Hand Out Survival

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.

1SURVIVAL ANALYSIS

Event

Dead, infection, relapsed, etc

Cured, improved, conception, discharged, etc

Smoking cessation, ect

Negative

Positive

Neutral

Time to event

General: event-free durationrelapse-free survival timeremission durationprogression-free survival

SURVIVAL TIME

Time to event

Failed: event occurred at time of observation

Censored: event have not occurred at time of observation

2CensoringEvent-free duration cannot be determined

Event have not yet occurred

Lost to follow-up

Competing event

Accrual, Follow-up, and EventID 2540 2541 2542 2543

Begin the study End of the study

Start of accrual End of accrual End of follow-up

123456

Recruitment period Follow-up period

Time since the beginning of the studyID

0 1 2 3 4

DeadDead

123456

48 months22 months14 months40 months26 months13 months

The data : >48 >22 14 40 >26 >13

3DATA

1 48 Still alive at the end of the study Censored2 22 Dead due to accident Censored3 14 Dead caused by the disease under investigation Dead4 40 Dead caused by the disease under investigation Dead5 26 Still alive at the end of the study Censored6 13 Lost to follow-up Censored

ID SURVIVAL TIME OUTCOME AT THE END EVENT(Months) OF THE STUDY

DATA

1 48 Censored2 22 Censored3 14 Dead4 40 Dead5 26 Censored6 13 Censored

ID TIME EVENT

1 48 02 22 03 14 14 40 15 26 06 13 0

ID TIME EVENT

ANALYSIS

1 48 02 22 03 14 14 40 15 26 06 13 0

ID TIME EVENT

Prevalence = 2/6

Incidence density = 2/163 person-months

Proportion of surviving at month t

Median survival time

5. listid time event

1. 1 48 02. 2 22 03. 3 14 14. 4 40 15. 5 26 06. 6 13 0

. stset time, failure(event)failure event: event ~= 0 & event ~= .

obs. time interval: (0, time]exit on or before: failure-------------------------------------------------------------------

6 total obs.0 exclusions

-------------------------------------------------------------------6 obs. remaining, representing2 failures in single record/single failure data

163 total analysis time at risk, at risk from t = 0earliest observed entry t = 0

last observed exit t = 48

. stsumfailure _d: event

analysis time _t: 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

. stquantNumber of deads = 2Number 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) |+---------------------------------------------------------------------------+

6. sts list, at(12, 24, 36, 48)failure _d: event

analysis time _t: timeBeg. Survivor Std.

Time Total Fail Function Error [95% Conf. Int.]-------------------------------------------------------------------------------

12 0 0 1.0000 . . .24 4 1 0.8000 0.1789 0.2038 0.969236 3 0 0.8000 0.1789 0.2038 0.969248 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 riskfailure _d: event

analysis time _t: time

Kaplan-Meier survival estimate

analysis time0 20 40 60

0.00

0.25

0.50

0.75

1.00 6

4

1

RESULTS

1 48 02 22 03 14 14 40 15 26 06 13 0

ID TIME EVENT

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

Time tj

Number at risk

nj

Dead dj

Censored qj

Survival probability S(tj)

13 6 0 1 6/6 = 1.0 14 6-1 = 5 1 0 [(5-1)/5]1.0 = 0.8 22 5-1 = 4 0 1 [(4-0)/4]0.8 = 0.8 26 4-1 = 3 0 1 [(3-0)/3]0.8 = 0.8 40 6-1 = 2 1 0 [(2-1)/2]0.8 = 0.4 48 2-1 = 1 0 1 [(1-0)/1]0.4 = 0.4

Kaplan-Meier survival curve

ID TIME DEAD 1 48 0 2 22 0 3 14 1 4 40 1 5 26 0 6 13 0

ID TIME DEAD 6 13 0 3 14 1 2 22 0 5 26 0 4 40 1 1 48 0

Data sorted by time Original Data

0.5

S(t) 1.0

0 13 14 22 26 40 48

Survi

val p

robab

ility

Time (months)

Survival analysis

Computing notes and Exercises

Summer School in Modern Methods in Biostatistics and EpidemiologyVeneto, Italy

2328 June, 2003

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

Contents

1 Survival analysis using Stata 2

2 Exercises using Stata 3

3 Splitting on two time scales and calculating SMRs 9

4 The Finnish Cancer Registry 12

References 12

1

1 Survival analysis using Stata

In order to analyse survival data it is necessary to specify (at a minimum) a variablerepresenting survival time and a variable specifying whether or not the event of interestwas observed (called the failure variable). Instead of specifying a variable representingsurvival time we can specify the entry and exit dates.

