Healthcare is the largest and fastest SPC adoption story of the last decade, and it happened for an unusually clear reason: the alternative was provably wrong. Board reports built on red-amber-green ratings and month-on-month comparisons generate a stream of explanations for variation that is pure noise. NHS England's Making Data Count programme exists to replace that with control charts, and it has trained well over a hundred organisations to do it.
The argument is the same one Shewhart made in 1924, in a setting where the cost of getting it wrong is measured in ambulance handovers.
A safety metric, before and after an intervention
Month
Limits frozen on the 16 months before the intervention. From month 17 the rate drops and stays down: four months below the lower process limit, and a sustained run below the mean.
On a red-amber-green report this is 24 months of noise: eleven months are "worse than last month" and nobody can say which of them mattered. The chart says one thing happened, it happened in month 17, and it is still happening.
Illustrative data — generated to behave like this process, not taken from anyone's plant. The limits, the signals and the capability indices above are computed at page load by the same engine that draws a customer's chart.
This is 24 months of inpatient falls per 1,000 occupied bed days. Look at what each reporting style would say about it.
A RAG report compares each month with the last one, or with a target. In these 24 months you get eleven "worse than last month" reds, some of which happened while the process was completely stable, and each of which cost somebody an explanation at a board meeting.
A control chart says one thing happened: from month 17 the rate dropped and has stayed down. Four months below the lower process limit, a sustained run below the mean, and everything before it is common-cause variation — the system doing what the system does. If a falls-prevention bundle went live in month 17, this is evidence it worked, and it is the only view of the data that could have shown that.
That asymmetry is the whole pitch for SPC in healthcare: it tells you when to act and when not to, and the second one saves more time than the first.
What gets charted
- Flow and access — 4-hour A&E performance, ambulance handover delays, referral-to-treatment waits, length of stay, discharge before noon, cancelled operations.
- Safety — inpatient falls, pressure ulcers, medication errors, VTE assessment compliance, hospital-acquired infections (C. difficile, MRSA bacteraemia).
- Demand and capacity — attendances, admissions, bed occupancy, theatre utilisation, DNA rates.
- Mortality and outcomes — SHMI and HSMR trends, readmission rates.
- Improvement projects — the before-and-after of any QI intervention, which is where SPC is at its most honest.
Almost all of it is a rate or a count over time, one point per week or month, which means one chart type dominates: the individuals chart, which the NHS calls an XmR chart.
Where the NHS conventions differ from the industrial ones
This matters if you are buying a tool built for factories, so here it is plainly.
| Making Data Count | Industrial / Nelson | |
|---|---|---|
| Chart | XmR for nearly everything | X̄-R, X̄-s, I-MR, attribute charts |
| Limits | Mean ±3σ from the mean moving range | Same maths |
| Shift rule | A run of six or seven points one side of the mean (trusts vary) | Nine points (Nelson rule 2) |
| Trend rule | Six or seven consecutive rising/falling | Six (Nelson rule 3) |
| Presentation | Variation and assurance icons per metric | Rule numbers and flagged points |
| Target | Drawn as a separate line; never a control limit | Specification limits, kept off subgrouped charts |
The run-length difference is real and we will not paper over it: this
platform evaluates Nelson's nine-in-a-row for a shift, not the NHS's six or
seven. On the chart above, the improvement is caught either way, but a
borderline shift will be flagged a couple of months later here than on a chart
drawn with the Making Data Count conventions. If exact NHS icon compliance for
a board pack is the requirement, the
NHS England SPC tool is free and does
exactly that, and the NHS-R community's NHSRplotthedots package does it in R.
Where this platform fits anyway
The NHS tools are excellent for producing a board pack once a month, by an analyst, in Excel or R. They are not a monitoring system. If what you want is charts that update themselves and shout when something changes, the trade looks different:
- The data arrives on its own — a CSV drop or an API call from your warehouse — instead of an analyst rebuilding a workbook every month.
- Limits are frozen with their baseline recorded, so "we re-based the limits after the service change in April" is a documented event rather than an unexplained step in a chart.
- Rules are evaluated as data lands, with an e-mail when something crosses the line. A ward-level metric that goes wrong on the 3rd does not wait for the month-end pack.
- A ward or department screen. The shop-floor display is a read-only, cookie-free wall view — it works as well outside a theatre suite as it does in a machine shop.
- A share link for a commissioner, a partner trust or a CQC visitor, with no account needed and no ability to change anything.
What we do not do
The NHS tools are better than we are at producing a board pack, and a few of their conventions are ones we do not reproduce:
- Making Data Count run lengths. We evaluate Nelson's nine-in-a-row rather than the six or seven MDC teaches, so a borderline shift is flagged a little later here. Worth knowing if your board is trained on the MDC charts.
- Variation and assurance icons. The blue and orange icon set is the visual grammar of an NHS board report, and we do not draw it.
- Funnel plots and p-charts. Cross-organisation comparison and proportion-with-a-varying-denominator metrics are not supported yet; counts (c-charts) and individuals charts are.
- Risk adjustment. VLAD and risk-adjusted CUSUM for surgical outcomes are specialist tools, and this is not one of them.
- Patient-identifiable data. Aggregate rates are the intended use. The platform is EU-hosted, tenanted and role-controlled, but anything identifiable deserves a DPIA first.
Try it with a metric you already publish
Board papers are public. Take one metric from your own last board pack, paste the monthly values into the chart generator — nothing is stored, nothing is sent anywhere — and see how many of the reds in that pack were noise. It is a five-minute experiment and it is usually the argument that wins the room.