Course :
Biostatistics
Modules :
Lecturer:
Period:
9.00-12.00
Date :
6 August 2007
SURVIVAL ANALYSIS
Event
Dead, infection, relapsed, etc
Negative
Positive
Neutral
Time to event
General: event-free duration
relapse-free survival time
remission duration
progression-free survival
SURVIVAL TIME
Time to event
Failed: event occurred at time of observation
Censored: event have not occurred at time of observation
Censoring
Event-free duration cannot be determined
Event have not yet occurred
Lost to follow-up
Competing event
2540
Begin the study
2541
2542
2543
End of the study
Start of accrual
End of accrual
End of follow-up
Recruitment period
Follow-up period
1
48 months
22 months
14 months
40 months
26 months
13 months
Dead
Dead
>22
14
40
>26
>13
DATA
ID
SURVIVAL TIME
(Months)
1
2
3
4
5
6
48
22
14
40
26
13
EVENT
Censored
Censored
Dead
Dead
Censored
Censored
DATA
ID
TIME
1
2
3
4
5
6
48
22
14
40
26
13
EVENT
ID
TIME
Censored
Censored
Dead
Dead
Censored
Censored
1
2
3
4
5
6
48
22
14
40
26
13
EVENT
0
0
1
1
0
0
ANALYSIS
ID
TIME
1
2
3
4
5
6
48
22
14
40
26
13
EVENT
0
0
1
1
0
0
Prevalence = 2/6
Incidence density = 2/163 person-months
Proportion of surviving at month t
Median survival time
. list
1.
2.
3.
4.
5.
6.
id
1
2
3
4
5
6
time
48
22
14
40
26
13
event
0
0
1
1
0
0
. stsum
failure _d:
analysis time _t:
event
time
|
incidence
no. of
|------ Survival time -----|
| time at risk
rate
subjects
25%
50%
75%
---------+--------------------------------------------------------------------total |
163
.0122699
6
40
40
.
. cii 163
0.0122699 , p
-- Poisson Exact -Variable | Exposure
Mean
Std. Err.
[95% Conf. Interval]
---------+------------------------------------------------------------|
163
.0122699
.0086762
.0014888
.0443053
. stquant
Number of deads = 2
Number of ties = 0
+---------------------------------------------------------------------------+
| 50 percentile
95% Confidence Interval
|
| survival time =
40.000
19.650
60.350
(Large Sample Approx.) |
|
14.000
48.000
(Brookmeyer-Crowley)
|
+---------------------------------------------------------------------------+
event
time
Beg.
Survivor
Std.
Time
Total
Fail
Function
Error
[95% Conf. Int.]
------------------------------------------------------------------------------12
0
0
1.0000
.
.
.
24
4
1
0.8000
0.1789
0.2038
0.9692
36
3
0
0.8000
0.1789
0.2038
0.9692
48
1
1
0.4000
0.2966
0.0114
0.8290
------------------------------------------------------------------------------Note: Survivor function is calculated over full data and evaluated at
indicated times; it is not calculated from aggregates shown at left.