In many statistical software programs (such as SAS), these variables must be specifiedevery time a new analysis is performed. In Stata, these variables are specified once usingthe 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 analysecause-specific survival with survival time in completed months (surv_mm) as the timevariable. Of the four possible values of status, we have specified that only code 1 indicatesan event (death due to melanoma). If we wanted to analyse observed survival (where alldeaths 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 Declare data to be survival-time datastsplit Split time-span recordsstdes Describe survival-time datastsum Summarize survival-time datasts Generate, graph, list, and test the survivor and cumulative

hazard functionsstir Report incidence-rate comparisonstrate Tabulate failure ratestptime Calculate person-time at risk and failure ratesstcox Estimate Cox proportional hazards modelstphtest Test of Cox proportional hazards assumptionstphplot Graphical assessment of the Cox prop. hazards assumptionstcoxkm Graphical assessment of the Cox prop. hazards assumptionstreg Estimate parametric survival models

Once the data have been stset we can use any of these commands without having tospecify the survival time or failure time variables. For example, to plot the estimatedcause-specific survivor function by sex and then fit a Cox proportional hazards modelwith sex and calendar period as covariates

. sts graph, by(sex)

. stcox sex year8594

2

2 Exercises using Stata

1. Using hand calculation (i.e. using a spreadsheet program or pen, paper, and a cal-culator) estimate the cause-specific survivor function for the sample of 35 patientsdiagnosed with colon carcinoma (see the table on page 11 of this handout) usingboth 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 wereconsidered to be events) using the Kaplan-Meier and actuarial methods; your task isto estimate the cause-specific survivor function (only deaths due to colon carcinomaare 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 func-tion) we must construct a new variable indicating whether the observation periodended with an event (the new variable is assigned code 1) or censoring (the newvariable is assigned code 0). We will call this new variable csr_fail (cause-specificfailure). The ltable command is not a standard Stata survival analysis (st) com-mand 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).


A listing of the Kaplan-Meier estimates is then obtained as follows

. sts list

To graph the Kaplan-Meier estimates

. sts graph

3

Note that we only have to stset the data once. You can also tell Stata to show thenumber 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 thecause-specific survivor function, using the Kaplan-Meier method with survival timein months, separately for each of the two calendar periods 19751984 and 19851994.

The following commands can be used

. use melanoma

. keep if stage == 1


. sts graph, by(year8594)

(a) Without making reference to any formal statistical tests, does it appear thatpatient survival is superior during the most recent period?

(b) The following commands can be used to plot the hazard function (instantaneousmortality rate):

. sts graph, hazard by(year8594)

At what point in the follow-up is mortality highest? Does this pattern seemreasonable)? Do you feel that there is there more information in this plotcompared to the plot of the survivor function?

4. In question 3 we studied plots of the survivor function for patients diagnosed withlocalised skin melanoma by calendar period of diagnosis. Use the log rank test todetermine whether there is a statistically significant difference in patient survivalbetween 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


(a) Plot estimates of the survivor function and hazard function by stage. Does itappear that stage is associated with survival?

(b) Estimate the mortality rates for each stage using, for example, the stratecommand. What are the units of the estimated rates?

(c) If you havent already done so, estimate the mortality rates for each stage per1000 person-years of follow-up.

(d) Study whether survival is different for males and females (by plotting the sur-vivor function or tabulating rates).

4

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 person-years for each category of hieng. Calculate (by hand) the ratio of the twoincidence rates.

(b) Use the command poisson to find the incidence rate ratio for the high energygroup compared to the low energy group and compare the estimate to the oneyou 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 muchabout how the CHD rate changes with total energy. It is a useful exploratorydevice, but to look more closely we need to group the total energy into perhaps3 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 range150002499, 2500 for values in the range 25002999, and 3000 for values in therange 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



(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, creatingthe indicators automatically with i.eng3.

5

7. Now lets study cause-specific survival using the melanoma data restricted to lo-calised stage (stage==1). The following commands can be used to load and stsetthe data.

. use melanoma, clear


. 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 tabulatemortality rates and model them using Poisson regression. We expect mortalityto 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 thisseem reasonable?

(c) Compare the plot of the estimated rates to a plot of the hazard rate as a functionof continuous time.

sts graph, hazard

Is the interpretation similar? Do you think it is sufficient to classify follow-uptime into annual intervals or might it be preferable to use, for example, narrowerintervals?

(d) Use Poisson regression to estimate incidence rate ratios as a function of follow-up time.

xi: streg i.fu, dist(exp)

Does the pattern of estimated incident rate ratios mirror the pattern you ob-served 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 thatthe effect of follow-up confounded by age, sex, and calendar period?

6

8. Using the melanoma data restricted to localised stage (stage==1), fit a Cox propor-tional hazards model (for cause-specific survival) with calendar period as the onlyexplanatory variable. The following commands can be used

. use melanoma



. stcox year8594

(a) Interpret the estimated hazard ratio, including a comment on statistical signif-icance.

