CA -- eLetters for Eysenbach, 53 (6) 356-371

QUICK SEARCH:

	Author:		Keyword(s):
Year:		Vol:		Page:

Electronic Letters to:

ARTICLES:
Gunther Eysenbach
The Impact of the Internet on Cancer Outcomes
CA Cancer J Clin 2003; 53: 356-371 [Abstract] [Full text] [PDF]

eLetters: Submit a response to this article

Electronic letters published:

In Reply: Gunther Eysenbach (7 June 2004)

Evaluating the Impact of Information on Cancer Outcome is Possible: Christian Weissenberger M.D., Sinje Jonassen, David Mueller, Jan Beranek-Chiu, Marcus Neumann, Gerlo Witucki (13 May 2004)

In Reply 7 June 2004

Gunther Eysenbach
Toronto General Hospital, Canada
Send letter to journal:
Re: In Reply

geysenba{at}uhnres.utoronto.ca Gunther Eysenbach

Dr. Weissenberger “totally disagrees” with my statement that the impact of the Internet on cancer outcomes cannot be proven in a randomized trial, only to concede later that a “randomization into Internet users and non-users is – for obvious reasons – impossible”, which was exactly my point.
He then continues to describe a non-randomized cohort study, which is actually what I advocated at the end of my article when I mentioned longitudinal studies. I did however also warn that results from non- randomized studies need to be adjusted to eliminate the influence of confounders, such as age, education, and socio-economic status (SES). These variables are associated with both Internet use (as the first part of my review showed) and also with cancer outcomes. Thus, they are obvious confounders. It is therefore misleading to show an unadjusted survival curve and suggest that “Internet use appears to be a prognostic factor for better overall survival”. One could probably also show an association between using a word processor and better cancer outcomes, much as people have shown an association between storks nests and birth rates as an example for confounding. Without adjusting for confounders such as baseline morbidity (very sick patients are unlikely to use the Internet and are likely to die sooner), sex and age (younger age being associated with both computer/Internet use and better survival) this figure is meaningless. As a minimum, Weissenberger should have stratified his population into groups with similar baseline morbidity, age and SES and tested whether within these groups Internet use makes a difference. Better yet, a multivariate regression model allows to simultaneously adjust for multiple confounding variables such as morbidity, cancer type, SES and demographic variables, and show the independent role of Internet use. This analysis is easy to do and there is no excuse for showing an unadjusted analysis.
Even if in the adjusted analysis the association between Internet use and better survival remains, one may still wonder whether Internet use is the cause or just a “symptom” of another underlying factor that leads to better outcomes, e.g. another psychosocial confounder (such as “fighting spirit”) that is also associated with increased Internet use. I do not want to be misunderstood – I am not arguing that Internet use has no influence on outcomes – on the contrary, my article makes a case for the opposite. However, this is not easy to show, and the letter of Weissenberger amply demonstrates one of the many fallacies.
Gunther Eysenbach, MD MPH Associate Professor Department of Health Policy, Management and Evaluation, University of Toronto; Senior Scientist, Centre for Global eHealth Innovation, Division of Medical Decision Making and Health Care Research; Toronto General Research Institute of the UHN, Toronto General Hospital, Canada

Evaluating the Impact of Information on Cancer Outcome is Possible 13 May 2004

Christian Weissenberger M.D.,
Radiooncologist
Dept. of Radiotherapy, University Hospital of Freiburg (Germany),
Sinje Jonassen, David Mueller, Jan Beranek-Chiu, Marcus Neumann, Gerlo Witucki
Send letter to journal:
Re: Evaluating the Impact of Information on Cancer Outcome is Possible

weissenb{at}mst1.ukl.uni-freiburg.de Christian Weissenberger M.D., et al.

Submitted: 12/3/03
Published: 4/14/04
To the Editor,

In his recent article in CA: Cancer Journal for Clinicians, Dr. Eysenbach gave an excellent summary of the current literature on Internet use and cancer disease. He emphasized that the Internet is likely to have substantial impact on patients' behavior, attitudes and conditions. However, although addressed in the title of his paper, the major question asked by patients and physicians remained unanswered: Do Internet-informed cancer patients live longer than others? Aware of the potential bias of - doubtlessly existing - confounders like education and socioeconomic status, the author is skeptical that the impact of the Internet use on cancer treatment outcome could ever be proven in a randomized trial.

In this point we totally disagree with the author. Modern treatment concepts in oncology take into account that treatment outcome is modulated by many factors including certain tumor properties like size or grading, the overall performance status of a patient, patterns of behavior, psychosocial parameters and many others (1). And among these factors the factor information becomes more and more important. This raises the question how to design a study evaluating the impact of Internet use on cancer treatment outcome. Confounders, mentioned by the author, are more or less confounders in every oncological study (2) and not typical for studies addressing the role of Internet use. In general certain preconditions have to be fulfilled to rule out any bias based on these confounders: First, - like in every clinical study - the patient cohort has to be as homogenous as possible, as large as possible and the follow-up interval has to be long enough to minimize the number of censored data in the survival analysis. Second, Internet use can only alter treatment outcome if it leads to an altered patient behavior like demanding new drugs (e.g. Trastuzumab in adjuvant setting), selection of different physicians or termination of harmful behavior (e.g. smoking). Therefore, Internet-use-dependent changes in patient behavior and its impact on treatment decisions should be considered as a primary study endpoint, while treatment outcome will serve as secondary endpoint. Finally, because a valid randomization into Internet users and non-users is - for obvious reasons - impossible, non-randomized studies on level of evidence 2a (according to principles of Evidence-based Medicine) should be performed to answer the urging question whether a well informed patient will have a survival benefit or not.

Using this approach we have already recruited more than 1,400 cancer patients from a single institution, which were characterized by both cancer specific and psychosocial data. After a median follow-up of three years, use of the Internet appears to be a prognostic factor for better overall survival (Fig. 1). Although long-term follow-up and multivariate testing have to be performed to validate this preliminary data, our study suggests that the impact of Internet use should not be underestimated in clinical praxis.

Survival analysis of Internet users and non-users (Kaplan-Meier)

Fig. 1. Survival rates of Internet users and non-users were evaluated by using Kaplan-Meier methods (p = 0.11). Chi-square tests (data not shown) confirmed equal distribution of tumor stages between both groups.

References

1. Lee SJ, Earle CC, Weeks JC. Outcomes research in oncology: history, conceptual framework, and trends in the literature. J Natl.Cancer Inst. 2000;92:195-204.
2. Deeks JJ, Dinnes J, D'Amico R, Sowden AJ, Sakarovitch C, Song F et al. Evaluating non-randomised intervention studies. Health Technol.Assess. 2003;7:iii-173.