. sts g, at risk
failure _d:
analysis time _t:
event
time
0.75
0.50
1
0.25
0.00
0
20
40
60
analysis time
RESULTS
ID
TIME
1
2
3
4
5
6
48
22
14
40
26
13
EVENT
0
0
1
1
0
0
Kaplan-Meier methods
Original Data
ID TIME
1 48
2 22
3 14
4 40
5 26
6 13
13
14
22
26
40
48
Number
at risk
nj
DEAD
0
0
1
1
0
0
ID TIME
6 13
3 14
2 22
5 26
4 40
1 48
Censored
qj
Dead
dj
6
6-1 = 5
5-1 = 4
4-1 = 3
6-1 = 2
2-1 = 1
0
1
0
0
1
0
1
0
1
1
0
1
DEAD
0
1
0
0
1
0
Survival probability
S(tj)
6/6 = 1.0
[(5-1)/5]1.0 = 0.8
[(4-0)/4]0.8 = 0.8
[(3-0)/3]0.8 = 0.8
[(2-1)/2]0.8 = 0.4
[(1-0)/1]0.4 = 0.4
Time
tj
0.5
13 14
22 26
Time (months)
40
48
Survival analysis
Computing notes and Exercises
http://www.pauldickman.com/teaching/veneto2003/
Contents
1 Survival analysis using Stata
12
References
12
Once the data have been stset we can use any of these commands without having to
specify the survival time or failure time variables. For example, to plot the estimated
cause-specific survivor function by sex and then fit a Cox proportional hazards model
with sex and calendar period as covariates
. sts graph, by(sex)
. stcox sex year8594
Note that we only have to stset the data once. You can also tell Stata to show the
number at risk or the number of censored observations on the Kaplan-Meier plot
. sts graph, atrisk
. sts graph, lost
3. For the patients diagnosed with localised skin melanoma, use Stata to estimate the
cause-specific survivor function, using the Kaplan-Meier method with survival time
in months, separately for each of the two calendar periods 19751984 and 19851994.
The following commands can be used
.
.
.
.
use melanoma
keep if stage == 1
stset surv_mm, failure(status==1)
sts graph, by(year8594)
(a) Without making reference to any formal statistical tests, does it appear that
patient survival is superior during the most recent period?
(b) The following commands can be used to plot the hazard function (instantaneous
mortality rate):
. sts graph, hazard by(year8594)
At what point in the follow-up is mortality highest? Does this pattern seem
reasonable)? Do you feel that there is there more information in this plot
compared to the plot of the survivor function?
4. In question 3 we studied plots of the survivor function for patients diagnosed with
localised skin melanoma by calendar period of diagnosis. Use the log rank test to
determine whether there is a statistically significant difference in patient survival
between the two periods. The following command can be used:
. sts test year8594
What do you conclude?
An alternative test is the generalised Wilcoxon, which can be obtained as follows
. sts test year8594, wilcoxon
5. Lets now read the melanoma data again, but study all stages.
. use melanoma, clear
. stset surv_mm, failure(status==1)
(a) Plot estimates of the survivor function and hazard function by stage. Does it
appear that stage is associated with survival?
(b) Estimate the mortality rates for each stage using, for example, the strate
command. What are the units of the estimated rates?
(c) If you havent already done so, estimate the mortality rates for each stage per
1000 person-years of follow-up.
(d) Study whether survival is different for males and females (by plotting the survivor function or tabulating rates).
6. Load the diet data and stset the data using time-on-study as the timescale.
. use diet, clear
. stset dox, id(id) fail(chd) origin(doe) scale(365.25)
(a) Use the strate command to tabulate CHD mortality rates per 1000 personyears for each category of hieng. Calculate (by hand) the ratio of the two
incidence rates.
(b) Use the command poisson to find the incidence rate ratio for the high energy
group compared to the low energy group and compare the estimate to the one
you obtained in the previous question:
. poisson chd hieng, e(y) irr
(c) Grouping the values of total energy into just two groups does not tell us much
about how the CHD rate changes with total energy. It is a useful exploratory
device, but to look more closely we need to group the total energy into perhaps
3 or 4 groups. In this example we shall use the cut points 1500, 2500, 3000, 4500.
(d) Use the commands
. egen eng3=cut(energy), at(1500, 2500, 3000, 4500)
. tabulate eng3
to create a new variable eng3 coded 1500 for values of energy in the range
150002499, 2500 for values in the range 25002999, and 3000 for values in the
range 30004500. These codes are called the levels of the variable.
(e) To find the rate for different levels of eng3 try
.
The option graph will show a graph of rate against levels of exposure.
.