(b) (This part is more theoretical and is not required in order to understand theremaining parts.)Stata reports a Wald test of the null hypothesis that survival is independentof calendar period. The test statistic (and associated P-value) is reported inthe table of parameter estimates (labelled z). Under the null hypothesis, thetest statistic has a standard normal (Z) distribution, so the square of the teststatistic will have a chi square distribution with one degree of freedom.Stata also reports a likelihood ratio test statistic of the null hypothesis thatnone of the parameters in the model are associated with survival (labelledLR chi2(1)). In general, this test statistic will have a chi-square distributionwith degrees of freedom equal to the number of parameters in the model. Forthe current model, with only one parameter, the test statistic has a chi squaredistribution with one degree of freedom.Compare these two test statistics with each other and with the log rank teststatistic (which also has a 21 distribution) calculated in question 4.Would you expect these test statistics to be similar? Consider the null andalternative hypotheses of each test and the assumptions involved with eachtest.

(c) Now include sex and age (in categories) in the model. The following commandcan be used

. xi: stcox sex year8594 i.agegrp

(d) Interpret the estimated hazard ratio for the parameter labelled Iagegr_2, in-cluding 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 effectof 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 theresults. The following commands can be used


. 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 variablesas in the previous question, but where the outcome is observed survival (i.e. all deathsare considered to be events).

. stset surv_mm, failure(status==1,2)

. keep if stage==1


(a) Interpret the estimated hazard ratio for the parameter labelled Iagegr_2, in-cluding a comment on statistical significance.

(b) On comparing the estimates between the observed and cause-specific survivalmodels 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 tochange 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 appearto 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 insurvival between males and females?

(c) Decide on a most appropriate model. Be sure to evaluate the proportionalhazards assumption.

11. For the melanoma data, plot the log cumulative hazard function for each calendarperiod. 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 thesedata?

(b) Does the appropriateness of the proportional hazards assumption have anyimplications for the log rank test?

(c) From the plot, estimate the hazard ratio for patients diagnosed 198594 to thosediagnosed 197584. You may find this easier if you plot the curves withoutsymbols; this can be done by including the s(ii) option to the stphplotcommand. In the next question we will estimate this hazard ratio using a Coxproportional hazards model.

8

3 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 calendarperiod as 1/1/1900 for convenience in setting the breaks.

3. Each subjects followup is now divided into small pieces corresponding to the agebands and calendar period bands the subject passes through. We can make tablesof 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 ageand calendar period, it is first necessary to break the followup into parts whichcorrespond to these age bands and calendar periods, as above. Since there are nosuitable rates for these data, we have used reference rates which are uniform for bothage and calendar period at 11/1000.

The reference rates are in the file ref. Before calculating the expected number ofcases 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 appro-priate reference rate to each record. The system variable merge takes the followingvalues:1 record in the master file but no match in ref2 record in ref but no match in the master file3 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 withno match in the master file

. drop if _merge==2

6. To calculate the expected number of cases, multiply the follow-up time for eachrecord 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, andis 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 Finlandduring 198594; followed-up until the end of 1995

Age Clinical dx date Surv. timeID Sex at dx stage mmyy mm yy Status1 male 72 Localised 2.89 2 0 Dead - other2 female 82 Distant 12.91 2 0 Dead - cancer3 male 73 Distant 11.93 3 0 Dead - cancer4 male 63 Distant 6.88 5 0 Dead - cancer5 male 67 Localised 5.89 7 0 Dead - cancer6 male 74 Regional 7.92 8 0 Dead - cancer7 female 56 Distant 1.86 9 0 Dead - cancer8 female 52 Distant 5.86 11 0 Dead - cancer9 male 64 Localised 11.94 13 1 Alive10 female 70 Localised 10.94 14 1 Alive11 female 83 Localised 7.90 19 1 Dead - other12 male 64 Distant 8.89 22 1 Dead - cancer13 female 79 Localised 11.93 25 2 Alive14 female 70 Distant 6.88 27 2 Dead - cancer15 male 70 Regional 9.93 27 2 Alive16 female 68 Distant 9.91 28 2 Dead - cancer17 male 58 Localised 11.90 32 2 Dead - cancer18 male 54 Distant 4.90 32 2 Dead - cancer19 female 86 Localised 4.93 32 2 Alive20 male 31 Localised 1.90 33 2 Dead - cancer21 female 75 Localised 1.93 35 2 Alive22 female 85 Localised 11.92 37 3 Alive23 female 68 Distant 7.86 43 3 Dead - cancer24 male 54 Regional 6.85 46 3 Dead - cancer25 male 80 Localised 6.91 54 4 Alive26 female 52 Localised 7.89 77 6 Alive27 male 52 Localised 6.89 78 6 Alive28 male 65 Localised 1.89 83 6 Alive29 male 60 Localised 11.88 85 7 Alive30 female 71 Localised 11.87 97 8 Alive31 male 58 Localised 8.87 100 8 Alive32 female 80 Localised 5.87 102 8 Dead - cancer33 male 66 Localised 1.86 103 8 Dead - other34 male 67 Localised 3.87 105 8 Alive35 female 56 Distant 12.86 108 9 Alive

