PubMed HealthSearch

PubMed · 1525263

Robust bivariate errors-in-variables regression and outlier detection.

Abstract

A bivariate regression model is introduced where both variables are subject to error. The structural regression line is equivariant against interchanging coordinates and permits bivariate calibration, i.e. the prediction of one variable by means of the other. Maximum likelihood and robust parameter are estimated, based on order statistics. Residual analysis and outlier detection are performed. The model is applied to the comparison of clinical chemical analytical methods.

Explore related subjects

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

U Feldmann. 1992. Robust bivariate errors-in-variables regression and outlier detection.. https://doi.org/10.1515/cclm.1992.30.7.405

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related citations

A three-dimensional analysis of plasma skimming at microvascular bifurcations.

This paper analyzes an important underlying mechanism for the discharge hematocrit reduction observed in microvessels, which refers to the plasma skimming from the cell-free layer near the parent tube wall in the presence of a side branch. The three-dimensional theory recently developed by the authors (Yan et al., 1991, J. Fluid Mech., in press) for treating the simple shear flow past a side branch tube in a plane wall with suction is first summarized and then extended to treat T bifurcations from parent vessels with an upstream Poiseuille flow. For unequal vessel bifurcations, a fundamental new dimensionless group, Q = 1/8(qb/qp)(Rp/Rb)3, is derived whose value determines the shape of the upstream capture tube of the plasma phase, when the partitioning qb/qp of the flow into the side branch and the ratio Rp/Rb of the radii of the parent and side branch vessels are varied. Closed form expressions are then presented for the three-dimensional fluid capture tube shape upstream of the bifurcation which are valid when Q greater than 1 or Q less than 0.2. Based on this theory and its modification for an upstream Poiseuille velocity profile, the separating surface shape, the critical minimum fractional flux for incipient cell capture, and the discharge hematocrit defect and its dependence on the flow rate are predicted. It is shown, furthermore, that for flows typical of the microcirculation, a single dimensionless number, P = 3 pi Q(Rb/gamma 2), with gamma being the cell-free layer thickness, can be defined whose value determines the discharge hematocrit defect that arises from plasma skimming. The minimum critical flow rate for any red cells to enter the side branch is then given by the criterion P = 1. Although this theory does not account for the cell screening effect arising from the hydrodynamic interaction between the cells and the tube walls, it leads to predictions which exhibit the same trends as the experimental observations and is able to explain the results of several seemingly contradictory microvascular experiments that have puzzled investigators in recent years.

Hematocrit

Apparent polycythaemia.

Patients with apparent polycythaemia are characterised by a raised packed cell volume (PCV; males above 0.51, females above 0.48) but normal red cell mass (RCM; less than 25% greater than predicted). Prediction and interpretation of RCM and PV should be based on height and weight, since the use of body weight alone is misleading. Patients with PCV values up to 0.60 may have apparent polycythaemia but only 18% have a reduced PV (relative polycythaemia). Therefore, the most common cause of the raised PCV is a change in RCM and/or PV within their normal ranges. The clinical associations and possible causes for the RCM/PV changes include male sex, obesity, dehydration, diuretics, smoking, hypertension, alcohol, arterial oxygen desaturation, renal disease and increased catecholamine levels. Retrospective studies of patients with apparent polycythaemia and information from other groups of polycythaemic patients suggest an increased risk of vascular occlusion, although other factors, such as hypertension and smoking, are also involved. Proposed management includes modification of possible underlying causes and examination for risk factors for vascular occlusion. In patients with PCV levels chronically above 0.54 venesection should be used, but patients with PCV values below this level should only be venesected if they are considered to be at risk of vascular occlusion. The suggested target value for PCV for venesected patients is 0.45 or below.

Hematocrit