An Introduction to Survival Analysis

Course :

Biostatistics

Modules :

An Introduction to Survival Analysis

Lecturer:

Bandit Thinkhamrop, PhD.

Period:

9.00-12.00

Date :

6 August 2007

Objectives : After completion of this modules, fellows should be able to:

1. describe concept of survival analysis
2. calculate and interpret survival probability from a given data set
3. interpret survival curve
4. describe situation where survival analysis could be applied
6. perform data analysis using most common command of STATA.
Study materials:
1. Slide
2. Published article that used survival analysis
Teaching and learning activities :
1. Read the materials in Parts 1 and 2 as listed above
2. Attend a brief lecture and demonstration of using computer
3. Practice using computer as instructed in the study material Part 1
4. Discuss the problem faced during the practice session
Evaluation:
Formative evaluation will be done by observation of participants' responses to
questions given while conducting the session as well as active participation and
discussion in all learning activities. Summative evaluation will be based on the
exercise.
Reference:
Hosmer, D.E., and Lemeshow, S. (1999). Applied survival analysis. New York.
John Wiley & Sons.
Kleinbaum, D.G. (1996). Survival analysis: A self-learning text. New York:
Springer-Verlag.
StataCorp. (1999). Stata statistical software: Release 6.0. College Station. TX:
Stata Corporation.

SURVIVAL ANALYSIS
Event
Dead, infection, relapsed, etc

Negative

Cured, improved, conception, discharged, etc

Positive

Smoking cessation, ect

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

Accrual, Follow-up, and Event

ID
1
2
3
4
5
6

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

Time since the beginning of the study

ID
1
2
3
4
5
6

1
48 months
22 months
14 months
40 months
26 months
13 months

The data : >48

Dead
Dead

>22

14

40

>26

>13

DATA
ID

SURVIVAL TIME
(Months)

1
2
3
4
5
6

48
22
14
40
26
13

OUTCOME AT THE END

OF THE STUDY

EVENT

Still alive at the end of the study

Dead due to accident
Dead caused by the disease under investigation
Dead caused by the disease under investigation
Still alive at the end of the study
Lost to follow-up

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

. stset time, failure(event)

failure event:
obs. time interval:
exit on or before:

event ~= 0 & event ~= .

(0, time]
failure

------------------------------------------------------------------6 total obs.

0 exclusions
------------------------------------------------------------------6 obs. remaining, representing
2 failures in single record/single failure data
163 total analysis time at risk, at risk from t =
0
earliest observed entry t =
0
last observed exit t =
48

. 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)
|
+---------------------------------------------------------------------------+

. sts list, at(12, 24, 36, 48)

failure _d:
analysis time _t:

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

Kaplan-Meier survival estimate

1.00

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

Incidence density = 1.2 per100 person-months

(95%CI: 0.1 to 4.4)
Proportion of surviving at 24 month = 80%
(95%CI: 20 to 97)
Median survival time = 40 Months
(95%CI: 14 to 48)

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

Kaplan-Meier survival curve

S(t)
1.0
Survival probability

Time
tj

Data sorted by time

0.5

13 14

22 26

Time (months)

40

48

Survival analysis
Computing notes and Exercises

Summer School in Modern Methods in Biostatistics and Epidemiology

Veneto, Italy
2328 June, 2003

http://www.pauldickman.com/teaching/veneto2003/

Contents
1 Survival analysis using Stata

2 Exercises using Stata

3 Splitting on two time scales and calculating SMRs

4 The Finnish Cancer Registry

12

References

12

Survival analysis using Stata

In order to analyse survival data it is necessary to specify (at a minimum) a variable

