Unsupervised machine learning method for indirect estimation of reference intervals for chronic kidney disease in the Puerto Rican population

Reference intervals (RIs) for clinical laboratory values are extremely important for diagnostics and treatment of patients. However, the determination of these ranges is costly and time-consuming. As a result, often different unverified RIs are used in practice for the same analyte and the same range is used for all patients despite evidence that the values are gender, age, and ethnicity dependent. Moreover, the abnormal flags are rudimentary, merely indicating if a value is within the RI. At the same time, clinical lab data generated in the everyday medical practice contains a wealth of information, that given the correct methodology, can help determine the RIs for each specific segment of the population, including populations that suffer from health disparities. In this work, we develop unsupervised machine learning methods, based on Gaussian mixtures, to determine RIs of analytes related to chronic kidney disease, using millions of routine lab results for the Puerto Rican population. We show that the measures are both gender and age dependent and we find evidence for normal age-related organ function deterioration and failure. We also show that the joint distribution of measures improves the diagnostic value of the lab results.

Androgens in women: Establishing reference intervals for dehydroepiandrostenedione sulphate and androstenedione on the Roche Cobas

Introduction

Immunoassays are the most common method in routine practice for measuring androgens in women. Study's aim was to establish new population specific indirect reference intervals (RI) for dehydroepiandrostenedione sulphate (DHEAS) and for new androstenedione test available on automated Roche Cobas electrochemiluminescent immunoassay method.

Materials and methods

From extracted laboratory records, testosterone, sex hormone binding globulin and follicle-stimulating hormone were used as reference tests to exclude possibly diseased women. After the data selection steps, the study included 3500 subjects for DHEAS and 520 for androstenedione aged 20-45 years. To evaluate the need for age partitioning, we calculated standard deviation ratio and bias ratio. For each hormone, 90% and 95% RIs were calculated with appropriate statistical method.

Results

Total age group (20-45 years) 95% RIs were: 2.77-11.50 µmol/L for DHEAS and 2.48-8.89 nmol/L for androstenedione. Age-stratified 95% RIs for DHEAS were: 3.65-12.76 µmol/L (20-25 years); 2.97-11.50 µmol/L (25-35 years) and 2.30-9.83 µmol/L (35-45 years). Age-stratified 95% RIs for androstenedione were: 3.02-9.43 nmol/L (20-30 years) and 2.23-7.75 nmol/L (30-45 years).

Conclusion

New RIs for DHEAS were slightly wider for age group 20-25 and 35-45, while the differences in the age group 25-35 years were more pronounced. Androstenedione RI showed significantly higher concentrations than the manufacturer's. Age-related decrease of androgens should be considered when calculating RIs. We propose population specific, age-stratified RIs for DHEAS and androstenedione on electrochemiluminescent method, which should improve test interpretation in women of reproductive age.

Harmonization of indirect reference intervals calculation by the Bhattacharya method

OBJECTIVES

The aim of this study was to harmonize the criteria for the Bhattacharya indirect method Microsoft Excel Spreadsheet for reference intervals calculation to reduce between-user variability and use these criteria to calculate and evaluate reference intervals for eight analytes in two different years.

METHODS

Anonymized laboratory test results from outpatients were extracted from January 1st 2018 to December 31st 2019. To assure data quality, we examined the monthly results from an external quality control program. Reference intervals were determined by the Bhattacharya method with the St Vincent's hospital Spreadsheet firstly using original criteria and then using additional harmonized criteria defined in this study. Consensus reference intervals using the additional harmonized criteria were calculated as the mean of four users' lower and upper reference interval results. To further test the operation criteria and robustness of the obtained reference intervals, an external user validated the Spreadsheet procedure.

RESULTS

The extracted test results for all selected laboratory tests fulfilled the quality criteria and were included in the present study. Differences between users in calculated reference intervals were frequent when using the Spreadsheet. Therefore, additional criteria for the Spreadsheet were proposed and applied by independent users, such as: to set central bin as the mean of all the data, bin size as small as possible, at least three consecutive bins and a high proportion of bins within the curve.

CONCLUSIONS

The proposed criteria contributed to the harmonization of reference interval calculation between users of the Bhattacharya indirect method Spreadsheet.

RIbench: A Proposed Benchmark for the Standardized Evaluation of Indirect Methods for Reference Interval Estimation

BACKGROUND

Indirect methods leverage real-world data for the estimation of reference intervals. These constitute an active field of research, and several methods have been developed recently. So far, no standardized tool for evaluation and comparison of indirect methods exists.

METHODS

We provide RIbench, a benchmarking suite for quantitative evaluation of any existing or novel indirect method. The benchmark contains simulated test sets for 10 biomarkers mimicking routine measurements of a mixed distribution of non-pathological (reference) values and pathological values. The non-pathological distributions represent 4 common distribution types: normal, skewed, heavily skewed, and skewed-and-shifted. To identify strengths and weaknesses of indirect methods, test sets have varying sample sizes and pathological distributions differ in location, extent of overlap, and fraction. For performance evaluation, we use an overall benchmark score and sub-scores derived from absolute z-score deviations between estimated and true reference limits. We illustrate the application of RIbench by evaluating and comparing the Hoffmann method and 4 modern indirect methods -TML (Truncated-Maximum-Likelihood), kosmic, TMC (Truncated-Minimum-Chi-Square), and refineR- against one another and against a nonparametric direct method (n = 120).

RESULTS

For the modern indirect methods, pathological fraction and sample size had a strong influence on the results: With a pathological fraction up to 20% and a minimum sample size of 5000, most methods achieved results comparable or superior to the direct method.

CONCLUSIONS

We present RIbench, an open-source R-package, for the systematic evaluation of existing and novel indirect methods. RIbench can serve as a tool for enhancement of indirect methods, improving the estimation of reference intervals.

Reference intervals of automated reticulocyte count and immature reticulocyte fraction in a pediatric population

INTRODUCTION

Reticulocytes are erythroid precursors that develop into mature erythrocytes, and they are an important tool to assess erythropoietic activity, as their count indicates the balance between the cells released from the bone marrow, their stage of maturity, and their rate of development into mature erythrocytes. Considering the described biological variability of the absolute reticulocyte count (ARC) and the immature reticulocyte fraction (IRF) and the limited information available on these hematological parameters in children, this study determined the reference intervals (RIs) of these parameters in a healthy pediatric population.

METHODS

A retrospective, observational, and analytical study was designed to establish RIs for the ARC and the IRF according to age and sex. An indirect sampling method was applied to a mixed database of complete blood counts from children aged 2 months to 18 years, using the truncated maximum likelihood indirect method for reference interval estimation. Percentiles were calculated to obtain bimodal RIs.

RESULTS

From a total of 190,812 samples, 6,814 were selected. Gender stratification was not necessary for the ARC and the IRF but they required partitioning into six and two age groups, respectively.

CONCLUSION

This study determined, by an indirect sampling method, RIs for the ARC and the IRF in a pediatric population according to age and sex.