Prostate alpha/beta revisited – an analysis of clinical results from 14 168 patients

Abstract

Purpose. To determine the dose response parameters and the fractionation sensitivity of prostate tumours from clinical results of patients treated with external beam radiotherapy. Material and methods. The study was based on five-year biochemical results from 14 168 patients treated with external beam radiotherapy. Treatment data from 11 330 patients treated with conventional fractionation have been corrected for overall treatment time and fitted with a logit equation. The results have been used to determine the optimum α/β values that minimise differences in predictions from 2838 patients treated with hypofractionated schedules. Results. Conventional fractionation data yielded logit dose response parameters for all risk groups and for all definitions of biochemical failures. The analysis of hypofractionation data led to very low α/β values (1–1.7 Gy) in all mentioned cases. Neglecting the correction for overall treatment time has little impact on the derivation of α/β values for prostate cancers. Conclusions. These results indicate that the high fractionation sensitivity is an intrinsic property of prostate carcinomas and they support the use of hypofractionation to increase the therapeutic gain for these tumours.

The interest in the radiobiology of prostate cancers has increased considerably since the publication of the initial report of Brenner and Hall [1] of a rather low α/β = 1.5 Gy for these tumours. Indeed, according to standard clinical knowledge, a higher fractionation sensitivity than that of normal tissues at risk would suggest that a departure from the conventionally fractionated schedules and the use of higher doses per fraction for these tumours could widen the therapeutic window, leading either to the same tumour control with less complications or to better tumour response for the same level of complications [2]. The original report of a low α/β value for prostate had been based on a comparison of results from low dose rate brachytherapy with those from high dose rate external beam radiotherapy. Not unexpectedly, its results have been disputed on various grounds, such as the relative biological effectiveness of brachytherapy radiation, tumour proliferation, heterogeneity of dose distributions or of tumour cell radiosensitivity, as reviewed by Dasu [3]. Nevertheless, the initial report of a low α/β value for prostates had been followed by an increasing number of publications suggesting that prostate carcinomas might indeed have a high fractionation sensitivity that would favour therapeutic hypofractionation [3–10]. Consequently, several clinical studies exploring the feasibility and effectiveness of hypofractionated schemes have been initiated in recent years [2]. Enough results have now matured to warrant a new evaluation of the clinically relevant α/β value for prostates and this is the aim of the present study.

Material and methods

Clinical studies reporting the outcome of prostate radiotherapy have been identified in the literature using standardised queries or tracking references from relevant papers, as well as in the personal reference database of the authors. From these, only results from studies employing external beam radiotherapy have been selected in order to avoid uncertainties introduced by comparisons with brachytherapy results that employ dose distributions with different heterogeneities. This has in fact been one of the reasons for questioning the original reports of a low α/β for prostate tumours [11]. Furthermore, only studies reporting biochemical tumour control probabilities (TCP) at five years have been retained to minimise heterogeneities. This strategy had allowed the identification of 25 studies encompassing 14 168 patients that have been used for analysis [12–37]. One of these publications reported the results of a hyperfractionated study employing two daily fractions of 1.2 Gy [17]. These results have not been included in the analysis, as it has been argued that incomplete repair between the two daily fractions might interfere with isoeffect calculations [7].

The results of the studies selected for analysis are summarised in Tables I–IV for two definitions of biochemical failure, ASTRO and Phoenix. The ASTRO definition of biochemical failure is three consecutive prostate specific antigen (PSA) rises after a nadir with backdating the date of failure. The Phoenix definition of biochemical failure is a rise by 2 ng/ml or more above the nadir PSA without backdating [38]. The ASTRO definition has been criticised for the uncertainties it introduces in patients receiving hormonal treatment, as well as for its dependence on backdating. Nevertheless, it has been extensively used in earlier publications and therefore it was included in a parallel analysis with the Phoenix definition. The results have been analysed not only according to the definition of biochemical failure, but also according to the fractionation schedule and the risk group where this information was available.