(f) Create your own indicator variables for the three levels of eng3 with
. tabulate eng3, gen(X)
(g) Check the indicator variables with
. list X1 X2 X3 if eng3==1500
. list X1 X2 X3 if eng3==2500
. list X1 X2 X3 if eng3==3000
(h) Use poisson to compare the second and third levels with the first, as follows:
. poisson chd X2 X3, e(y) irr
(i) Use poisson to compare the first and third levels with the second.
(j) Use xi: poisson to compare the second and third levels with the first, creating
the indicators automatically with i.eng3.
7. Now lets study cause-specific survival using the melanoma data restricted to localised stage (stage==1). The following commands can be used to load and stset
the data.
.
.
.
.
(a) Later we will use Cox regression to analyse these data. For now we will tabulate
mortality rates and model them using Poisson regression. We expect mortality
to depend on time since diagnosis so we need to split the data by this timescale.
We will restrict our analysis to mortality up to 10 years following diagnosis.
stsplit fu, at(0(1)10) trim
(b) Now tabulate (and produce a graph of) the rates by follow-up time.
strate fu, per(1000) graph
Mortality appears to be quite low during the first year of follow-up. Does this
seem reasonable?
(c) Compare the plot of the estimated rates to a plot of the hazard rate as a function
of continuous time.
sts graph, hazard
Is the interpretation similar? Do you think it is sufficient to classify follow-up
time into annual intervals or might it be preferable to use, for example, narrower
intervals?
(d) Use Poisson regression to estimate incidence rate ratios as a function of followup time.
xi: streg i.fu, dist(exp)
Does the pattern of estimated incident rate ratios mirror the pattern you observed in the plots?
(e) Now control for age, sex, and calendar period.
xi: streg i.fu i.agegrp year8594 sex, dist(exp)
What conclusions can you draw from the fitted model? Is there evidence that
the effect of follow-up confounded by age, sex, and calendar period?
8. Using the melanoma data restricted to localised stage (stage==1), fit a Cox proportional hazards model (for cause-specific survival) with calendar period as the only
explanatory variable. The following commands can be used
.
.
.
.
use melanoma
keep if stage == 1
stset surv_mm, failure(status==1)
stcox year8594
(a) Interpret the estimated hazard ratio, including a comment on statistical significance.
(b) (This part is more theoretical and is not required in order to understand the
remaining parts.)
Stata reports a Wald test of the null hypothesis that survival is independent
of calendar period. The test statistic (and associated P-value) is reported in
the table of parameter estimates (labelled z). Under the null hypothesis, the
test statistic has a standard normal (Z) distribution, so the square of the test
statistic will have a chi square distribution with one degree of freedom.
Stata also reports a likelihood ratio test statistic of the null hypothesis that
none of the parameters in the model are associated with survival (labelled
LR chi2(1)). In general, this test statistic will have a chi-square distribution
with degrees of freedom equal to the number of parameters in the model. For
the current model, with only one parameter, the test statistic has a chi square
distribution with one degree of freedom.
Compare these two test statistics with each other and with the log rank test
statistic (which also has a 21 distribution) calculated in question 4.
Would you expect these test statistics to be similar? Consider the null and
alternative hypotheses of each test and the assumptions involved with each
test.
(c) Now include sex and age (in categories) in the model. The following command
can be used
. xi: stcox sex year8594 i.agegrp
(d) Interpret the estimated hazard ratio for the parameter labelled Iagegr_2, including a comment on statistical significance.
(e) Is the effect of calendar period strongly confounded by age and sex? That is,
does the inclusion of sex and age in the model change the estimate for the effect
of calendar period?
(f) Perform a Wald test of the overall effect of age and interpret the results.
. test _Iagegrp_1 _Iagegrp_2 _Iagegrp_3
(g) Perform a likelihood ratio test of the overall effect of age and interpret the
results. The following commands can be used
.
.
.
.
9. Now fit a model to the localised melanoma data, with the same explanatory variables
as in the previous question, but where the outcome is observed survival (i.e. all deaths
are considered to be events).
