Night lighting, falls rates and a premature rollout
You are the quality manager of a 120-bed residential aged care home with three 40-bed units. In January, Unit A installed a new night-lighting system; Unit B made no change; Unit C installed lighting in corridors only. Over the previous quarter (October–December), falls per 1,000 occupied bed-days were 5.1 in A, 5.4 in B and 5.2 in C. In the January–March quarter they were 6.4 in A, 5.3 in B and 5.0 in C. Night-round audit adherence was 62% in A, 88% in B and 79% in C. Unit A recorded three fractures after the change, compared with one in the previous quarter. In March alone, Unit A's rate was 5.9. The chief executive says: 'Falls are coming down in March, so let's roll out the lighting to all units.'
1. Assess whether the data support the claim that the lighting reduced falls. Include the March figure.
Read the worked answer
No. Unit A's rate rose from 5.1 to 6.4 per 1,000 bed-days, a 25% increase, and fractures rose from one to three. The March figure of 5.9 is still above the 5.1 baseline and is a single month with a small denominator-one or two falls can move it substantially. Units B and C were roughly stable (5.4→5.3; 5.2→5.0), which argues against a facility-wide seasonal explanation but does not prove harm from lighting. Night-round adherence was much lower in A (62%) than in B (88%) or C (79%), so the change is confounded by a known falls-prevention behaviour. The claim that 'falls are coming down in March' cherry-picks the latest point and ignores the baseline.
2. Identify three confounders or data-quality issues and explain how each could change the interpretation.
Read the worked answer
Confounders and data issues:
- Adherence: lower night-round compliance in A could explain the rise; without adjustment, lighting and staffing behaviour are entangled.
- Seasonality and case mix: activity, acuity and dementia mix may differ by unit and quarter.
- Reporting or denominator: a new installation may increase reporting, or occupied bed-days may change; small numbers make rates unstable.
- Implementation disruption: installation itself may create glare, shadows or temporary corridor clutter.
Each could change the interpretation: if adherence fell in A, the lighting may be irrelevant; if reporting rose, the true rate may be unchanged.
3. What would you recommend before any rollout? Give a specific analysis and a decision rule.
Read the worked answer
Before rollout: keep A, B and C as comparison groups; audit night-round adherence and staffing by unit; obtain 12 months of data; analyse falls as counts with occupied bed-days as the denominator, using a rate ratio or Poisson model with confidence intervals. Adjust for adherence and unit. Predefine success: a statistically credible reduction-for example, a rate ratio below 0.8 with a 95% confidence interval excluding 1-sustained for two quarters, with no increase in fractures. Decision rule: if the confidence interval includes 1 or adherence remains below 80%, do not roll out; fix adherence first. Report the March figure as one data point, not a trend.