Table I. Clinical results from conventionally fractionated studies reporting five-year biochemical control according to the ASTRO definition.

Reference	No. of patients	Total dose (Gy)	Dose per fraction (Gy)	Treatment duration (days)	Biochemical control rate	Risk group
Horwitz et al. [12]	939	74.0	2.0	51	0.70	Mixed
Dearnaley et al. [14]	62	74.0	2.0	51	0.71	Mixed
Dearnaley et al. [14]	64	64.0	2.0	44	0.59	Mixed
Kupelian et al. [15]	310	78.0	2.0	53	0.78	Mixed
Lukka et al. [16]	470	66.0	2.0	45	0.58	Mixed
Valdagni et al. [17]	161	74.0	2.0	51	0.70	Mixed
Peeters et al. [19]	331	68.0	2.0	46	0.64	Mixed
Peeters et al. [19]	333	78.0	2.0	53	0.74	Mixed
Yeoh et al. [22]	109	64.0	2.0	44	0.64	Mixed
Eade et al. [24]	43	69.0	2.1	45	0.60	Mixed
Eade et al. [24]	552	72.5	2.1	47	0.68	Mixed
Eade et al. [24]	568	77.5	2.1	51	0.76	Mixed
Eade et al. [24]	367	81.0	2.1	53	0.84	Mixed
Beckendorf et al. [35]	153	70.0	2.0	49	0.73	Mixed
Beckendorf et al. [35]	153	80.0	2.0	56	0.81	Mixed
Kupelian et al. [15]	68	78.0	2.0	53	0.93	Low
Valdagni et al. [17]	75	74.0	2.0	51	0.85	Low
Cahlon et al. [26]	100	86.4	1.8	66	0.99	Low
Goldner et al. [29]	117	70.0	2.0	49	0.74	Low
Kupelian et al. [15]	84	78.0	2.0	53	0.79	Intermediate
Valdagni et al. [17]	58	74.0	2.0	51	0.61	Intermediate
Cahlon et al. [26]	192	86.4	1.8	66	0.79	Intermediate
Goldner et al. [29]	199	70.0	2.0	49	0.66	Intermediate
Kupelian et al. [15]	158	78.0	2.0	53	0.72	High
Valdagni et al. [17]	28	74.0	2.0	51	0.65	High
Cahlon et al. [26]	186	86.4	1.8	66	0.72	High
Goldner et al. [29]	83	74.0	2.0	51	0.50	High

Table II. Clinical results from hypofractionated studies reporting five-year biochemical control according to the ASTRO definition.

Reference	No. of patients	Total dose (Gy)	Dose per fraction (Gy)	Treatment duration (days)	Biochemical control rate	Risk group
Livsey et al. [13]	705	50.0	3.1	22	0.56	Mixed
Lukka et al. [16]	466	52.5	2.6	28	0.53	Mixed
Yeoh et al. [22]	108	55.0	2.8	28	0.66	Mixed
Kupelian et al. [25]	770	70.0	2.5	38	0.82	Mixed
Livsey et al. [13]	181	50.0	3.1	22	0.82	Low
Kupelian et al. [25]	262	70.0	2.5	38	0.95	Low
Livsey et al. [13]	247	50.0	3.1	22	0.56	Intermediate
Kupelian et al. [25]	216	70.0	2.5	38	0.85	Intermediate
Livsey et al. [13]	277	50.0	3.1	22	0.39	High
Kupelian et al. [25]	292	70.0	2.5	38	0.68	High

Table III. Clinical results from conventionally fractionated studies reporting five-year biochemical control according to the Phoenix definition.

