LAUSR

Unauthorized reproduction of this article is prohibited Pediatric Critical Care Medicine www.pccmjournal.org 503 In this issue of Pediatric Critical Care Medicine, Hassan et al (1) report on their ad hoc analysis of hemoglobin (Hb) levels collected as part of the PROphylaxis against ThRombosis prACTice (PROTRACT) study (2). This report benefits from the original’s multinational cross-sectional setting, including 59 PICUs in the United States, Canada, Spain, Portugal, Singapore, Australia, and New Zealand. Of the 2,900 patients included in the PROTRACT study, 2,389 were included due to available Hb values. Data were collected for Hb within the previous 7 days, but for RBC transfusions only within the previous 24 hours. Because chronology of Hb levels and transfusion data were not available, transfusion thresholds could not be induced. Although the median Hb for the entire population was 11 g/dL, the main findings of interest were the associations between geography, Hb levels, and RBC transfusion status. Patients in Spain and Portugal combined were more likely to have higher Hb levels than in other geographic regions, in both univariate and multivariate analyses. RBC transfusions were also more likely to occur in patients in Spain and Portugal (31.1% vs 13.8%, 6.9%, and 11.7% in Singapore/ Australia/New Zealand, Canada, and the United States, respectively, p < 0.001). Whether this was due to differences in medical conditions could not be determined. Many patients were in more than one diagnosis group, rendering “diagnosis group” unable to be included as a variable in a regression model. How do the findings of Hassan et al (1) fit into the current literature? To answer this, we must consider the study population. For most epidemiologic studies, the patient population is straightforward. Period prevalence studies like those by Demaret et al (3) and Bateman et al (4) evaluate all the patients within a specified time period. Bateman et al (4) further limited their population to those with PICU length of stay (LOS) longer than 48 hours. These study populations are readily understood by clinicians. A point prevalence study with a single measurement also represents a simple population. All the patients present at the time of study are evaluated, but the likelihood of finding a particular patient is proportional to their PICU LOS. In other words, patients with longer LOS are more likely to be found in a point prevalence study than those with shorter LOS, assuming random distribution. With repeated measurements, as in the current report, the math becomes more complicated. The calculations to follow are not readily found in either textbooks or the literature. Consequently, we will dwell briefly on this topic here. We set about determining mathematically the effects of patient LOS and repeated sampling on the probability of including an individual patient in a point prevalence study. For purposes of calculation, we assumed that 1) LOS was randomly distributed and 2) no patient stay was sampled more than once. The second assumption would hold true if assessments were repeated at intervals greater than the maximum LOS in the PICU. The total probability that an individual patient would be present in the PICU during one of repeated measurements can be described generically by this equation: