Ranked 45th of 48: Reading a League Table Honestly
A statewide quality collaborative publishes 30-day all-cause readmission rates for 48 hospitals as a ranked league table. Your 180-bed hospital appears at 16.1% (412 index discharges) against a state rate of 13.9%, placing you 45th of 48. The newspaper ran a story naming the bottom five. Your payer mix includes 34% dual-eligible patients and 12% patients with limited English proficiency, and the hospital holds a safety-net designation. Internally, two service lines drive the count: joint replacement at 6.2% (n=118) and congestive heart failure at 24.8% (n=96). The board chair has asked for an explanation and a corrective action plan by next week. A physician leader has suggested simply excluding CHF patients from future internal reporting because 'those patients are just complex.'
1. Is the 45th-place ranking statistically meaningful? Show the reasoning.
Read the worked answer
Compute the interval before you compute the action plan.
For your hospital: p = 0.161, n = 412. Standard error = √(0.161 × 0.839 / 412) ≈ 0.0181. A 95% interval is 16.1% ± 3.5 points, i.e. 12.6% to 19.7%.
The state rate of 13.9% sits comfortably inside that interval. Your hospital is not statistically distinguishable from the state average.
Do the same for the state comparison: at n = 412, the 2-standard-deviation band around 13.9% runs roughly from 10.5% to 17.3%. You are inside the funnel.
Therefore the rank is noise. With 48 hospitals of varying size, the bottom five in any given year is partly a sorting of confidence intervals, not a sorting of quality. Demonstrate this by computing how many hospitals are statistically different from the state mean - typically far fewer than a ranked table implies - and by showing whether your rank moved over the prior three years. Rank volatility across years is the clearest proof that the ranking is unstable.
Second finding: the two service lines tell different stories. Joint replacement at 6.2% (n=118, 95% CI ≈ 1.8%–10.6%) is genuinely good. CHF at 24.8% (n=96, CI ≈ 16.2%–33.4%) is high but the interval is enormous. The aggregate is driven by case mix, not by uniform poor performance.
2. How do you report this to the board without either inducing panic or appearing to make excuses?
Read the worked answer
Lead with the measure's uncertainty, then move immediately to what is actionable. Boards accept 'the number is noisy' only if it is followed by 'and here is what we are doing anyway.'
Structure the report in four slides:
- The number and its interval. 'Our observed rate is 16.1%; the plausible range given our volume is 12.6%–19.7%; the state rate is inside that range.' Show the funnel plot, not the league table.
- Decomposition. Show the rate by service line and by payer/social-risk stratum. CHF accounts for the gap; joint replacement is a strength. Present the CHF interval alongside its point estimate so the board sees the precision problem.
- Case-mix context, stated carefully. Show the dual-eligible and LEP proportions against the collaborative average without claiming they explain the gap - the evidence on social risk adjustment is contested and the board should hear that it is contested.
- Actions with owners and dates. A transitional-care intervention for CHF (pharmacist reconciliation, 7-day follow-up, daily-weight remote monitoring), with pre-specified metrics and a 6-month readout.
Do not lead with 'the ranking is wrong.' Lead with the interval and the decomposition, and let the ranking's fragility be the conclusion the board draws. Align the internal corrective plan with the collaborative's specification so the next published number can be read as improvement.
3. The physician leader proposes excluding CHF patients from internal reporting. How do you respond, and what is the defensible alternative?
Read the worked answer
Decline the exclusion, and say plainly why: removing the patients with the worst outcomes from the denominator is measure manipulation. It would also be visible to the collaborative and the payer, and it converts a statistical argument (the number is noisy) into an integrity argument (we don't want to be measured on our hardest patients). The credibility lost there is much larger than the rate itself.
Defensible alternatives, in order of preference:
- Stratified reporting. Continue to report the all-cause, all-patient rate as the headline and add service-line and risk-stratum views. This is the standard approach and keeps the measure comparable.
- Risk adjustment per specification. Apply the steward's risk model exactly as written. If your hospital believes social risk should be added, argue for it through the collaborative's measure-review process with data - not by changing your own internal denominator.
- Condition-specific improvement. Reduce CHF readmissions through a defined care redesign; that is the only path that changes both the aggregate and the narrative.
- If the CHF population is genuinely different (e.g., a disproportionate share of patients discharged from SNF or with no caregiver), show that with a stratified comparison and ask the collaborative to add the stratum - a public, transparent ask.
The test to apply to any proposed exclusion: would we be comfortable publishing the reason for it alongside the number? If not, it is not a measurement decision.