|
Advertisement |
|
|
||
 |
|
|||||
|
|||||
|
 |
||||||
|
||||||
|
||||||
|
© 2003 American Society for Clinical Oncology
Effectiveness of Adjuvant Fluorouracil in Elderly Colon Cancer Patients: The Internal and External Validity of Nonrandomized Research Design
1 Istituto di Ricerche Farmacologiche "Mario Negri", Milan, Italy To the Editor: The effectiveness of several therapeutic interventions in clinical practice, although suggested by well-conducted randomized controlled trials (RCTs) and systematic reviews and meta-analyses, is often taken as granted but it is seldom demonstrated. This is particularly true for the case of most chemotherapy regimens, as RCTs are carried out in selected samples, rarely reflecting the real-world setting. It has well documented, for example, that patients 65 years of age or older are underrepresented in cancer-treatment trials.1 In addition, most of the new anticancer drugs (or new combinations of old drugs) seem to offer limited advantages over existing preparations, at least in terms of survival,2 thus introducing the question whether the general population of cancer patients not involved in clinical trials would gain from the small benefit documented in the RCTs. Such an issue has become more relevant, as the current Food and Drug Administration (FDA) and European Medicinal Evaluation Agency (EMEA) attitude to anticipate an earlier than ideal point along the drug approval path may lead to the marketing of drugs that are not effective or that are not safe. For these reasons, the article by Iwashyna and Lamont3 is welcomed, as it gives empirical evidence about the effectiveness of adjuvant fluorouracil (FU)-based regimens in elderly patients with stage III colon cancer. The value of the article also relies on the fact that it is the result of well-applied statistical techniques on population-based cohort data from administrative databases, thus ensuring the external validity of the findings produced. Similar exercises were recently published by others for non–small-cell lung cancer and coronary artery disease4,5 and are to be interpreted in the context of the controversial issue of the value of observational studies for assessment of treatment effect.6 As a matter of fact, authors have applied state-of-the-art of methods now available to take into account, by design or by statistical adjustment, the selection bias introduced by the observational nature of the data. Nonetheless, a few points pertaining both the internal and external validity may be raised. Briefly, the authors produced a propensity score for FU treatment using multivariable logistic regression analysis: A full, nonparsimonious model with 67 preselected variables was developed to produce the probability that a patient would be receiving the drug, and then such probabilities were used to perform a matched analysis to estimate the effect on survival. The "internal" validity of the approach was supported by the value of the c statistic that describes the discrimination of the model (0.83) and by sensitivity analyses that confirmed the robustness of results when a few relevant variables, such as comorbidity, were intentionally omitted by the predictive models (with a change of the hazard ratio from 0.73 to 0.59). Both figures, the c value and the change in the estimated effect of FU after excluding comorbidity, indicate a particular caution in the interpretation of the findings. First, as there is not single universally accepted measure of the performance of a model, usually several statistics are estimated and reported, with the c statistic being only one of those recommended. A reader would be more confident about the results’ validity if indicators of concordance and discordance (such as the Somers’ D) and other summary statistics such as the R2-type and the goodness-of-fit measures were also given.7 It is unlikely that a model with 67 variables produces a c statistic much lower than that one reported by the authors. Second, a 20% change in the estimated benefit of FU when omitting just one variable, although important in such a population for its direct and indirect effect on survival, indicates that the model is not immune to the action of confounders or that bias is not taken into account in their analysis. The magnitude of the effect of removing comorbidity is compatible with the magnitude of the confidence intervals of the hazard ratio estimated by the full model. As to the external validity of the findings, it should be mentioned that the generalizability of results is assured only for the United States context, as the propensity approach is based on the hypothesis that it balances only for the covariates that were used to construct the score.8 This means that, for example, present results cannot be applied to the European context, where other determinants are probably involved in the physicians’ and patients’ discussions of whether to use or not to use adjuvant FU after curative surgery. Finally, in the discussion, the authors report that, "results suggest that adjuvant 5-FU would have benefited those untreated patients in the sample." We think that this statement, which has important implications for community physicians, should be more supported by data and findings, as what shown is not enough to support such a conclusion. To be able to agree with the authors’ point of view on this particular aspect, we would need to know the comparison of survival (ie, the hazard ratio) across the strata based on the propensity to receive chemotherapy, from the lowest to the highest probabilities to receive chemotherapy. An example of such a way to evaluate the potential effectiveness of the treatment within strata of patients who are similar in terms of covariates is present in the literature.4 Only the stratification of patients based on their propensity to receive chemotherapy will show whether patients with less propensity for treatment appear to realize the same benefits as those with greater propensity. In conclusion, this study has added further evidence about the value of outcome research in oncology but has also given another example of the dependence of observational approaches on statistics. REFERENCES
1. Hutchings LF, Unger JM, Crowley JJ, et al: Underrepresentation of patients 65 years and older in cancer-treatment trials. N Engl J Med 341:2061–2067, 1999
2. Garattini S, Bertelè V: Efficacy, safety and cost of new anticancer drugs. Br Med J 325:260–271, 2002
3. Iwashyna TJ, Lamont EB: Effectiveness of adjuvant fluorouracil in clinical practice: A population-based cohort study of elderly patients with stage III colon cancer. J Clin Oncol 20:3992–3998, 2002
4. Earle CC, Tsai JS, Gelber RD, et al: Effectiveness of chemotherapy for advanced lung cancer in the elderly: Instrumental variable and propensity analysis. J Clin Oncol 19:1064–1070, 2001 5. Gum PA, Thamilasaran M, Watanabe J, et al: Aspirin use and all-cause mortality among patients being evaluated for known or suspected coronary artery disease. A propensity analysis. J Am Med Assoc 286:1187–1194, 2002
6. Concato J, Shan N, Horwitz RI: Randomized, controlled trials, observational studies and the hierarchy of research design. N Engl J Med 342:1887–1892, 2000 7. Hosmer DW, Lemeshow S. Applied logistic regression. New York, NY, Wiley, 1989
8. Joffe MM, Rosembaum PR: Invited commentary: propensity score. Am J Epidemiol 150:327–333, 1999
Related Article
This article has been cited by other articles:
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|
|||||||||||
|
|||||||||||
|
|
Copyright © 2003 by the American Society of Clinical Oncology, Online ISSN: 1527-7755. Print ISSN: 0732-183X
|