. stset surv_mm, failure(status==1,2)
. keep if stage==1
. xi: stcox sex year8594 i.agegrp
(a) Interpret the estimated hazard ratio for the parameter labelled Iagegr_2, including a comment on statistical significance.
(b) On comparing the estimates between the observed and cause-specific survival
models it appears that only the parameters for age have changed substantially.
Can you explain why the estimates for the effect of age would be expected to
change more than the estimates of the effect of sex and period?
10. Use Cox regression to model the cause-specific survival of patients with skin melanoma
(including all stages).
(a) First fit the model with sex as the only explanatory variable. Does there appear
to be a difference in survival between males and females?
(b) Is the effect of sex confounded by other factors (e.g. age, stage, subsite, period)?
After controlling for potential confounders, does there appear to a difference in
survival between males and females?
(c) Decide on a most appropriate model. Be sure to evaluate the proportional
hazards assumption.
11. For the melanoma data, plot the log cumulative hazard function for each calendar
period. The following command can be used
. stphplot, by(year8594) c(ll) xlabel ylabel
(a) Do you think that a proportional hazards assumption is appropriate for these
data?
(b) Does the appropriateness of the proportional hazards assumption have any
implications for the log rank test?
(c) From the plot, estimate the hazard ratio for patients diagnosed 198594 to those
diagnosed 197584. You may find this easier if you plot the curves without
symbols; this can be done by including the s(ii) option to the stphplot
command. In the next question we will estimate this hazard ratio using a Cox
proportional hazards model.
To study how rates vary jointly on two-time scales we need to split the records twice.
1. Start by breaking the followup for the diet data into 5 year age bands
. use diet, clear
. stset dox, fail(chd) origin(dob) entry(doe) scale(365.25) id(id)
. stsplit ageband, at(30(5)70) trim
2. Now break these new records into 5 year calendar period bands using
. stsplit period, after(time=d(1/1/1900)) at(50(5)80) trim
. replace period=period+1900
. list id ageband period in 1/20
Note that we have used the second syntax for stsplit and set the origin for calendar
period as 1/1/1900 for convenience in setting the breaks.
3. Each subjects followup is now divided into small pieces corresponding to the age
bands and calendar period bands the subject passes through. We can make tables
of deaths and person-years by ageband and period with
. gen _y=_t - _t0 if _st==1
. table ageband period, c(sum _d)
. table ageband period, c(sum _y) format(%5.1f)
4. To make a table of rates per 1000 by ageband and period, try
. gen obsrate=_d/_y*1000
. table ageband period [iw=_y], c(mean obsrate) format(%5.1f)
5. To calculate the expected cases for a cohort, using reference rates classified by age
and calendar period, it is first necessary to break the followup into parts which
correspond to these age bands and calendar periods, as above. Since there are no
suitable rates for these data, we have used reference rates which are uniform for both
age and calendar period at 11/1000.
The reference rates are in the file ref. Before calculating the expected number of
cases it is necessary to add the reference rates to the expanded data with
. sort ageband period
. merge ageband period using ref
This is a matched merge on age band and calendar period and will add the appropriate reference rate to each record. The system variable merge takes the following
values:
1 record in the master file but no match in ref
2 record in ref but no match in the master file
3 record in the master file with a match in ref
9
. tab _merge
should show mostly 3s with some 2s but no 1s. You can now drop the records with
no match in the master file
. drop if _merge==2
6. To calculate the expected number of cases, multiply the follow-up time for each
record by the reference rate for that record:
. gen e=_y*refrate/1000
. list id e _d if in 1/20
The SMR is the ratio of the total observed cases to the total number expected, and
is most easily obtained with
. strate, smr(refrate) per(1000)
or with
. smrby _d e
7. To calculate the SMR for the high and low energy groups,try
. strate hieng, smr(refrate) per(1000)
or
. smrby _d e, by(hieng)
10
11
The Finnish Cancer Registry is population-based and covers the whole of Finland (population 5.1 million). The Registry was established in 1952, with 1953 being the first calendar
year with complete registration. The Registry obtains information from many different
sources: hospitals and other institutions with inpatient beds, physicians working outside
hospitals, dentists, and pathological and cytological laboratories. The Finnish Cancer
Registry also receives copies of all death certificates where cancer is mentioned. Notification of new cancer cases to the Cancer Registry is mandatory by law. If the reported
information is deficient or contradictory, requests are sent to informants in order to ensure
accuracy in the following areas: patient details, the primary site of the tumour, and the
date of diagnosis.