11

4 The Finnish Cancer Registry

The Finnish Cancer Registry is population-based and covers the whole of Finland (popula-tion 5.1 million). The Registry was established in 1952, with 1953 being the first calendaryear with complete registration. The Registry obtains information from many differentsources: hospitals and other institutions with inpatient beds, physicians working outsidehospitals, dentists, and pathological and cytological laboratories. The Finnish CancerRegistry also receives copies of all death certificates where cancer is mentioned. Notifi-cation of new cancer cases to the Cancer Registry is mandatory by law. If the reportedinformation is deficient or contradictory, requests are sent to informants in order to ensureaccuracy in the following areas: patient details, the primary site of the tumour, and thedate of diagnosis.

The diseases registered at the Finnish Cancer Registry include, in addition to all clearlymalignant neoplasms, carcinoma in situ lesions (except those of the skin), all neoplasms ofthe intracranial space and spinal canal irrespective of their malignancy, benign papillomasof the urinary organs, semimalignant tumours of the ovary, basal cell carcinomas of theskin, and cases of polycythaemia vera and myelofibrosis. Various check-ups have shownthat the coverage of the Cancer Registry file is almost complete with respect to cancercases diagnosed in the Finnish population [1, 2]. All independent primary neoplasms inthe same person are registered separately. When evaluating whether a new tumour is anindependent cancer or a recurrence, attention is focused on, among other aspects, the timeinterval between the tumours, histology, and knowledge of the general behaviour of eachcancer type. In principle, multiple metachronous tumours in the same organ (e.g., in thecolon or skin) are registered separately, especially when they have different histologies.However, each case is evaluated individually and a primary site code multiple canceris 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 befound in the annual incidence publications [3].

The Finnish Cancer Registry has kindly provided data on patients diagnosed with skinmelanoma and colon carcinoma in Finland during 19751994 with follow-up to 31 Decem-ber 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 population-based cancer registry. Experience in Finland. Acta Oncologica 1994;33:365369.

[2] Hakulinen T. Health care system, cancer registration and follow-up of cancer patientsin Finland. In: Berrino F, Sant M, Verdecchia A, Capocaccia R, Hakulinen T, Este`ve J,eds., Survival of Cancer Patients in Europe: The EUROCARE Study , IARC ScientificPublications No. 132. Lyon: International Agency for Research on Cancer, 1995; 5354.

[3] Finnish Cancer Registry. Cancer Incidence in Finland 1995 . Cancer Society of FinlandPublication No. 58. Helsinki: Cancer Society of Finland, 1997.

[4] Dickman PW, Hakulinen T, Luostarinen T, Pukkala E, Sankila R, Soderman B, TeppoL. Survival of cancer patients in Finland 1955-1994. Acta Oncologica 1999;38 (Suppl.12):1103.

12

EXERCISE 1. The data set id drug cured time 1 0 1 6 2 0 1 7 5 0 1 12 3 0 1 12 4 0 1 12 6 0 1 13 8 0 1 14 7 0 1 14 9 0 0 30 10 0 0 50 11 1 1 1 12 1 1 1 13 1 1 2 15 1 1 2 14 1 1 2 16 1 1 3 17 1 1 3 18 1 1 4 19 1 0 15 20 1 0 16 2. Steps for data analysis . stset time, failure( cured) failure event: cured ~= 0 & cured ~= . obs. time interval: (0, time] exit on or before: 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: cured analysis time _t: 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: cured analysis time _t: time . stcox drug, noshow nolog Cox regression -- Breslow method for ties No. of subjects = 20 Number of obs = 20 No. of failures = 16 Time at risk = 219 LR chi2(1) = 2.35 Log likelihood = -38.788082 Prob > chi2 = 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 ------------------------------------------------------------------------------

Kaplan-Meier survival estimates, by drug

analysis time0 20 40 60

0.00

0.25

0.50

0.75

1.00

drug 0drug 1

Papers

Nicotine nasal spray with nicotine patch for smokingcessation: randomised trial with six year follow upThorsteinn Blondal, Larus Jon Gudmundsson, Ingileif Olafsdottir, Gunnar Gustavsson, Ake Westin

