Systematic review with meta-analysis of the epidemiological evidence relating smoking to COPD, chronic bronchitis and emphysema

Attention was restricted to epidemiological studies published before 2007 on COPD, CB or emphysema, providing relative risk (RR) estimates for one or more defined "major indices" (ever, current or ex smoking compared with never smoking) or "dose-related indices" (amount smoked, age of starting to smoke, pack-years smoked, duration of smoking or duration of quitting). Throughout this paper, we use the term RR to include its various estimators, including the odds ratio and the hazard ratio.

Studies were excluded if in children or adolescents, or in subjects at especially high risk of respiratory disease (e.g. workers in risky occupations), selected as having co-existing diseases or conditions, or from atypical populations likely to have a highly unusual prevalence of smoking or disease. Also excluded were uncontrolled case studies, and studies of disease exacerbation or undiagnosed disease, of symptom-free subjects, or where the only results were adjusted for symptoms or precursors of disease.

Searching was carried out in phases. Initially, 1407 potentially relevant papers, published up to 2002, were derived by AJT from an unpublished project which used the MeSH terms chronic bronchitis and symptoms, emphysema, lung function, genetic determinants, mortality, adults and smoking. Subsequently, additional Medline searches were conducted in 2006 by AJT and in 2008 by BAF, using the MeSH term "Pulmonary disease, chronic obstructive". Papers were also sought from in-house files on smoking and health, and references cited in papers obtained. Publications before 2007 were considered, with no restriction on language or on peer-reviewed journals. Reasons for rejection were recorded.

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. DICKIN or CHEN3), based on the principal author's name, and distinguishing multiple studies by the same author. Occasionally, an original study was split into separate studies (e.g. where follow-up periods differed by sex).

Some studies were noted as having overlaps or links with other studies. To minimize problems in meta-analysis arising from double-counting of cases, these links were divided into three types, as shown in Additional file 2. The first involved no such double-counting, while the second included studies with minor overlap, which could not be disentangled, and which it was decided to ignore. The third type contains sets of studies which probably or definitely overlap. Here the set member containing the most valuable data (e.g. largest study size 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.

For each study, relevant information was entered onto a study database and a linked RR database. The study database contains a record for each study, describing relevant publications, sexes considered, age range, location, timing, length of follow-up, whether principal or subsidiary, overlaps or links with other studies, study design, populations studied, major study weaknesses, outcome definitions, numbers of cases and subjects, types of controls and matching factors used in case-control studies, confounding variables, and availability of results for each smoking index. The RR database holds 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 outcome, the sex and the analysis type (prevalence or incidence). Smoking exposure is defined by status (ever, current or ex), product (any, cigarettes, cigarettes only) and similar information about the unexposed base. 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 exposed and unexposed cases, and, for unadjusted results, numbers of exposed and unexposed members of the comparison group. The RR itself and its lower and upper 95% confidence limits (LCL and UCL) are always recorded, with the odds ratio chosen if available for a prevalence analysis and the relative risk (or hazard ratio if provided) for an incidence analysis. These may be as reported, or derived by various means (see below), with the method of derivation noted.

For each outcome RRs were entered relating to defined combinations of smoking index (major or dose-related), confounders adjusted for, and sex, as described below.

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. [12]), 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. The more commonly used methods are summarized below, fuller details being given in Additional file 1.

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 ℓ strata (e.g. from a 2 × 2 × ℓ table) using fixed-effect meta-analysis [13], giving an estimate adjusted for the stratifying variable. Combining non-independent RRs. The Hamling et al method [14] 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 diseases, each relative to a single control or disease-free 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.

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 ([15]).

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 an order of preference defined for the meta-analysis. Orders of preference 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 for cigarette smoking for females), the most appropriate RR is chosen within each sex. 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 orders of preference are involved, the sequence of implementation may affect the selection, so preferences for the most important aspects, usually concerning smoking, are implemented first.

Fixed-effect and random-effects meta-analyses were conducted using the methods of Fleiss and Gross [13], with heterogeneity quantified by H, the ratio of the heterogeneity chisquared to its degrees of freedom. For all meta-analyses, Egger's test of publication bias [16] was also conducted.

