Full details of the methods used are described in Additional file
: Methods, and are summarized below. Throughout this paper, we use the term relative risk (RR) to include its various estimators, including the odds ratio and the hazard ratio.
Inclusion and exclusion criteria
Attention was restricted to epidemiological prospective or case–control studies published up to and including 1999, which involved 100 lung cancers or more, and which provided RR estimates for one or more defined major, cigarette-type or dose-related smoking indices. The “major indices” compare ever, current or ex smoking with never or non-current smoking, and refer to smoking of any product, cigarettes, pipes, cigars and combinations, or of specific types of cigarette. The “cigarette type indices” compare smokers of different types of cigarette – filter with plain, manufactured with handrolled and mentholated with non-mentholated. The “dose-related indices” concern amount smoked, age of starting to smoke, duration of smoking, duration of quitting, tar level, butt length or fraction smoked. Pack-years was not considered as it was felt more important to separate effects of extent and duration of exposure. Uncontrolled case studies were not included. There were no further exclusion criteria.
Between 1997 and 2001 potentially relevant papers were sought from Medline and Emtree searches, from British Library monthly bulletins, from files on smoking and health accumulated over many years by P N Lee Statistics and Computing Ltd, and from references cited in papers obtained, until ultimately no paper examined cited a paper of possible relevance not previously examined.
Identification of studies
Relevant papers were allocated to studies, noting multiple papers on the same study, and papers reporting on multiple studies. Each study was given a unique reference code (REF) of up to 6 characters (e.g. COMSTO or LUBIN2), based on the principal author’s name and distinguishing multiple studies by the same author.
Some studies were noted as having overlaps with other studies. To minimize problems in meta-analysis arising from double-counting of cases, overlapping studies were divided into two categories, as shown in Additional file
: Studies. The first category involved minor overlap, which could not be disentangled, and which it was decided to ignore. The second category contains sets of studies which probably or definitely overlap. Here the set member containing the most comprehensive data (e.g. largest number of cases or longest follow-up) was called the ‘principal study’, other members being ‘subsidiary studies’ only considered in meta-analyses where the required RR was unavailable from the principal study.
Relevant information was entered onto a study database and two linked RR databases. Data entry was carried out in two stages. In 1997–2002, data were entered on the first RR database for the major smoking indices, cigarette type indices, and amount smoked. In 2009–2010, data were entered on the second RR database for the remaining dose-related indices.
The study database contains a record for each study, describing the following aspects: relevant publications; study title; study design; sexes considered; age range, race(s) and other details of the population studied; location; timing and length of follow-up; whether principal or subsidiary, with details of overlaps or links with other studies; number of cases and extent of histological confirmation; number of controls or subjects at risk; types of controls and matching factors used in case–control studies; use of proxy respondents, interview setting and response rates; confounding variables considered; availability of results by histological types; and availability of results for all smoking indices (including those indices not considered here, such as pack-years).
The RR databases hold the detailed results, typically containing multiple records for each study. Each record is linked to the relevant study and refers to a specific RR, recording the comparison made and the results. This record includes the sex, age range, race, lung cancer type, and (for prospective studies) the follow-up period. The smoking exposure of the numerator of the RR is defined by the smoking status (ever, current or ex), smoking product (e.g. any, cigarettes, cigarettes only, pipes only) and cigarette type (e.g. any, mainly hand-rolled cigarettes, filter cigarettes only, mentholated cigarettes). Similar information is recorded about the denominator of the RR. For dose-related indices, the level of exposure is recorded. The source of the RR is also recorded, as are details on adjustment variables. Results recorded include numbers of cases for the numerator and denominator, and, for unadjusted results, numbers of controls, persons at risk or person-years at risk. The RR itself and its lower and upper 95% confidence limits (LCL and UCL) are always recorded. These may be as reported, or derived by various means (see below), with the method of derivation noted.
Identifying which RRs to enter
RRs were entered relating to defined combinations of lung cancer type, smoking index (major, cigarette type or dose-related), confounders adjusted for, and strata, as described below.