representing survival time and a variable specifying whether or not the event of interest
was observed (called the failure variable). Instead of specifying a variable representing
survival time we can specify the entry and exit dates.
In many statistical software programs (such as SAS), these variables must be specified
every time a new analysis is performed. In Stata, these variables are specified once using
the stset command and then used for all subsequent survival analysis (st) commands
(until the next stset command). For example
. use melanoma
. stset surv_mm, failure(status==1)
The above code shows how we would stset the skin melanoma data in order to analyse
cause-specific survival with survival time in completed months (surv_mm) as the time
variable. Of the four possible values of status, we have specified that only code 1 indicates
an event (death due to melanoma). If we wanted to analyse observed survival (where all
deaths are considered to be events) we could use the following command
. stset surv_mm, failure(status==1,2)
Some of the Stata survival analysis (st) commands relevant to this course are given below.
Further details can be found in the manuals or online help.
stset
stsplit
stdes
stsum
sts
stir
strate
stptime
stcox
stphtest
stphplot
stcoxkm
streg

Declare data to be survival-time data

Split time-span records
Describe survival-time data
Summarize survival-time data
Generate, graph, list, and test the survivor and cumulative
hazard functions
Report incidence-rate comparison
Tabulate failure rate
Calculate person-time at risk and failure rates
Estimate Cox proportional hazards model
Test of Cox proportional hazards assumption
Graphical assessment of the Cox prop. hazards assumption
Graphical assessment of the Cox prop. hazards assumption
Estimate parametric survival models

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

Exercises using Stata

1. Using hand calculation (i.e. using a spreadsheet program or pen, paper, and a calculator) estimate the cause-specific survivor function for the sample of 35 patients
diagnosed with colon carcinoma (see the table on page 11 of this handout) using
both the Kaplan-Meier method (up to at least 30 months) and the actuarial method
(at least the first 5 annual intervals).
In the lectures we estimated the observed survivor function (i.e. all deaths were
considered to be events) using the Kaplan-Meier and actuarial methods; your task is
to estimate the cause-specific survivor function (only deaths due to colon carcinoma
are considered events) using the same data.
2. Use Stata to confirm the results you obtained in question 1. After starting Stata,
you will first have to specify the data set you wish to analyse, that is
. use colon_sample
In order to use the Stata ltable command (life table estimates of the survivor function) we must construct a new variable indicating whether the observation period
ended with an event (the new variable is assigned code 1) or censoring (the new
variable is assigned code 0). We will call this new variable csr_fail (cause-specific
failure). The ltable command is not a standard Stata survival analysis (st) command and does not require that the data be stset.
. generate csr_fail=0
. replace csr_fail=1 if status==1
The following command will give the actuarial estimates
. ltable surv_yy csr_fail
Alternatively, we could use
. ltable surv_mm csr_fail, interval(12)
Before most Stata survival analysis commands can be used (ltable is an exception)
we must first stset the data using the stset command (see Section 1).
. stset surv_mm, failure(status==1)
A listing of the Kaplan-Meier estimates is then obtained as follows
. sts list
To graph the Kaplan-Meier estimates
. sts graph

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
.

strate eng3, per(1000)

The option graph will show a graph of rate against levels of exposure.
.

strate eng3, per(1000) graph

(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.
.
.
.
.

use melanoma, clear

keep if stage == 1
gen id=_n /* we need to generate an ID variable */
stset surv_mm, failure(status==1) scale(12) id(id)

(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
.
.
.
.

xi: stcox sex year8594 i.agegrp

est store A
stcox sex year8594
lrtest A

Compare your findings to those obtained using the Wald test.

7

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.

Splitting on two time scales and calculating SMRs

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

Table 1: A random sample of 35 patients diagnosed with colon carcinoma in Finland

