This Article Full Text (PDF) Alert me when this article is cited Alert me if a correction is posted Services Email this article to a colleague Similar articles in this journal Similar articles in Web of Science Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Request Permissions Citing Articles Citing Articles via Web of Science (3) Citing Articles via Google Scholar Google Scholar Articles by Reves, J. G. Search for Related Content PubMed PubMed Citation Articles by Reves, J. G.

---

EDITORIAL

Lessons on Learning About Learning Curves

Department of Anesthesiology, Duke University Medical Center, Durham, North Carolina

Address correspondence and reprint requests to Joseph G. Reves, MD, Department of Anesthesiology, Box 3094, Duke University School of Medicine, Durham, NC 27710.

The introductionof anything new in anesthesiology practice, whether it is an anesthetic drug, a new piece of technology, or a new practice concept, has three common results. The first is the direct benefit and the second is the cost of the item. (Cost may be documented not solely in financial terms, but there may be undesirable clinical effects or iatrogenic sequelae.) The third, much more subtle consequence of the introduction of anything new is the learning of how to use what’s new to obtain maximum benefit. Conceptually, with anything new, over infinite time, cumulative knowledge about the innovation ultimately becomes absolute (100%.)

The learning curve represents the assimilation of this new knowledge (Figure 1). The shape of the learning curves (sigmoidal, bimodal, exponential, etc.) and the time over which optimal learning occurs are what is important about the learning curves. The shapes of these learning curves are dependent on two fundamental factors, the amount of knowledge and the ease of learning. It is understood and accepted that the degree of knowledge about any innovation at the time of introduction will change (increase) over time, meaning an expanded knowledge base exists the longer a new drug, device, or concept is applied. This expanded knowledge might be described as a moving target of learning that is inevitable and inherent in the learning curve. Of the almost limitless array of possible learning curves, four are depicted in Figure 1. Each innovation has a unique learning curve. An idealized learning curve (such as A) has a quick learning phase and no late surprises (as in D) that prolong the learning.

View larger version (11K): [in this window] [in a new window]   Figure 1. Four types of learning curves are drawn. Curve A describes a rapid exponentially early learning that is later followed by slower learning and, as in curve C, reaches maximum relatively early. Curve B describes learning that is slow at first, then rapid, and then followed by slow again. Curve C shows a constant rapid learning that reaches a maximal point relatively early. Curve D shows rapid learning, followed by slow, followed by rapid again, and finally slow. Of all the curves, A is best, although C is certainly preferable to B and D. Where vertical axis is percent of cumulative knowledge, t is time, and is infinity.

If one sets out to describe a learning curve, there are some essential features required to properly describe the curve. The most important is to decide what knowledge is to be learned. In the case of a new opioid in the practice of anesthesia, one would focus learning on the safety, efficacy, methods of administration, and side effects of the drug. It is vital to document how quickly the anesthesiologist can learn to safely use the drug to accomplish the therapeutic goals. Importantly, it is critical to design the study to span a time sufficiently long enough to actually measure the learning. Obviously, it is impractical to measure at the beginning and again after infinite experience, but it is crucial that the study time be sufficiently long to encompass a significant portion of the learning. Another factor in study design is to have an adequate sample size to extrapolate the results to a larger universe of clinicians. Finally, as with any study design, the fewer the confounding variables the better. For example, if one is studying the use of a new opioid, and multiple other drugs known to influence analgesia, as well as side effects of the drug, are also concurrently used, then the information about the opioid alone will be less valid. However, the modern practice of anesthesia involves the use of multiple drugs and any study that included only a single drug would be almost meaningless within the context of usual clinical practice: Thus, some confounding from other drugs is expected and desirable.

The study by Joshi et al. (1) unfortunately has many limitations in describing the true nature of the remifentanil learning curve. Some relate to study design, some to data available to the investigators, and some to interpretation of the data. The first difficulty is in the selection of endpoints of learning. A number of these outcome variables chosen were for the most part controlled by the remifentanil use protocol. Thus, nothing is learned because the anesthesiologists were told exactly how and how much remifentanil to give. It is not surprising that wake-up times, remifentanil doses, and use of other drugs did not change over the study time: they were prescribed by protocol. One key outcome that was omitted was the degree of subjective pain experienced by the patients. One of the domains in which learning was demonstrated was in the use of adjuvant analgesics, and the crucial question regarding this outcome is did postoperative analgesia per se change over time in the patients. Another fundamental problem in the study is that the learning curve consisted of a data set taken from the initial 3 and final 3 patients in a series of only 10 patients in the experience of 190 clinicians. It would have been more instructive to describe the learning curve by sampling the first 3 uses and 3 subsequent uses separated in a series of perhaps 10, 25, and 100 consecutive patients. To sample multiple times over a longer experience without a strict protocol would describe the shape of the learning curve best.

The investigators could only use data available to them. It is unfortunate that, to demonstrate the value of the five hours of education, half the physicians had been randomly given the full five hours of education and the other half a brief 15- to 30-minute instruction. This methodology would allow us to determine whether the intensive education led to the learning or whether the results were a product of either the drug characteristics (easy to learn to use) or the strict adherence to the protocol (learning was controlled by protocol).

Despite these deficiencies, we can draw valid conclusions. First, using a strict protocol and five hours of education with a new, unique drug permits very rapid adaptation and relatively uniform clinical results. The fact that fewer total adverse events occurred in the last three patients compared to the first three probably means that some subtle learning was taking place early in the use of remifentanil. The investigators document change in the outpatient use of adjuvant analgesics and this shows that learning about postoperative analgesia was occurring. What we learn best, however, is that in future studies a better understanding of how to describe a learning curve must contain measuring the important knowledge domains of learning, sampling over a long enough time to adequately describe the curve, and ideally, sampling multiple times to accurately fit a curve. With properly designed studies, we can determine just how close to ideal the learning curve of an anesthesia innovation is, a fact that will influence its safe, effective, and timely adoption.

References

1. Joshi GP, Jamerson BD, Roizen MF, et al. Is there a learning curve associated with the use of remifentanil? Anesth Analg 2000; 91: 1049–55.[Abstract/Free Full Text]

This Article Full Text (PDF) Alert me when this article is cited Alert me if a correction is posted Services Email this article to a colleague Similar articles in this journal Similar articles in Web of Science Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Request Permissions Citing Articles Citing Articles via Web of Science (3) Citing Articles via Google Scholar Google Scholar Articles by Reves, J. G. Search for Related Content PubMed PubMed Citation Articles by Reves, J. G.