Radiotherapy and Oncology 48 (1998) 165–173

Effect of gap length and position on results of treatment of cancer of the larynx
in Scotland by radiotherapy: a linear quadratic analysis

A. Gerald Robertson a, Chris Robertson b ,*, Concetta Perone b, Kathy Clarke c, John Dewar d,
M.H. Elia e, David Hurman f, R. Hugh MacDougall g, Hosni M.A. Yosef a
a
Beatson Oncology Centre, West Glasgow Hospitals University Trust, Glasgow, G11 6NT, UK
b
Division of Epidemiology and Biostatistics, European Institute of Oncology, via Ripamonti 435, 20141 Milano, Italy
c
Information and Statistics Division, Trinity Park House, Edinburgh, EH5 3SQ, UK
d
Department of Radiotherapy, Ninewells Hospital and Medical School, Dundee, DD1 9SY, UK
e
Department of Radiotherapy and Oncology, Raigmore Hospital, Inverness, IV2 3UJ, UK
f
Department of Radiotherapy, Aberdeen Royal Infirmary, Aberdeen, AB9 2ZD, UK
g
Department of Clinical Oncology, Western General Hospital, Edinburgh, EH4 2XU, UK

Received 28 July 1997; revised version received 23 March 1998; accepted 27 March 1998

Abstract

Purpose: This paper reports on the analysis of the effect of the length and position of unplanned gaps in radiotherapy treatment schedules.
Materials and methods: Data from an audit of the treatment of carcinoma of the larynx are used. They represent all newly diagnosed
cases of glottic node-negative carcinoma of the larynx between 1986 and 1990, inclusive, in Scotland that were referred to one of the five
Scottish Oncology Centres for primary radical radiotherapy treatment. The end-points are local control of cancer of the larynx in 5 years
and the length of the disease-free period. The local control rates at ≥5 years, Pc, were analyzed by log linear models and Cox proportional
hazard models were used to model the disease-free period.
Results: Unplanned gaps in treatment are associated with poorer local control rates and an increased hazard of a local recurrence through
their effect on extending the treatment time. A gap of 1 day is potentially damaging but the greatest effect is at treatment extensions of 3 or
more days, where the hazard of a failure of local control is increased by a factor of 1.75 (95% confidence interval 1.20–2.55) compared to
no gap. The time factor for the actual time was imprecisely estimated at 2.7 Gy/day with a standard error of 13.2 Gy/day. Among those
cases who had exactly one gap resulting in a treatment extension of 1 day, there is no evidence that gap position influences local control
(P = 0.17). The treatment extension as a result of the gap is more important than the position of the gap in the schedule.
Conclusions: Gaps in the treatment schedule have a detrimental effect on the disease-free period. A gap has a slightly greater effect than
an increase in the prescribed treatment time. Any gap in treatment is potentially damaging. The position of the gap in the schedule was
shown to be not important.  1998 Elsevier Science Ireland Ltd. All rights reserved

Keywords: Linear quadratic model; Time effects; Gap length; Gap position; Carcinoma of the larynx

1. Introduction gaps occur in these schedules for a variety of reasons, such

Radiotherapy schedules for the treatment of cancer of the
larynx have evolved with a planned gap at the weekend.
Some of the common schedules include, for example,
52.5 Gy in 20 fractions over 28 days, 60 Gy in 25 or 30
fractions over 35 or 42 days, 55 Gy in 16 fractions over 21
days and 50 Gy in 20 fractions over 28 days. Unplanned
* Corresponding author.
as public holidays, machine servicing, transport failure,
machine breakdowns and patient illness. If the gap is long
then the treatment may be altered to take into account the
delay in treatment completion.
The outcome for various patient groups from centres in
North America and Europe since 1980 have shown that
unplanned interruption of treatment resulting in the prolon-
gation of the treatment time can reduce the local control rate
and hence the cure rate [1,21,22]. The conclusion from

