Since the discovery that puerperal fever is associated with poor hand hygiene, the care that is given by obstetric providers has been implicated in infections in childbirth. However, in modern times with sterile techniques, the question may be asked: Is the health care still responsible for causing infections at the time of delivery? Goff et al in this issue of the American Journal of Obstetrics and Gynecologists have taken an administrative database and authored a thought-provoking article on hospital variation in childbirth infections. Taking into account preexisting patient characteristics and mode of delivery, Goff et al compare rates of hospital infection at the time of delivery. They show that there is significant variation in outcomes after adjustment for patient characteristics. Women who delivered at a hospital at the 75th percentile of infection rates had a 2-fold risk of experiencing an infection compared with women who delivered at a hospital at the 25th percentile. The authors also explored hospital characteristics that are associated with higher rates of infectious morbidity.

Comparing outcomes in health care has become big business. Hospitals and doctors constantly find themselves being compared with each other and to national standards. Implicit in these comparisons is that providers with better outcomes provide higher quality care. Yet, there are 3 major sources of variation in outcomes: preexisting patient characteristics, quality of care, and random variation. As with any equation, if you can control 2 of the variables, you can better understand the third. Comparisons of outcomes without adjustments for patient characteristics can be misleading because differences in outcomes may be due to factors outside the health care system’s control. Before hospital outcomes can be compared meaningfully as a marker of hospital quality, one must account for the differences that exist before the health care encounter and are not able to be influenced by the health care system.

The ability to make comparisons of health care outcomes, while accounting for differences in patient characteristics and random variation, has been made easier as large uniform datasets from multiple hospitals have become available. The weakness of this approach is that large datasets, which usually are collected for other purposes such as insurance billing or collection of vital statistics, often lack the detail that is needed to risk adjust comprehensively for patient characteristics.

An administrative data source is only one weakness in the risk-adjustment process. Even with “perfect” data collection, some patient variables that affect outcomes will never be able to be measured comprehensively. As with many things in medicine, the inability to produce a perfect analysis does not mean that risk adjustment is not worth doing. For example, in this analysis, it would have been potentially useful to adjust for gestational age at delivery, but this information was not available. Nevertheless, despite the fact that gestational age was not available, this study still suggests that marked interhospital variation in peripartum infection exists and may be related to differences in obstetric care that is being provided.

It is important to realize that measuring health care outcomes is only the first step in quality improvement. Having a measure allows one to determine whether quality improvement efforts are helping, not making a difference, or hurting. Risk-adjusted outcomes provide the necessary first step in quality improvement so that we can be as sure as possible that the variation we are seeing is a result of the health care system and not of population differences among hospitals.

Much is known about the methods to improve quality. Benchmarking is a process by which hospitals are provided with data about where they stand compared with other hospitals. By noting the variation between hospitals, we can look at practices at the hospitals with the “best risk-adjusted outcomes” and try to emulate their practices. Then, by following risk-adjusted outcomes, we can determine whether outcomes are improving. This method has been used effectively both in and out of obstetrics to improve quality of care in a community.

In addition to the risk-adjusted infectious outcomes, Goff et al provide some macro level insight into other hospital characteristics that are associated with risk-adjusted infectious outcomes. For example, high volume and inner city hospitals tend to have higher rates of infection. Although some of this may be explained by unmeasured patient risk, these characteristics provide a starting place to look at how practices differ between large and small hospitals and urban vs rural practices. Future exploration in this area needs to understand how staffing patterns, surgical techniques, and labor management practices influence infection rates.

We have come a long way in quality improvement since doctors recognized the need to wash their hands. There may still be much to learn about infection control in childbirth because it is clear that some hospitals have better rates of infection than others. The analysis that Goff et al provide is a step forward in understanding how the hospitals with the best outcomes achieve those outcomes and allowing others to begin to emulate best practices.