“A randomized clinical trial of radiation therapy versus thermoradiotherapy in stage IIIB cervical carcinoma” of Yoko Harima et al. (2001): multiple biases and no advantage of hyperthermia

Dear Sir,

In 2001, the International Journal of Hyperthermia (IJH) published a trial of Harima et al. “A randomised clinical trial of radiation therapy (RT) vs. thermoradiotherapy (TRT) in stage IIIB cervical carcinoma” [1]. This “classical” RCT is a cornerstone evidence of the efficacy of hyperthermia (HT) in the treatment of locally advanced cervical carcinoma (LACC). As such, it was reprinted in the IJH in 2009 [2] accompanied with a “hyperthermia classic commentary” of the key researchers Y. Harima and S. Sawada. They concluded that “our study clearly demonstrated that TRT can result in beneficial effects for complete response and/or local relapse-free survival” [3]. Preparing a systematic review of hyperthermia in the treatment of LACC, [4] we performed a thorough data appraisal, including a comprehensive bias analysis and recalculation of each parameter reported in the original trials (if possible). Here, we present the results of the bias analysis of this RCT.

A sample of 40 subjects (20:20) was calculated as sufficient for the 80% statistical power of the study based on the difference in complete response rates (CRR) of 80% in the TRT arm vs. 50% in the RT arm, using a two-sided chi-square test. This calculation is incorrect. The actual power of the difference is from 38% (Yates’s correction) to 50% (Pearson chi-square test), so that the trial is critically underpowered. The chi-square test provides the 80% power using these parameters only at a sample size of more than 80 subjects. This was recognised by the researchers in their next study, where the sample size, calculated using the same parameters, was estimated as 88 subjects [5].

The significance of the observed difference in CRR, which is a main evidence confirming the advantage of TRT in this trial, was incorrectly calculated. The difference (80% [16/20] in the TRT arm vs. 50% [10/20] in the RT arm) was reported as statistically significant with p = 0.048 by Fisher’s exact test. This is correct for the one-sided Fisher’s exact test [6]. Since the calculation of the sample size was based on the two-sided chi-square test, only two-sided tests are suitable. One cannot use two-sided and one-sided tests at the same time, because they use different statistical hypotheses. The two-sided Fisher’s exact test returns the actual p-value of 0.0958 [6]. An alternative Yates’s correction for small samples also returns p = 0.0974 [7]. So, the reported p-value is not valid, and the local response did not attain the significance level.

The actual use of the one-sided Fisher’s exact test instead of the implied two-sided test is concealed. It is never mentioned and can only be revealed by recounting the data, thus misleading the reader on the significance of the difference in CRR.

There is an alarming discrepancy in the reporting of the number of deaths. In the text description of the results, [8] six deaths were reported in the TRT arm (excluding a case of not cancer-related death), with an overall survival (OS) of 65%, although eight events are clearly seen on the KM-plot of OS, and the OS of 58.2% is reported elsewhere.

In conclusion, this underpowered trial hardly showed an advantage of TRT. The actual difference in CRR was not statistically significant (p = 0.096), and there was no difference in the OS (p > 0.3). Multiple discrepancies and shortcomings make the trial unreliable.

Disclosure statement

No potential conflict of interest was reported by the author.