Weekend hospital admissions associated with increased risk of death
"Don’t get admitted to the hospital on a weekend."
Listen, and you might hear that rueful advice muttered under the breath of a frustrated physician unable to get a prompt Saturday consultation or procedure, the requested specialist being at another hospital, or maybe her son’s soccer game.
The weekend effect -- patients admitted on a weekend are more likely to die or suffer other adverse outcomes, because of decreased staffing or other resources -- has long been postulated, but is it real? A new large meta-analysis suggests yes, and that it's a global phenomenon.
Authors analyzed 97 studies including over 51 million patients around the world, in whom the day of the week of admission could be identified. The patients hospitalized on a weekend had a 19% relative increased risk for death, compared to those hospitalized on weekdays.
Other statistical trends in the data appeared to support the weekend mortality effect:
Decreased staffing levels and longer time to interventions on the weekends were associated with a higher mortality for weekend patients.
In centers that did not vary staffing levels on weekends, there was no increased mortality seen on weekends.
It was impossible to rule out the possibility that weekend patients are somehow sicker or more likely to die (e.g., trauma victims) than patients admitted on weekdays, but this seemed unlikely. The analysis combined data from countries around the world, increasing its statistical power but also potentially limiting its conclusions’ strength in the U.S.
The findings are in line with prior analyses of the risk of weekend hospital admissions, but at double the magnitude:
A 2010 meta-analysis suggested that patients admitted to the ICU on a weekend (but not overnight during the week) were more likely to die (odds ratio 1.08).
A 2016 analysis suggested that “off-hour” (nights and weekends) admissions for 28 different conditions increased risk of death, with an odds ratio of 1.1 overall.
Authors argue that even if the observed excess risk of death is roughly halved to 10%, this could represent 25,000 preventable deaths each year, which would make weekend hospitalization the #8 cause of death in the U.S.
This would be remarkable if true, and is not as incredible as the figure of 440,000 U.S. deaths per year due to medical errors, trumpeted by other hospital safety gurus to NPR in 2013. Johns Hopkins researchers gave NPR their relatively sedate estimate of 250,000 annual deaths by medical error, which would still make medical errors the #3 cause of death in the U.S. behind heart disease and cancer.
Reality check: the key phrase to note in the methodologies in most of the scare-mongering medical error death studies -- ever since the Institute of Medicine's 1999 figure of 100,000 deaths per year served as the kick-off event to the media's and researchers' bidding-up of estimates of death by doctor -- is "associated with."
Oversimplifying somewhat, their methodologies search for events like dialysis or infections occurring in the hospital. The techniques are standardized so that large data sets can be analyzed systematically by relatively cheap data analysts or computer programs. No physician adjudicates the tallied deaths, necessarily.
If an adverse event is found, and the patient turns out to have died, DING! That can be recorded as a death associated with a medical error. The fact that the patient was an 88 year old with metastatic lung cancer whose death was vanishingly unlikely to have been caused by lisinopril being given two hours late, won't be considered in most such analyses.
When the study gets published, the news media report the deaths were "caused by" or "due to" the medical errors, not "associated with." That fundamental error goes uncorrected by the researchers (that would make for a boring interview). Voila! A public health crisis is produced, along with advertising revenue for the media outlet, and recognition for the researchers.
A researcher basing her career on alternate methodologies that conclude preventable hospital deaths aren't really that high, or that 98% of the resulting deaths occur in people with a staggering disease burden in their actuarial last 6 months of life, will have a short CV, and no calls from NPR.
I did these reviews using the IHI's global trigger tool for all deaths for six months at a hospital where I worked. The IHI tool flagged multiple deaths as associated with adverse events. Curious, I looked in the charts. The patients -- virtually every one -- were in their 80s or 90s and admitted in septic shock, and the "errors" had nothing to do with their deaths.
The sicker a patient is (and thus more likely to die), the more "errors" in care that these tools will find, and the less likely any "error" will be genuinely contributory to his death. Many, if not most such "medical errors" are epiphenomena of severe illness and the complexities of care during hospitalization, particularly critical care.
The "weekend effect" meta-analysis is more interesting, powerful and concerning, because it doesn't share this methodologic bias toward a large effect or spuriously impute causation. The day of the week shouldn't have any effect on one's hospital survival, let alone a ~10-20% relative increase in mortality risk.
On the other more-reassuring hand, in large data sets of U.S. patients, analyses of those admitted on the weekend for heart failure, stroke, or out-of-hospital cardiac arrest found no increased mortality overall compared to patients hospitalized on the weekdays for those conditions.
If there is a weekend effect, it will be expensive to erase. Physician and ancillary staffing on the weekends is reduced for an obvious reason: health workers are people, with families and friends and lives outside of work, which they need to nurture to remain resilient enough to consistently provide good care for hospitalized patients.
For U.S. hospitals to pay the tab and do the necessary browbeating and cultural re-engineering to entice full-strength staff back into the hospital on weekends would probably require a high-profile catastrophic event, or more conclusive data than have yet been presented.