Lung cancer type
Results were entered for all lung cancer, for Kreyberg I (as originally presented, or by combining squamous, small and large cell carcinoma) and Kreyberg II (as originally presented, or by combining adenocarcinoma and others not in Kreyberg I), and for squamous, small, and large cell carcinoma and for adenocarcinoma separately. Additionally, the following groups were constructed if not originally presented: all lung cancer or nearest equivalent, but at least squamous cell carcinoma and adenocarcinoma; squamous cell carcinoma or nearest equivalent; adenocarcinoma or nearest equivalent.
Major and cigarette type smoking indices
The intention was to enter RRs comparing current smokers, ever smokers or ex smokers with never or non smokers. Near-equivalent definitions were accepted when stricter definitions were unavailable, so that, for example, never smokers could include occasional smokers (or exceptionally, light smokers), while current smokers could include, and ex-smokers exclude, recent quitters. RRs were to be entered relating to smoking of defined products and, when the product related to cigarette smoking, to defined cigarette types (see also Additional file
: Methods). If available, results (for each of current, ex and ever smoking) were entered for five comparisons: any product vs. never any product, cigarettes vs. never any product, cigarettes only vs. never any product, cigarettes vs. never cigarettes, and cigarettes only vs. never cigarettes (and also for five equivalent comparisons for current vs non smoking). Here “cigarettes” ignores whether other products (i.e. pipes and cigars) are also smoked, while “cigarettes only” excludes mixed smokers. Additionally, when the numerator related to the smoking of filter, handrolled or mentholated cigarettes, RRs were entered with the denominator defined as relating to plain, manufactured or non-mentholated smokers respectively.
Dose-related smoking indices
RRs were entered for seven measures: amount smoked, age of starting, duration of smoking, duration of quitting, tar level, butt length and fraction smoked. RRs were expressed relative to never smokers (or near equivalent), if available, or relative to non smokers otherwise. For duration of quitting, RRs were also expressed relative to current smokers. Except for amount smoked, further RRs were entered, restricted to smokers, and expressed relative to the level expected to have the lowest risk (e.g. shortest duration or latest age started).
Confounders adjusted for
For case–control studies, results were entered adjusted for the greatest number of potential confounding variables for which results were available, and also unadjusted (or adjusted for the smallest number of confounders). For prospective studies, results were entered adjusted for age and the greatest number of confounders, and for age only or age and the smallest number of confounders, with unadjusted results entered only if no age-adjusted results were available. These alternative RRs are subsequently referred to as “most-adjusted” and “least-adjusted”. For dose-related RRs restricted to smokers, results with “most adjustment” but without adjustment for other aspects of smoking were also entered if available.
Three strata were considered – sex, age and race. Results were entered for males and females separately when available, with combined sex results only entered when sex-specific results were not available. Results were entered for all ages combined and for individual age groups, and for all races and for individual racial groups.
Derivation of RRs
Adjusted RRs and their 95% CIs were entered as provided, when available. Unadjusted RRs and CIs were calculated from their 2
2 table, using standard methods (e.g.
]), noting any discrepancies between calculated values and those provided by the author. Sometimes the 2
2 table was constructed by summing over groups (e.g. adding current and ex smokers to obtain ever smokers) or from a percentage distribution. Various other methods were used as required to provide estimates of the RR and CI. Some more commonly used methods are summarized below, fuller details being given in Additional file
Correction for zero cell
If the 2
2 table has a zero cell, 0.5 was added to each cell, and the standard formulae applied.
Combining independent RRs
RRs were combined over
strata (e.g. from a 2 × 2 ×
table) using fixed-effect meta-analysis
], giving an estimate adjusted for the stratifying variable.
Combining non-independent RRs
The Hamling et al. method
] was used (e.g. to derive an adjusted RR for ever smokers from available adjusted RRs for current and ex smokers, each relative to never smokers, or to combine adjusted RRs for several histological types, each relative to a single control group).
Estimating CI from crude numbers
If an adjusted RR lacked a CI or p-value but the corresponding 2
2 table was available, the CI was estimated assuming that the ratio UCL/LCL was the same as for the equivalent unadjusted RR.
