Q: What should be included on a corrective action document? My company has never had one before, and I want to make sure all of the necessary elements are present.
A: Whether you’re sending a reply to a customer or establishing a general procedure for corrective action within your company, the content for the document is about the same.
The document should begin with a scope statement that identifies what types of issues or inputs will be subject to the new corrective action process. For example, external inputs are customer complaints or observations from your field reps, while internal inputs are manufacturing incidents that occurred recently or observations from managers who are dissatisfied with certain performance metrics.
While the initial thought might be that the scope of the document system would pertain only to quality problems, it could also be used for safety, cost and delivery issues. It also could be extended to your suppliers. In that situation, you would ask your suppliers to address certain issues with the prescribed method and report back to you.
The meat of the corrective action system should be based on a popular Six Sigma approach to improvement: define, measure, analyze, improve and control (DMAIC). A standard form with these steps could be created.
These steps are most easily understood when couched in the following scenario:
- Define—The customer finds an unacceptable level of defects in shipment number 65432.
- Measure—2.4% of the items contain defect X, and 0.9% of the items contain defect Y.
- Analyze—A series of designed experiments in the manufacturing process enable identification of four factors that contribute to the appearance of these defects. The details are logged in technical reports 37172 and 37173, and optimal process settings (along with an acceptable operating window for each setting) are identified.
- Improve—As a baseline, process performance is monitored for two weeks before the new settings are put in place. The changes are made, and results are monitored for two more weeks. The data indicate a reduction of 50% for each of the two targeted defects.
- Control—The process windows are captured in a standard operating procedure, the controllers on the equipment are programmed to integrate the new settings, and the operators are trained on the acceptable ranges. Further, the inspection procedures in the work center are amended to include descriptions of the defects. Those amendments are added to the routine audit forms.
An optional, but recommended, topic for your document is identification of a tracking system. You may also want to assign one person or more to oversee the topic of corrective action, establish a database of events being addressed, and verify the improvements and controls that are identified. Sign-off may be required to close out a corrective action.
Peter E. Pylipow
Senior design excellence engineer
Vistakon—Johnson and Johnson Vision Care
For more information
- Niemann, Craig A., "A Clearer Picture with RCA," Quality Progress, May 2008, pp. 64-65.
James J., Lee N. Vanden Heuvel, Donald K. Lorenzo and Laura O. Jackson, "Cause and Effect," Quality Progress, February 2009, pp. 38-44.
Q: Is it acceptable to implement dock-to-stock at a medical-device manufacturer without being in violation of 21 CFR 820.80? I would like to propose it to my team, but I want to make sure we will be in compliance with ISO 13485:2003 and with the U.S. Food and Drug Administration.
Alajuela, Costa Rica
A: Dock-to-stock typically means placing incoming product directly into stock. This is not allowed unless special precautions are taken, including segregating the product and placing it in quarantine until the incoming product is inspected, tested or otherwise verified as conforming to specified requirements.
Verification can be obtained via defined procedures that specify the means of verifying that shipments have the proper identity and are complete, undamaged and received in accordance with specifications. The procedures should also include provisions for verifying that incoming products are accompanied by supporting documentation, such as certificates of analysis or test results.
In general, the manufacturer has the burden of establishing a high degree of confidence that the supplied product meets requirements. Typically, that confidence is based on supplier evaluations and controls, past inspection history, in-plant rejection history or customer complaints, and needs to be directly related to the risks associated with the incoming product.
Senior vice president of quality assurance and regulatory affairs
Q: In a recent installment of Statistics Roundtable ("In a Certain Way," March 2009), Christine Anderson-Cook wrote, "The 95% confidence interval implies that if we repeated the procedure of collecting a sample many times, the resulting intervals will include the true population mean length 95% of the time."
That, of course, is consistent with the way we’re taught how to interpret confidence intervals. But I’m perplexed by her subsequent statement: "This is not the same as saying there is a 95% chance the population length is contained in [13.45, 14.91]."
From her first statement, if I were to repeatedly construct confidence intervals and put each confidence interval into a bin, 95% of the confidence intervals in this bin will contain the population length. If I were to draw one confidence interval randomly from this bin, the probability of me drawing a confidence interval that would contain the population length is 95%.
My question is whether that corresponds to "there is a 95% chance that a single interval thus constructed contains the population length."
I was also wondering if it was possible to make any conjecture about the probability that the true value is inside or outside the constructed confidence interval for a single study. Is it 50/50?
Alex T. Lau
Process analytics and blending specialist
Engineering Services Canada
A: I think your questions are at the heart of a very common confusion about confidence intervals and how they should be interpreted. Let me attempt to clarify.
Suppose we have a large population of lightbulbs, and we are told they have 95% reliability, meaning 95% of the bulbs will illuminate when we plug them in. We know something about the population of lightbulbs, which is helpful. If we are handed one lightbulb, however, our problem has fundamentally changed—either that particular bulb will work, or it won’t. We think it is much more likely it will work than not, but we still are not sure about the status of that bulb.
A working bulb means our confidence interval is correct and includes the true (but unknown) value. For a population of confidence intervals—in this case, repeating the same process of drawing a sample and then calculating that interval based on the data—we know that on average the true value will fall within the interval 95% of the time. But, for a particular study with a single set of data that matches the single bulb that has been handed to us, the interval will either be right or wrong.
Because we don’t know the true value, which is the case in almost any realistic situation, we don’t know if this is one of the many cases in which the true value falls within the interval or if it’s one of the rare cases in which it doesn’t. The big difference is whether we are talking about a population (or long-run probability) of confidence intervals or dealing with a single study with a single data set in which the answer is right or wrong.
As for your second question, there is a 95% probability that any particular confidence interval will contain the true value, because this is the long-run average of correct intervals based on the underlying theory. Perhaps another analogy would help explain.
Suppose that for a very simple lottery, there are two outcomes: a 95% chance you will lose and a 5% chance you will win. You hold a single ticket in your hand. That ticket is either a winner or a loser, but you are much more likely to be holding a loser (95 times out of 100).
Going back to the lightbulb example, you conducted a single study and have a single confidence interval that most likely will contain the true value, meaning the bulb will most likely work. But, if you’re handed the true value (this would require some luck, because usually this value is unknown), either the interval would be correct and contain the true value, or it would be incorrect.
Los Alamos National Laboratory
Los Alamos, NM