0167-8140/98/$19.00  1998 Elsevier Science Ireland Ltd. All rights reserved

PII S0167-8140 (98 )0 0038-3
166 A.G. Robertson et al. / Radiotherapy and Oncology 48 (1998) 165–173
recent overviews of these studies [2,7,8] is that a break in
treatment of about 1 week is associated with an absolute
reduction in local control rates of 10–12%; hence a break of
1 day may reduce the control rate by around 1.4%.
Linear quadratic modelling analyses of data from hospital
records of patients treated in different overall times for car-
cinoma of the larynx have estimated the time factor for
tumour control, expressed as the extra dose per additional
day of overall treatment time to maintain the same level of
tumour control. About 0.6–0.7 Gy/day, using 2 Gy frac-
tions, was required to maintain the same level of tumour
control after a lag period of about 3–4 weeks [11,21,26,27,
31,33].
The above analyses focused on the overall length of treat-
ment time and did not specifically consider unplanned gaps.
In the pooled analysis of Robertson et al. [28], the effect of
unplanned gaps were specifically estimated independently
of the planned treatment time. This analysis revealed small
but important reductions in the probability of local control
at ≥2 years associated with an increased treatment time.
From these results, one would conclude that for a specified
schedule, where dose and fraction number are specified, any
gap is potentially damaging.
This paper reports on an audit of the treatment of patients
with glottic node-negative carcinoma of the larynx and
assesses the impact of gaps on the radiotherapy treatment
schedule. This differs from previous analyses as the posi-
tion, length and reason for each gap in the schedule are
known. Thus, we know that a gap in treatment occurred
and can then see its effect on the prolongation of the total
treatment time. In the previous study [28], the treatment
extension was estimated on the basis of knowledge of the
dose and fractions administered. Thus, the current data are
more reliable for the estimation of the direct effect of a gap
as we are able to restrict the analysis to patients in whom the
treatment schedule was not modified to compensate for the
gap.
2. Materials
A database of all newly diagnosed cases of carcinoma of
the larynx between 1986 and 1990, inclusive, in Scotland
that were referred to one of the five Oncology Centres for
primary radical radiotherapy treatment has been assembled.
These patients were identified via the Scottish Cancer
Registration Scheme and Oncology Centre Registrations,
thus ensuring a complete coverage of all cases. Radiother-
apy was the primary treatment for all patients. Only 3% of
patients had any prior surgery and in those cases radiother-
apy commenced within 42 days of surgery.
We use data from all patients with clinically node-nega-
tive squamous cell glottic cancer of the larynx, yielding a
total of 629 patients. The 5-year follow-up analysis uses 352
patients. The patient loss occurs with patients who were not
followed up for the full 5 years but who did not have a
recurrence or larynx cancer death within the period of obser-
vation. As the data represent all cases of laryngeal cancer
where the primary treatment was by radiotherapy in Scot-
land between 1986 and 1990, inclusive, and as data collec-
tion began on 1 July 1994, all patients had a potential
follow-up time of at least 3.5 years. This means that a little
bias will be present in the analysis of the 5-year rates, which
is not present in the survival analysis or in the analysis of 3-
year rates [16,24]. The estimates of the effect of gaps was
similar for the 3- and 5-year rates and so we present the
results only for the latter rates.
3. Methods
Mathematical models are used to estimate the effect of
delays on the completion of the treatment. This enables the
estimation of the tumour doubling times and the biological
effect of the radiotherapy treatment, as well as the effect of
the gaps on the schedule. Survival analysis models are used
to assess the impact of tumour and treatment characteristics
on the disease-free period.
The disease-free period is the primary end-point of the
study and is defined as the time from the start of treatment
until recurrence of the tumour in the same site for those
patients with a recurrence, or the time from the start of
treatment until death from cancer of the larynx for those
patients with no recurrence but who died with the disease.
There were only four patients with a nodal recurrence but
with no local recurrence in the same site who did not die
from cancer of the larynx. They are treated as censored in
this analysis. Patients who had no recurrence, or who did not
die from cancer of the larynx are also censored. Their dis-
ease-free period is taken to be the time from the beginning
of treatment until they were last seen for patients who are
still alive, or until death for patients who died without can-
cer of the larynx.
The analysis of the 5-year local control rates is based on
the same logistic regression model as in Robertson et al.
[27]. The linear quadratic model with the addition of treat-
ment time relating the probability of local control, Pc, to the
treatment schedule is
ln( − ln(Pc )) = ln(M0 ) − F(ad + bd 2 ) + lT
= m − aD − bDd + lT (1)
where F is the number of fractions, d is the dose per frac-
tion, T is the total treatment time and D is the total dose.
The parameters of the model which have to be estimated
are m, the intercept term, which is related to the effective
clonogen number per tumour M0, a and b, which estimate
the linear and quadratic effects of dose, respectively, and l,
which is the rate of tumour repopulation. The model is a
generalized linear model based on a Poisson distribution
[4] for the number of clonogenic cells remaining after
radiotherapy [27].
 A.G. Robertson et al. / Radiotherapy and Oncology 48 (1998) 165–173 167

It would not be appropriate to use Eq. (1) to assess the h(t)

