IHP 604 Module 7 Data Analysis Paper Example

Signal or Noise? Control Chart Analysis of Diabetes Results at Crestline Medical Group

[Student Name]

Southern New Hampshire University

IHP 604: Healthcare Quality and Improvement

Module Seven Paper

[Instructor Name]

[Date]

The organization, setting and figures below are a composite written as a model document. No real employer, client, colleague or patient is described.

Signal or Noise? Control Chart Analysis of Diabetes Results at Crestline Medical Group

Six months after Crestline's Eastside clinic began nurse care management for patients above an A1c of 9%, its poor-control rate had fallen from 36% to 30%. Leaders want to know whether that drop is real or the kind of month-to-month wobble any clinic shows. This paper answers that question with control charts, compares the 18 clinics fairly and checks for harm.

Why Averages Mislead

Comparing one number before a change with one number after hides everything that happened in between. A drop could reflect a random low month, a trend that began before the change or a seasonal pattern. Mohammed et al. (2001), analyzing high-profile failures in British health care, argued that Walter Shewhart's distinction between common-cause variation, the natural noise of a stable process, and special-cause variation, a signal of something new, is essential for anyone judging performance. Treating noise as a signal leads to blame or false credit; treating a signal as noise misses real problems.

Choosing the Right Chart

Mohammed et al. (2008) wrote tutorial notes for health care staff on plotting basic control charts. They explained that the chart type depends on the data: p charts for proportions, such as the share of patients above a threshold; u charts for rates per unit of exposure; c charts for counts of events in equal periods; and individuals charts for continuous measures. They recommended building limits from a baseline of about 20 or more points where possible and warned against recalculating limits every time new data arrive. Because Eastside's measure is a monthly proportion with a fairly stable denominator, a p chart fits.

Calculating the Limits

Eastside's baseline covers eighteen months before the change, with about 780 adults with diabetes each month. The average proportion above 9% was 0.36. The standard error of a proportion is found by multiplying 0.36 by 0.64, dividing by 780 and taking the square root, which gives 0.0172. The upper and lower control limits sit three standard errors from the center line: 0.36 plus or minus 0.0516, or 30.8% to 41.2%. During the baseline, every month stayed inside those limits without any suspicious runs, meaning the process was stable, if stuck at an undesirable level.

What Happened After the Change

Care management began in month nineteen. The next nine months read 35%, 34%, 33%, 32%, 31%, 30%, 30%, 29% and 29%. Two signals appear. First, from month nineteen onward every point sits below the center line, and a streak of at least eight months in a row beneath the average counts as a special-cause signal under standard chart rules, a threshold Eastside crossed in month twenty-six. Second, the months at 30% and 29% fall below the lower limit of 30.8%. Both rules point to a real shift rather than noise.

Table 1. Eastside Clinic Poor-Control Rate After Care Management Began

Note. Selected months; baseline center line and limits from months 1-18.

Was It the Care Management?

A control chart shows that something changed, not what caused it. Three checks support attributing the shift to care management. The process measure, patients contacted within 14 days of a high result, rose from 31% to 82% in the first two months, just before the outcome began to move. The other 17 clinics, which had no new program, showed no signal over the same months. And no other major change, such as a new laboratory or payer rule, occurred at Eastside during the period.

Comparing Clinics Without a League Table

Crestline's earlier rankings ordered the 18 clinics from best to worst, implying that each position meant something. Mohammed and colleagues warned that most differences in such tables are common-cause variation. A p chart across clinics, with limits that widen for smaller clinics, gives a fairer picture. Using the group average of 31%, a clinic with 640 patients has limits of 25.5% to 36.5%, and a clinic with 910 has limits of 26.4% to 35.6%. Only two clinics fall outside their limits: the best, at 19%, and the worst, at 44%. The other 16 are statistically indistinguishable from the group average.

What the Clinic Chart Means for Action

The chart tells leaders to study the best clinic to learn what it does differently and to support the worst with focused help, while treating the other 16 as one system to be improved together rather than ranked. That is a very different message from last year's table, which singled out several clinics whose position reflected little more than chance.

Checking for Harm

Intensifying treatment raises the risk of low blood sugar. Monthly emergency visits and admissions for hypoglycemia across the group are counts in equal periods, so a c chart applies. The baseline average was 6.2 events a month; the upper limit is 6.2 plus three times its square root, or about 13.7, and the lower limit is zero. Since the pilot began, monthly counts have ranged from 3 to 9, all within limits and with no runs above the center. There is no signal of harm so far.

Strengths and Limits of Control Charts

Thor et al. (2007) systematically reviewed the use of statistical process control in health care improvement, finding it in use from laboratories to wards to primary care, where it helped teams understand variation, judge the effect of changes and monitor performance. They also noted challenges, including the need for training, adequate data and careful choice of chart and rules, and that charts alone do not explain why a change occurred. Crestline's analysis reflects both sides: the charts separate signal from noise clearly, but attribution still depends on process data and knowledge of context.

Limitations of This Analysis

Nine post-change points are enough to detect a shift but not to judge whether it will hold. The poor-control measure counts missing tests, so part of the drop may reflect more testing rather than better control; separating the two components will be the next step. Seasonal effects cannot yet be ruled out with fewer than two full years of data.

Conclusion

Eastside's drop from 36% to about 30% is a real shift by two standard rules, supported by process data and the absence of change elsewhere, with no signal of harm. Comparing clinics on a control chart rather than a ranking shows two true outliers and a large middle group best improved together. The next milestone will extend care management and continue charting monthly.

References

Mohammed, M. A., Cheng, K. K., Rouse, A., & Marshall, T. (2001). Bristol, Shipman, and clinical governance: Shewhart's forgotten lessons. The Lancet, 357(9254), 463-467. https://doi.org/10.1016/S0140-6736(00)04019-8

Mohammed, M. A., Worthington, P., & Woodall, W. H. (2008). Plotting basic control charts: Tutorial notes for healthcare practitioners. Quality and Safety in Health Care, 17(2), 137-145. https://doi.org/10.1136/qshc.2004.012047

Thor, J., Lundberg, J., Ask, J., Olsson, J., Carli, C., Pukk Härenstam, K., & Brommels, M. (2007). Application of statistical process control in healthcare improvement: Systematic review. Quality and Safety in Health Care, 16(5), 387-399. https://doi.org/10.1136/qshc.2006.022194