Download PDF
Methods Inf Med 1988; 27(03): 125-132
DOI: 10.1055/s-0038-1635532
Original Article
Schattauer GmbH

Three Graphic Representations to Aid Bayesian Inference

Drei Darstellungen in graphischer Form zur Unterstützung des Bayesschen Schlusses
W. G. Cole
1   (From the Section on Medical Information Science, University of California at San Francisco, USA)
› Author Affiliations
Further Information

Publication History

Publication Date:
17 February 2018 (online)

    Permissions and Reprints

Summary

Visual representation may help physicians and patients interpret laboratory results, for example by aiding Bayesian reasoning. This paper is concerned with the psychological and formal properties of such visual representations. One popular way to present laboratory results is via signal detection curves. These curves represent many parameters of a laboratory test including parameters, such as distribution variance, that are not typically known. Such curves can be seriously misleading.

Two alternative representations are suggested. Probability maps represent only the three laboratory test parameters most likely to be known: sensitivity, specificity and prevalence, and thus avoid the problems of the richer signal detection curves. Probability maps, however, do not remind the user of why there are false positives and false negatives nor of the nature of the criterion for positivity. Detection bars, a third type of representation, are a compromise between signal detection curves and probability maps.

Die visuelle Darstellung kann Ärzten und Patienten bei der Interpretation von Laboratoriumsergebnissen behilflich sein, zum Beispiel durch die Unterstützung Bayesscher Schlußfolgerungen. Diese Arbeit behandelt die psychologischen und formalen Eigenschaften solcher visueller Darstellungen. Eine populäre Form der Darstellung von Laboratoriumsergebnissen ist via Signaldetektionskurven. Diese Kurven stellen viele Parameter eines Labortests dar einschließlich Parameter wie Verteilungsvarianz, die nicht allgemein bekannt sind. Solche Kurven können ernstlich irreführend sein.

Zwei alternative Darstellungen werden vorgeschlagen. Wahrscheinlichkeitskarten stellen nur die drei Labortestparameter dar, die mit größter Wahrscheinlichkeit bekannt sind: Sensitivi-tät, Spezifität und Prävalenz und umgehen so die Probleme reichhaltigerer Signaldetektionskurven. Jedoch erinnern Wahrscheinlichkeitskarten den Benutzer weder daran, warum es falschpositive und falschnegative Ergebnisse gibt, noch an den Charakter des Kriteriums für Positivität. Messbalken, eine dritte Form der Darstellung, stellen einen Kompromiß zwischen Signaldetektionskurven und Wahrscheinlichkeitskarten dar.

Key-Words:

Bayesian Inference - Diagnosis - Graphics - Laboratory Testing - Visual Representation - Computer-aided Decision

Schlüssel-Wörter:

Bayesscher Schluß - Diagnose - Graphik - Labortests - visuelle Darstellung - computerunterstützte Entscheidung
 

  • REFERENCES

  • 1 Beck J. R, Shultz E. K. The Use of Relative Operating Characteristic (ROC) Curves in Test Performance Evaluation. Arch. Path. Lab. Med 1986; 110: 13-20.
    PubMedGoogle Scholar
  • 2 Casscells W, Shoenberger W, Grayboys T. Interpretation by Physicians of Clinical Laboratory Results. N. Engl. J. Med 1978; 299: 999-1000.
    CrossrefPubMedGoogle Scholar
  • 3 Cleveland W. S. The Elements of Graphic Data. Monterey, California: Wadsworth; 1985
    Google Scholar
  • 4 Cleveland W. S, McGill R. Graphical Perception: Theory, Experimentation, and Application to the Development of Graphical Methods. J. Amer. Stat. Ass 1984; 79: 531-554.
    CrossrefPubMedGoogle Scholar
  • 5 Green D. M, Swets J. A. Signal Detection Theory and Psychophysics. New York: Wiley; 1966
    Google Scholar
  • 6 Greenes R. A. Interactive Microcomputer-Based Graphical Tools for Physician Decision Support. Med. Dec. Making 1983; 3: 15-21.
    CrossrefPubMedGoogle Scholar
  • 7 Greenes R. A. Computer-Aided Diagnostic Strategy Selection. Rad. Clinics of N. Amer 1986; 24: 105-120.
    PubMedGoogle Scholar
  • 8 Gullick H. D, Schauble M. K. SD Unit System for Standardized Reporting and Interpretation of Laboratory Data. Am. J. Clin. Path 1972; 57: 517-525.
    CrossrefPubMedGoogle Scholar
  • 9 Johnson-Laird P. A. Mental Models. Cambridge Press; 1983
    Google Scholar
  • 10 Politser P. E. Explanations of Statistical Concepts: Can They Penetrate the Haze of Bayes? Meth. Inform. Med 1984; 23: 99-108.
    PubMedGoogle Scholar
  • 11 Pollack M. A, Greenes R. A. A Pictorial Simulation Construction Kit for Enhancing Knowledge-Based Learning. In Salamon R, Blum B, Jørgensen M. (Eds.) MEDINF0 86. Amsterdam: Elsevier; 1986: 887-890.
    Google Scholar
  • 12 Rossing R. G, Hatcher W. E. A Graphic Method for the Evaluation of Diagnostic Tests. Meth. Inform. Med 1980; 19: 149-156.
    PubMedGoogle Scholar
  • 13 Sisson J. C, Schoomaker E. B, Ross J. C. Clinical Decision Analysis: The Hazard of Using Additional Data. J. Amer. Med. Ass 1976; 236: 1259-1263.
    CrossrefPubMedGoogle Scholar
  • 14 Swets J. A, Picket R. M. Evaluation of Diagnostic Systems. New York: Academic Press; 1982
    Google Scholar
  • 15 Tversky A, Kahneman D. Evidential Impact of Base Rates. In Kahneman D, Slovic P, Tversky A. (Eds.) Judgment under Uncertainty: Heuristics and Biases. Cambridge Press; 1982: 153-160.
    Google Scholar