Reference	No. of patients	Total dose (Gy)	Dose per fraction (Gy)	Treatment duration (days)	Biochemical control rate	Risk group
Lukka et al. [16]	470	66.0	2.0	45	0.62	Mixed
Williams et al. [21]	1458	63.2	2.1	42	0.47	Mixed
Williams et al. [21]	703	66.0	2.0	45	0.65	Mixed
Dearnaley et al. [23]	422	74.0	2.0	51	0.71	Mixed
Dearnaley et al. [23]	421	64.0	2.0	44	0.60	Mixed
Eade et al. [24]	43	69.0	2.1	45	0.70	Mixed
Eade et al. [24]	552	72.5	2.1	47	0.81	Mixed
Eade et al. [24]	568	77.5	2.1	51	0.83	Mixed
Eade et al. [24]	367	81.0	2.1	53	0.89	Mixed
Zelefsy et al. [27]	358	70.2	1.8	53	0.61	Mixed
Zelefsy et al. [27]	471	75.6	1.8	58	0.74	Mixed
Zelefsy et al. [27]	741	81.0	1.8	63	0.85	Mixed
Zelefsy et al. [27]	477	86.4	1.8	66	0.82	Mixed
Aizer et al. [28]	352	75.6	1.8	58	0.75	Mixed
Leborgne and Fowler [30]	130	78.0	2.0	53	0.90	Mixed
Buyyounouski et al. [32] reporting on Pollack et al. [20]	50	76.0	2.0	52	0.79	Mixed
Beckendorf et al. [35]	153	70.0	2.0	49	0.80	Mixed
Beckendorf et al. [35]	153	80.0	2.0	56	0.86	Mixed
Williams et al. [21]	219	63.2	2.1	42	0.77	Low
Williams et al. [21]	232	66.0	2.0	45	0.88	Low
Cahlon et al. [26]	100	86.4	1.8	66	0.98	Low
Aizer et al. [28]	80	75.6	1.8	58	0.85	Low
Goldner et al. [29]	117	70.0	2.0	49	0.81	Low
Leborgne and Fowler [30]	56	78.0	2.0	55	0.98	Low
Alicikus et al. [34]	49	81.0	1.8	63	0.98	Low
Williams et al. [21]	627	63.2	2.1	42	0.55	Intermediate
Williams et al. [21]	246	66.0	2.0	45	0.66	Intermediate
Zelefsy et al. [27]	155	70.2	1.8	53	0.67	Intermediate
Zelefsy et al. [27]	171	75.6	1.8	58	0.81	Intermediate
Zelefsy et al. [27]	332	81.0	1.8	63	0.88	Intermediate
Zelefsy et al. [27]	191	86.4	1.8	66	0.85	Intermediate
Aizer et al. [28]	155	75.6	1.8	58	0.82	Intermediate
Goldner et al. [29]	199	70.0	2.0	49	0.80	Intermediate
Leborgne and Fowler [30]	66	78.0	2.0	55	0.84	Intermediate
Alicikus et al. [34]	89	81.0	1.8	63	0.89	Intermediate
Williams et al. [21]	612	63.2	2.1	42	0.27	High
Williams et al. [21]	225	66.0	2.0	45	0.41	High
Zelefsy et al. [27]	156	70.2	1.8	53	0.40	High
Zelefsy et al. [27]	234	75.6	1.8	58	0.61	High
Zelefsy et al. [27]	176	81.0	1.8	63	0.66	High
Zelefsy et al. [27]	186	86.4	1.8	66	0.71	High
Aizer et al. [28]	117	75.6	1.8	58	0.62	High
Goldner et al. [29]	83	74.0	2.0	51	0.60	High
Leborgne and Fowler [30]	8	78.0	2.0	55	0.87	High
Arcangeli et al. [31]	85	80.0	2.0	54	0.62	High
Alicikus et al. [34]	32	81.0	1.8	63	0.70	High

Table IV. Clinical results from hypofractionated studies reporting five-year biochemical control according to the Phoenix definition.