ln[ ] = − aD − bDd + lp P + lg G (5)
effect of gap length on the probability of local control, as T h0 (t)
is the observed treatment time which contains information
both on planned treatment time and any gaps which where the baseline hazard for an individual who has D0 Gy
occurred. Thus, we replace total time in Eq. (1) by planned in F0 fractions over T0 days is denoted by h0(t). The linear
treatment time, P, and the total length of any gaps in and quadratic assumptions about the relationships between
treatment, G. the log hazard and dose and time are checked by fitting a
model which uses dose groups, time groups and gap length
ln( − ln(Pc )) = m − aD − bDd + lp P + lg G (2) groups.
Effects are also included to control for the T stage and the hijk (t)
ln[ ] = Di + Pj + Gk (6)
treatment centre. The total dose D = Nd is measured in Gy h0 (t)
and in the log-log linear model Eq. (2) the variables are The same reference categories are used as for Eq. (3). Thus,
centred around their midpoints. Dose is centred on D0 = 54 h0(t) is the hazard for a patient with 55–58 Gy in 4 weeks
Gy, the number of fractions is centred on F0 = 20, the gap with no gap.
length is centred on 0 days and the treatment time is centred All the statistical analysis was carried out using SAS [32].
on T0 = 28 days. These figures correspond closely to the All statistical significance tests are based on differences in
means for most of the centres, with the exception of Glas- log likelihoods which follow x2 distributions in large sam-
gow. With this centring, exp{−exp{m}} represents the ples.
probability of local control for a patient on a 54–20–28
schedule with no gap in treatment. Any significance tests
are based on the differences of deviances which follow a x2 4. Results
distribution in large samples [19].
Implicit in these models is an assumption of linearity with There are 629 cases who were node-negative and with the
respect to time and a quadratic assumption with respect to primary tumour originating in the glottis. There are 321 T1
dose. This assumption is checked using cases (51%), 216 T2 cases (34%) and 78 T3 cases (12%). As
there are only 14 T4 cases, they are combined with the T3
ln( − ln(Pc )) = m + Di + Pj + Gk (3) cases in the modelling analysis. There was no statistical
In this equation, which is a direct extension of Eq. (2), we evidence using interaction tests that the effects of dose
use groups for dose, planned time and treatment prolonga- and time depended upon T stage (data not shown). The
tion as a result of gaps. The reference category for dose is doses ranged from 43 to 70 Gy with 39% of cases receiving
55–58 Gy, for planned treatment time it is 4 weeks and for 60 Gy. Patients were treated with between 15 and 41 frac-
treatment extension it is no extension. Thus, exp{−exp{m}} tions; 44% of patients were treated with 20 fractions and
represents the probability of local control for a patient trea- 32% were treated with 25 fractions. The planned treatment
ted with 55–58 Gy in 4 weeks with no gap. time ranged from 12 to 49 days; 40% of patients were trea-
As there is complete information on the length of the ted over 4 weeks, 32% were treated over 5 weeks and 23%
local control period (or disease-free period), survival analy- were treated over 6 weeks. The actual time ranged from 12
sis can be carried out to estimate the effect of the gap length to 56 days.
on the hazard of a local failure, i.e. a local recurrence or a Overall, 46% of cases had a gap of at least 1 day in the
larynx cancer death. The disease-free curves are estimated treatment, other than the planned gaps at the weekend
using the Kaplan–Meier method [17,18] and the log-rank
test [10] is used to test for differences in these curves. Table 1
The Cox proportional hazards regression model [3] is Number of gaps by the number of days lost as a result of the gap (not
used to estimate the effect of the gap length on the hazard including weekends) (frequency and percentage)
of a local recurrence, adjusting for the effects of other treat- No. of No gap Gap length (days) Total
ment and tumour variables. The hazard of a local recurrence gaps
is the instantaneous probability of a local recurrence (or 1 2–3 4+
treatment failure) at a time t after the beginning of treat- 0 292 (100) – – – 292
ment, given that the patient does not have any local recur- 1 1 (0.6)a 94 (55.3) 60 (35.3) 15 (8.8) 170
rence. It is denoted by h(t) and the proportional hazards 2 – – 87 (81.3) 20 (18.7) 107
model for the hazard of a local recurrence corresponding 3 – – 21 (46.7) 24 (53.3) 45
4 – – – 9 (100) 9
to the linear quadratic model is written as 5 – – – 5 (100) 5
6 – – – 1 (100) 1
h(t)
ln[ ] = − aD − bDd + lT (4) Total 293 94 168 74 629
h0 (t)
Values in parentheses are percentages.
and its extension to include effects for the gap length. a
Treatment finished early for this patient after fraction 26 due to illness.
168 A.G. Robertson et al. / Radiotherapy and Oncology 48 (1998) 165–173

Table 2
Number of days lost as a result of the gap (not including weekends) by the number of days by which the total treatment time was extended (frequency and
percentage)

Gap length No gap Total treatment time extension (days) Total

(days)
1–2 3–4 5–7 8+

No gap 292 (99.7) 1 (0.3)a – – – 293

1 2 (2.1)a 71 (75.5) 21 (22.3) – – 94
2–3 2 (1.2)a 77 (45.8) 53 (31.5) 36 (21.4) – 168
4+ 5 (6.7)b 0 (0) 5 (6.7) 40 (54.0) 24 (32.4) 74
Total 301 149 79 76 24 629

Values in parentheses are percentages.

a
Treatment rescheduled to compensate for a gap.
b
All patients had a planned split treatment.

(Table 1). Of the 170 cases with exactly one gap, the length
of the gap was more than 1 day in 65 instances. Further-
more, a gap of 1 day can have a greater effect on the pro-
longation of the treatment time (Table 2). Ninety-four cases
had one gap of exactly 1 day but in 21 cases this resulted in
an increase in the treatment time of 2–4 days. This is an
effect of the weekend where patients due to end treatment
on a Friday have their treatment extended by 3 days as a
result of a gap of 1 day. In this analysis we use the extension
to the treatment time as the main variable to assess the effect
of a gap.
There are some apparent inconsistencies in Tables 1 and 2
which arise because some patients had their treatment
rescheduled as a result of the gaps. This most commonly
occurs because the dose per fraction in the remaining frac-
tions is increased, or extra fractions are given. In some cases
the treatment time was not extended but the dose and num-
ber of fractions were adjusted from that planned. These
patients are excluded from the modelling analysis below,
leaving 613 patients, as we wish to estimate the effect of
gaps in the absence of a correction to the schedule.
ences in the hazard of a recurrence associated with the
different centres and so there is no need to take centre
differences into account. The only tumour characteristic
to be included in the Cox proportional hazards model is
the T stage. Most of the patients are T1 or T2 and an
increasing T stage is associated with an increased hazard
of a failure of local control.
In eqns (4) and (5) in Table 3, the coefficient of Dd was
not significantly different from 0 and so was not considered.
This means that we are unable to estimate the a/b ratio from
these data as we are constraining b = 0. There is evidence
that longer total treatment times are associated with a
greater hazard of a recurrence (P = 0.028). The tumour
doubling time is estimated as ln(2)/g = 22.4 days (95% CI
2.2–42.6 days). The maximal time factor, which is applic-
able using very small fractions, is the extra dose in Gy
required per day to counteract the effect of protracting the
treatment time. This is given by g/a = 2.69 Gy/day with a
standard error of 13.2 Gy/day for the actual time (Eq. (4))