A series of meta-analyses was conducted for each of the three main outcomes. 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. These characteristics may include sex, continent, national cigarette type (blended, Virginia), start year of study, publication year, study type, lowest age included, highest age included, presence of study weakness, outcome subtype, how asthma was taken into account, use of a bronchodilator, study size (number of cases), analysis type (prevalence, onset), smoking product (any, cigarettes, cigarettes only), unexposed base (never any product, never cigarettes), smoking results available (ever smoking, current smoking, both), number of adjustment variables, whether the RR was adjusted for sex, age or for other factors, and how the RR and CI were derived. In this univariate approach, differences in fixed-effect estimates by level of a characteristic were tested for significance using an F-test which compared variation between and within levels of the characteristic considered. Additional file 1 fully defines the levels of each characteristic considered, and which characteristics are considered in each meta-analysis. It also details all the meta-analyses conducted, and describes the layout and notation used in the meta-analyses and associated forest and funnel plots.

For each selected outcome and exposure, separate meta-analyses were conducted based on most-adjusted and least-adjusted RRs.

For the major smoking indices, four broad types of meta-analysis were conducted: A ever smoking, B current smoking, C ever smoking (but using current smoking RRs if ever smoking RRs are not available) and D ex smoking. In each type, RRs for the "main analysis" were selected in the following order of preference: firstly for smoking of any product vs. never smoked any product, then for smoking of cigarettes (or of cigarettes only) vs. never smoked any product, and then for smoking of cigarettes vs. never smoked cigarettes, accepting RRs vs. near-equivalents to never smokers only when RRs vs. never smokers were unavailable. A variant analysis used a different order of preference, so that RRs for cigarette smoking were preferred. In type C meta-analyses, a further variant analysis preferred RRs for current smoking to those for ever smoking. Other variant analyses restricted attention to specific subtypes of outcome (e.g. for COPD, whether the definition was based on mortality, on lung function criteria only, or on other definitions).

For the dose-related indices, meta-analyses were conducted for: E amount smoked, F age of starting to smoke, G pack-years, H duration of smoking, I duration of quitting compared to never smokers (or long-term ex smokers), and J duration of quitting compared to current smokers (or short-term quitters). 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. 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 RRs from 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 key value schemes used were: 5, 20 & 45 and 1, 10, 20, 30, 40 & 999 for amount smoked; 26, 18 & 14 for age of starting to smoke; 5, 20 & 45 and 1, 10, 20, 30 & 999 for pack-years; 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 (with 999 indicating an open-ended category). The second approach involves meta-analysing RRs comparing the highest vs. lowest categories of exposure. Though this approach generally includes RRs from all studies, whereas the key-value approach does not, the highest and lowest categories compared may vary markedly by study.

For COPD and CB 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 "basic model" was fitted which included eight categorical variables (sex, continent, outcome subtype, how asthma was taken into account, smoking product, unexposed base group, adjustment for age, and adjustment for factors other than age or sex) and also midpoint age, a continuous variable estimated from the age range of the population. 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 (national cigarette type, publication year, study type, presence of a study weakness, use of a bronchodilator, study size, smoking results available for the study, method of derivation of the RR and CI and analysis type), with the significance of adding each characteristic to the basic model estimated by an F-test based on the increase in deviance. Alternative formulations of some basic variables were also tested; see also Additional file 1.

For each outcome, and for ever smoking and current smoking, pairs of corresponding RR and CI estimates within the same study for males and for females were used to carry out meta-analyses of the sex ratio. Pairs of corresponding least-adjusted or most-adjusted RRs were also identified. Unlike the sex-specific pairs, these pairs were non-independent and the variance of their ratio cannot readily be calculated. Here the numbers of pairs where the most-adjusted/least-adjusted ratio exceeded or did not exceed 1 were compared by the sign test, with separate meta-analyses also conducted for the least-adjusted and most-adjusted members. Similar methods were also used to compare non-independent pairs of RRs for current smokers of cigarettes only and for current smokers of cigarettes ignoring other products.

All data entry and most statistical analysis 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 Excel 2003.

Some 218 relevant studies were identified, based on information from 298 papers.

For the 2,150 papers rejected, reasons are summarized in Table 1, with further details of the searching, including a flow diagram, shown in Additional file 1. Many papers had multiple reasons for rejection, the counts in Table 1 relating only to the first listed reason which applied. A Reference Manager file is available on request which, for each rejected publication, gives its reference and the reasons for rejection.

Table 2 presents selected details of the 218 studies while Table 3 gives the distribution of their major characteristics. Additional file 2 gives fuller descriptions of the studies, including overlapping and linked studies, medical and other exclusions, detailed definitions of disease outcomes, and fuller distributions.