Reference	No. of patients	Total dose (Gy)	Dose per fraction (Gy)	Treatment duration (days)	Biochemical control rate	Risk group
Lukka et al. [16]	466	52.5	2.6	28	0.58	Mixed
Higgins et al. [18]	300	52.5	2.6	26	0.43	Mixed
Kupelian et al. [25]	770	70.0	2.5	38	0.83	Mixed
Leborgne and Fowler [30]	89	61.3	3.1	28	0.88	Mixed
Buyyounouski et al. [32] reporting on Pollack et al. [20]	50	70.2	2.7	36	0.83	Mixed
Rene et al. [33]	129	66.0	3.0	31	0.98	Mixed
Higgins et al. [18]	37	52.5	2.6	26	0.74	Low
Kupelian et al. [25]	262	70.0	2.5	38	0.94	Low
Leborgne and Fowler [30]	29	61.3	3.1	28	0.96	Low
Rene et al. [33]	93	66.0	3.0	31	1.00	Low
Freeman and King [36]	41	36.3	7.3	5	0.93	Low
Higgins et al. [18]	103	52.5	2.6	26	0.56	Intermediate
Kupelian et al. [25]	216	70.0	2.5	38	0.83	Intermediate
Leborgne and Fowler [30]	45	61.3	3.1	28	0.84	Intermediate
Rene et al. [33]	36	66.0	3.0	31	0.92	Intermediate
Higgins et al. [18]	160	52.5	2.6	26	0.31	High
Kupelian et al. [25]	292	70.0	2.5	38	0.72	High
Leborgne and Fowler [30]	15	61.3	3.1	28	0.85	High
Arcangeli et al. [31]	83	62.0	3.1	28	0.76	High
Quon et al. [37]	97	67.5	2.7	35	0.83	High

The linear quadratic (LQ) formalism has been used for isoeffective calculations between treatment schedules with various fractionations and durations [39,40]. Thus, the equivalent dose with standard 2 Gy per fraction (EQD2) of a treatment schedule delivering a total dose D in fractional doses d is given by Equation 1.

where α/β is the fractionation sensitivity of the tumour, d_prolif is the dose equivalent of proliferation, T is the treatment duration and T_delay is the time delay before the onset of proliferation. Thames et al. [41] have recently reported clinically relevant dose equivalents of proliferation (mixed risks 0.24 Gy/day, low risk 0.38 Gy/day, intermediate risk 0.14 Gy/day and high risk 0.28 Gy/day) and time delays (T_delay = 7 weeks) for prostate tumours and these have been used in the present analysis.

Results from clinical studies employing conventional fractionation have been fitted with a logit expression (Equation 2) in order to determine the dose response parameters D₅₀ and γ.

The next step of the analysis was to use the logit parameters to calculate the predicted TCP for each of the hypofractionation schedules under the assumption of various α/β values in Equation 1 and then to calculate the sum of the squares of the differences between the TCP reported in each study and the predicted TCP. This allows the determination of the α/β values leading to the least square deviation between results from conventional fractionation and results from hypofractionation.

Results

Figure 1 shows the dose response curves obtained from schedules with conventional fractionation. The logit parameters for these curves are presented in Table V. It can be seen that the two definitions of biochemical failures lead to different parameters, indicating that they are not equivalent and separate comparisons should be carried out in each group. Furthermore, dose response curves for mixed risks patients are shallower than curves for individual risk groups as the increased patient heterogeneity in this group influences the slope of the dose response curve [42]. Another interesting aspect from the logit analysis of the conventional fractionation schedules is that neglecting the correction for overall treatment time in Equation 1 leads to somewhat shallower dose response curves, but without a significant change of the position of the dose response curve (Table V).

Figure 1. Logit dose response curves obtained from schedules with conventional fractionation. Left panels – ASTRO definition of failure. Right panels – Phoenix definition of failure. a-b mixed risks, c-d low risk, e-f intermediate risk, g-h high risk.

Table V. Logit parameters for the dose response curves in Figure 1 and derived α/β values minimising the sum of squares in Figure 2.

