ABSTRACT
Introduction: Indicators that assess relationships among leukocytes may inform more and/or earlier than those measured in isolation. Method(s): Blood leukocyte differential counts collected from 101 Mayo Clinic COVID-19 patients were related to later outcomes following two approaches: (i) as unstructured data (e.g., lymphocyte percentages) and (ii) as data structures that assess intercellular interactions. Analyzing the same primary data, it was asked whether information contents differed among methods and/or when two sets of structured indicators are used. Result(s): While unstructured data did not distinguish survivors from non-survivors (Fig. 1, rectangle A), one data structure (here identified with letters expressed in italics) exhibited one perpendicular inflection that differentiated two patient groups (B). Two survivor-related observations were also distinguished from the remaining data points (B). A second data structure also revealed a single line of observations and a perpendicular data inflection (C), while more (four) patient groups were identified (D). Four validations were conducted: (i) increasing mortality levels among contiguous data subsets (0, 7.1, 16.2, or 44.4%) suggested construct validity (D);(ii) internal validity was indicated because 22 of the 45 survivors detected by the first data structure were also captured by the second one;(iii) the analysis of patients that differed in address, co-morbidities and other aspects supported external validity;and (iv) quasi non-overlapping data intervals predicted statistical validity (E, F). The structured approach also uncovered new and/ or dissimilar information: different leukocyte-related ratios explained the clusters identified in these analyses (E, F). Conclusion(s): Structured data may yield more information than methods that do not assess multicellular interactions. Possible applications include daily, longitudinal, and personalized analysis of hospital data.