during 198594; followed-up until the end of 1995
Age Clinical
dx date Surv. time
ID Sex
at dx stage
mmyy mm
yy Status
1
male
72
Localised
2.89
2
0 Dead - other
2
female
82
Distant
12.91
2
0 Dead - cancer
3
male
73
Distant
11.93
3
0 Dead - cancer
4
male
63
Distant
6.88
5
0 Dead - cancer
5
male
67
Localised
5.89
7
0 Dead - cancer
6
male
74
Regional
7.92
8
0 Dead - cancer
7
female
56
Distant
1.86
9
0 Dead - cancer
8
female
52
Distant
5.86
11
0 Dead - cancer
9
male
64
Localised
11.94
13
1 Alive
10 female
70
Localised
10.94
14
1 Alive
11 female
83
Localised
7.90
19
1 Dead - other
12 male
64
Distant
8.89
22
1 Dead - cancer
13 female
79
Localised
11.93
25
2 Alive
14 female
70
Distant
6.88
27
2 Dead - cancer
15 male
70
Regional
9.93
27
2 Alive
16 female
68
Distant
9.91
28
2 Dead - cancer
17 male
58
Localised
11.90
32
2 Dead - cancer
18 male
54
Distant
4.90
32
2 Dead - cancer
19 female
86
Localised
4.93
32
2 Alive
20 male
31
Localised
1.90
33
2 Dead - cancer
21 female
75
Localised
1.93
35
2 Alive
22 female
85
Localised
11.92
37
3 Alive
23 female
68
Distant
7.86
43
3 Dead - cancer
24 male
54
Regional
6.85
46
3 Dead - cancer
25 male
80
Localised
6.91
54
4 Alive
26 female
52
Localised
7.89
77
6 Alive
27 male
52
Localised
6.89
78
6 Alive
28 male
65
Localised
1.89
83
6 Alive
29 male
60
Localised
11.88
85
7 Alive
30 female
71
Localised
11.87
97
8 Alive
31 male
58
Localised
8.87 100
8 Alive
32 female
80
Localised
5.87 102
8 Dead - cancer
33 male
66
Localised
1.86 103
8 Dead - other
34 male
67
Localised
3.87 105
8 Alive
35 female
56
Distant
12.86 108
9 Alive

11

The Finnish Cancer Registry

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

2. Steps for data analysis

. stset

time, failure( cured)

failure event:
obs. time interval:
exit on or before:

cured ~= 0 & cured ~= .

(0, time]
failure

-----------------------------------------------------------------------------20 total obs.

0 exclusions
-----------------------------------------------------------------------------20 obs. remaining, representing
16 failures in single record/single failure data
219 total analysis time at risk, at risk from t =
0
earliest observed entry t =
0
last observed exit t =
50

. 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

. sts g, by( drug)

failure _d:
analysis time _t:

cured
time

Kaplan-Meier survival estimates, by drug

1.00

0.75

0.50

0.25

drug 1

0.00
0

drug 0

20

40

60

analysis time

. stcox drug, noshow nolog

Cox regression -- Breslow method for ties
No. of subjects =
No. of failures =
Time at risk
=
Log likelihood

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

Nicotine nasal spray with nicotine patch for smoking

cessation: randomised trial with six year follow up
Thorsteinn Blondal, Larus Jon Gudmundsson, Ingileif Olafsdottir, Gunnar Gustavsson, Ake Westin

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

treatment with nicotine by high dose nicotine patches

gave different results.57 In the Collaborative European
Antismoking Evaluation study, evidence in favour of
dose-response relations was found, but increasing the
duration of patch use did not improve abstinence
rates.7 Few studies on smoking cessation have looked at
the effect of combining two or more methods of
providing nicotine, but those that are available seem to
suggest a higher efficacy through using more or different ways of providing nicotine.810
The long term results of nicotine replacement
therapy have not been studied extensively. One recent
follow up study of a large multicentre trial on nicotine
patches reported abstinence rates of up to 20% for 4
years in the patch group versus 7% in the placebo
group.11 12 The respective abstinence rates for 1 year
were 29% and 10%. In another study, the abstinence
rates for patch and placebo for 3 years were 10% and
3% respectively.13 In a recently published study on nicotine nasal spray, the sustained abstinence rates for 3.5
years were 15% for the nasal spray and 6% for placebo.14
We evaluated the efficacy of using a nicotine patch
for 5 months with a nicotine nasal spray for 1 year on
smoking cessation. We studied the effects of a higher
level of nicotine substitution thus obtained and of
using a flexible method of providing nicotine with the
fixed method of a patch. We followed up the
participants after 6 years to assess the rates of
abstinence from smoking after the prescription for
nicotine had stopped.