AbstractObjective To evaluate the efficacy of using a nicotinepatch for 5 months with a nicotine nasal spray for 1year.Design Placebo controlled, double blind trial.Setting Reykjavik health centre.Subjects 237 smokers aged 2266 years living in oraround Reykjavik.Interventions Nicotine patch for 5 months withnicotine nasal spray for 1 year (n = 118) or nicotinepatch with placebo spray (n = 119). Treatment withpatches included 15 mg of nicotine for 3 months,10 mg for the fourth month, and 5 mg for the fifthmonth, whereas nicotine in the nasal spray wasavailable for up to 1 year. Both groups receivedsupportive treatment.Main outcome measure Sustained abstinence fromsmoking.Results The log rank test for 6 years (2 = 8.5,P = 0.004) shows a significant association betweenabstinence from smoking and type of treatment.Sustained abstinence rates for the patch and nasalspray group and patch only group were 51% v 35%after 6 weeks (P = 0.011 (2), 95% confidence interval1.17% to 3.32%), 37% v 25% after 3 months(P = 0.045, 1.01% to 3.08%), 31% v 16% after 6months (P = 0.005, 1.27% to 4.50%), 27% v 11% after12 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 ratesshow that the combination of using a nicotine patchfor 5 months with a nicotine nasal spray for 1 year is amore effective method of stopping smoking thanusing a patch only. The low percentage of participantsusing the nasal spray at 1 year, and the few relapsesduring the second year, suggest that it is not costeffective to use a nasal spray for longer than 7 monthsafter stopping a patch.

IntroductionIn controlled clinical trials of nicotine replacementtherapy, 1 in 5 smokers remained abstinent after 1 yearcompared with 1 in 10 smokers who abstained in selfadministered cessation.1 Better treatments for smokingcessation are clearly needed. In several controlled studies, the value of treatment with nicotine only has beenproved,24 but recent studies assessing the efficacy of

treatment with nicotine by high dose nicotine patchesgave different results.57 In the Collaborative EuropeanAntismoking Evaluation study, evidence in favour ofdoseresponse relations was found, but increasing theduration of patch use did not improve abstinencerates.7 Few studies on smoking cessation have looked atthe effect of combining two or more methods ofproviding nicotine, but those that are available seem tosuggest a higher efficacy through using more or different ways of providing nicotine.810

The long term results of nicotine replacementtherapy have not been studied extensively. One recentfollow up study of a large multicentre trial on nicotinepatches reported abstinence rates of up to 20% for 4years in the patch group versus 7% in the placebogroup.11 12 The respective abstinence rates for 1 yearwere 29% and 10%. In another study, the abstinencerates for patch and placebo for 3 years were 10% and3% respectively.13 In a recently published study on nicotine nasal spray, the sustained abstinence rates for 3.5years were 15% for the nasal spray and 6% for placebo.14

We evaluated the efficacy of using a nicotine patchfor 5 months with a nicotine nasal spray for 1 year onsmoking cessation. We studied the effects of a higherlevel of nicotine substitution thus obtained and ofusing a flexible method of providing nicotine with thefixed method of a patch. We followed up theparticipants after 6 years to assess the rates ofabstinence from smoking after the prescription fornicotine had stopped.

Subjects and methodsProtocolIn November 1991 we started recruiting smokers fromReykjavik and the surrounding towns (total populationabout 150 000; smoking prevalence 34%) by advertisements in local papers and on television. We evaluatedall respondents by interview at the baseline assessment.To be eligible for the study, respondents had to be aged2169 years and had to have smoked at least one cigarette per day for >3 years. We excluded those with ahistory of recent myocardial infarction, severe nasalallergy, or skin disease, and those who used smokelesstobacco, were currently misusing alcohol, or werepregnant or breast feeding.

We arranged contact with the participants on 10occasions during the study and an additional follow upafter 6 years. The first contact was the baseline

Reykjavik HealthCare Centre,Baronstigur 47,101 Reykjavik andNational UniversityHospital, IcelandThorsteinn Blondal,chest physicianLarus JonGudmundsson,researchphysiotherapistIngileif Olafsdottir,research nurse

Pharmacia andUpjohn, Box 941S251 09Helsingborg,SwedenGunnar Gustavsson,clinical researchmanagerAke Westin,biostatistician

Correspondence to:Dr [email protected]

BMJ 1999;318:2859

285BMJ VOLUME 318 30 JANUARY 1999 www.bmj.com

assessment conducted 36 weeks before the participants were to stop smoking. This appointment was followed by an instructional meeting one day before theparticipants were to stop smoking. We divided the 239participants into 12 heterogeneous subgroups of1725 subjects each. The participants in each subgroupwere allocated either patch and nasal spray or patchand placebo spray. The participants attended four supportive group meetings 1, 8, 15, and 22 days after stopping smoking. All participants were scheduled forindividual follow up at 6 weeks, and at 3, 6, 12, and 72months. No participants were lost to follow up duringthe first year; those who relapsed, however, werecontacted only by telephone at the 1 year follow up.After 6 years all participants who were abstinent at 1year were contacted again by telephone. At that timetwo participants had died of cancerone in each treatment group. Participants abstinent at the 6 year followup came to the clinic for measurement of carbon monoxide concentrations.