The diseases registered at the Finnish Cancer Registry include, in addition to all clearly
malignant neoplasms, carcinoma in situ lesions (except those of the skin), all neoplasms of
the intracranial space and spinal canal irrespective of their malignancy, benign papillomas
of the urinary organs, semimalignant tumours of the ovary, basal cell carcinomas of the
skin, and cases of polycythaemia vera and myelofibrosis. Various check-ups have shown
that the coverage of the Cancer Registry file is almost complete with respect to cancer
cases diagnosed in the Finnish population [1, 2]. All independent primary neoplasms in
the same person are registered separately. When evaluating whether a new tumour is an
independent cancer or a recurrence, attention is focused on, among other aspects, the time
interval between the tumours, histology, and knowledge of the general behaviour of each
cancer type. In principle, multiple metachronous tumours in the same organ (e.g., in the
colon or skin) are registered separately, especially when they have different histologies.
However, each case is evaluated individually and a primary site code multiple cancer
is also available for some organs. The International Classification of Diseases Volume 7
(ICD-7) is used at the Finnish Cancer Registry. Further details of the registry can be
found in the annual incidence publications [3].
The Finnish Cancer Registry has kindly provided data on patients diagnosed with skin
melanoma and colon carcinoma in Finland during 19751994 with follow-up to 31 December 1995. A detailed description and analysis of these data is given in Dickman et al.
(1999) [4].
References
[1] Teppo L, Pukkala E, Lehtonen M. Data quality and quality control of a populationbased cancer registry. Experience in Finland. Acta Oncologica 1994;33:365369.
[2] Hakulinen T. Health care system, cancer registration and follow-up of cancer patients
in Finland. In: Berrino F, Sant M, Verdecchia A, Capocaccia R, Hakulinen T, Est`eve J,
eds., Survival of Cancer Patients in Europe: The EUROCARE Study, IARC Scientific
Publications No. 132. Lyon: International Agency for Research on Cancer, 1995; 5354.
[3] Finnish Cancer Registry. Cancer Incidence in Finland 1995 . Cancer Society of Finland
Publication No. 58. Helsinki: Cancer Society of Finland, 1997.
[4] Dickman PW, Hakulinen T, Luostarinen T, Pukkala E, Sankila R, Soderman B, Teppo
L. Survival of cancer patients in Finland 1955-1994. Acta Oncologica 1999;38 (Suppl.
12):1103.
12
EXERCISE
1. The data set
id
1
2
5
3
4
6
8
7
9
10
11
12
13
15
14
16
17
18
19
20
drug
0
0
0
0
0
0
0
0
0
0
1
1
1
1
1
1
1
1
1
1
cured
1
1
1
1
1
1
1
1
0
0
1
1
1
1
1
1
1
1
0
0
time
6
7
12
12
12
13
14
14
30
50
1
1
2
2
2
3
3
4
15
16
failure event:
obs. time interval:
exit on or before:
. stsum, by(drug)
failure _d:
analysis time _t:
cured
time
|
incidence
no. of
|------ Survival time -----|
drug
| time at risk
rate
subjects
25%
50%
75%
---------+--------------------------------------------------------------------0 |
170
.0470588
10
12
12
14
1 |
49
.1632653
10
2
3
4
---------+--------------------------------------------------------------------total |
219
.0730594
20
3
12
14
cured
time
0.75
0.50
0.25
drug 1
0.00
0
drug 0
20
40
60
analysis time
20
16
219
-38.788082
Number of obs
20
LR chi2(1)
Prob > chi2
=
=
2.35
0.1252
-----------------------------------------------------------------------------_t |
_d | Haz. Ratio
Std. Err.
z
P>|z|
[95% Conf. Interval]
---------+-------------------------------------------------------------------drug |
2.231263
1.1555
1.550
0.121
.8086128
6.156883
------------------------------------------------------------------------------
Papers
Abstract
Objective To evaluate the efficacy of using a nicotine
patch for 5 months with a nicotine nasal spray for 1
year.
