gms | German Medical Science

49. Jahrestagung der Deutschen Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie (gmds)
19. Jahrestagung der Schweizerischen Gesellschaft für Medizinische Informatik (SGMI)
Jahrestagung 2004 des Arbeitskreises Medizinische Informatik (ÖAKMI)

Deutsche Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie
Schweizerische Gesellschaft für Medizinische Informatik (SGMI)

26. bis 30.09.2004, Innsbruck/Tirol

Use of baseline information in drug development – experience in industry

Meeting Abstract (gmds2004)

Suche in Medline nach

Kooperative Versorgung - Vernetzte Forschung - Ubiquitäre Information. 49. Jahrestagung der Deutschen Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie (gmds), 19. Jahrestagung der Schweizerischen Gesellschaft für Medizinische Informatik (SGMI) und Jahrestagung 2004 des Arbeitskreises Medizinische Informatik (ÖAKMI) der Österreichischen Computer Gesellschaft (OCG) und der Österreichischen Gesellschaft für Biomedizinische Technik (ÖGBMT). Innsbruck, 26.-30.09.2004. Düsseldorf, Köln: German Medical Science; 2004. Doc04gmds122

Die elektronische Version dieses Artikels ist vollständig und ist verfügbar unter:

Veröffentlicht: 14. September 2004

© 2004 Burger.
Dieser Artikel ist ein Open Access-Artikel und steht unter den Creative Commons Lizenzbedingungen ( Er darf vervielfältigt, verbreitet und öffentlich zugänglich gemacht werden, vorausgesetzt dass Autor und Quelle genannt werden.



Clinical studies today include a fair amount of information not only on treatment outcome but also on baseline medical conditions of patients before starting experimental treatment. Such baseline data are sampled for different reasons, e.g:

• To describe medical history for patients with unexpected severe adverse events

• To describe the patient population enrolled in the study in order to facilitate evaluation of study results in the context of existing literature

• To learn more about disease and important prognostic factors, especially with regard to anticipated treatment outcome

• To reduce variability of study endpoints by adjusting statistical models for prognostic factors

• To evaluate potential bias introduced in the overall study evaluation by imbalances between treatment groups in prognostic factors

• To evaluate consistency of treatment effect across various subgroups

The first three objectives require a relatively large amount of data to be collected compared to the quit limited amount of statistics used in the analysis. Physical examinations, ECG assessments, data on disease history and other data are sampled to be able to examine later the relationship to treatment of unexpected adverse reactions of patients. Baseline data to describe the treated population like medical staging of the disease, previous treatments and demographic parameters are usually summarized only descriptively to facilitate the evaluation of the studied patient population. Furthermore, in recent years biomarker and genetic parameters became more and more part of the regular sampling of baseline data as well. These data often form the basis of a sort of risk management in case the experimental treatment works only under certain conditions. This is especially of importance in case of disease targeted therapy.

For statisticians, prognostic parameters at baseline are usually of higher interest as a large part of exploratory analyses performed for the final analysis deal with these factors and their impact on study results. Prognostic baseline conditions by definition influence disease prognosis of a patient. Differences in disease prognosis between treatment groups can introduce bias into the study evaluation leading to false conclusions. Therefore, it is important to control such prognostic factors not only at baseline at the stage of randomization but to adjust the primary analysis of study outcome later as well for such factors. In case these parameters are carefully selected for their prognostic value, this would reduce variability of treatment outcome in the final adjusted analysis and by that increase the power of the study. Careful selection of these factors is however crucial and there is a history of studies in which factors selected for stratification at baseline turn out to be of no or little prognostic value at the analysis stage. Non-prognostic factors in the primary model do not increase but rather decrease the power, especially when using non-linear models like Cox regression models. The selection of such parameters is therefore an important exercise which should not be left over to medical colleagues.

Is the selection of prognostic factors for stratification and/or for the primary model already often a difficult exercise, then the post hoc selection of important factors impacting on study results later on is sometimes even a real dilemma. On the one hand, there is the wish to check all possible prognostic factors in the study population for their comparability between treatment groups and their impact on study results in order to rule out any potential bias in the study, on the other hand there is no accepted method available how to do this in an objective manner. One of many unsuccessful compromises is to leave the list of factors vague in the protocol and analysis plan but rather to specify the selection process in detail.

There is one further aspect of baseline information going beyond potential bias and power considerations. Is an application for a market authorization especially based on one single pivotal study, this study has to show a compelling and robust treatment effect consistent across various subgroups within the study. Latter is of specific interest when examining the validity of study outcome for the general disease population and when assessing how likely the experiment could be repeated with similar outcome. This leads to subgroup analyses with results displayed often in form of forest plots. Such analyses could strengthen and weaken the overall study conclusions depending on the heterogeneity of the treatment effect. In case of heterogeneity, study results may be questionable for the overall population. Extreme cases of heterogeneity can even lead to an inconclusive study.