4.1. Survival analysis

Among the 613 node-negative glottic patients, 152 had

tumour recurrence. There was no evidence of any statisti-
cally significant differences among the disease-free curves
of the five centres (log-rank x2 = 4.74, 4 d.f., P = 0.32).
There were significant differences among the disease-free
curves for the T stages (P = 0.0001). The local control rates
at 5 years for T1, T2 and T3 are estimated as 82, 72 and
46%, respectively. The disease-free curves in Fig. 1 show
that patients with unplanned gaps in the treatment schedule
have a significantly worse outcome than those with no gaps
if the gap leads to an extension of treatment time by more
than 2 days (P = 0.03). While the recurrence-free period for
those with a gap leading to a treatment extension of 1 or 2
days is less than for no gaps, the difference is not statisti- Fig. 1. Disease-free curves by gap length. The curves are truncated at 5
years as this is the planned follow-up period. The only events after 5 years
cally significant (P = 0.62). were two larynx cancer deaths. These occurred just after 5 years in one
The parameter estimates of the Cox regression models patient with no gap and after 7 years in one patient in the 1–2 days gap
are presented in Table 3. There were no significant differ- group. One hundred ninety-seven individuals were still at risk at 5 years.
 A.G. Robertson et al. / Radiotherapy and Oncology 48 (1998) 165–173 169

Table 3

Parameter estimates (standard errors) for the logistic regression and Cox regression models fitted to estimate the effect of treatment extension

Logistic regression estimates Cox regression estimates for the hazard

for the 5-year local control rates of a failure of local control

Eq. (1) Eq. (2) Eq. (3) Eq. (4) Eq. (5) Eq. (6)

Intercept −1.136 (0.160) −1.179 (0.162) −1.529 (0.279) – – –

T stage
T1 0 (0) 0 (0) 0 (0) 0 (0) 0 (0) 0 (0)
T2 0.508 (0.196) 0.499 (0.197) 0.540 (0.199) 0.504 (0.193) 0.494 (0.194) 0.526 (0.198)
T3T4 1.224 (0.219) 1.233 (0.220) 1.259 (0.224) 1.363 (0.206) 1.365 (0.206) 1.302 (0.211)
Dose (Gy) 0.016 (0.028) 0.005 (0.029) 0.011 (0.029) 0.006 (0.030)
Actual time 0.029 (0.014) 0.031 (0.014)
Planned time 0.012 (0.018) 0.025 (0.017)
Treatment extension 0.077 (0.031) 0.045 (0.027)
Dose group (Gy)
40+–55 0.215 (0.273) −0.096 (0.257)
55+–58 0 (0) 0 (0)
58+–61 0.202 (0.385) −0.351 (0.378)
61+ 0.156 (0.421) −0.009 (0.405)
Planned time (weeks)
3 0.751 (0.435) 0.008 (0.409)
4 0 (0) 0 (0)
5 0.093 (0.363) 0.237 (0.363)
6+ 0.161 (0.325) 0.451 (0.319)
Extension (days)
No gap 0 (0) 0 (0)
1–2 0.201 (0.231) 0.199 (0.225)
3–4 0.749 (0.244) 0.682 (0.232)
5+ 0.550 (0.248) 0.502 (0.243)
−2 log likelihood 35.07 37.87 45.45 45.93 46.22 56.03
Degrees of freedom 4 5 11 4 5 11

