European Urology

We thank Dr. Fizazi for his thoughtful comments [1]. As noted by Dr. Fizazi, bias can be common in nonrandomized studies and, in many instances, can markedly limit the ability to derive scientifically correct conclusions upon which to base patient care. Many statistical approaches have been devised to account for potential biases, but in general most of these methods require that a bias be recognized and data on confounding variables be measured and collected before it can be adjusted and corrected. This is consistent with the common assumption that many observational studies are biased. To overcome this limitation in our study, we used instrumental variable analysis (IVA) techniques [2].

Studies have shown that IVA techniques can effectively minimize biases in observational studies [3] and [4]. IVA captures the random component of patient–physician treatment choice (eg, some patients in a particular geographic area will receive primary androgen deprivation therapy [PADT] solely because it is more popular or readily available in that area) and uses it to balance treatment groups with respect to measured, unmeasured, or unknown confounding effects.

IVA methods using Surveillance Epidemiology and End Results (SEER)–Medicare data have yielded results very similar to randomized trials for surgery in localized prostate cancer [5] and radiotherapy with and without adjuvant androgen deprivation therapy in nonmetastatic prostate cancer patients [6]. These experiences contrast with results using multivariable or propensity score analytic methods, which cannot adjust for unmeasured or unknown confounding variables, contain residual bias, and have yielded results significantly different from those of randomized clinical trials.

In our study, IVA effectively balanced tumor characteristics, as shown in Table 2 [2]. This balance occurred without specific adjustment or identification of any of the characteristics. Because IVA captures “natural randomization,” the observed balance occurred as it would in a randomized study. Nonetheless, it is important to recognize that despite randomization, even prospective randomized clinical trials can contain residual baseline imbalances, and a confirmatory randomized study (although very likely a practical challenge in this situation) would represent the best way to establish the robustness of our results. Short of a randomized study, however, we believe that Dr. Fizazi's [1] concerns about bias may be somewhat less of an issue, and the main conclusion of our study [2], that early treatment with PADT does not delay the receipt of subsequent palliative therapies, may be useful for physicians and patients who need to make important clinical decisions before prospective randomized data are available.

Conflicts of interest

The authors have nothing to disclose.