Valganciclovir in Glioblastoma, Selection Bias, and Flawed Conclusions


Get Permission

William T. Curry, MD

As reviewed in this issue of The ASCO Post, Söderberg-Nauclér et al from the Karolinska Institute have written a provocative letter to The New England Journal of Medicine suggesting that long-term administration of valganciclovir (Valcyte), a drug that targets cytomegalovirus (CMV), improves overall survival in patients with glioblastoma.1

Prospective Study

This is a highly accomplished group in the field of virology. Their letter serves as a companion piece to a randomized study of valganciclovir combined with standard surgery, radiation, and chemotherapy for patients with newly diagnosed glioblastoma, published recently in the International Journal of Cancer.2

In this prospective trial, 42 patients with CMV-positive glioblastoma were randomly assigned in double-blind fashion to receive either valganciclovir or placebo in addition to standard therapy. The primary endpoints were tumor volume at 3 and 6 months; secondary endpoints, for which the study was not powered, were progression-free survival and overall survival. There were no significant differences in tumor volume at either timepoint. Also, progression-free survival (5.6 vs 5.5 months) and overall survival (17.9 vs 17.4 months) were similar between these intent-to-treat groups.

After the initial 6-month study period, patients could either continue taking valganciclovir or, if in the placebo group, to cross over and initiate valganciclovir treatment. Patients taking valganciclovir for longer than 6 months, regardless of the point of initiation, lived longer than patients receiving either short-term (< 6 months) or no valganciclovir.

Selection Bias

This is an extremely flawed conclusion, for the simple fact that good performance score patients and surviving patients, for that matter, are more likely to take any drug for a longer period of time, so the selection bias is heavy. The futility of this sort of analysis is detailed by Anderson, Cain, and Gelber in a methodology review in the Journal of Clinical Oncology.3 However, based on this “generated” hypothesis, the group from the Karolinska Institute treated 28 more patients (in addition to the ones on the protocol) with valganciclovir. These 50 patients form the substrate for the letter to The New England Journal of Medicine.

This report seems to corroborate the initial finding in the randomized trials. Compared to historical controls, patients receiving valganciclovir for at least 6 months had longer median overall survival. Among these patients, those who took valganciclovir continuously after the initial 6 months had the longest median survival—56 months.

Continued selection bias compels us to interpret the authors’ conclusions with circumspection. Most importantly, patients were only eligible for valganciclovir if they were in good neurologic condition and had stable disease after the completion of surgery and initial chemoradiation—no such selection criteria were applied to the “controls.” Immediately, “valganciclovir treatment” selects for a prognostically favorable population. Hints of this reality can be seen in the patient details, which are provided in the supplementary data. For instance, the control patients have lower Karnofsky performance scores, and they represent the only cohort for which the well-validated recursive partitioning analysis class data are not provided. The fact that patients within the treatment cohort must achieve certain exclusive criteria prior to enrollment unavoidably drives “immortal time bias”4 and shifts the Kaplan-Meier curves from the outset.

Furthermore, 30 of the 50 valganciclovir-treated patients are taken from the randomized trial; the 20 treated under compassionate use “had good neurological function and either approached their doctor and requested valganciclovir or learned about this option from their treating physician while discussing possible treatment options, including other clinical trials open for recruitment.” Both clinical trial participation and socioeconomic and educational status have been associated with improved survival in cancer, and the treatment cohort here is likely enriched for both.

Premature Inference

Although there is some evidence suggesting that CMV-infection of glioblastoma cells drives an oncogenic phenotype,5 it is premature and perhaps wrong to infer that valganciclovir treatment improves outcomes for these patients. It is not clear that a properly designed and powered randomized trial is justified at this point. However, if the authors are so motivated, they should be able to answer the question of whether long-term valganciclovir treatment impacts survival in glioblastoma patients in a relatively straightforward—though resource-intense—manner, with the help of a biostatistician. ■

Dr. Curry is Associate Professor and Attending Neurosurgeon in the Department of Neurosurgery, Massachusetts General Hospital and Harvard Medical School, Boston.

Disclosure: Dr. Curry reported no potential conflicts of interest.

References

1. Söderberg-Nauclér C, et al: N Engl J Med 369:985-986, 2013.

2. Stragliotto G, et al: Int J Cancer 133:1204-1213, 2013.

3. Anderson JR, et al: J Clin Oncol 26:3913-3915, 2008.

4. Suissa S: Am J Epidemiol 167:492-499, 2008.

5. Dziurzynski K, et al: Clin Cancer Res 17:4642-4649, 2011.


Advertisement

Advertisement



Advertisement