Every hospital in America is one line. It crosses five CMS questions that all describe the same twenty minutes at the end of a stay. Watch where the lines are on top of each other — and where they aren't.
Today's newsletter argues that the thing you noticed in a hallway years ago — small, specific, unactionable at the time — was the raw material, and that the habit you actually lost was writing it down. Here is what that looks like at national scale, in a dataset that has been quietly recording since 2006.
CMS mails a survey to a random sample of every hospital's discharged patients. Five of its questions describe the same handoff. They differ only in how closely they look.
On the left, the bundle is a rope. 85.5% of patients say yes, they were given information about what to do at home — and the spread between the best hospital in the country and the worst is 30 points, with over half of all hospitals packed inside 10 points of the top. That measure is on Care Compare. It is part of the star rating. It cannot distinguish between almost any two American hospitals, because almost all of them answer it the same way.
On the right, the rope has come apart. 45.1% — a coin flip — and 1.7× the spread. The question is Q15: before giving you any new medicine, how often did hospital staff describe possible side effects in a way you could understand?
Somebody sat in a room and wrote in a way you could understand into a federal survey instrument. That clause is the entire difference between a measure that sees nothing and a measure that sees a forty-point range. It is also, exactly, the kind of sentence a clinician mutters in a hallway and doesn't write down.
Switch the colour to by readmission. If the sharper question saw something real, the red lines — hospitals that readmit more heart-failure patients than expected — should separate out on the right-hand axis. They don't, much. They separate slightly more on the left.
Against 30-day heart-failure excess readmission, the blunt composite carries r = −0.22. The sharp side-effects question carries −0.09. Backwards from what the specificity story predicts — and both are weak enough (4.9% and 0.8% of variance) that neither is a lever.
Worse for the sharp question: its signal will not settle. Exactly 0.00 among hospitals with fewer than 100 heart-failure discharges; −0.14 in the 100–299 band; −0.03 above 700. Raise the volume floor cumulatively and it wanders — −0.096, −0.071, −0.034, then back to −0.101 on the last 149 hospitals. That is the shape of a small-denominator artifact, not a finding. The composite stays between −0.19 and −0.27 at every threshold.
Two separate claims live in today's essay: that the specific thing you noticed is real, and that it matters. This chart supports the first and declines to support the second. Both are worth knowing before you build on it.
The scarce part was never the measurement. It was knowing which twenty words were worth measuring.