Failure definition	Risk group	Patients with conventional fractionation	Patients with hypofractionation	EQD2 corrected for overall treatment time			EQD2 not corrected for overall treatment time
				Logit parameters			Logit parameters
				D50 (Gy)	γ	Derived α/β (Gy)	D50 (Gy)	γ	Derived β/β (Gy)
ASTRO	Mixed	4615	2049	59.0 ( ± 2.0)	1.04 ( ± 0.17)	1.7 (1.4–2.2)	58.7 ( ± 2.0)	0.99 ( ± 0.16)	1.7 (1.4–2.2)
	Low	360	443	67.1 ( ± 2.3)	6.42 ( ± 3.78)	0.6 (0.5–1.0)	65.7 ( ± 1.4)	3.98 ( ± 0.91)	0.6 (0.4–0.6)
	Intermediate	533	463	64.7 ( ± 5.4)	1.59 ( ± 0.93)	1.2 (0.7–3.8)	63.8 ( ± 6.0)	1.37 ( ± 0.78)	1.1 (0.7–3.2)
	High	455	569	71.9 ( ± 1.7)	3.85 ( ± 2.21)	1.1 (0.6–2.8)	70.7 ( ± 3.9)	1.88 ( ± 1.24)	1.6 (0.6–0.8)
Phoenix	Mixed	7889	1804	61.7 ( ± 1.1)	1.90 ( ± 0.26)	1.6 (1.2–2.2)	61.8 ( ± 1.1)	1.76 ( ± 0.22)	1.5 (1.1–1.9)
	Low	853	462	59.1 ( ± 3.9)	3.48 ( ± 1.85)	1.0 (0.7–3.8)	56.3 ( ± 4.4)	2.27 ( ± 0.93)	1.3 (0.7–3.4)
	Intermediate	2231	400	61.8 ( ± 1.3)	2.53 ( ± 0.49)	1.3 (0.9–1.7)	61.6 ( ± 1.2)	2.19 ( ± 0.34)	1.9 (1.2–2.6)
	High	1914	647	68.1 ( ± 1.1)	2.28 ( ± 0.53)	1.7 (1.1–2.4)	69.3 ( ± 1.1)	2.01 ( ± 0.41)	1.2 (0.7–1.7)

Figure 2 presents the variation of the sum of squares of differences between reported results for hypofractionated schedules and the predicted TCP for various assumptions regarding the α/β value for the individual risk groups considered. It can be seen that for all risk groups the sum of squares is minimised for low α/β values. The optimal α/β values for which the minimum of the sum of squares is achieved for each risk group considered are presented in Table V. These results show that the correction for overall treatment time has a small impact on the derivation of α/β values for prostates.

Figure 2. Variation of the sum of squares with assumptions regarding the α/β value. Left panels – ASTRO definition of failure. Right panels – Phoenix definition of failure. a-b mixed risks, c-d low risk, e-f intermediate risk, g-h high risk. Solid curves – with correction for overall treatment time, dashed curves – without correction for overall treatment time.

Figure 3 shows the relation between dose response curves for conventional fractionation and results from clinical studies employing non-standard fractionation for the optimal α/β values.

Figure 3. Relation between dose response curves for conventional fractionation and results from clinical studies employing non-standard fractionation converted with α/β values leading to minimum values in Figure 2. Left panels – ASTRO definition of failure. Right panels – Phoenix definition of failure. a-b mixed risks, c-d low risk, e-f intermediate risk, g-h high risk.