Data entry and checking
Master copies of all the papers in the study file were read closely, with relevant information highlighted to facilitate checking. Where multiple papers are available for a study, a principal publication was identified, although details described only in other publications were also recorded. Preliminary calculations and data entry were carried out by one author and checked by another, and automated checks of completeness and consistency were also conducted. RR/CIs underwent validation checks
Meta-analyses conducted – overview
A pre-planned series of meta-analyses was conducted for various smoking indices for each of the three main outcomes (all lung cancer, squamous cell carcinoma, and adenocarcinoma) and also for some indices for two other outcomes (large cell carcinoma and small cell carcinoma). Nearest equivalent definitions are allowed for the three main outcomes, with the terms “squamous” and “adeno” used subsequently to distinguish these results from those specifically for these cell types. Each meta-analysis was repeated, based on most-adjusted RRs and on least-adjusted RRs. For each meta-analysis conducted, combined estimates were made first for all the RRs selected, then for RRs subdivided by level of various characteristics, testing for heterogeneity between levels.
Selecting RRs for the meta-analyses
All meta-analyses are restricted to records with available RR and CI values. The process of selecting RRs for inclusion in a meta-analysis must try to include all relevant data and to avoid double-counting. For a given analysis (e.g. of current cigarette smoking), several definitions of RR may be acceptable (e.g. cigarette smoking, or cigarette only smoking), so, for studies with multiple RRs, the one to be used is determined by a preference order defined for the meta-analysis. Preference orders may be required for smoking status, smoking product, the unexposed base, and extent of confounder adjustment. As the definitions of RR available may differ by sex (e.g. a study may provide RRs for any product smoking for males, but only for cigarette smoking for females), the RRs chosen for each sex may not necessarily have the same definition. Sexes combined results are only considered where sex-specific results are not available. Similarly RRs from a subsidiary study are only used where eligible RRs are unavailable from the principal study. When multiple preference orders are involved, the sequence of implementation may affect the selection, so preferences for the most important aspects, usually concerning smoking, are implemented first.
Carrying out the meta-analyses
Fixed-effect and random-effects meta-analyses were conducted using the method of Fleiss and Gross
], with heterogeneity quantified by H, the ratio of the heterogeneity chisquared to its degrees of freedom, which is directly related to the statistic I2
] by the formula I2
100 (H-1)/H. For all meta-analyses, Egger’s test of publication bias
] was also included.
Meta-analyses were conducted in various sets (A to N) corresponding to the sub-sections of the results section of the paper. A full list of the analyses is given in Additional file
The major smoking indices
For the major smoking indices, the first four sets of meta-analyses relate to: A ever smoking, B current smoking, C ever smoking (but with current smoking used if ever smoking not available), referred to subsequently as “ever/current” smoking, and D ex smoking. In what is referred to as the main analysis in each set, smoking of any product is preferred by selecting RRs in the following preference order: 1. smoking of any product vs. never smoked any product; 2. smoking of cigarettes vs. never smoked any product, 3. smoking of cigarettes only vs. never smoked any product; 4. smoking of cigarettes vs. never smoked cigarettes; 5. smoking of cigarettes only vs. never smoked cigarettes; with options 6–10 the same as options 1–5 except that “never smoked” is replaced by “never smoked near equivalent”. A variant analysis prefers cigarette smoking (by changing the preference order to 4, 5, 2, 3, 1, 9, 10, 7, 8, 6). In meta-analyses of type C, a further variant analysis reverses the preference so current smoking results are preferred to those for ever smoking, referred to subsequently as “current/ever” smoking. Other variant analyses are based on RRs for specified age ranges.
A further set of meta-analyses, E, concerns smoking of pipes and/or cigars (but not cigarettes), referred to subsequently as smoking of “pipes/cigars only”, smokers of pipes only, smokers of cigars only, and smokers of cigarettes and pipes/cigars (“mixed” smokers). Separate meta-analyses were conducted for ever smoking, current smoking, ever/current smoking, current/ever smoking and ex smoking.
The cigarette type indices
Meta-analyses were conducted, in set F, for only filter vs. only plain, ever filter vs. only plain, only filter vs. ever plain, handrolled vs. manufactured, and mentholated vs. non-mentholated. These were only conducted for ever/current smoking, and preferring RRs for cigarettes over RRs for cigarettes only. The analyses with only filter as the numerator used the preference order of filter only, always, mainly, both, equally, and ever, while the analyses with ever filter as the numerator used the reverse preference. Similar preference orders applied to the denominators. The analyses of handrolled vs. manufactured cigarettes used the preference order of any, both, mainly, and only for handrolled, and only ever, only current, any and ever for manufactured.