The outcome measure was the duration ofsustained abstinence. Participants were considered tobe smokers if they had, after stopping smoking, taken asingle puff of a cigarette, used other forms of tobacco,used a nicotine drug other than that prescribed, had acarbon monoxide concentration of >10 ppm, or werelost to follow up. No participants were classified assmokers solely on the basis of the nonattendance partof the smoking status criteriathat is, no subject waslost at any follow up. The time to relapse was calculatedas the number of days from stopping smoking to theday of starting smoking. Carbon monoxide concentrations were measured at baseline and at all subsequentcontacts, including follow up at 6 years, with a EC50monitor (Bedfont Technical Instruments, Sittingborne,Kent).

Statistical analysesWe based the number of participants required for theefficacy analysis on a significance level of 5% using aone tailed test, a power of 90%, and there being 55% ofparticipants in the patch and nasal spray group and35% in the patch and placebo group; 105 participantswere needed in each treatment group. To allow forwithdrawals and protocol violations, however, weaimed to analyse 120 participants. We used two sidedprobability tests in all comparisons for more conclusiveresults, and we considered a P value of < 0.05 assignificant. All tests were performed with spsssoftware.15

We used the 2 statistic to compare the proportionsof abstainers in both groups at the various follow uptimes. To calculate the proportion of participantsremaining abstinent over time, we compared thesurvival of the two groups using the KaplanMeiermethod. Whenever distribution of data made itpossible, we carried out comparisons of continuousvariables between the groups using parametrical t tests,and we used the MannWhitney rank sum test for datathat were nonnormally distributed.

AssignmentOn the day before the participants were due to stopsmoking, they were allocated their treatment bycomputer generated randomisation code at a localpharmacy. The nasal spraysnicotine or placebowere taken from boxes labelled A or B, but the bottlesthemselves were unlabelled. The pharmacy staff wereblinded to the content of the bottles. To prevent switching of treatments among participants and to help protect blinding, the same treatment was on four separateoccasions dispensed to four couples. The staff of thesmoking clinic had no knowledge of the treatmentassigned to each participant. A total of 239 subjectswere randomised. Two subjects were excluded early inthe 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 attendmeetings, thus leaving a total of 237 participants.

MaskingNasal sprays were dispensed in identical brown bottlescontaining a colourless solution of either nicotine orblack pepper oleo resin (piperine). The nicotine spraydelivered 0.5 mg of drug per dose. The randomisationcode was kept at the pharmacy during the trial and notbroken until the data entry and analysis werecompleted. Blinding among participants was successful. At the 1 year follow up we found no significantrelation between type of treatment and the participants responses, which proved they had been unableto guess their treatment. From previous experience,one of the authors (TB) knew beforehand that participants using nasal spray for more than 36 months weremore likely to have been assigned the nicotine nasalspray rather than placebo nasal spray.

ResultsTable 1 shows the baseline characteristics of theparticipants, and figure 1 illustrates the flow of the participants through the trial.

Figure 2 shows the results of a KaplanMeiersurvival analysis for 6 years with the proportion of participants abstinent from smoking as the survivalvariable. 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 thepercentage of participants abstinent from smoking atthe various follow up times. After 6 years, 1 out of 6participants was still abstinent in the treatment groupcompared 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 andthree in the treatment group. During the next 4 years

Table 1 Baseline characteristics of participants. Values are mean (SD) unless statedotherwise

VariablesPatch 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 (2362) 43 (2266)

Tobacco (g/day) 25.6 (15.7) 25.0 (10.9)

Mean Fagerstrom test of nicotinedependence (110)

5.7 5.7

Mean cotinine concentration (ng/ml) (range) 378 (1071138) 341 (37765)

Carbon monoxide (ppm) 24.6 (12.30) 24.7 (10.70)

Mean body weight (kg) (range) 72 (40119) 71 (48118)

No of participants with history of treatmentfor alcoholism (%)

19 (16.1) 22 (18.5)

Papers

286 BMJ VOLUME 318 30 JANUARY 1999 www.bmj.com

one participant relapsed and one died in the patchonly group, and nine relapsed and one died in thetreatment group.

Table 3 shows the proportion of participants usingthe spray among those still abstinent at the various follow up times, and the number of doses used in thetreatment group and patch only group. Participantsassigned the placebo spray used fewer doses andstopped taking the spray earlier than those assignedthe nicotine nasal spray. At the 5 year follow up, 2 outof 22 (9.1%) participants in the treatment group wereusing nicotine chewing gum occasionally.

