Readmission Dashboard Rankings for the Board

The chart treats noisy observed rates as if they were precise measurements and then imposes a rank order that the data cannot support. With 22 discharges, 4 readmissions gives a rate of 18.2% but a very wide confidence interval, roughly 5% to 40%; that hospital is not meaningfully distinguishable from one at 10%. Ranking seven hospitals with overlapping intervals creates false contrasts and invites the board to infer real quality differences that may be random fluctuation. The measure itself is flawed: including transfers and planned readmissions inflates rates inconsistently, and incomplete out-of-network ascertainment biases three hospitals downward or upward depending on their patient mix. No risk adjustment means that hospitals serving older or sicker populations are compared unfairly. Red and green imply statistical signal and moral judgment at once, and 'name and shame' is likely to encourage gaming, such as avoiding readmission coding or diverting patients, rather than improving care. The chart also hides uncertainty from the board, which is a reporting failure in its own right.

Replace the ranked bar chart with a funnel plot of risk-standardized readmission rates against expected readmissions or volume, with 95% and 99.8% control limits. Show each hospital's point estimate with a confidence interval, and display the system average as a reference line. Suppress cells with fewer than 11 events or use three-year rolling rates for small hospitals and services so that the smallest hospital's 4/22 is not presented as a stable annual rate. Use a statistical process control chart, such as a u-chart, for monthly monitoring within each hospital, and annotate known data-completeness problems. Report both crude and risk-standardized rates, with the readmission algorithm excluding planned readmissions and transfers. Flag a hospital only when its risk-standardized rate falls outside the control limits after adjustment and after confirming data quality; do not flag on rank order or on a single year. Pre-specify these rules in the dashboard specification so that the display is not redesigned after the results are known.

Explain that the old chart answered 'who has the highest number?' but the board needs to know 'is any hospital's performance different from what we would expect, given its patients and volume?' A small hospital with 22 discharges can swing from 5% to 40% by chance; ranking it against a 150-discharge hospital is not a fair comparison. The new display shows each hospital as a dot with a range, and a hospital is flagged only when its result is outside the range we would expect from normal variation. That does not mean the others are perfect; it means the data do not yet show a real difference. Also explain that the measure itself is being cleaned up, because transfers and planned readmissions were inflating rates and out-of-network follow-up was incomplete. Emphasize that the purpose is improvement, not blame: flagged hospitals will get a review of discharge processes and follow-up, and the board will see trends over time rather than a single-year league table. Offer to show the funnel plot and the control chart side by side, and invite questions about uncertainty rather than suppressing it.