Design Placebo controlled, double blind trial.
Setting Reykjavik health centre.
Subjects 237 smokers aged 22-66 years living in or
around Reykjavik.
Interventions Nicotine patch for 5 months with
nicotine nasal spray for 1 year (n = 118) or nicotine
patch with placebo spray (n = 119). Treatment with
patches included 15 mg of nicotine for 3 months,
10 mg for the fourth month, and 5 mg for the fifth
month, whereas nicotine in the nasal spray was
available for up to 1 year. Both groups received
supportive treatment.
Main outcome measure Sustained abstinence from
smoking.
Results The log rank test for 6 years (2 = 8.5,
P = 0.004) shows a significant association between
abstinence from smoking and type of treatment.
Sustained abstinence rates for the patch and nasal
spray group and patch only group were 51% v 35%
after 6 weeks (P = 0.011 (2), 95% confidence interval
1.17% to 3.32%), 37% v 25% after 3 months
(P = 0.045, 1.01% to 3.08%), 31% v 16% after 6
months (P = 0.005, 1.27% to 4.50%), 27% v 11% after
12 months (P = 0.001, 1.50% to 6.14%), and 16% v 9%
after 6 years (P = 0.077, 0.93% to 4.72%).
Conclusions Short and long term abstinence rates
show that the combination of using a nicotine patch
for 5 months with a nicotine nasal spray for 1 year is a
more effective method of stopping smoking than
using a patch only. The low percentage of participants
using the nasal spray at 1 year, and the few relapses
during the second year, suggest that it is not cost
effective to use a nasal spray for longer than 7 months
after stopping a patch.
Introduction
In controlled clinical trials of nicotine replacement
therapy, 1 in 5 smokers remained abstinent after 1 year
compared with 1 in 10 smokers who abstained in self
administered cessation.1 Better treatments for smoking
cessation are clearly needed. In several controlled studies, the value of treatment with nicotine only has been
proved,24 but recent studies assessing the efficacy of
BMJ VOLUME 318
30 JANUARY 1999
www.bmj.com
Reykjavik Health
Care Centre,
Baronstigur 47,
101 Reykjavik and
National University
Hospital, Iceland
Thorsteinn Blondal,
chest physician
Larus Jon
Gudmundsson,
research
physiotherapist
Ingileif Olafsdottir,
research nurse
Pharmacia and
Upjohn, Box 941
S-251 09
Helsingborg,
Sweden
Gunnar Gustavsson,
clinical research
manager
Ake Westin,
biostatistician
Correspondence to:
Dr Blondal
thorsteinn.
blondal@hr.is
BMJ 1999;318:2859
Papers
assessment conducted 3-6 weeks before the participants were to stop smoking. This appointment was followed by an instructional meeting one day before the
participants were to stop smoking. We divided the 239
participants into 12 heterogeneous subgroups of
17-25 subjects each. The participants in each subgroup
were allocated either patch and nasal spray or patch
and placebo spray. The participants attended four supportive group meetings 1, 8, 15, and 22 days after stopping smoking. All participants were scheduled for
individual follow up at 6 weeks, and at 3, 6, 12, and 72
months. No participants were lost to follow up during
the first year; those who relapsed, however, were
contacted only by telephone at the 1 year follow up.