For the first 3 months, 91% or more of abstinentparticipants in either group used patches. During thefollowing 2 months the participants discontinuedpatch use successively, and after 5 months the

proportion was 25% to 30%. During the first 3 monthsonly, the mean number of patches containing 15 mg ofnicotine used by abstinent participants was significantly greater in the patch only group (P = 0.038) thanin 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 thepatch only group. Four participants with higher thanaverage cotinine substititution concentrations (compared with mean blood concentrations at baseline)were still using nicotine nasal spray after 12 months. Ofthose four, one relapsed during the second year andtwo during the third year; the fourth participantremained abstinent throughout the study.

At the various follow up times during the first 3months, the incidence of side effects (a yes or noanswer) from patches ranged from 7% to 25% andwere most often graded as mild on a scale from mildto moderate to severe. The most common side effectswere itchiness and skin irritation. At the various followup times during the first 3 months, the incidence ofside effects from nicotine nasal spray ranged from5% to 25% and were most often graded as mild or

Registered patients meeting entry criteria (n=239)

Patch and nicotinespray (n=120)

Followed up for 1 year(n=118)

ExcludedNon-compliance (n=1)

Illness (n=1)

Patch and placebospray (n=119)

Followed up for 1 year(n=119)

Randomised (n=239)

Death during observationperiod (n=1)

Death during observationperiod (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 participantsabstinent at 6 year follow up

(n=19)

Clinic visit of participantsabstinent at 6 year follow up

(n=10)

Fig 1 Flow of participants through trial

72Months after stopping smoking

Prop

ortio

n ab

stin

ent (

%)

604836241200

20

40

60

n=37

n=32

n=19 n=13 n=10

n=19

Patch and nicotine sprayPatch and placebo spray

80

100

Fig 2 KaplanMeier survival curves showing difference in abstinencerates between participants allocated nicotine patch and nicotine nasalspray and those allocated nicotine patch only. Number of relapsesduring the 5 years was 12 v 2 in the treatment and patch onlygroups respectively (two abstinent participants died during the 6year follow up period; one in each treatment group)

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

Follow upPatch and nicotine

spray (n=118)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: percentagescalculated as 19/117 in treatment group and 10/118 in patch only group.

Table 3 Number of smokers abstaining at each follow up and percentage of abstainersusing spray. Doses of self reported nasal spray (1 mg)

PeriodPatch 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

Period

Patch and nicotine spray Patch and placebo spray

Users* Mean% of baseline

cotinine Users* Mean% of baseline

cotinine

Baseline 111 378 100 111 341 100

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 8 226 59.8 0 0 0

12 months 4 495 131 0 0 0

*Those abstinent in respective treatment groups and using prescribed nicotine as either patch with nicotinenasal spray or patch only. At baseline, users refers to smokers.Percentages at each follow up derived from mean cotinine concentration at baseline.

Papers

287BMJ VOLUME 318 30 JANUARY 1999 www.bmj.com

moderate, but at follow up on day 1 after stoppingsmoking, 7 out of 22 participants with side effectsgraded theirs as severe. The most common side effectwas nasal irritation.

DiscussionOur study shows that from day 15 after stoppingsmoking the use of a nicotine patch with a nicotinenasal spray is significantly more effective at stoppingsmoking than using a nicotine patch with placebospray (fig 1, table 2). The difference in abstinence rateswas double at 6 months and triple at 1 year, and aftera further 5 years without nicotine replacementtherapy the difference between treatment groups wasdouble.

The participants in the treatment group hadreceived 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 toconsider changes in self image. The almost triple difference in abstinence rates after 1 year (P = 0.001) canprobably be explained by higher levels of substitutionduring the first 5 months (table 4), and particularly soduring the remaining 7 months when only thetreatment group had access to the nictoine spray, evenif 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 ismore decisive can not be decided from the design ofour study. Studies on doseresponse relations ofnicotine substitution have given different results.7 1618

The combination of a nicotine patch and nicotinenasal spray may have been successful not only becauseof the high level of substitution (table 4) but alsobecause of the opportunity to respond quickly to thesmokers need.

Within the treatment group, 32 of the 118participants (27%) were abstinent from smoking at 1year and only 4 out of 32 were still using the nicotinenasal spray at that time. These four were, however,using high doses of nicotine throughout the study.

The 6 year abstinence rate in the treatment groupwas 16% versus 9% in the patch only group, a findingthat shows the long term efficacy of treatment.