and g/a = 6.94 Gy/day with a standard error of 32.4 Gy/day
for the extension time (Eq. (5)). The maximal time factors
are large and very imprecisely estimated as the standard
error of the coefficient of dose (a) in Table 3 is relatively
large and the coefficient is small and not significantly dif-
ferent from 0.
From Eq. (6) in Table 3, which is the most general of the
models, there is evidence that treatment extension influ-
ences the hazard of a failure of local control (x2 = 9.63, 3
d.f., P = 0.02), but that dose (x2 = 2.26, 3 d.f.) and planned
treatment time (x2 = 2.37, 3 d.f.) do not. Stratifying by T
stage, we find that the effect of treatment extension is not
significant for T1 but it is for T2 and for T3/T4. However,
the pattern of the estimates is largely the same and there is
no evidence that there are statistically significant differences
in the effect of treatment extension across the three T stage
groups (x2 = 6.21, 6 d.f.). The test statistic for a deviation
from linearity in the effect of treatment extension based on
Eq. (6) is x2 = 2.21, 2 d.f., P = 0.33. This means that Eq. (5)
with only a linear effect for treatment extension is a suffi-
ciently good representation of the hazard of a local recur-
rence and no quadratic term in treatment extension is
required.
No discernible linear effect of dose or planned treatment
time can be detected for this group of patients. One of the
reasons for this is that those patients with very high doses
(greater than 61 Gy) have poorer local control compared to
patients with lower doses. In order to see a significant linear
effect of dose or planned treatment time some sub-optimal
schedules, i.e. very low doses or long planned times, would
have to be prescribed and this is not the case. The schedules
used in the centres are all optimal in the sense that there are
no schedules which have a significantly poorer outlook than
others.
There is no evidence that gaps had more or less effect in
different regimes. It is true that gaps are more likely to occur
in 6 week schedules than in 4 week schedules just because
there is more opportunity. As far as this analysis goes, a gap
has an adverse effect whatever the planned regime.
In the survival analysis of the disease-free period we have
evidence that unplanned gaps in the treatment schedule
which lead to an extension of the treatment time by at
least 3 days are associated with an increased hazard of a
local failure. From Eq. (6), a treatment extension of 3 or 4
days is associated with an increased log (hazard) of 0.68
(0.23) and an extension of 5 or more days is associated
with an increased log (hazard) of 0.50 (0.24), adjusting for
T stage, dose and planned treatment time. This is the same
order of magnitude as the hazard of a local recurrence of a
T2 tumour compared to a T1 tumour. Pooling the treatment
170 A.G. Robertson et al. / Radiotherapy and Oncology 48 (1998) 165–173

