Quality improvement methods in child health — branching viva

Opening (candidate)

My first job is to reframe this from a poster-and-reaudit into a genuine quality improvement project. I would use the Model for Improvement: define a specific, measurable, time-bound aim; choose a balanced set of outcome, process and balancing measures; establish a reliable baseline run chart; then test change ideas with small PDSA cycles before scaling anything. I would secure a senior sponsor and protected time, build a multidisciplinary team including nursing, pharmacy and a family partner, and disaggregate the data by subgroup so an equity gap does not hide behind an improving average. [1] [8]

Branch A — The aim

Examiner: Write me the aim statement. [1]

Candidate: Reduce the proportion of infants on this unit receiving antibiotics beyond 48 hours without a proven infection from 40% to 20% within 6 months. That is population-specific, measurable, has a target and a deadline. 'Educate everyone and reaudit in a year' fails every one of those tests — it has no target, no timeline and no test of change. [1] [8]

Branch B — Measures

Examiner: What will you measure, and what is the trap? [8]

Candidate: A balanced set. The outcome measure is the proportion of infants on antibiotics beyond 48 hours without proven infection. The process measure is the proportion of courses with a documented 48-hour stop-or-review decision. The balancing measure is the re-start rate — infants with a subsequently proven infection whose antibiotics were stopped at 48 hours — plus length of stay, because the whole point of stewardship is to reduce unnecessary exposure without under-treating real sepsis. The trap is measuring only the process (did we do the checklist?) and never the outcome (did we help?), and forgetting the balancing measure so that under-treatment hides behind a falling exposure rate. [1] [13]

Branch C — PDSA

Examiner: How do you actually test a change? [2]

Candidate: I would generate change ideas from a driver diagram — for example, a 48-hour stop-review checklist, a pharmacist-led review, and a decision-support tool in the electronic record. Then I test one small: Plan the test on one team for one shift, defining exactly what data I will collect and what I predict will happen. Do it. Study the result against the prediction and gather staff feedback. Act — adopt, adapt or abandon. Only if it works do I scale to more shifts, then the whole unit. The point of small testing is that if it fails, it fails cheaply and safely, and I learn the local fit before imposing anything. A big-bang rollout of an untested change is the opposite of this. [2] [8]

Branch D — Reading the data

Examiner: Show me you can read a run chart. [9]

Candidate: I plot the measure over time and draw the median of the baseline phase. Special-cause variation breaks a rule: a shift of six or more consecutive points on one side of the median, a trend of five or more all rising or falling, a run of five or more identical values off the median, or an astronomical outlier. Common-cause variation is the noise within those limits — I must not chase it, because tweaking a stable process makes it worse. A Shewhart control chart adds calculated upper and lower control limits for tighter detection, but the run chart is the everyday tool. The systematic review by Thor established how statistical process control is applied across healthcare to tell signal from noise. [9]

Branch E — Collaboratives and evidence

Examiner: Single-centre improvement is slow. What is the collaborative model, and how strong is the evidence? [5]

Candidate: A collaborative — the Breakthrough Series model — has many teams learn together over months using shared data and methods to spread a known improvement faster than any single site could. The Vermont Oxford Network NIC/Q projects are the paediatric exemplar: Horbar showed multi-centre neonatal collaboratives can move outcomes, and Dukhovny more recently showed collaboratives can reduce unnecessary antibiotic exposure in newborns. But I must be an honest appraiser: the methods literature is cautious. Taylor's systematic review found PDSA is widely used but often poorly applied, and before-after designs are weak because regression to the mean and secular trends confound. So I would want time-series data with a clear baseline, a balancing measure, and disaggregated subgroups before claiming success, and I would report with SQUIRE so others can appraise and adopt it. [5] [8] [10] [13]

Close

Define a specific aim, build a balanced measure set with an outcome and a balancing measure, establish a baseline run chart, test change ideas with small PDSA cycles, scale only what the data show works, sustain with ownership and ongoing data, and report with SQUIRE. If I can join or seed a collaborative, I spread the improvement faster — but I read the evidence critically, because a before-after success can be regression to the mean, and an improving average can hide an equity gap. [1] [8] [10]