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

Subjects and methods

Protocol
In November 1991 we started recruiting smokers from
Reykjavik and the surrounding towns (total population
about 150 000; smoking prevalence 34%) by advertisements in local papers and on television. We evaluated
all respondents by interview at the baseline assessment.
To be eligible for the study, respondents had to be aged
21-69 years and had to have smoked at least one cigarette per day for >3 years. We excluded those with a
history of recent myocardial infarction, severe nasal
allergy, or skin disease, and those who used smokeless
tobacco, were currently misusing alcohol, or were
pregnant or breast feeding.
We arranged contact with the participants on 10
occasions during the study and an additional follow up
after 6 years. The first contact was the baseline
285

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

Patch and nicotine spray

(n=118)

Patch and placebo spray

(n=119)

No of men

43

35

No of women

75

84

Mean age (years) (range)

41 (23-62)

43 (22-66)

Tobacco (g/day)

25.6 (15.7)

25.0 (10.9)

5.7

5.7

Mean Fagerstrom test of nicotine

dependence (1-10)
Mean cotinine concentration (ng/ml) (range)

378 (107-1138)

341 (37-765)

Carbon monoxide (ppm)

24.6 (12.30)

24.7 (10.70)

Mean body weight (kg) (range)

72 (40-119)

71 (48-118)

19 (16.1)

22 (18.5)

No of participants with history of treatment

for alcoholism (%)

286

We used the 2 statistic to compare the proportions

of abstainers in both groups at the various follow up
times. To calculate the proportion of participants
remaining abstinent over time, we compared the
survival of the two groups using the Kaplan-Meier
method. Whenever distribution of data made it
possible, we carried out comparisons of continuous
variables between the groups using parametrical t tests,
and we used the Mann-Whitney rank sum test for data
that were non-normally distributed.
Assignment
On the day before the participants were due to stop
smoking, they were allocated their treatment by
computer generated randomisation code at a local
pharmacy. The nasal spraysnicotine or placebo
were taken from boxes labelled A or B, but the bottles
themselves were unlabelled. The pharmacy staff were
blinded to the content of the bottles. To prevent switching of treatments among participants and to help protect blinding, the same treatment was on four separate
occasions dispensed to four couples. The staff of the
smoking clinic had no knowledge of the treatment
assigned to each participant. A total of 239 subjects
were randomised. Two subjects were excluded early in
the trial without breaking the randomisation code.
They had been assigned the patch and nicotine spray.
One was unable to attend meetings because of illness,
and the other refused to use the drugs and attend
meetings, thus leaving a total of 237 participants.
Masking
Nasal sprays were dispensed in identical brown bottles
containing a colourless solution of either nicotine or
black pepper oleo resin (piperine). The nicotine spray
delivered 0.5 mg of drug per dose. The randomisation
code was kept at the pharmacy during the trial and not
broken until the data entry and analysis were
completed. Blinding among participants was successful. At the 1 year follow up we found no significant
relation between type of treatment and the participants responses, which proved they had been unable
to guess their treatment. From previous experience,
one of the authors (TB) knew beforehand that participants using nasal spray for more than 3-6 months were
more likely to have been assigned the nicotine nasal
spray rather than placebo nasal spray.

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

Registered patients meeting entry criteria (n=239)

Randomised (n=239)
Patch and nicotine
spray (n=120)

Patch and placebo

spray (n=119)

Excluded
Non-compliance (n=1)
Illness (n=1)
Followed up for 1 year
(n=118)

Followed up for 1 year

(n=119)

Death during observation

period (n=1)

Death during observation

period (n=1)

Followed up for 6 years

(n=117)

Followed up for 6 years

(n=118)

Completed trial
(n=117)

Completed trial
(n=118)

Clinic visit of participants

abstinent at 6 year follow up
(n=19)

