NU803 Unit 7 Run Chart Interpretation Example

Ten Weeks Below the Line: A Run Chart Interpretation of Hyperglycemic Monitored Days After Timed Glucose Checks Began for Patients on Steroids, Naming the Rule Behind Each Call

[Student Name]

Purdue University Global

NU803: Change Management and Clinical Data Analysis

Unit 7 Run Chart Interpretation

[Instructor Name]

[Date]

The units and weekly data are composite; figures are bracketed and computed consistently.

The Chart

The chart plots hyperglycemic monitored days per 100 monitored days for [16] baseline weeks and [12] implementation weeks, with the two education weeks omitted. The baseline median, [30.8], is drawn across the whole chart. Baseline weekly values ranged from [27.5] to [34.2]. Implementation values were [38.5] and [31.9] in weeks one and two, then [19.2], [17.5], [15.9], [15.2], [17.3], [15.0], [15.7], [14.1], [15.0] and [14.6] in weeks three to twelve.

The Rules Applied

Four rules were set in advance, drawn from published guidance on run charts (Perla et al., 2011). A shift needs at least six points in a row, all above or all below the median. A trend is five or more consecutive points all rising or all falling. Too few or too many runs is judged against a table of expected limits for the number of useful points. An astronomical point is one so obviously different that it would be noticed without a rule. Points falling exactly on the median are not counted when counting shifts and runs.

Simulations of run chart rules have shown that some produce more false signals than others as series lengthen, and recommended using the length of the longest run and the number of median crossings, with limits that depend on the number of points (Anhøj & Olesen, 2014). That approach was used as a check on the rules above.

Shift: Present

From implementation week three to week twelve, ten consecutive points fall below the baseline median. This meets the shift rule. Using the longest-run approach, the limit for [26] useful points is about eight; a run of ten exceeds it.

The signal begins in week three, not week one, which is what a bundle that takes time to reach every patient should look like.

Trend: Absent

No sequence of five or more points rises or falls continuously. The fall from week two to week three is large, but it is a single step, followed by values that level off around [15]. A trend rule would not capture that pattern, and the shift rule does.

Runs: Within the Expected Range

Excluding the two baseline points that sit exactly on the median, the series has [26] useful points and [12] runs. For [26] points, the expected range is roughly [9] to [19] runs, so the number of runs does not by itself signal change. Most of the runs occur during the baseline, where values cross the median frequently, as a stable process should.

The Tallest Point: Examined, Not Declared

Week one's value of [38.5] is the highest on the chart and above every baseline week. It is tempting to call it astronomical and treat it as a sign that the bundle made things worse at first. It was not declared astronomical, for two reasons. It is only [4.3] points above the highest baseline week, not dramatically outside the range, and it rests on only [39] monitored days, far fewer than the [72] to [83] of later weeks, because the order-set trigger was still being corrected in week one. The small denominator makes the week's rate unstable. Examining the underlying days, discussed in the seminar reflection, showed that three patients accounted for most of that week's hyperglycemic days.

Alternative Explanations Checked

Three events around go-live could have produced a similar shift, so each was checked. The hospitalist group added [one] new physician in implementation week [four]; the shift began a week earlier, and the new physician covered only [one] unit. The mix of steroid indications could have changed, for example with fewer high-dose pulses; the share of steroid days at [60] mg or more in prednisone terms was [21] percent at baseline and [23] percent during implementation, essentially unchanged. And monitoring itself changed the denominator: many more days became monitored days during implementation, including days in patients without diabetes, who may be less likely to exceed 250 mg/dL. Restricting the chart to patients with diabetes, who were monitored in both periods, still showed a shift beginning in week [four], which suggests that the change is not only an effect of monitoring more patients.

What the Chart Supports

The chart supports a conclusion that hyperglycemic monitored days fell and stayed lower from the third implementation week onward, a change unlikely to reflect random variation. Taken alone, it cannot show that the bundle caused the change, since other events could coincide with go-live. The companion charts help: pre-dinner checks rose above [80] percent by week three and pre-lunch checks by week five, after the lunch wave moved, and escalations followed by scheduled insulin rose from [34] percent at baseline to [71] percent. A change in outcome that follows changes in the processes the bundle targets is more credible than one that appears alone (Provost & Murray, 2022).

Next Steps

The chart will continue weekly for at least [12] more weeks. A new median, [15.5], calculated from implementation weeks three to twelve, will be used for the sustainment phase, so that any drift back toward baseline can be detected by the same rules.

References

Anhøj, J., & Olesen, A. V. (2014). Run charts revisited: A simulation study of run chart rules for detection of non-random variation in health care processes. PLoS ONE, 9(11), Article e113825. https://doi.org/10.1371/journal.pone.0113825

Perla, R. J., Provost, L. P., & Murray, S. K. (2011). The run chart: A simple analytical tool for learning from variation in healthcare processes. BMJ Quality & Safety, 20(1), 46-51. https://doi.org/10.1136/bmjqs.2009.037895

Provost, L. P., & Murray, S. K. (2022). The health care data guide: Learning from data for improvement (2nd ed.). Jossey-Bass.