The dose-related smoking indices
For the dose-related indices, sets of meta-analyses were conducted for: G amount smoked, H age of starting to smoke, I duration of smoking, J duration of quitting compared to never smokers (or long-term ex smokers), K duration of quitting compared to current smokers (or short-term quitters), L tar level, and M butt length or fraction smoked (taking short butt length as being equivalent to a large fraction smoked). For any measure, a study typically provides a set of non-independent RRs for each dose-category, expressed relative to a common base. To avoid double-counting only one was included in any one meta-analysis. Two approaches were adopted. The first involves specifying a scheme with a number of levels of exposure (“key values”), then carrying out meta-analyses for each level in turn, expressed relative to never smokers. For an RR to be allocated to a key value, its dose-category has to include that key-value and no other. Schemes with a few, widely spaced, key values tend to involve more studies, whereas schemes with more key values, closely spaced, involve RRs from fewer studies, but ones with dose categories more closely clustered around the key value. The sets of key values used (with 999 indicating an open-ended category) were 5, 20, 45 and 1, 10, 20, 30, 40, 999 for amount smoked; 26, 18, 14 and 30, 26, 22, 18, 14, 10 for age of starting to smoke; 20, 35, 50 and 5, 20, 30, 40, 50, 999 for duration of smoking; 12, 7, 3 and 20, 12, 3 for duration of quitting vs. never; and 3, 7, 12 and 3, 12, 20 for duration of quitting vs. current. No key value analysis was conducted for tar level, or for butt length/fraction smoked. The second approach (not conducted for amount smoked) involves meta-analysing of RRs for the highest compared with the lowest categories of exposure within smokers available for each study.
While full multivariable analysis of the data is considered beyond the scope of this report, meta-regression analyses were also carried out using the sets of RRs selected for the main meta-analyses for ever smoking and for current smoking. Following preliminary meta-regressions (not shown), a “fixed model” was fitted to examine the effect on the results of six different categorical variables (sex, location, start year of study, major study type, number of lung cancer cases and number of adjustment factors). Note that the number of lung cancer cases (in the study as a whole), which is referred to subsequently as “number of cases”, is used as an indicator of study size. The significance of each of these variables was estimated by an F-test based on the increase in deviance resulting from its exclusion from the basic model. A list of secondary variables was also defined (relating to more detailed aspects of location, outcome, study type and confounder adjustment, national cigarette tobacco type, the product smoked, the denominator used in the RR, use of proxy respondents, whether the study required 100% histological confirmation of lung cancer, whether the population studied worked in risky occupations, the age of the subjects, and the derivation of the RR) with the significance of adding each characteristic to the fixed model estimated by an F-test based on the increase in deviance. Fuller details are given in Additional file
Additional tests of the relationship of lung cancer risk to various characteristics of interest were based on corresponding pairs of RR and CI estimates within the same study for the same definition of outcome and exposure, and deriving the ratio of the two RRs. Where the pairs involved independent sets of subjects, the variance of the ratio was also derived, and meta-analyses of the ratio were conducted. Where the pairs involved non-independent sets of subjects the numbers of ratios greater and less than 1 were compared using the sign test. Tests of independent pairs related to sex (males vs. females), age (oldest vs. youngest age group) and race (white people vs. non-white or black people). Tests of non-independent pairs related to level of adjustment (most-adjusted vs. least-adjusted), and to comparisons of product smoked (mixed smokers vs. cigarette only smokers, and vs. smokers of pipes/cigars only). Tests were always carried out for all lung cancer and ever/current smoking. For sex, additional analyses were conducted for current and for ever smoking, for squamous and adeno, and also within level of amount smoked. For level of adjustment, two sets of analyses were run. The first, relating to RRs for ever/current smoking were based on the most-adjusted/least-adjusted ratio, while the second, for highest vs. lowest RRs for age of starting to smoke, duration, years quit and tar level, compared RRs that were most- or least-adjusted for other aspects of smoking.
All data entry and most statistical analyses were carried out using ROELEE version 3.1 (available from P.N. Lee Statistics and Computing Ltd, 17 Cedar Road, Sutton, Surrey SM2 5DA, UK). Some analyses were conducted using Quattro Pro 9 or Excel 2003.