A thorough understanding of the techniques underlying prognostic modelling, the potential shortcomings of prognostic models, and their applicability to dermatology are important for both clinicians and researchers.
Objective:
This article distinguishes between prognostic and explanatory (causal) models, discusses some of the techniques used in developing a prognostic model, and addresses the importance of model generalizability.
Conclusion:
Prognostic models may be useful to the clinician, but must be used with care. It is important to critically appraise prognostic models and to assure that they are relevant to the population of interest. The ability to critically appraise prognostic models is predicated on a thorough understanding of the techniques used in their development and evaluation.
Get full access to this article
View all access options for this article.
References
1.
SennertD. The art of chirurgery. London, 1661.
2.
BoringCCSquiresTSTongT. Cancer statistics, 1994. CA Cancer J Clin1993; 43:7–26.
3.
BalchCMSoongSJMuradTM. A multifactorial analysis of melanoma. Surgery1979; 86:343–351.
4.
DayCLJrMihmMCJrLewRA. Prognostic factors for patients with clinical stage I melanoma of intermediate thickness (1.51–3.99 mm). A conceptual model for tumour growth and metastasis. Ann Surg1982; 195:35–43.
5.
SchuchterLSchultzDJSynnestvedtM. A prognostic model for predicting 10-year survival in patients with primary melanoma. Ann Intern Med1996; 125:369–375.
6.
RothmanKJGreenlandS. Modern epidemiology. 2nd Ed.Philadelphia: Lippincott-Raven, 1998.
7.
WatsonJHSoxHCNeffRK. Clinical prediction rules: Applications and methodological standards. N Engl J Med1985; 313:793–799.
8.
HosmerDWLemeshowS. Applied logistic regression. New York: John Wiley & Sons, 1989.
9.
TuJV. Advantages and disadvantages of using artificial neural networks versus logistic regression for predicting medical outcomes. J Clin Epidemiol1996; 49:1225–1231.
10.
Baxt”WG. Use of an artificial neural network for the diagnosis of myocardial infarction. Ann Intern Med1991; 115:843–848.
11.
PatilSHenryJWRubenfireM. Neural network in the clinical diagnosis of acute pulmonary embolism. Chest1993; 104:1685–1689.
12.
BreimanLFriedmanJHOlshenRA. Classification and regression trees. Belmont, CA: Wadsworth International, 1984.
13.
CrichtonNJHindeJPMarchiniJ. Models for diagnosing chest pain: Is CART helpful?Stat Med1997; 16:717–727.
14.
YarnoldPRSoltysikRCBennettCL. Predicting in-hospital mortality of patients with AIDS-related Pneumocystis carinii pneumonia: An example of hierarchically optimal classification tree analysis. Stat Med1997; 16:1451–1463.
15.
PiloteLMillerDPCaliffRM. Determinants of the use of coronary angiography and revascularization after thrombolysis for acute myocardial infarction. N Engl J Med1996; 335:1198–1205.
16.
SelkerHPGriffithJLPatilS. A comparison of performance of mathematical predictive methods for medical diagnosis: Identifying acute cardiac ischemia among emergency department patients. J Investig Med1995; 43:468–476.
17.
HosmerDWLemeshowS. Goodness of fit tests for multiple logistic regression model. Commun Statist Theor Math1980; 10: 1043–1069.
18.
HlatkyMACaliffRMHarrellFEJr. Clinical judgement and therapeutic decision making. J Am Coll Cardiol1990; 15:1–14.
19.
HlatkyMAMarkDBHarrellFEJr. Rethinking sensitivity and specificity. Am J Cardiol1987; 59:1195–1198.
20.
HarrellFEJrLeeKLMarkDB. Multivariable prognostic models: Issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med1996; 15:361–387.
21.
HarrellFEJrLeeKLCaliffRM. Regression modeling strategies for improved prognostic prediction. Stat Med1984; 3:143–152.
22.
ClarkeJRHaywardCZ. A scientific approach to surgical reasoning. I. Diagnostic accuracy—sensitivity, specificity, prevalence, and predictive value. Theor Surg1990; 5:129–132.
23.
ClarkeJRO'DonnellTFJr. A scientific approach to surgical reasoning. III. What is abnormal? Test results with continuous values and receiver operating characteristic (ROC) curves. Theor Surg1991; 6:45–51.
24.
LastJM, ed. A dictionary of epidemiology. 3d Ed.New York: Oxford University Press, 1995.
25.
Murphy-FilkinsRTeresDLemeshowS. Effect of changing patient mix on the performance of an intensive care unit severity-of-illness model: How to distinguish a general from a specialty intensive care unit. Crit Care Med1996; 24:1968–1973.
26.
ArkesHRDawsonNVSperoffT. The covariance decomposition of the probability score and its use in evaluating prognostic estimates. SUPPORT Investigators. Med Decis Making1995; 15:120–131.
27.
DolanJGBordleyDRMushlinAI. An evaluation of clinicians' subjective prior probability estimates. Med Decis Making1986; 6:216–223.
28.
YatesFJ. External correspondence: Decompositions of the mean probability score. Organizational Behavior and Human Performance1982; 30:132–156.
29.
RedelmeirerDABlochDAHickamDH. Assessing predictive accuracy: How to compare Brier scores. J Clin Epidemiol1991; 44:1141–1146.
30.
CampbellDTStanleyJC. Experimental design and quasi-experimental designs for research. Boston: Houghton Mifflin, 1963.
31.
ClarkWHJrElderDEGuerryD4th. Model predicting survival in stage I melanoma based on tumour progression. J Natl Cancer Inst1989; 81:1893–1904.