After 6 years all participants who were abstinent at 1
year were contacted again by telephone. At that time
two participants had died of cancerone in each treatment group. Participants abstinent at the 6 year follow
up came to the clinic for measurement of carbon monoxide concentrations.
The outcome measure was the duration of
sustained abstinence. Participants were considered to
be smokers if they had, after stopping smoking, taken a
single puff of a cigarette, used other forms of tobacco,
used a nicotine drug other than that prescribed, had a
carbon monoxide concentration of >10 ppm, or were
lost to follow up. No participants were classified as
smokers solely on the basis of the non-attendance part
of the smoking status criteriathat is, no subject was
lost at any follow up. The time to relapse was calculated
as the number of days from stopping smoking to the
day of starting smoking. Carbon monoxide concentrations were measured at baseline and at all subsequent
contacts, including follow up at 6 years, with a EC50
monitor (Bedfont Technical Instruments, Sittingborne,
Kent).
Statistical analyses
We based the number of participants required for the
efficacy analysis on a significance level of 5% using a
one tailed test, a power of 90%, and there being 55% of
participants in the patch and nasal spray group and
35% in the patch and placebo group; 105 participants
were needed in each treatment group. To allow for
withdrawals and protocol violations, however, we
aimed to analyse 120 participants. We used two sided
probability tests in all comparisons for more conclusive
results, and we considered a P value of < 0.05 as
significant. All tests were performed with spss
software.15
Table 1 Baseline characteristics of participants. Values are mean (SD) unless stated
otherwise
Variables
No of men
43
35
No of women
75
84
41 (23-62)
43 (22-66)
Tobacco (g/day)
25.6 (15.7)
25.0 (10.9)
5.7
5.7
378 (107-1138)
341 (37-765)
24.6 (12.30)
24.7 (10.70)
72 (40-119)
71 (48-118)
19 (16.1)
22 (18.5)
286
Results
Table 1 shows the baseline characteristics of the
participants, and figure 1 illustrates the flow of the participants through the trial.
Figure 2 shows the results of a Kaplan-Meier
survival analysis for 6 years with the proportion of participants abstinent from smoking as the survival
variable. The results of the log rank test (2 = 8.5,
P = 0.004) show a significant association between abstinence and type of treatment. Table 2 shows the
percentage of participants abstinent from smoking at
the various follow up times. After 6 years, 1 out of 6
participants was still abstinent in the treatment group
compared with 1 out of 12 in the patch only group.
During the second year after the start of the study,
one participant relapsed in the patch only group and
three in the treatment group. During the next 4 years
BMJ VOLUME 318
30 JANUARY 1999
www.bmj.com
Papers
Excluded
Non-compliance (n=1)
Illness (n=1)
Followed up for 1 year
(n=118)
Completed trial
(n=117)
Completed trial
(n=118)
100
Patch and nicotine spray
Patch and placebo spray
80
60
n=37
40
n=32
P value (2)
1 day
88.1 (104)
82.4 (98)
0.210
15 days
70.3 (83)
52.1 (62)
0.004
43 days
50.8 (60)
34.5 (41)
0.011
3 months
37.3 (44)
25.2 (30)
0.045
6 months
31.4 (37)
16.0 (19)
0.005
1 year
27.1 (32)
10.9 (13)
0.001
6 years*
16.2 (19)
8.5 (10)
0.08
*Two subjects died, one in each treatment group, between 1 and 6 years of follow up: percentages
calculated as 19/117 in treatment group and 10/118 in patch only group.
