Yes trends analysis are typically required in many different settings and I agee with Dan O'Leary "Rolled up metrics can be misleading." I also agree and would add caution about plotting "a rate, not counts, since the denominator can change over time. Counts could be misleading." This is generally true for most, but not all, manufacturing process situations in my experience. My caution is _do NOT confuse_ issues where single (or low) count events are critical signals for "trending" and should not be subject solely to the same "plotting a rate" philosophy. First, consider Pareto charts to include bar graphs about individual values and cumulative total line graph.
Next, remember, Careful trends analyses are especially important for all data. For example, clinical data where individual data point and trends analyses may differ greatly from manufacturing process data analyses. For clinical data, knowing clinical benefits/risks and clinical safety performance data are required. Reporting trends in individual datapoint or small numbers among thousands of data points is critical for things like serious, unexpected adverse event related to the product . This is an art (the practice of medicine) and not really a "rate" per se.
------------------------------
Joy Frestedt PHD, CPI, RAC, FRAPS, FACRP
President and CEO
Frestedt Incorporated (www.frestedt.com)
Saint Louis Park MN
United States
612-219-9982
jf@frestedt.com------------------------------
Original Message:
Sent: 19-Aug-2022 09:43
From: Dan O'Leary
Subject: Trending and Analysis of Data (Monitoring and Measurement)
GHTF/SG3/N18:2010 (attached) has a good explanation of how to set up the system. (It also explains why the acronym "CAPA" is bad terminology.)
In general, trending implies a time series, in which you observe and record the data on some "calendar" basis, usually monthly. It is best if you plot a rate, not counts, since the denominator can change over time. Counts could be misleading.
The first idea is get a picture of what is happening, so the observation period should be long enough to be instructive. Trending implies that you will also identify a trend such as an increasing rate, a decreasing rate, or an flat rate. You should be able to pick out the rate by eye. A more quantitative approach calculates the linear regression and displays it and its equation on the fraph. (This is quite easy with MS Excel.)
A common approach is to to set warning limits and action limits to help you identify a trend. When you get a signal (warning or action) it suggests the process trend is unfavorable and you should do something. The first something is to investigate the reason for the signal and set the system back to its prior acceptable state.
In my experience people are too quick to invoke the corrective action process, especially when only a correction will fix the problem. In many companies the corrective action process is overly compicated and doesn't achieve the intended result – eliminate the cause of the nonconformance.
Rolled up metrics can be misleading because they can hide a problem. The reason for doing all of this analysis is to identify problems and take care of them early. In the best case you deal with signal from the warning limits before they read the action limits.
------------------------------
Dan O'Leary CQA, CQE
Swanzey NH
United States
Original Message:
Sent: 18-Aug-2022 10:58
From: Anonymous Member
Subject: Trending and Analysis of Data (Monitoring and Measurement)
This message was posted by a user wishing to remain anonymous
My colleagues, my boss, and I are currently in a standoff over the best practices around trending and analysis of data in terms of non-conformances, complaints, and process data (ie., FPY, efficiency, etc.). The main disagreement is centered around trending over the time and determining when something needs to be elevated to a Corrective Action.
My questions are: Is trending overtime required/recommended? How do you determine if a CA is required? Do your answers change if the rolled up metric is green?