Health Data Analytics carries 26 scored items, and it's the category candidates most often under-prepare. You don't need to be a statistician. You need to design a sound measure, pick the right chart and tell a real signal from noise. This guide builds each skill with a hands-on tool.
Types of measures
Donabedian's framework sorts measures into structure (capacity to deliver care), process (what was done) and outcome (what happened to the patient). Improvement teams add balancing measures, which watch for unintended harm elsewhere, like readmissions rising after you shorten length of stay.
Anatomy of a well-built measure
| Example: timely follow-up after heart failure discharge | |
|---|---|
| Numerator | Patients with a follow-up visit within 7 days of discharge |
| Denominator | Heart failure patients discharged home during the period |
| Exclusions | Discharged to hospice, left against medical advice, transferred or died |
| Data source | EHR discharge and scheduling records |
| Target and threshold | Target 80%; below 65% triggers review |
| Frequency | Monthly, displayed on a run chart |
Exam trap
If a rate suddenly improves, check whether the definition, exclusions or data source changed before celebrating.
Sampling methods
- Simple random: every case has an equal chance of selection.
- Systematic: every nth case from a list, with a random start.
- Stratified: random samples from subgroups (units, payers, age groups) so each is represented.
- Convenience: whatever is easy to get. Fast, but prone to bias.
Descriptive statistics
Mean, median and mode describe the center of your data; standard deviation and range describe spread. For skewed data like length of stay, where a few long stays pull the average up, the median is the better summary. Paste some numbers to see the difference.
Which statistical test?
| Tool | Use it to | Example |
|---|---|---|
| t-test | Compare the means of two groups | Average wait time before vs. after a change |
| ANOVA | Compare means of three or more groups | Average LOS across four units |
| Chi-square | Test association between categorical variables | Readmission (yes/no) by discharge disposition |
| Correlation (r) | Measure strength and direction of a linear relationship, from −1 to +1 | Nurse staffing and fall rates |
| Regression | Predict an outcome from one or more variables | Predicting readmission risk |
Remember: correlation is not causation, and a p-value below 0.05 is the conventional threshold for statistical significance.
Practice with 500+ CPHQ questions
Scenario-based, mapped to the NAHQ content outline, with an explanation for every answer. Your first 50 are free.
Statistical process control
Every process varies. Common cause variation is the natural, random noise built into a stable process; reacting to it (tampering) makes things worse. Special cause variation comes from something unusual and should be investigated. Control charts separate the two with a centerline and upper and lower control limits, typically set at three standard deviations.
Common special cause rules
- A single point outside a control limit.
- A shift: a run of consecutive points on one side of the centerline (8 is common; some rule sets use 7 or 9).
- A trend: 6 or more consecutive points steadily rising or falling.
- Two of three consecutive points near a control limit (beyond two standard deviations, same side).
Rule sets vary slightly by source. Know the concept behind each rule rather than a single number.
Which control chart?
1. What kind of data?
Choose a data type to begin.
Sensitivity, specificity and predictive values
These appear in questions about screening tools and data validation. Change the counts to see how each value moves.
| Measure | Formula | Answers the question |
|---|---|---|
| Sensitivity | TP ÷ (TP + FN) | Of people who have it, how many test positive? |
| Specificity | TN ÷ (TN + FP) | Of people who don't, how many test negative? |
| PPV | TP ÷ (TP + FP) | If positive, how likely is it real? |
| NPV | TN ÷ (TN + FN) | If negative, how likely is it truly negative? |
Dashboards, scorecards and benchmarks
A dashboard shows current performance on a few key measures, often with status colors, for quick monitoring. A scorecard tracks performance against targets over a longer period and is often tied to strategy. Tailor both to the audience: boards need a handful of strategic measures, while unit managers need operational detail.
Benchmarking compares your performance with an external reference, such as national databases or high-performing peers. Make sure the comparison group uses the same definitions and is risk-adjusted where appropriate.
A unit manager sees one month's hand hygiene compliance fall from 91% to 86%. The control chart shows the point is within limits and no other rule is triggered. What is the best response?
With no special cause signal, the drop is consistent with common cause variation. Reacting to it as if it were special (tampering) wastes resources and can increase variation. If overall performance is below goal, improving the system as a whole is a separate decision.
A hospital wants to track falls per 1,000 patient days. Patient days differ each month, and a single patient can fall more than once. Which control chart should be used?
Falls are counts of events (defects), a patient can have more than one, and the area of opportunity (patient days) varies. That combination calls for a u-chart. A c-chart assumes a constant area of opportunity; p- and np-charts are for yes/no "defective" data.
Frequently asked questions
The content outline names measures of central tendency, standard deviation, correlation, regression and the t-test, plus statistical process control including common and special cause variation, control charts and trend analysis.
Expect some straightforward calculations such as rates, percentages, sensitivity and specificity. Check the current candidate handbook for what the testing platform provides.
Both plot data over time. A control chart adds statistically calculated upper and lower control limits, which let you separate common cause variation from special cause variation.
Use a p-chart for the proportion of units with a problem (defectives), such as the percent of patients who fell. Use a u-chart for the rate of defects per unit when a unit can have more than one, such as falls per 1,000 patient days.
Figures are from NAHQ's 2024 CPHQ Detailed Content Outline (Revision 2b) and 2024 Domestic Candidate Handbook unless noted. Fees and policies change; the current handbook on nahq.org is the controlling source.