extensions of 3 or more days, we find that the increase in the Thus, the estimate of the effect of a gap does not depend
hazard for an extension of 3 or more days is by a factor of crucially on how the effect of the different schedules are
e0.56 = 1.75 (95% CI 1.20–2.55). This corresponds to a 1.6% modelled. The effect of treatment extension is to reduce the
loss in local control per day on a 5-year local control rate of local control rate by 0.9% per day at a 90% 5-year rate. This
90%. is slightly smaller, but still comparable, to the figure calcu-
lated from the survival analysis.
4.2. Modelling local control rates at 5 years
4.3. Gap position
Adjusting for T stage, there are no centre differences in
the 5-year local control rates (x2 = 5.46, 4 d.f.). In eqns (1) The distribution of the position of the first gap is given in
and (2) the signs of the parameters associated with the effect Table 4. This analysis can only be performed on patients
of dose, D and Dd, are negative in Section 3. This means who have exactly one gap, as patients with multiple gaps are
that all estimated parameters are expected to be positive. more likely to have the first gap at the beginning of the
Positive estimates for dose imply that an increase in the dose schedule. Each schedule was divided into three on the
is associated with an increase in the local control rates, basis of the position of the gap with reference to the number
whereas an increase in the values of the other variables is of fractions prescribed. For those with one gap there is not a
associated with a reduction in the local control rates. This is uniform distribution of gap position over the treatment per-
not the case with Eq. (3) where positive values are asso- iod (P = 0.05) and there are slightly more gaps in the middle
ciated with a decrease in the local control rates for dose, third and slightly fewer in the first third.
planned time and treatment extension. The parameter esti- Among those with a gap of exactly 1 day, there is no
mates of these models are presented in Table 3. evidence of any different hazard of a failure of local control
At 5-year follow-up, in Eq. (3) there is no significant associated with gap position (x2 = 1.85, 2 d.f.). Surpris-
effect of dose (x2 = 0.65, 3 d.f.) or planned treatment time ingly, there is clear evidence that among patients with no
(x2 = 2.65, 3 d.f.). There is clear evidence that a treatment gap or one gap of 1 day in the schedule, those with a gap
extension is associated with a significant reduction in the have a significantly greater hazard of a failure of local con-
local control rate (x2 = 10.88, 3 d.f., P = 0.01). trol (P = 0.005) (Table 5). This is unexpected in view of the
From Eq. (1), there is some evidence that longer treat- results in Table 3. It is principally driven by cases within the
ment times are associated with a reduction in the local con- lowest dose group or in stage T1. Outside of these areas
trol rates. The estimated effect of an increase of 1 day in the there is no evidence of an effect of a 1-day gap. This is
treatment time is to increase ln(−ln(Pc)) by 0.029 (0.014) possibly just a subgroup effect but may indicate that T1
units (P = 0.04), i.e. to reduce Pc. Using Eq. (2), the esti- patients or low dose patients are particularly susceptible to
mated effect of an extension to the treatment time of 1 day the effects of a gap.
as a result of an unplanned gap is 0.078 (0.031) (P = 0.01). Within the model for the local control rates at 5 years
In Eq. (2), the split of the total treatment time between there is slight evidence (P = 0.07) that even a gap of 1 day is
planned time and treatment extension due to an unplanned associated with an increased probability of a failure of local
gap reveals that there is some evidence that the extension is control (Table 5). There is no evidence that the position of
the more important of the time effects, based on the magni- the gap has any major effect on the local control rate
tude of the effects and their standard errors (P = 0.08). (x2 = 1.92, 2 d.f.). The parameter estimates suggest that
From Eq. (1), the maximal time factor for the actual time is the greatest effect comes with gaps in the last third of the
estimated as g/a = 1.76 Gy with a standard error of 3.97 Gy. schedule but the information is not very reliable.
The a/b ratio is not precisely estimated in view of the very
small estimated value for b, which led to the omission of Dd
from the models. Also, ln(2)/g is an estimate of the tumour
doubling time and it is 24.1 days (95% CI 1.3–47.1 days). Table 4
The analysis of gap length was extended to use the treat- Number of gaps by the position of the first gap
ment schedule rather than the effects of dose and planned
No. of No gap Third of schedule Total
time to model treatment schedule effects. Scrutiny of the gaps
prescribed days, prescribed fractions and prescribed dose First Second Final
enabled the identification of 13 treatment schedules which
0 292 (100) – – – 292
represent the main treatment regimes. There were no sig- 1 1 (0.6)a 44 (25.9) 70 (41.2) 55 (32.4) 170
nificant differences in local control rates over the 13 sche- 2 – 58 (54.2) 44 (41.1) 5 (4.7) 107
dules (x2 = 7.61, 12 d.f.), but there was evidence that a gap 3 – 32 (71.1) 13 (28.9) – 45
in the treatment was associated with a decrease in the local 4+ – 13 (86.7) 2 (13.3) – 15
control rates (P = 0.01, 1 d.f.). The linear effect of a treat- Total 293 147 129 60 629
ment extension of 1 day was estimated as 0.080 (0.031) per Values in parentheses are percentages.
a
day and is comparable with the estimate in Table 3 (Eq. (2)). Treatment finished early for this patient after fraction 26 due to illness.
 A.G. Robertson et al. / Radiotherapy and Oncology 48 (1998) 165–173
Table 5
Parameter estimates (standard errors) to assess the effect of gap position for those patients with only one gap of 1 day
Logistic regression estimates
for the 5-year local control rates
Gap of 1 day
Intercept −1.203 (0.200)
T stage
T1 0 (0)
T2 0.354 (0.256)
T3T4 1.137 (0.279)
Dose (Gy) 0.015 (0.040)
Planned time −0.008 (0.023)
One gap 0.477 (0.261)
Gap position
First third
Second third
Final third
− 2 log likelihood 17.55
Degrees of freedom 5 7
5. Discussion
This analysis has demonstrated that long gaps in the treat-