In one study of nicotine patches, continuous selfreported abstinence rate at 4 years follow up was 20%for a patch with 21 mg of nicotine, 7% for placebo, andintermediate for patches with 14 mg and 7 mg of nicotine.11 Our study confirms that by providing nicotine inseveral combinations, abstinence rates at 6 years can bedouble those of using a single method of nicotinereplacement. It must still be acknowledged, however,that 7 out of 10 smokers relapse within the first year ofstopping smoking.

We thank Pharmacia and Upjohn for supplying the nicotinepatches, nicotine sprays, and placebo and for measuring thecotinine 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 thestudy and wrote the protocol. IO and TB handled the group sessions and follow ups and helped collect the data. TB, LJG, andAW carried out the data analysis. The paper was written by TBand LJG. TB will act as guarantor for the paper.

Funding: Pharmacia and Upjohn provided the drugs andplacebo.

Competing interests: TB was a consultant for Pharmacia andUpjohn, and GG and AW are employed by Pharmacia andUpjohn.

1 Benowitz NL. Treating tobacco addictionnicotine or no nicotine? NewEngl J Med 1997;337:12301.

2 Fiore MC, Smith SS, Jorenby DE, Baker TB. The effectiveness of the nicotine patch in smoking cessation. JAMA 1994;271:19407.

3 Silagy C, Mant D, Fowler G, Lancaster T. Nicotine replacement therapyfor smoking cessation. In: Cochrane Collaboration. Cochrane Library.Issue 4. Oxford: Update Software, 1998.

4 Tonnesen P, Fagerstrom KO. Nicotine replacement. In: Richmond R, ed.Interventions for smokers,an international perspective. Baltimore: William andWilkins, 1994.

5 Dale LC, Hurt RD, Offord KP, Lawson GM, Croghan IT, Schroeder DR.Highdose nicotine patch therapypercentage of replacement andsmoking cessation. JAMA 1995;274:13538.

6 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;274:134752.

7 Tnnesen P, Paoletti P, Gustavsson G, Russell MA, Saracci R, Gulsvik A, etal. Higher dosage nicotine patches increase oneyear smoking cessationrates: results from the European CEASE trial. Eur Respir J 1999 (in press).

8 Kornitzer M, Boutsen M, Dramaix M, Thijs J, Gustavsson G. Combineduse of nicotine patches and gum in smoking cessation: a placebocontrolled clinical trial. Prev Med 1995;24:417.

9 Fagerstrom KO, Schneider NG, Lunell E. Effectiveness of nicotine patchand nicotine gum as individual versus combined treatments for tobaccowithdrawal symptoms. Psychopharmacol (Berl) 1993;111:2717.

10 Puska P, Korhonen HJ, Vartiainen E, Urjanheimo EL, Gustavsson G,Westin A. Combined use of nicotine patch and gum compared with gumalone in smoking cessation: a clinical trial in North Karelia. Tobacco Control 1995;4:2315.

11 Daughton DM, Hatsukami D, Repsher L, Fortmann S, Glover E, Lichtenstein E, et al. The smoking cessation efficacy of varying doses of nicotinepatch four years post quit day. Am J Resp Crit Care Med 1997;155:A549.

12 Transdermal nicotine study group. Transdermal nicotine for smokingcessation. JAMA 1991;226:31338.

13 Mikkelsen KL, Tonnesen P, Norregaard J. Threeyear outcome of twoand threeyear sustained abstainers from a smoking cessation study withnicotine patches. J Smok Relat Disord 1994;5:95100.

14 Stapleton JA, Sutherland G, Russell MAH. How much does relapse afterone year erode effectiveness of smoking cessation treatments? Long termfollow up of randomized trial of nicotine nasal spray. BMJ1998;316:8301.

15 Green SB, Salkind NJ, Akey TM. Using SPSS for windows. Analyzing andunderstanding data. New Jersey: Prentice Hall, 1997.

16 Hughes JR. Treatment of nicotine dependence. Is more better? JAMA1995;274:13901.

17 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 doubleblind study with nicotine patch. Eur Respir J1996;9:64351.

18 Sachs DPL, Benowitz NL. Individualizing medical treatment for tobaccodependence. Eur Respir J 1996;9:62931.

(Accepted 30 November 1998)

Key messages

x Combined methods of nicotine replacementtherapy have a potential advantage over onemethod because of high levels of substitution

x Nicotine patches release nicotine slowly, butnicotine nasal spray delivers nicotine morerapidly, enabling the smoker to respond quicklyto any smoking urges

x Treatment with a patch and nicotine nasalspray was significantly more effective thanpatch and placebo from day 15 after stoppingsmoking

x Using a patch for 5 months with a nicotinenasal spray for 1 year provides a more effectivemeans of stopping smoking than using a patchonly

x It is not cost effective to use a nicotine nasalspray for longer than 7 months after stopping apatch

Papers

288 BMJ VOLUME 318 30 JANUARY 1999 www.bmj.com

Documents

Hand Out Survival