One of my medical school classmates recently posted a Washington Post article on Doximity, a “social media” site for physicians. The article caught my attention. I think it’s worthy of some comment. The author of the Post article is a brilliant young physician, Dhruv Khullar; his subject was a recent paper in JAMA by Denson, Jensen, Saag, et al. titled “Association Between End-of-Rotation Resident Transition in Care and Mortality Among Hospitalized Patients.”
The primary finding from this large, retrospective, multicenter observational study was, “end-of-rotation house staff transition in care was associated with significantly higher in-hospital mortality.” The headline for the Post article was “When a new team takes over your care at the hospital, it can be a precarious time.” The change in titles itself is a pretty dangerous transition.
After 14 paragraphs describing the intellectual and emotional uncertainties inherent in teaching-hospital house staff transitions, Khullar finally wrote, “Patients who remain hospitalized during transitions may simply be sicker.” An editorial raising this issue, and the data in the JAMA publication from a restricted analysis that showed markedly lower mortality after attempting to adjust for the degree of illness, at last got some attention in the text.
Here’s my point. There is a critical feature that links all three pieces: the research paper, the editorial, and the Post article; they all acknowledge that the basic premise of the work is flawed, but all three then go on to propose solutions to the undefined “transition problem” anyway.
Why do I suggest that the analysis is flawed? I’m not a statistician, but after decades of reviewing medical papers, I know enough to realize that this retrospective study was not a statistical problem of tossing fair dice. Patients who are or are not candidates for discharge from hospital on any given day differ in many ways. The study population was not randomized, and as the researchers commented, “the differences observed in patient populations might represent a direct consequence of clinical decisions made because of an upcoming transition rather than confounding. That is, if clinicians try to discharge as many patients as possible prior to transitioning off service, but have more difficulty discharging complex and long-stay patients, the average severity or complexity of patients exposed to transitions in care could be increased.” [Italics mine]
An ad hoc attempt to adjust for these issues with an alternative approach to analysis demonstrated that the findings were highly dependent on unmeasured factors. As the researchers noted, “The increased 30-day and 90-day mortality risks observed in the main analysis suggest that the delayed discharge of these complicated patients following transition could be detrimental… The alternative analysis, however, did not demonstrate these findings, [italics mine] which could be related to the noted differences between analyses.”
So, I would like to pose a really interesting question. Why did the authors (and the reviewers and editors) of all three pieces perceive a problem and then propose solutions to it in spite of the facts that the “adjusted” data suggested that the effect of house staff transition was not nearly as great as initially suspected, and that both the authors and the editorialists realized that, “Patients who remain hospitalized during a change in personnel on the inpatient service are likely different than (sic) those who are discharged”?
For the answer, I suggest we ask Nobel Prize winner, Daniel Kahneman. In his book, Thinking, Fast and Slow, Kahneman pointed out, “even compelling causal statistics will not change long-held beliefs…” In the transition-of-care publications authors, reviewers, and editors cling to the long-held belief that doctors can somehow overpower the poor prognosis of old age, frailty, and severe disease by improving the mechanics of healthcare delivery.
In our society, this belief encourages us to think about developing better hand-off forms for interns and residents instead of thinking about how we might really care for patients.
All’s not lost! These observational data do suggest a testable hypothesis. With an upcoming transition in care, rather than pushing early discharge for patients who are basically going to do well, a more productive approach might be to focus effort on recognizing the sicker patients and making appropriate arrangements for them, for example, earlier initiation of specialty consultations, social services for discharge planning, or consideration of hospice or palliative care transfers. In other words, one could randomize two large groups to either usual care (perhaps with the undeniably attractive hand-off forms) or pre-transition intervention starting a week or so before transitions and focused on appropriate discharge planning for sicker patients.
But then again, shouldn’t that be happening now?
P.S. The bird pictured is a blue-footed booby that Katherine photographed in the Galápagos Islands. I included it to make the point that blue-footed boobies do, indeed, have blue feet. This is directly related to the clinical concept that sicker patients do, indeed, do worse.