ment time in excess of 2 days are associated with a poorer
local control. In this analysis, and also in that of Duncan et
al. [6], who report similar but less extensive results, treat-
ment gaps were identified as opposed to using excess treat-
ment time over planned treatment time according to
specified schedules. The data are unselected using all
cases from a variety of centres both large and small, reflect-
ing the practice in the real world, as opposed to some large
second referral centre. There was no evidence of any differ-
ences in local control rates between the centres.
Within the linear quadratic model for local control rates
and the linear quadratic model for the hazard of a local
recurrence, there is evidence that long treatment times are
associated with a poorer outcome. The results for actual
treatment time are stronger than the results for the planned
treatment time. However, long gaps in treatment are cer-
tainly bad, but there is evidence that even a gap of 1 day
can have a deleterious effect. This is so when the gap leads
to a treatment extension of 3 or more days as a result of an
extra weekend creeping into the schedule. This analysis has
demonstrated that a schedule in which there is a treatment
extension of 3 days or more has significantly poorer local
control than a schedule completed in the planned time.
In the log linear modelling of the 5-year local control
rates, the coefficients associated with the quadratic dose
term, b in eqns (1) and (2), were not significantly different
from 0 and were imprecisely estimated. Omission of this
term from the models serves to increase the precision of
the other estimates but has no effect on the overall conclu-
sions. The same is true in the proportional hazards survival
analysis in eqns (4) and (5).
Hansen et al. [9] report that the tumour differentiation is
important in the effect of extensions to the treatment time
171
Cox regression estimates for the hazard
of a failure of local control
Gap position Gap of 1 day Gap position
−1.190 (0.208) – –
0 (0) 0 (0) 0 (0)
0.349 (0.256) 0.301 (0.258) 0.316 (0.258)
1.124 (0.278) 1.365 (0.275) 1.338 (0.277)
0.022 (0.040) 0.038 (0.042) 0.039 (0.042)
−0.006 (0.023) −0.002 (0.022) 0.000 (0.023)
0.718 (0.256)
0.126 (0.444) 0.641 (0.408)
0.173 (0.329) 0.349 (0.361)
0.405 (0.394) 0.874 (0.437)
15.63 27.92 26.07
5 7
only in moderately and well differentiated tumours among
advanced supraglottic and pharyngeal cancers. We were
unable to replicate these results as 28% of cases did not
have tumour differentiation recorded and 78% of those
with differentiation recorded were moderately or well dif-
ferentiated. There have been two recent reports of rando-
mized trials of accelerated versus conventional radiotherapy
in head and neck cancers [15,23]. Both report a high inci-
dence of normal tissue effects in the accelerated arm com-
pared to conventional radiotherapy. This suggests that
shortening of the total treatment time may not be acceptable.
Consequently, the evidence that gaps in conventional sche-
dules are detrimental means that greater attention needs to
be paid to maintaining the existing schedules without treat-
ment time extensions. There have been other reports of
increased local control with accelerated fractionation [14]
and of beneficial results from CHART [5], pointing to
tumour repopulation as the key to radiation failure. Our
study supports these conclusions.
Gaps at the beginning and end of the schedule have been
found to be potentially more harmful than gaps in the mid-
dle of the schedule [30]. Herrmann et al. [13] found that
gaps in the second half of the schedule were more harmful
than gaps at the beginning of the schedule. We find that gap
position is not important.
In Herrmann et al. [13], over half of the 192 patients had a
gap of 4 or more days and one-quarter had a gap of 2 or more
weeks. These gaps are longer than in our data (Tables 1 and
2) where there were treatment extensions of greater than 4
days for 16% of patients and 1 week or more for only 4% of
cases. Details of the number of gaps for each patient were
not presented [13] but most had just one gap which was
longer than observed in our Scottish data. The results of
Skladowski et al. [30] were based on 971 patients of
whom 76% had exactly one gap and only 12% had no
gap. The mean duration of a gap was 8.2 days, which is
172 A.G. Robertson et al. / Radiotherapy and Oncology 48 (1998) 165–173
much greater than we observed. In many cases, the one gap
was a planned gap designed to reduce acute tissue reactions.
In both of these studies, the position of the gap is impor-
tant when the delay due to the gap is long. In our study, we
are looking at the effect of gaps of a relatively short duration
in an environment where just under half of the patients have
no gap. Thus, our study should not be thought of as a repli-
cate of the previous two studies but more as providing infor-
mation in areas which are not covered by these two studies.
We find no evidence that gaps at the beginning or end of
the schedule are potentially more harmful than those in the
middle of the schedule. Thus, a gap is a serious adverse
event in a radiotherapy schedule irrespective of its position
and more importantly, even when the treatment extension
resulting from the gap is as short as 3 days; these gaps were
discarded in the analysis of Skladowski et al. [30].
Furthermore, it is not the length of the gap itself which is
the only cause for concern, as the weekend can turn a 1 day
gap into a 3 day gap. Our analysis shows that while the
effect of a 1 day treatment extension is not statistically
significantly different from the effect of no extension, a 3
day extension most certainly is. Consequently, attempts
should be made to minimize the occurrence of gaps in the
radiotherapy schedule. If gaps occur then some attempt
should be made to minimize the effect and methods which
do not increase the total treatment time are to be preferred
[12].
Acknowledgements
The data on the treatment of cancer of the larynx by
radiotherapy were collected under the auspices of the Scot-
tish Oncology and Radiology Audit Group as part of a study
funded by the Clinical Resource and Audit Group (GRAG),
Scottish Home and Health Department, whose help is grate-
fully acknowledged. It is a pleasure to acknowledge the