n=19
20
n=19 n=13
0
Follow up
12
24
36
n=10
48
60
72
Months after stopping smoking
30 JANUARY 1999
www.bmj.com
Mean (SD)
Mean (SD)
1 day
104 (89)
10 (6)
98 (92)
11 (7)
15 days
83 (77)
11 (7)
62 (68)
9 (6)
22 days
75 (64)
10 (7)
57 (54)
7 (6)
43 days
60 (52)
11 (7)
41 (39)
6 (5)
3 months
44 (39)
11 (8)
30 (20)
4 (1)
4 months
44 (23)
17 (9)
30 (3)
5 (0)
5 months
40 (28)
16 (9)
24 (4)
5 (0)
6 months
37 (24)
14 (9)
19 (5)
5 (0)
12 months
32 (13)
22 (9)
13 (0)
Users*
Mean
% of baseline
cotinine
100
Users*
Mean
% of baseline
cotinine
111
378
111
341
1 day
94
227
60.1
90
170
49.9
15 days
61
171
45.2
56
128
37.5
43 days
31
192
50.8
38
136
39.9
3 months
16
191
50.5
27
88
25.8
6 months
226
59.8
12 months
495
131
100
*Those abstinent in respective treatment groups and using prescribed nicotine as either patch with nicotine
nasal spray or patch only. At baseline, users refers to smokers.
Percentages at each follow up derived from mean cotinine concentration at baseline.
287
Papers
moderate, but at follow up on day 1 after stopping
smoking, 7 out of 22 participants with side effects
graded theirs as severe. The most common side effect
was nasal irritation.
Key messages
x Combined methods of nicotine replacement
therapy have a potential advantage over one
method because of high levels of substitution
Discussion
Our study shows that from day 15 after stopping
smoking the use of a nicotine patch with a nicotine
nasal spray is significantly more effective at stopping
smoking than using a nicotine patch with placebo
spray (fig 1, table 2). The difference in abstinence rates
was double at 6 months and triple at 1 year, and after
a further 5 years without nicotine replacement
therapy the difference between treatment groups was
double.
The participants in the treatment group had
received an incentive over the patch only group by having access to nicotine nasal spray during the first year,
thus alleviating smoking urges and giving them time to
consider changes in self image. The almost triple difference in abstinence rates after 1 year (P = 0.001) can
probably be explained by higher levels of substitution
during the first 5 months (table 4), and particularly so
during the remaining 7 months when only the
treatment group had access to the nictoine spray, even
if the spray was not always used daily (table 3).
The results suggest an increased efficacy in prevention of relapse with more intake of nicotine or by combining different nicotine replacement therapies.812
Whether the quantitative or the qualitative aspect is
more decisive can not be decided from the design of
our study. Studies on dose-response relations of
nicotine substitution have given different results.7 1618
The combination of a nicotine patch and nicotine
nasal spray may have been successful not only because
of the high level of substitution (table 4) but also
because of the opportunity to respond quickly to the
smokers need.
Within the treatment group, 32 of the 118
participants (27%) were abstinent from smoking at 1
year and only 4 out of 32 were still using the nicotine
nasal spray at that time. These four were, however,
using high doses of nicotine throughout the study.
The 6 year abstinence rate in the treatment group
was 16% versus 9% in the patch only group, a finding
that shows the long term efficacy of treatment.
In one study of nicotine patches, continuous self
reported abstinence rate at 4 years follow up was 20%
for a patch with 21 mg of nicotine, 7% for placebo, and
intermediate for patches with 14 mg and 7 mg of nicotine.11 Our study confirms that by providing nicotine in
several combinations, abstinence rates at 6 years can be
double those of using a single method of nicotine
replacement. It must still be acknowledged, however,
that 7 out of 10 smokers relapse within the first year of
stopping smoking.
We thank Pharmacia and Upjohn for supplying the nicotine
patches, nicotine sprays, and placebo and for measuring the
cotinine concentrations, Karl Olov Fagerstrm for discussions,
Rannveig Gunnarsdottir for dispensing the drugs and placebo,
and Asgeir Hallgrimsson for managing data collection.
Contributors: TB and GG conceived the original idea for the
study and wrote the protocol. IO and TB handled the group sessions and follow ups and helped collect the data. TB, LJG, and
AW carried out the data analysis. The paper was written by TB
and LJG. TB will act as guarantor for the paper.
288
10
11
12
13
14
15
16
17
18
30 JANUARY 1999
www.bmj.com