Clinic visit of participants

abstinent at 6 year follow up
(n=10)

proportion was 25% to 30%. During the first 3 months

only, the mean number of patches containing 15 mg of
nicotine used by abstinent participants was significantly greater in the patch only group (P = 0.038) than
in the treatment group.
Table 4 shows the level of substitution. In the treatment group the level of substitution ranged from 45%
to 60% for the first 3 months versus 26% to 50% in the
patch only group. Four participants with higher than
average cotinine substititution concentrations (compared with mean blood concentrations at baseline)
were still using nicotine nasal spray after 12 months. Of
those four, one relapsed during the second year and
two during the third year; the fourth participant
remained abstinent throughout the study.
At the various follow up times during the first 3
months, the incidence of side effects (a yes or no
answer) from patches ranged from 7% to 25% and
were most often graded as mild on a scale from mild
to moderate to severe. The most common side effects
were itchiness and skin irritation. At the various follow
up times during the first 3 months, the incidence of
side effects from nicotine nasal spray ranged from
5% to 25% and were most often graded as mild or

Proportion abstinent (%)

Fig 1 Flow of participants through trial

Table 2 Percentage (number) of participants abstinent from smoking at follow up

100
Patch and nicotine spray
Patch and placebo spray
80

60

n=37

40

n=32

Patch and placebo

spray (n=119)

P value (2)

Odds ratio (95% CI)

1 day

88.1 (104)

82.4 (98)

0.210

1.59 (0.77 to 3.30)

15 days

70.3 (83)

52.1 (62)

0.004

2.18 (1.28 to 3.72)

43 days

50.8 (60)

34.5 (41)

0.011

1.97 (1.17 to 3.32)

3 months

37.3 (44)

25.2 (30)

0.045

1.76 (1.01 to 3.08)

6 months

31.4 (37)

16.0 (19)

0.005

2.40 (1.27 to 4.50)

1 year

27.1 (32)

10.9 (13)

0.001

3.03 (1.50 to 6.14)

6 years*

16.2 (19)

8.5 (10)

0.08

2.09 (0.93 to 4.72)

*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

Patch and nicotine

spray (n=118)

Follow up

12

24

36

n=10
48
60
72
Months after stopping smoking

Fig 2 Kaplan-Meier survival curves showing difference in abstinence

rates between participants allocated nicotine patch and nicotine nasal
spray and those allocated nicotine patch only. Number of relapses
during the 5 years was 12 v 2 in the treatment and patch only
groups respectively (two abstinent participants died during the 6
year follow up period; one in each treatment group)

one participant relapsed and one died in the patch

only group, and nine relapsed and one died in the
treatment group.
Table 3 shows the proportion of participants using
the spray among those still abstinent at the various follow up times, and the number of doses used in the
treatment group and patch only group. Participants
assigned the placebo spray used fewer doses and
stopped taking the spray earlier than those assigned
the nicotine nasal spray. At the 5 year follow up, 2 out
of 22 (9.1%) participants in the treatment group were
using nicotine chewing gum occasionally.
For the first 3 months, 91% or more of abstinent
participants in either group used patches. During the
following 2 months the participants discontinued
patch use successively, and after 5 months the
BMJ VOLUME 318

30 JANUARY 1999

www.bmj.com

Table 3 Number of smokers abstaining at each follow up and percentage of abstainers

using spray. Doses of self reported nasal spray (1 mg)
Period

Patch and nicotine

spray (n=118)

Mean (SD)

Patch and placebo

spray (n=119)

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)

Table 4 Cotinine substitution concentrations

Patch and nicotine spray
Period
Baseline

Users*

Mean

Patch and placebo spray

% 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

x Nicotine patches release nicotine slowly, but

nicotine nasal spray delivers nicotine more
rapidly, enabling the smoker to respond quickly
to any smoking urges
x Treatment with a patch and nicotine nasal
spray was significantly more effective than
patch and placebo from day 15 after stopping
smoking
x Using a patch for 5 months with a nicotine
nasal spray for 1 year provides a more effective
means of stopping smoking than using a patch
only
x It is not cost effective to use a nicotine nasal
spray for longer than 7 months after stopping a
patch

