What do I use
Hospitals and health facilities
Clinical audit, quality improvement, bed and staffing decisions, infection control, service evaluation.
Your question, and what answers it
The question every manager asks and almost nobody answers correctly. A control chart separates ordinary variation from a real change, so you stop reacting to noise.
In ZStat: open your study, go to Analysis, and choose control chart.
Whether the service can meet a standard, as opposed to whether it happened to last month.
In ZStat: open your study, go to Analysis, and choose process capability.
With a post-hoc test, and with case mix in mind before anybody is blamed for a difference.
In ZStat: open your study, go to Analysis, and choose one way anova.
With the base rate reported beside the accuracy, because a model that always says "no" looks excellent when most patients are not readmitted.
In ZStat: open your study, go to Analysis, and choose logistic regression.
Length of stay is time-to-event data, including the patients still admitted when you counted.
In ZStat: open your study, go to Analysis, and choose kaplan meier.
For infection or admission clusters by location, with the multiple-testing problem stated rather than hidden.
In ZStat: open your study, go to Analysis, and choose getis ord.
Before and after, with the effect size, and with the honest caveat that anything else that changed at the same time is confounded with it.
In ZStat: open your study, go to Analysis, and choose independent t.
What goes wrong in hospitals and health facilities
- Comparing raw outcome rates between units without case mix. The unit that takes the sickest patients will always look worst.
- Reacting to a single month above target. Most such movements are ordinary variation, and acting on them makes the process less stable, not more.
- Treating routinely collected patient data as exempt from data protection because it was collected for care. The purpose has changed, and that is exactly what the law is about.