad paper. It is a careful study that describes something real — that young adults with severe sepsis are scattered across two systems, that the ones in children’s units are overwhelmingly childhood-disease survivors, and that nobody has a comorbidity index that fits them. It simply cannot answer the question its headline number appears to answer, and it is honest enough to say so. Papers like that are the ones worth taking to a journal club, because the argument is not about the result. It is about what the result is allowed to mean.
Can you actually get the paediatric notes?
A twenty-year-old with a childhood-onset condition arrives in your adult unit at two in the morning. Somewhere there are fifteen years of records describing exactly how this person deteriorates, what their baseline saturation is, which antibiotic they reacted to, and what their family was told last time.
Have you ever got hold of them inside an admission? What did it take — a phone call, a fax, a shared record, a parent with a folder? Or is the honest answer that you have never tried, because there is no route and everyone knows it? That answer is as useful as the other one, and I suspect it is commoner. Replies from regional and rural units especially welcome, where the children’s hospital is several hundred kilometres away.
The button opens a reply straight to me. If your device blocks it, write to icuandyou@icloud.com instead.
Read something that changed what you do?
This column runs on what people bring to it. Tell me the paper and, more to the point, tell me what it changed — a journal club is not a reading list, it is an argument about whether something ought to alter practice. Papers that failed to convince you are just as welcome, and a good deal rarer in print.
If you would rather write the appraisal yourself, I will publish it here with your name on it — or without, entirely as you prefer. Trainees especially: if you have already appraised something for a teaching session, most of the work is done.
How you are credited is entirely your call — by name, by initials, or not at all — whether you write the piece or I do, and whether I quote you or simply say who sent something in; nothing of yours goes up with your name on it unless you have said yes.
The button opens a reply straight to me. If your device blocks it, write to icuandyou@icloud.com instead.