Hand Hygiene: 94% Reported Versus What the Product Data Show
A 400-bed hospital reports hand hygiene compliance of 94% from about 120 direct observations per month. Most observations are conducted by unit managers on day shift. Night shift accounts for roughly 40% of patient days but only 8% of observations. Alcohol-based hand rub purchases equal 22 mL per patient day, against an internal benchmark of 40 mL. MRSA bacteremia has increased from 4 to 9 cases year over year. The board dashboard displays the 94% as green. The infection preventionist is uneasy but has been told the observation data are 'the standard.' The CMO asks whether the hospital has a hand hygiene problem.
1. Why might the 94% be misleading, and what data would you triangulate?
Read the worked answer
The 94% measures what observers recorded, not necessarily what happens when no one is watching.
Plausible biases:
- Hawthorne effect: staff recognize the observer and the observation moment.
- Observer bias: unit managers observe their own staff; social desirability and hierarchy suppress noncompliance recording.
- Sampling bias: 8% of observations on night shift, which covers 40% of patient days. Night shift may differ in staffing, urgency, and glove use.
- Denominator and definition drift: are all five moments observed, or only before patient contact?
Triangulate:
- Hand rub consumption per patient day (22 vs. 40 mL) - a structural signal that actual use is lower than reported.
- Soap and paper towel usage where relevant.
- Unit-level and shift-level observation rates.
- MRSA bacteremia trend and cluster investigation.
- Short covert audits by trained observers external to the unit, with inter-rater reliability checks.
No single measure is truth. Converging evidence from consumption, outcomes, and independent observation is what supports a conclusion.
2. How would you redesign the measurement system without destroying staff trust?
Read the worked answer
Move from a compliance score to a measurement system with transparency about its limits.
- Stratify sampling by shift, unit, and role so night shift is represented proportionally.
- Use trained observers who do not manage the units they observe, and validate inter-rater agreement.
- Add consumption data as a continuous, low-burden structural measure; trend it monthly.
- Add covert or automated audits sparingly and disclose that they occur - secrecy about methods breeds distrust, but unannounced sampling is legitimate.
- Report uncertainty: show sample size and confidence intervals so a 94% based on 120 observations is not treated as precise.
- Separate measurement from punishment. Tell staff that observation data will be used for system improvement and coaching, not individual discipline, unless there is reckless behavior.
Pilot the new method on two units and compare with the old method before scaling. Expect the number to fall; prepare leadership for that so a more honest 78% is not treated as a failure of the staff.
3. What should you tell the board, and what actions should follow?
Read the worked answer
Tell the board the dashboard is not measuring what it claims to measure, and explain why in plain language.
- The 94% comes from a sample that overrepresents day shift and is collected by unit managers. It is a compliance audit with known social-desirability bias.
- Consumption at 22 mL per patient day is roughly half the internal benchmark, and MRSA bacteremia has risen. Those two signals do not prove noncompliance, but they are inconsistent with 94%.
- Recommendation: change the dashboard to show a measurement-quality flag and multiple indicators - observation rate with sample size, consumption, and MRSA bacteremia - rather than a single green number.
Actions: redesign sampling, validate observers, run a night-shift-focused improvement test, and investigate the MRSA increase for common sources. Re-baseline after 60–90 days. The board should hear 'we are improving how we measure and we will report a more reliable number,' not a defense of 94%. The credibility cost of a false green exceeds the discomfort of a corrected number.