Funding: Pharmacia and Upjohn provided the drugs and

placebo.
Competing interests: TB was a consultant for Pharmacia and
Upjohn, and GG and AW are employed by Pharmacia and
Upjohn.
1
2
3

10

11

12
13

14

15
16
17

18

Benowitz NL. Treating tobacco addictionnicotine or no nicotine? New

Engl J Med 1997;337:1230-1.
Fiore MC, Smith SS, Jorenby DE, Baker TB. The effectiveness of the nicotine patch in smoking cessation. JAMA 1994;271:1940-7.
Silagy C, Mant D, Fowler G, Lancaster T. Nicotine replacement therapy
for smoking cessation. In: Cochrane Collaboration. Cochrane Library.
Issue 4. Oxford: Update Software, 1998.
Tonnesen P, Fagerstrom KO. Nicotine replacement. In: Richmond R, ed.
Interventions for smokers, an international perspective. Baltimore: William and
Wilkins, 1994.
Dale LC, Hurt RD, Offord KP, Lawson GM, Croghan IT, Schroeder DR.
High-dose nicotine patch therapypercentage of replacement and
smoking cessation. JAMA 1995;274:1353-8.
Jorenby DE, Smith SS, Fiore MC, Hurt RD, Offord KP, Croghan IT, et al.
Varying nicotine patch dose and type of smoking cessation counseling.
JAMA 1995;274:1347-52.
Tnnesen P, Paoletti P, Gustavsson G, Russell MA, Saracci R, Gulsvik A, et
al. Higher dosage nicotine patches increase one-year smoking cessation
rates: results from the European CEASE trial. Eur Respir J 1999 (in press).
Kornitzer M, Boutsen M, Dramaix M, Thijs J, Gustavsson G. Combined
use of nicotine patches and gum in smoking cessation: a placebocontrolled clinical trial. Prev Med 1995;24:41-7.
Fagerstrom KO, Schneider NG, Lunell E. Effectiveness of nicotine patch
and nicotine gum as individual versus combined treatments for tobacco
withdrawal symptoms. Psychopharmacol (Berl) 1993;111:271-7.
Puska P, Korhonen HJ, Vartiainen E, Urjanheimo E-L, Gustavsson G,
Westin A. Combined use of nicotine patch and gum compared with gum
alone in smoking cessation: a clinical trial in North Karelia. Tobacco Control 1995;4:231-5.
Daughton DM, Hatsukami D, Repsher L, Fortmann S, Glover E, Lichtenstein E, et al. The smoking cessation efficacy of varying doses of nicotine
patch four years post quit day. Am J Resp Crit Care Med 1997;155:A549.
Transdermal nicotine study group. Transdermal nicotine for smoking
cessation. JAMA 1991;226:3133-8.
Mikkelsen KL, Tonnesen P, Norregaard J. Three-year outcome of twoand three-year sustained abstainers from a smoking cessation study with
nicotine patches. J Smok Relat Disord 1994;5:95-100.
Stapleton JA, Sutherland G, Russell MAH. How much does relapse after
one year erode effectiveness of smoking cessation treatments? Long term
follow up of randomized trial of nicotine nasal spray. BMJ
1998;316:830-1.
Green SB, Salkind NJ, Akey TM. Using SPSS for windows. Analyzing and
understanding data. New Jersey: Prentice Hall, 1997.
Hughes JR. Treatment of nicotine dependence. Is more better? JAMA
1995;274:1390-1.
Paoletti P, Fornai E, Maggiorelli F, Puntoni R, Viegi G, Carrozzi L, et al.
Importance of baseline cotinine plasma values in smoking cessation:
results from a double-blind study with nicotine patch. Eur Respir J
1996;9:643-51.
Sachs DPL, Benowitz NL. Individualizing medical treatment for tobacco
dependence. Eur Respir J 1996;9:629-31.

(Accepted 30 November 1998)

BMJ VOLUME 318

30 JANUARY 1999

www.bmj.com