Critical appraisal of two Box-Cox formulae for their utility in determining reference intervals by realistic simulation and extensive real-world data analyses.

BACKGROUND: The reference interval (RI) is defined as the central 95 % range of reference values (RVs) from healthy individuals. The ideal method for determining RIs is to transform RV distribution into Gaussian and estimate its 95 % range parametrically. One-parameter Box-Cox formula (1pBC) is widely used for correcting skewness (Sk) or kurtosis (Kt) in data distribution. However, 1pBC is not popular for computing RIs due to its unreliability in Gaussian transformation. While its two-parameter version (2pBC) is not used due to a challenge in fitting power (λ) and shift (α) parameters simultaneously. In this study, technical issues in fitting both formulae are assessed, and an alternative algorithm for successful use of 2pBC is proposed. METHODS: For fitting 1pBC, optimal λ was determined by stepwise linear search. For 2pBC, optimal [λ, α] combination was pursued in two ways: by grid search of λ and α (2pBCgrid) or by using the grid search but keeping α-range close to the reference distribution (2pBCopt). Their accuracy and precision in determining RIs were compared by generating power-normal distributions simulating RVs of 23 major chemistry analytes. Additionally, their practical utilities were compared by analyzing 776 real-world datasets comprising test results of 25 analytes that were obtained from the global multicenter RV study of IFCC. For comparison, the performance of nonparametric method was evaluated in both settings. RESULTS: For analytes with not-much-skewed distributions, unbiased estimation of RIs was accomplished by all methods. Nevertheless, when reference distributions are located far from zero, λ estimated by1pBC and 2pBCgrid fluctuated widely, which was attributable to virtually flat goodness-of-fit profile for [λ, α]. For highly skewed distributions, 1pBC caused bias in estimating RI and λ due to remotely peaked goodness-of-fit profile. Real-world data analyses revealed that 2pBCopt and 1pBC achieved Gaussian transformation (|Sk|<0.1 and |Kt|<0.3) in 82.4 % and 66.9 % among 776 datasets, respectively. Fitting bias signified by Kt<-0.4 was more common to 1pBC. The practical utility of 2pBCopt was unbiased prediction of analyte-specific distribution-shape (λ). Whereas nonparametric method gave highly variable RIs for both simulated and real-world datasets. CONCLUSIONS: 2pBCopt is suitable for calculating RIs and grasping distribution-shape from the estimated λ.

[Accuracy of Gender Distinction Based on Basic Laboratory Test Results: Comparison between Conventional Logistic Regression Analysis and a Weighted Log-Likelihood Estimation].

Feasibility of gender distinction based on laboratory tests was explored by employing test results for 21 basic analytes which were obtained from 1,944 healthy Japanese (males = 856; females = 1,088: age = 20-64) in a multicenter reference interval (RI) study conducted in 2009. Two methods for the distinction were examined. One was logistic regression analysis with stepwise selection of analytes representing gender differences. The estimated probability (EP) of the sample's being from a female was calculated from the regression model. The other was the weighted-average log-likelihood (WALL) method, which is based on matching of an arbitrary set of lab test results with those of a given diagnostic category. In this study, WALLs belonging to distributions of male test results were computed with SDR (SD representing gender-difference divided by SD comprising RI) set as the weight for each analyte. WALL was computed by inclusion of either all analytes(WALL1), those with SDR ≥ 0.3(WALL2), or SDR ≥ 0.5(WALL3). The performance of EP and WALLs was evaluated as accuracy (% correct distinction) and area under the ROC curve (AUC). Because SDR decreased with age 46, EP and WALLs were computed for all-ages, age ≤ 45 and age ≥ 46. Analytes chosen in the logistic model were not necessarily those with high SDR. In all-age analysis, accuracy and AUC for EP vs. WALL1 were 94.2% and 0.986 vs. 91.1% and 0.973; respectively; in age ≤ 45 analysis, 95.1% and 0.990 vs. 93.2% and 0.981; in age 46 analysis, 92.6%, and 0.980 vs. 90.5% and 0.973. WALL2 showed slightly better performance in any age group. WALL analysis showed a little less performance than the conventional logistic regression analysis, but it has superior properties for the binary categorization with its nonparametric nature and flexibility in the selection of test items.

Evaluating the benefit of introducing medical clerks as transcriptionists to assist physicians in outpatient clinics: a quantitative analysis of medical records by counting characters.

This study evaluates the effects of the medical clerks introduced to reduce physicians' workloads in outpatient clinics by assisting with their documentation processes (e.g., the production of electronic medical records (EMRs)). The volume of information written in narrative text in EMRs from 2007 (pre-introduction of medical clerks) to 2012 (post-introduction) was measured by counting Japanese characters. The total number of medical records for analysis was 1,577. The average number of characters in EMRs increased from before the introduction of medical clerks to afterwards regardless of the types of documents (subjective or objective data) or visits (first or second visits). We conclude that introducing medical clerks improves the quantity of outpatients' medical records and that such a character-counting method is useful for evaluating the benefit of the introduction of medical clerks to assist physicians.

A novel weighted cumulative delta-check method for highly sensitive detection of specimen mix-up in the clinical laboratory.

BACKGROUND: We sought to detect specimen mix-up by developing a new cumulative delta-check method applicable to a mixture of test items with heterogeneous units and distribution patterns. METHODS: The distributions of all test results were successfully made Gaussian using power transformation. Values were then standardized into z-score (zx) based on reference interval (RI) so that limits of RI take zx=±1.96. To find a weight for summing absolute value of delta between current and previous zx (Dz), we evaluated the distribution of Dz. Its central portion was always regarded as Gaussian despite the presence of symmetrical long tails. Thus, an adjusted SD (aSD) representing the center was estimated with an iterative method. By setting 1/aSD2 as a weight factor, we computed a weighted mean of Dz as an index for specimen mix-up (wCDI). RESULTS: The performance of wCDI was evaluated, using a model laboratory database consisting of 32 basic test items, by a simulation study generating artificial cases of mix-up. When wCDI was computed from three commonly ordered test sets consisting of 6-9 items each, its diagnostic efficiency in detecting the artificial cases was 0.937-0.967 expressed as area under ROC curves (AUC). When the performance of wCDI was evaluated simply by the number of test items (p) included in the computation, AUC gradually increased from 0.944 (p=5) to 0.976 (p=8). However, when p≥10, AUC stayed at approximately 0.98. CONCLUSIONS: wCDI was proven to be highly effective in uncovering cases of specimen mix-up. The diagnostic efficiency of wCDI depends only on the number of test items included in the computation.