Discussion

The fractionation sensitivity of prostates has been a hot topic of debate for the past decade as it has the potential to change the fractionation paradigm in radiation therapy. Current radiotherapy practice assumes that tumours have lower fractionation sensitivity in comparison to late reacting normal tissues, which led to the use of many small fractions to ensure the therapeutic differential between tumours and normal tissues. This certainly is the case for many tumours with an active proliferation pattern. Prostate tumours however have slow proliferation and this raised the question whether their fractionation sensitivity is comparable to or even higher than those of late reacting normal tissues at risk which would in turn favour fractionated patterns employing a few large fractions. Such patterns would be beneficial both to the patients that would face shorter treatments and the radiotherapy departments that could increase their patient throughput. However, the reversal of the fractionation pattern is unsettling given the clinical examples of hypofractionation from other sites that led to an unacceptable rate of late reactions [43]. This is why the initial report of Brenner and Hall [1] of a low α/β value for prostate and all the subsequent proposals have been closely scrutinised for factors that might influence the derivation of the clinically relevant fractionation sensitivity for these tumours. Thus, dose and inter-patient heterogeneity, treatment duration and proliferation, relative biological effectiveness and other factors have been proposed as possible causes of interferences with isoeffect calculations [3]. While attempts were made to carefully account for these factors, it was generally agreed that best results would probably be obtained from randomised studies that minimise the number of differences between the arms of comparison. However, retrospective analyses of data from many patients may offer some advantages, including reducing stochastic fluctuations and the availability of more dose levels that could be used to derive dose response curves.

This study used this latter approach to analyse the results of more than 14 000 patients, making it the largest such analysis this far. Nevertheless, the use of a retrospective analysis approach may be subject to heterogeneity issues and other limitations that need to be taken into account when generalising the applicability of the results. Thus, comparing outcome data across trials increases heterogeneity in the datasets since patients and treatments in different subgroups are not the same. Differences may include risk stratification of the patients from various studies, the availability of adjuvant hormonal therapy that is given to some prostate cancer patients, duration of the follow-up, etc. Indeed, it has previously been shown that androgen suppression therapy may lead to better local control in patients undergoing radiotherapy [44] and therefore better results could be expected from studies employing androgen deprivation in some or all of their patients in comparison to studies refraining from hormonal treatment. However, it has recently been shown that this is not expected to affect the α/β derivation [10]. The duration of the follow-up and its impact on the reported results is another aspect that has to be carefully considered. Thus, Vicini et al. [45] showed that too short follow-ups might lead to the overestimation of the apparent efficacy of the treatments for the ASTRO definition of biochemical failure that relies strongly on backdating. This observation has later been used by Diez et al. [46] to explain the apparently steeper dose response curves that are obtained from non-randomised studies. Thus, they argued that higher dose series tend to have shorter follow-up as they are later introduced and reported and this translates into an overestimation of their efficacy. For these reasons, this study used as many studies as were available and analysed separately the results reported for the two definitions of biochemical failure.

The size and composition of the patient subgroups are other sources of heterogeneity that could influence the results. Thus, controlling a tumour is a process governed by probabilities and therefore subject to stochastic fluctuations, hence results of larger studies are probably less affected by noise than those of smaller studies. The majority of the patients included in the present study (11 330/14 168) were treated with conventional fractionation and this reflects the slow and careful introduction of hypofractionated schedules in clinical trials. The bias towards conventionally fractionated schedules might also translate into somewhat larger confidence intervals ( ± one standard deviation) for the fractionation sensitivities. Risk stratification of the large groups is another aspect that increases the heterogeneity of the analysed data. Indeed, we have earlier shown that the slope of the TCP curves is influenced by heterogeneity in radiosensitivity or population size [42]. Thus, assuming that the various risk groups correspond to tumours with various sizes or intrinsic radiosensitivities that makes them progressively more difficult to control with a given radiation dose, joining all risk groups in an analysis would inevitably lead to shallower dose response curves, as confirmed by the results in Table V where dose response curves for patients with mixed risks are generally shallower than similar curves for each risk group. Furthermore, analysing results originating from patient groups with different compositions (e.g. one with more high-risk patients and one with more low-risk patients) increases the heterogeneity of mixed risks results. It should be noted that stratified data according to risk represents a considerable fraction of the analysed data, but some heterogeneity remains even in the individual risk groups since the criteria for stratification is based on intervals of values for the risk factors. The effect of this type of heterogeneities could only be reduced in randomised studies that control the composition of arms used for comparisons. Nevertheless, the fractionation sensitivity was similar across the various risk groups considered with only a slight correlation between α/β and risk group. This is in agreement with previous findings in smaller patient numbers [10], as well as with analyses of proliferation patterns in individual risk groups [41].

