To the Editor:-Referring to the article of Badner et al., Dr. Mangano encourages assessment of the value of perioperative observations and interventions for predicting quality of life, event-free survival, and cost. Badner et al. furthers this goal by relating postoperative signs and symptoms to longer-term risk of myocardial infarction (MI) and associated mortality. This valuable data set could provide even further insight into the risk profile for MI and MI-related mortality among surgical patients by a more complete analysis of the available information.
Badner et al. confirm that postoperative heart rate is strongly associated with postoperative MI (PMI). For reasons that are not clear, however, lower opioid use was more strongly associated with PMI than was blood pressure, history of angina and MI, and previous bypass surgery. Only height, weight, age, and nitrate use were predictive among the other patient characteristics that were investigated.
The use of multivariable statistical models could be helpful in further developing and interpreting risk profiles. For example, only heart rate measured on days 1 and 2 after surgery were associated with PMI. This does not imply, however, that elevated heart rate before or after these days is unimportant. If preoperative heart rates are also available, we could develop models that would evaluate the combined effects of heart rate before surgery and change in heart rate during time after surgery and of the absolute postoperative heart rate. In this way, we could determine the features of the time course of heart rate that most contribute to risk-an analysis that might be of interest in itself, but also useful for suggesting therapeutics to reduce risk (see Mangano et al.)
If the data set and number of endpoints were larger, the development of risk profiles could make use of multivariable regression models to evaluate simultaneously the relative contributions of demographics, clinical characteristics, and medical history to risk of MI, death, or both (see Marshall et al. ). But, even in a smaller data set, we could evaluate the combined effect of a few characteristics in this way. Such analysis might start by investigating the relationships among the predictors of PMI and death, by determining whether patients with higher heart rates received less opioid. Then, multivariate models could help in identifying which characteristics independently predict adverse outcomes. For example, if higher heart rate is associated with less opioid use, then predictive models that include both variables would help to determine whether the opioid use reduced death because of its effect on heart rate or independently of that effect.
Such models could also help in the development of a definition or definitions for MI that best predict future cardiac disease outcomes. Although the authors found no difference in cardiac outcome in the four definitions of MI presented, more sophisticated statistical methods might help to further define MI according to the prediction of subsequent events. Such definitions might help in risk stratification. Such analyses must be based on all surgical information, not just MI-related deaths, because deaths from competing risks may result in biased analyses unless properly incorporated into statistical models. The reason is that deaths from causes indirectly related to MI eliminate the risk of MI for the patient.
Development and validation of risk profiles for MI and other adverse surgical outcomes may necessitate larger databases for adequate precision. The valuable data set of Badner et al. might be combined with other sources of information to ensure adequate power for these investigations.
Robert S. Litwack, M.D.
Chief; Anesthesiology Service; McGuire VA Medical Center; Richmond, Virginia;Litwack.Robert_S@Richmond.VA.Gov
Victor DeGruttola, Sc.D.
Professor of Biostatistics; Harvard School of Public Health; Boston, Massachusetts
(Accepted for publication July 7, 1998.)