Sepsis Bundle Data That Jump, a Night-Shift Gap, and Twelve Proposed PDSA Cycles
A 520-bed academic medical center reports monthly 3-hour sepsis bundle compliance: January 68% (n=140), February 71% (n=132), March 64% (n=151), April 79% (n=147), May 66% (n=139), June 72% (n=145). Stratified data show the emergency department night shift (19:00–07:00) at 52% versus 78% on days. The sepsis coordinator wants to launch 12 PDSA cycles simultaneously across the house. The CMIO wants an EHR alert redesign released next month. Nurse managers argue the abstracted compliance number is wrong because it uses the time a lactate result was documented rather than when the specimen was drawn, and because triage-to-bed delays shift the start of the clock. Risk-adjusted sepsis mortality O/E is 0.91. The chief nursing officer has committed two hours per week of each nurse manager's time and wants "a plan by Friday." The quality committee meets monthly and has asked whether the April jump proves the new order set worked.
1. Is the April value of 79% a signal of improvement? What analysis would you complete before the team spends effort, and how should the night-shift gap be treated?
Read the worked answer
No. With a monthly denominator around 145 and a baseline near 70%, the 95% confidence interval on a monthly proportion is roughly plus or minus 7 percentage points. April's 79% overlaps the confidence intervals of every other month. Plot the data on a p-chart with variable control limits (subgroup size changes monthly) and apply standard rules - a shift of six or more points on one side of the centerline, a trend of six, or a point beyond three sigma. Six months oscillating between 64% and 79% is a stable system at about 70%: common cause. The quality committee should be told that the order set has not yet demonstrated an effect, and that reacting to individual monthly points would be tampering.
Stratification is where the real signal sits. A sustained 52% versus 78% split by shift is far too large to be noise; aggregate three months of shift-level data to get adequate denominators, compute intervals, and confirm the pattern holds.
Before any intervention, validate the measure. The nurse managers are describing a specification problem: abstraction timing versus specimen collection, and clock start at triage versus arrival. Audit 20 charts against source timestamps and the written specification, quantify the disagreement, and fix the definition. Intervening on data you know is unreliable wastes the CNO's two hours a week and destroys credibility when the number later moves for definitional reasons.
2. How do you sequence the 12 PDSA cycles and the EHR alert redesign, and which balancing measures do you track?
Read the worked answer
Build a driver diagram and a Pareto of which bundle element actually fails - antibiotic order-to-administration time, cultures before antibiotics, lactate collection, fluid documentation. If night-shift failures are dominated by pharmacy delivery time and physician order timing, an alert redesign does not touch the dominant failure mode and should wait.
Run two to three PDSA cycles at a time, sequenced against the dominant failure mode, on one unit and one shift, each with a stated aim and a two-week ramp. Queue the rest with named owners and dates so the coordinator's energy goes somewhere visible rather than into twelve simultaneous, unattributable changes. Keep a change log so you can tell which intervention moved the number.
Treat the alert as an intervention to be tested, not released. Measure current alert volume and override rate, pilot with a small user group, and compare against a control area. Alert fatigue is a realistic failure mode here.
Balancing measures: antibiotic days of therapy and appropriateness, ED length of stay, alert override rate and false-positive rate, nurse-reported workload, Clostridioides difficile rates, and evidence of documentation gaming such as edited timestamps. Add a process measure - time from triage to antibiotic order - so you see movement before the composite bundle rate does.
3. What do you put in front of the quality committee next month, and what review rule do you commit to?
Read the worked answer
One annotated control chart rather than a table of monthly percentages, with the centerline and limits, plus a plain-language statement of common versus special cause. Alongside it, the shift-stratified view, the measure-validation status and its expected completion date, and the list of PDSA cycles with owners.
State the review rule explicitly: act on special cause when it appears, and revisit the night-shift gap if it persists three consecutive months after validation. Commit to reporting the validated measure going forward and to annotating any definition change on the chart so the committee can see why a level moved.
Explicitly answer the question they asked: the April jump is within expected variation and does not yet demonstrate that the order set worked; here is what would constitute evidence. That answer protects the committee from over-reacting later and protects the improvement team from having a real effect dismissed as noise. Avoid ranking tables or red-amber-green status grids for a measure in common-cause variation - they invite exactly the single-point reaction you are trying to prevent.