Data mining for prothrombin time and international normalized ratio reference intervals in children

Reference intervals (RIs) help physicians in differentiating healthy from sick individuals. The prothrombin time (PT) and International normalized ratio (INR) fluctuate in coagulation pathway defects and have interlaboratory variability due to the instrument/reagent used. As direct method is difficult in children, we chose an indirect data mining method for the determining PT/INR RIs. The indirect method overcomes the substantial financial and logistic challenges, and ethical restrictions in children, moreover, allows partitioning in more fine-grained age groups. Prothrombin Time/INR measurements performed in patients aged birth-18 years between January 2013 and December 2020, were retrieved from laboratory management system of the Aga Khan Hospital. Reference intervals were computed using an indirect KOSMIC algorithm. The KOSMIC package function on the assumption that the non-pathologic samples follow a Gaussian distribution (after Box-Cox transformation of the data), following an elaborate statistical process to isolate distribution of physiological samples from mixed dataset. A total of 56,712 and 52,245 values were retrieved for PT and INR respectively. After the exclusion of patients with multiple specimens obtained during the study period, RIs were calculated for 37,356 (PT) and 37,192 (INR) children with stratification into 9 age groups. A comparison of 2.5th and 97.5th percentile results with those of established RIs from SickKids Handbook of Pediatric Thrombosis and Hemostasis demonstrated good agreement in between different age groups. This study supports data mining as an alternate approach for establishing PT/INR RIs, specifically in resource-limited settings. The results obtained are specific to studied population and instrument/reagent used. The study also allows understanding of fluctuations in coagulation pathways with increasing age and hence better clinical decision-making based on PT and INR results.