A sequential method for real-time detection of life-threatening signals in serum potassium level after myocardial infarction

Clinical guidelines recommend maintaining potassium levels between 4.0 and 5.0 mEq/L in patients with acute myocardial infarction (MI). These guidelines are based on studies that found associations between crossing of absolute potassium limits and mortality. This article investigates a different approach: we hypothesized that a change in the potassium level may be a harbinger of short survivability, rather than crossing of absolute boundaries. Our objectives were to (1) examine whether a change in the mean potassium level can predict in-hospital death in complicated MI hospitalizations (30 days or more) and (2) formulate a framework for on-line detection of such changes in patients hospitalized with MI. The study included 195 patients who were hospitalized from 2002 to 2014. We found evidence that a change in the mean has the ability to distinguish between survivors and nonsurvivors. A threshold for raising an alarm was specified by plotting a receiver operating characteristic (ROC) curve and choosing the best combination of sensitivity and specificity. The method detected 64% of the patients who eventually died, while wrongly alerting for 25% of the survivors. Almost all of the true alarms were given in a time that would have been reasonable enough to address the cause of the disorder.