The impact of these sources of heterogeneities could be qualitatively assessed by comparing α/β values obtained in this study from dose response curves with different slopes. Thus, the results in Table V show that similar α/β values were obtained for various risk groups as well as from dose response curves with different slopes, suggesting that heterogeneities have a small impact on the derivation of fractionation sensitivity. Furthermore, the derived α/β values are similar, irrespective of the definition of biochemical failure used for the clinical results. This indicates that the fractionation sensitivity is not an artefact of the definition of failure and that uncertainties associated with the ASTRO definition [38] have little impact on the derivation of the ratio α/β. Nevertheless, failure definitions must be taken into account when analysing clinical data as mixing definitions increases the heterogeneity and decreases the statistical significance of the results.

The logit parameters determined in this study for conventional fractionation schedules are somewhat different from those reported in previous studies [47,48]. The difference could be caused by the inclusion of more results from dose escalation studies in the present analysis that could either decrease the uncertainty at high doses, leading to more relevant dose response parameters, or reflect the heterogeneities mentioned above. Nevertheless, correcting the clinical data sets for overall treatment time during lengthy schedules leads to steeper dose response curves thus indicating further decreases of the uncertainties. Furthermore, dose response curves for the Phoenix definition of biochemical failure are steeper than those for the ASTRO definition, as was also found by Diez and colleagues [46].

The results of this study (α/β = 1–2 Gy) are very close to the initial derivation of Brenner and Hall [1] and other subsequent reports [4–5,8–10] indicating that the practical impact of the potentially adverse factors identified earlier could be smaller than was initially thought. This includes the effect of proliferation of prostate tumour cells, as similar values were obtained when the correction for overall treatment time was taken into account or was neglected. This may also depend on the small values for the dose equivalents of proliferation reported by Thames et al. [41], but these are in agreement with other studies showing variable impact of the overall treatment time for various risk groups [49,50]. It should also be mentioned that calculations have been performed based on the estimated treatment durations, as real treatment times were not available for every patient. Given the low dose equivalents of proliferation, this aspect is however not expected to have a strong impact on the reported results.

The α/β values found in this analysis are generally lower than the fractionation sensitivities of normal tissues previously reported [3]. This suggests that hypofractionation for prostate tumours might bring a therapeutic advantage, especially if combined with better conformal techniques that limit the irradiation of normal tissues. However, there are other reasons for which extreme hypofractionation might fail to bring the expected gain and these should be taken into account when designing the therapeutic schedules. Thus, too few fractions might create problems if hypoxic cells exist in the tumour, as few reoxygenation opportunities would be offered to the cells [51]. Also, too short treatment schedules might prevent compensatory proliferation in acutely reacting tissues, thus increasing the severity of the normal tissue reactions [52]. Nevertheless, studies employing extreme hypofractionations have been initiated and when results will mature the impact of these aspects could be evaluated.

Conclusions

This study presents the largest analysis this far of the fractionation sensitivity for prostate tumours based on clinical results from patients treated with external beam radiotherapy. The analysis yielded very low α/β values for all risk groups and for all definitions of biochemical failures. These results therefore appear to justify the earliest publications that pointed out that α/β values might be as low as approximately 1.5 Gy [1,4]. They also indicate that the high fractionation sensitivity is an intrinsic property of prostate carcinomas and they support the use of hypofractionation to increase the therapeutic gain for these tumours.

Declaration of interest: The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the paper.