CPHQ Health Data Analytics: What the Exam Really Tests

Twenty-six scored items sit here, second only to performance and process improvement at 27. Together those two are 53 of the 125 scored questions, just over 42% of the exam, which is why they are worth studying first. The full weighting is in the CPHQ exam blueprint.

The domain covers the measurement side of quality work: defining measures, sampling and collecting data, displaying it over time, interpreting variation, adjusting for case mix, benchmarking, and reporting to people who will decide something from it. It is not a statistics paper, and there is no requirement to derive a control limit by hand. What it wants is judgement about data.

That is why it catches people. A physician reads a forest plot every week and still loses marks here, because clinical training teaches interpretation of research data, while quality work asks you to interpret your own operational data as it arrives, in small batches, with no control group and no p-value to hide behind.

Every process varies. Common cause variation is the inherent fluctuation of a stable process; special cause variation comes from outside that system, an assignable event. The whole domain turns on this distinction, because the correct response is opposite in each case. Special cause means investigate that occurrence. Common cause means the process is doing what it was designed to do, and if the result is unacceptable you redesign the process rather than chase the point. Treating common cause variation as a special cause is tampering, and it usually makes performance worse.

A run chart plots your measure over time against a centre line, normally the median, and is read with rules based on runs above and below it. It needs few data points, assumes nothing about the distribution, and is the right first tool for a small improvement project. A control chart, or Shewhart chart, adds control limits calculated from the data itself, usually at three standard deviations either side of the mean. It needs more data, separates signal from noise more powerfully, and can tell you whether a process is stable, which a run chart cannot.

Control limits are not targets, and not specification limits set by a regulator or contract. A process can be perfectly stable and still perform at a level nobody would accept.

Both chart types come with a small set of pattern tests: a point outside the control limits, a run of consecutive points on one side of the centre line, a sustained trend, too few or too many runs, and points clustered in the outer zones. For control charts these are commonly called the Western Electric or Nelson tests; run charts use a smaller set of shift, trend, run-count and astronomical-point rules. The number of rules and the thresholds differ between textbooks and software, so learn the logic of each pattern rather than memorising a numbered list from one source. [VERIFY: which set of run-chart and control-chart rules the current HQ Solutions text and NAHQ prep materials use, and how many rules each set contains]

These three appear more often than candidates expect, usually inside a stem about reporting. A proportion is part of a whole, so its numerator sits inside the denominator, and it reads as a percentage. A ratio compares two quantities where the numerator is not part of the denominator. A rate counts events over a defined population or exposure period, which is why it carries a multiplier: falls per 1,000 patient days. Getting the denominator right is the actual test.

Quality work rarely audits everything, so you sample. Know simple random, systematic, stratified and convenience sampling, and what each risks. Convenience sampling is not wrong in itself, but it is not generalisable, and a stem describing an audit of whichever notes were on the desk tests whether you notice.

Risk adjustment accounts for case mix so outcomes can be fairly compared between units or hospitals, and its usual output is an observed-to-expected ratio. Benchmarking is the comparison itself: internal over time, competitive against similar organisations, or against a best-in-class standard. The examinable point is that an unadjusted comparison between populations of different severity is evidence of nothing.

Know structure, process and outcome measures, and know that balancing measures exist to catch the harm your improvement causes elsewhere. A measure without an operational definition, a stated numerator and denominator, inclusion and exclusion rules and a named data source is not a measure; it is an opinion with a number attached. Dashboards are examined as a communication problem: the right measures, at the right frequency, shown over time rather than as a single figure.

Four tools carry most of the marks. Learn what each is for, what data it needs and what it cannot tell you.

On control charts, the type follows the data: proportions of a defined denominator on a p-chart, counts over a varying area of opportunity on a u-chart, counts over a constant area of opportunity on a c-chart, and individual continuous values on an individuals chart with a moving range. Matching data type to chart type is a fair exam expectation; deriving the limits is not.

These are original items written in the style of the exam, not retired questions. Read the last sentence of each stem first and decide what is being asked before looking at the options.

A quality team plots monthly catheter-associated urinary tract infection rates on a control chart. For 14 months the points vary within the limits with no unusual pattern. In month 15, shortly after two wards were merged, one point falls above the upper control limit. What is the most appropriate next action?

Correct answer: 2. A single point beyond a control limit is a special cause signal, and the response is to investigate that occurrence, here the ward merger. Option 1 recalculates limits to absorb the signal, which hides it; limits are recomputed only after a deliberate, sustained process change is confirmed. Option 3 is the response to common cause variation at an unacceptable level, and redesigning a stable process on one point is tampering. Option 4 discards the signal you already have, and the assignable cause gets harder to find with time.

In one month a hospital records 24 inpatient falls across units reporting 6,000 patient days. The quality committee wants the figure expressed so it can be compared with an external benchmark that accounts for exposure. Which measure should be reported?

Correct answer: 1. A rate expresses events over a defined period of exposure, and patient days capture both how many patients there were and how long each was exposed. Option 2 is a proportion using admissions as the denominator, so it ignores length of stay and a unit with long-stay patients looks falsely safe. Option 3 is a true ratio but answers a different question, describing injury severity rather than exposure-adjusted frequency. Option 4 is a raw count, which moves with occupancy and cannot be benchmarked externally.

A cardiac surgery unit's crude mortality is above the published national figure. The surgeons state that their patients are sicker because the unit takes tertiary referrals. The board asks which analysis will settle the question. What should the quality director recommend?

Correct answer: 3. Risk adjustment is the method for comparing outcomes across populations with different case mix, and the observed-to-expected ratio is its standard output. Option 1 removes the very cases in dispute and introduces selection bias, which is not risk adjustment. Option 2 reduces random error but leaves case-mix confounding untouched; a larger biased estimate is still biased. Option 4 is a legitimate internal benchmark, but it cannot answer a question about the national comparison.

Analytics is the domain the others borrow from, which is why its questions appear wearing other clothes. In performance and process improvement, every PDSA cycle needs a measure and a chart before you can claim a change was an improvement, and the annotated run chart is the standard evidence. In quality review and accountability, the same skills reappear as professional practice evaluation: small denominators, comparisons between individual practitioners, and the risk adjustment that makes those comparisons defensible.

Patient safety leans on it too. Incident reporting produces counts with an unknown denominator and a known reporting bias, and recognising that more reported events can mean a safer reporting culture rather than a more dangerous ward is an analytics judgement, not a safety one.

Work through the checklist with your own service's data rather than a textbook example, because the exam asks you to judge messy operational data. Then test judgement rather than recall: our free CPHQ practice test includes analytics items written to the same application standard as the three above, and NAHQ's content outline at nahq.org confirms the weighting used here. The video lesson accompanying this article sits with the rest of the domain series in our CPHQ study video library.