13 Adding Value to Hematocrit Testing Using Analytics. (11th January 2018)

• Record Type: Journal Article
• Authors: Simonson, Paul; Mathias, Patrick; Sabath, Daniel
• Is Part Of: American journal of clinical pathology. Volume 149(2018)Supplement 1
• Page Start: S170
• Page End: S170
• Publication Date: 2018-01-11
• Subjects: Diagnosis, Laboratory -- Periodicals; Pathology -- Periodicals; 616.07
• DOI: 10.1093/ajcp/aqx149.382
• Languages: English
• ISSNs: 0002-9173

Abstract

Abstract: Diagnosis of blood loss is multifaceted and relies heavily on laboratory testing. By refining the analysis and reporting of common laboratory data, the laboratory can further help clinicians to diagnose and characterize blood loss in an efficient and timely manner. Hematocrit results are commonly available due to the multitude of indications for ordering. We first examined the hematocrit characteristics at a population level with simple descriptive statistics, stratified by patient setting. Our institution measured 358, 080 hematocrits during a one-year period. When plotted as a histogram, these values comprised a clearly bimodal distribution. Separating hematocrits by inpatient vs outpatient produced two unimodal distributions, with modes occurring at 26% and 39%, respectively. A hematocrit of 26% is the usual transfusion threshold for red cells at our institution. Because we have a relatively high number of inpatients who are transfusion dependent, the mode of the inpatient hematocrits is not surprising. Oncology, cardiology, OB/GYN, and the transplant services contributed most to low hematocrit values. Delta values, defined as the difference between a pair of sequential hematocrit values, are used by our laboratory to alert personnel to possible instrument and sampling errors, or actual large drops in patient hematocrits. We found that two standard deviations below the mean delta value (–8%) corresponded well with the threshold already used by our lab to raise flags. While useful, delta values fail to communicate the rate of change ("hematocrit velocity"), which can conceivably be used to identify dropping hematocrit before a large drop has occurred. Velocities can be simply calculated as v = (h2 – h1)/Δt. By default, most instruments report hematocrit in 1% increments; such reporting contributes significantly to velocity error for small time intervals, Δt. After eliminating time intervals <1 h, the velocity distribution, which was not distributed normally, was found to be centered nearly at zero with an interquartile range of –0.042%/h to 0.015%/h and a standard deviation of 0.27%/h (6.6%/day). Delta values and hematocrit velocities were correlated with transfusions to estimate probability of transfusion as a function of recent delta value or velocity. Transfusions most commonly occurred for hematocrits of 26%, delta values of –2%, and velocities of –0.1%/h. With big data comes big responsibility to maximize the utility of commonly measured values. We have demonstrated how characterization of hematocrits and derived values (delta values, velocities) can be used to set thresholds for flagging laboratory values, estimate rate of hematocrit loss, and correlate with subsequent transfusion. Further characterization of hematocrit based on patient population and refined derived values might be used to guide hematocrit measurement frequency and predicting need for specific transfusion blood products. Such questions will be the subject of future work.