Patient-specific analysis of sequential haematological data by multiple linear regression and mixture distribution modelling
✍ Scribed by C. E. McLaren; E. L. Kambour; G. J. McLachlan; H. C. Lukaski; X. Li; G. M. Brittenham; G. D. McLaren
- Publisher
- John Wiley and Sons
- Year
- 2000
- Tongue
- English
- Weight
- 120 KB
- Volume
- 19
- Category
- Article
- ISSN
- 0277-6715
No coin nor oath required. For personal study only.
✦ Synopsis
Automated storage and analysis of the results of serial haematologic studies are now technically feasible with present-day laboratory instruments and devices for data storage and processing. In current practice, physicians mentally compare a laboratory result with previous values and use their clinical judgement to determine the signiÿcance of any change. To provide a statistical basis for this process, we describe a new approach for the detection of changes in patient-speciÿc sequential measurements of standard haematologic laboratory tests. These methods include hierarchical multiple regression modelling, with a weighted minimum risk criteria for model selection, to choose models indicating changes in mean values over time. This study is the ÿrst to analyse sequential patient-speciÿc distributions of laboratory measurements, utilizing mixture distribution modelling with systematic selection of starting values for the EM algorithm. To evaluate these statistical methods under controlled conditions, we studied 11 healthy human volunteers who were depleted of iron by serial phlebotomy to iron-deÿciency anaemia, then treated with oral iron supplements to replete iron stores and correct the anaemia. Application of sequential patient-speciÿc analyses of haemoglobin, haematocrit, and mean cell volume showed that signiÿcant departures from past values could be identiÿed, in many cases, even when values were still within the population reference ranges. Additionally, for all subjects sequential alterations in red blood cell volume distributions during development of iron-deÿciency anaemia could be characterized and