Veli Kairisto 22.11.1995
University of Turku / Medical Faculty / Dept. of Clinical Chemistry
Annales Universitatis Turkuensis D203, 128 pages, 1995
Distributor of the Thesis: Turku University Library, FIN-20500 Turku, Finland
(fax: +358-21-3335050)

Abstract
List of original communications
Supervisors, reviewers and the opponent



Production of Reference Values from Patient Data

Abstract

Availability of reference values is the basic requirement for the clinical use of laboratory tests. Reference values produced from patients correspond closely to the needs of clinical test interpretation provided that the patients are reliably grouped into different diagnostic categories. Hospital database information has previously been used mainly for epidemiologic and administrative purposes and it has not been customary to use the data in patient care. This study was undertaken to study the possibilities of using routinely available computerized patient data for the production of reference values.

In the different parts of the study reference values were produced from patient data for the diagnosis of pheochromocytoma, for the diagnosis of myocardial damage, for the characterization of usual red blood cell values in hospitalized patients, and for the detection of random analytical or preanalytical errors in a multichannel hematology analyzer. By the used methods it was possible to obtain representative and large reference sample groups, which facilitated, for example, the observation of previously unreported significant dependencies of urinary normetanephrine excretion, and serum LD-1 and CK-MB activities on age.

Red blood cell values can be affected by several different diseases and treatments. A set of exclusion criteria based on hospital discharge diagnoses was developed. In the construction of quality control limits for the detection of random errors in a multichannel hematology analyzer it was not necessary to remove the illness-related values, on the contrary, those were considered necessary for the correct estimation of the quality control limits. Physiological interdependence of the red blood cell values was utilized for the detection of aberrant results. Using reference data obtained from the patients a bivariate reference region was constructed for the detection of most unusual combinations of results and possible random errors.

In the detection of myocardial damage, quite as in several other clinical uses of laboratory tests, it is important to observe not only single test results but also the changes between consecutively collected samples. Using the cardiac enzyme data as a model a new indirect method was developed for the estimation of reference change limits from routine patient data. Previously indirect methods have not been used for the estimation of reference changes. The proposed algorithm was included in the software developed during the study.

The availability of several sets of reference values produced from patient databases would make it increasingly important to understand and use properly such concepts as decision limit, clinical sensitivity and clinical specificity, predictive values, and ROC-curves. In the second part of the study a computer program with graphical user interface was constructed. Illustrative display of reference distributions forms the core of the software. Numerical data is not shown unless requested. Relevant numbers can be "picked up" from the graph by simple mouse operations. The program includes several previously published methods for the treatment of reference values. The optimization of bin widths for illustrative graphical presentation of reference distributions and ROC-curves and the indirect method for the estimation of health-related change limits are previously unpublished features developed during this study.

As a summary of this study it can be concluded that hospital databases can be used for the production of clinically useful reference values. However, the possibilities for reference value production are in some cases still severely limited due to insufficiencies in the computerized source data. Knowledge about this should encourage physicians to improve the contents of the databases further. Improved accuracy in the classification of hospitalized reference patients would benefit the clinical use of laboratory tests and the diagnostic process.



LIST OF ORIGINAL COMMUNICATIONS

The thesis was based on the following papers:

  1. Kairisto V, Koskinen P, Mattila K, Puikkonen J, Virtanen A, Kantola I, Irjala K. Reference intervals for 24-h urinary normetanephrine, metanephrine, and 3-methoxy- 4-hydroxymandelic acid in hypertensive patients. Clin Chem 1992; 38: 416-20.
  2. Kairisto V, Hänninen K-P, Leino A, Pulkki K, Peltola O, Näntö V, Voipio-Pulkki L-M, Irjala K. Generation of reference values for cardiac enzymes from hospital admission laboratory data. Eur J Clin Chem Clin Biochem 1994; 32: 789-796.
  3. Kouri T, Kairisto V, Virtanen A, Uusipaikka E, Rajamäki A, Finneman H, Juva K, Koivula T, Näntö V. Reference intervals developed from data for hospitalized patients: computerized method based on combination of laboratory and diagnostic data. Clin Chem 1994; 40: 2209-2215.
  4. Kairisto V, Kouri T, Rajamäki A, Virtanen A, Uusipaikka E, Näntö V. Quality control of multichannel hematology analyzers - bivariate on-line comparison of MCV and MCH values for the detection of random errors. Am J Clin Pathol 1992; 97: 645-651.
  5. Kairisto V, Virtanen A, Uusipaikka E, Voipio-Pulkki L-M, Näntö V, Peltola O, Irjala K. Method for determining reference changes from patients´ serial data: example of cardiac enzymes. Clin Chem 1993; 39: 2298-2304.
  6. Kairisto V, Poola A. Software for illustrative presentation of basic clinical characteristics of laboratory tests - GraphROC for Windows. Scand J Clin Lab Invest 1995; 55 (Suppl. 222): 43-60.


Supervised by

Associate Professor Veikko Näntö
M.D., Department of Clinical Chemistry, University of Turku
and
Associate Professor Esa Uusipaikka
Ph.D., Department of Statistics, University of Turku

Reviewed by

Associate Professor Aimo Ruokonen
M.D., Department of Clinical Chemist,y, University of Oulu
and
Consultant Per Winkel
M.D., Department of Clinical Biochemistry, University of Copenhagen

Opponent

Professor Ralph Gräsbeck
M.D., Department of Medical Biochemistry, University of Kuwait
and Department of Clinical Chemistry, University of Helsinki