contribution of the staff in the centres who helped with
the collection of the data.
References
[1] Barton, M.B., Keane, T.J., Gadalla, T. and Maki, E. The effect of
treatment time and treatment interruption on tumour control follow-
ing radical radiotherapy of laryngeal cancer. Radiother. Oncol. 23:
137–143, 1992.
[2] Bentzen, S.M. Radiobiological considerations in the design of clin-
ical trials. Radiother. Oncol. 32: 1–11, 1994.
[3] Cox, D.R. Regression models and life tables (with discussion). J. R.
Stat. Soc., Ser. B 74: 187–220, 1972.
[4] Cox, D.R. and Snell, E.J. The Analysis of Binary Data, 2nd edn.
Chapman and Hall, London, 1990.
[5] Dische, S., Saunders, M., Barrett, A., Harvey, A., Gibson, D. and
Parmar, M. A randomised multicentre trial of CHART versus con-
ventional radiotherapy in head and neck cancer. Radiother. Oncol.
44: 123–136, 1997.
[6] Duncan, W., MacDougall, R.H., Kerr, G. and Downing, D. The
adverse effects of treatment gaps in the outcome of radiotherapy
for laryngeal cancer. Radiother. Oncol. 41: 203–207, 1996.
[7] Fowler, J.F. and Lindstrom, M.J. Loss of local control with prolon-
gation in radiotherapy. Int. J. Radiat. Oncol. Biol. Phys. 23: 457–
467, 1992.
[8] Fowler, J.F. and Chappell, R. Effect of overall time and dose on the
response of glottic carcinoma of the larynx to radiotherapy (letter).
Eur. J. Cancer 30A: 719–721, 1994.
[9] Hansen, O., Overgaard, J., Hansen, H.S., et al. Importance of overall
treatment time for the outcome of radiotherapy of advanced head and
neck carcinoma: dependency on tumor differentiation. Radiother.
Oncol. 43: 47–51, 1997.
[10] Harrington, D.P. and Fleming, T.R. A class of rank test procedures
for censored survival data. Biometrika 69: 553–566, 1982.
[11] Hendry, J.H., Roberts, S.A., Slevin, N.J., Keane, T.J., Barton, M.B.
and Agren-Cronqvist, A. Influence of radiotherapy treatment time on
control of laryngeal cancer: comparisons between centres in Man-
chester, UK and Toronto, Canada. Radiother. Oncol. 31: 14–22,
1994.
[12] Hendry, J.H., Bentzen, S.M., Dale, R.S., et al. A modelled compar-
ison of the effects of using different ways to compensate for missed
treatment days in radiotherapy. Clin. Oncol. 8: 297–307, 1996.
[13] Herrmann, T., Jakubek, A. and Trott, K.R. The importance of the
timing of a split in radiotherapy of squamous cell carcinomas of the
head and neck. Strahlenther. Onkol. 170: 545–549, 1994.
[14] Horiot, J.C., Bontemps, P., van den Bogaert, W., et al. Accelerated
fractionation (AF) compared to conventional fractionation (CF)
improves loco-regional control in the radiotherapy of advanced
head and neck cancers: results of the EORTC 22851 randomised
trial. Radiother. Oncol. 44: 111–121, 1997.
[15] Jackson, S.M., Weir, L.M., Hay, J.H., Tsang, V.H. and Durham, J.S.
A randomised trial of accelerated versus conventional radiotherapy
in head and neck cancer. Radiother. Oncol. 43: 39–46, 1997.
[16] Jung, S.H. Regression analysis for long-term survival rate. Biome-
trika 83: 227–232, 1996.
[17] Kalbfleisch, J.D. and Prentice, R.L. The Statistical Analysis of Fail-
ure Time Data. Wiley, New York, 1980.
[18] Kaplan, E.L. and Meier, P. Non-parametric estimation from incom-
plete samples. J. Am. Stat. Assoc. 53: 457–481, 1958.
[19] McCullagh, P. and Nelder, J.A. Generalised Linear Models, 2nd edn.
Chapman and Hall, London, 1989.
[20] Maciejewski, B., Taylor, J.M.G. and Withers, H.R. Alpha/beta value
and the importance of size of dose per fraction for late complications
in the supraglottic larynx. Radiother. Oncol. 7: 323–326, 1986.
[21] Maciejewski, B., Preuss-Bayer, G. and Trott, K.R. The influence of
the number of fractions and of overall treatment time on local control
and late complications rate in squamous cell carcinoma of the larynx.
Int. J. Radiat. Oncol. Biol. Phys. 9: 321–328, 1983.
[22] Maciejewski, B., Withers, H.R., Taylor, J.M.G. and Hliniak, A. Dose
fractionation and regeneration in radiotherapy for cancer of the oral
cavity and oropharynx. Part 1. Tumour dose response and regenera-
tion. Int. J. Radiat. Oncol. Biol. Phys. 16: 831–843, 1989.
[23] Maciejewski, B., Skladowski, K., Pilecki, B., et al. Randomised
clinical trial on accelerated 7 days per week fractionation in radio-
therapy for head and neck cancer. Preliminary report on acute toxi-
city. Radiother. Oncol. 40: 137–145, 1996.
[24] Oakes, D.A. Note of the Kaplan Meier Estimator. Am. Statist. 47:
39–40, 1993.
[25] Overgaard, J., Hjelm-Hansen, M., Johansen, L.V. and Andersen, A.P.
Comparison of conventional and split-course radiotherapy as primary
treatment in carcinoma of the larynx. Acta Oncol. 27: 147–152,
1988.
[26] Roberts, S.A., Hendry, J.H., Brewster, A.E. and Slevin, N.J. The
influence of radiotherapy treatment time on the control of laryngeal
cancer: a direct analysis of data from two British Institute of Radio-
therapy trials to calculate the lag period and the time factor. Br. J.
Radiol. 67: 790–794, 1994.
 A.G. Robertson et al. / Radiotherapy and Oncology 48 (1998) 165–173
[27] Robertson, A.G., Robertson, C., Boyle, P., Symonds, R.P. and Whel-
don, T.E. The effect of differing radiotherapeutic schedules on the
response of glottic carcinoma of the larynx. Eur. J. Cancer 29A:
501–510, 1993.
[28] Robertson, C., Robertson, A.G., Hendry, J.H., et al. The effect of
overall treatment time and unplanned gaps on local control of carci-
noma of the larynx treated by radiotherapy in four centres. Int. J.
Radiat. Oncol. Biol. Phys. 40: 319–329, 1998.
[29] Sambrooke, D. Clinical trial of a modified (split course) technique of
X-ray therapy in malignant tumours. Clin. Radiol. 36: 369–372,
1968.
[30] Skladowski, K., Law, M.G., Maciejewski, B. and Steel, G.G. Planned
173
and unplanned gaps in radiotherapy: the importance of gap position
and gap duration. Radiother. Oncol. 30: 109–120, 1994.
[31] Slevin, N.J., Hendry, J.H., Roberts, S.A. and Agren-Cronqvist, A.
The effect of increasing the treatment time beyond three weeks on
the control of T2 and T3 laryngeal cancer using radiotherapy. Radio-
ther. Oncol. 24: 215–220, 1992.
[32] Statistical Analysis System version 6.11. SAS Institute, Cary, NC,
1996.
[33] Withers, H.R., Taylor, J.M.C. and Maciejewski, B. The hazard of
accelerated tumor clonogen repopulation during radiotherapy. Acta
Oncol. 27: 131–146, 1988.