Of the 218 studies, 193 are classified as principal, 20 (10.4%) of these being case-control studies, 39 (22.7%) prospective, and 134 (69.4%) cross-sectional. The other 25 studies are classified as subsidiary. Ninety-three principal studies are of COPD only, 63 of CB only, nine of emphysema only, with 28 providing results for multiple outcomes. In total, information is available on COPD for 133 studies (116 principal), CB for 101 (87 principal) and emphysema for 28 (26 principal). Of the principal studies, only 9 (10.3%) are prospective for CB, compared with 35 (30.2%) for COPD and 7 (26.9%) for emphysema. There are no case-control studies for emphysema.

Of the 193 principal studies, 145 (75.1%) provide results for both sexes, 42 (21.8%) for males only, and six (3.1%) for females only. Ninety-six (49.7%) of the studies include subjects under 30 (at baseline for prospective studies), while only 24 (12.4%) are restricted to subjects aged 50 or more. Subjects aged 80 or more are included by 103 (53.6%), while only 20 (10.3%) are restricted to subjects aged 60 or less. Fifty-six (29.0%) principal studies were conducted in USA or Canada, with 32 (16.6%) in Scandinavia, 32 (16.6%) in Asia, 23 (11.9%) in the UK, 17 (8.8%) in Western Europe, 14 (7.3%) in other parts of Europe, eight (4.1%) in South or Central America and seven (3.6%) elsewhere. Four (2.1%) were carried out in more than one of these areas. Of the 159 principal studies where the start year is given, 76 (47.7%) started before 1980. For 26 (13.5%) of the 193 studies a major study weakness is noted. Most commonly this is a failure to clarify, or to state at all, how study subjects were selected (studies ALESSA, ANDER3, COCCI, ITABAS, MOLLER, SHIMUR, ZIETKO). Other more commonly occurring weaknesses include use of unrepresentative samples which oversampled smokers (DEJONG, DETORR, JENSEN), those with respiratory disease (VOLLM1, VOLLM2) or those with occupational exposure (PETO, PRATT), and the use of controls that systematically differ from cases and controls in various ways (BROGGE, DEAN1, LUNDB2, STERLI). These weaknesses are described more fully in the footnotes to Table 2.

Most principal studies provide some results compared to never smokers, 146 (75.6%) for current smokers, 134 (69.4%) for ex smokers and 158 (81.8%) for ever smokers. Dose-response data are commonly available by amount smoked (77 studies, 39.9%) and by pack-years (58, 30.1%), but less so by age of starting to smoke (17, 8.8%), duration of smoking (12, 6.2%) or duration of quitting (18, 9.3%).

Of the 116 principal studies of COPD, outcome is based on ICD codes in 29 (25.0%), and lung function only in 59 (50.9%). The GOLD criteria are used in 27 (23.2%) studies, with MRC, ATS, ERS or BTS criteria used in 12 (10.3%). In 69 (59.5%) studies the subjects' asthma status is ignored, in 18 (15.5%) all asthmatics subjects are excluded, and in 14 (12.1%) the disease definition includes asthma. Only 19 (16.4%) of the 116 principal studies mention conducting spirometry after use of a bronchodilator. The outcome is based on prevalence in 79 (68.1%) principal studies, mortality in 28 (24.1%) and incidence in 10 (8.6%). In the principal studies, the median number of subjects is 2,033, and of cases 131 (range 13 to 32,822).

Of the 87 principal studies of CB, the outcome is based on symptoms (not lung function) in 59 (67.8%), and on ICD in only six (6.9%). Other studies use self-report, a doctor diagnosis, or other definitions. The MRC questionnaire is used in 21 (24.1%). Asthmatics are excluded totally from six (6.9%) studies, with asthmatics excluded only from the controls in three (3.4%). The outcome is based on prevalence in 78 (89.7%) of the principal studies, mortality in six (6.9%) and incidence in three (3.4%). The median number of subjects is 2,826, and of cases 193.5 (range 2 to 4,769).

Of the 26 principal studies of emphysema, the outcome is based on visual comparison in 10 (38.5%), on diagnosis in seven (26.9%), on ICD codes in five (19.2%) and on other sources including self-report in four (15.4%). Asthmatics are excluded in two (7.7%) studies. The outcome is based on prevalence in 19 (73.1%) of the studies, on mortality in five (19.2%) and on incidence in two (7.7%). The median number of subjects is 2,433, and of cases 96.5 (range 2 to 1384).

A total of 3,538 RRs are entered, 1,578 for COPD, 1,689 for CB and 271 for emphysema, the number recorded per study varying from 1 to 211. Some 675 relate to subsidiary studies. Table 4 summarizes the distribution of various characteristics of the RRs by outcome, by study type for the principal studies, and overall. For fuller distributions of the RRs, referred to as necessary below, see Additional file 3.

Of the 2,863 RRs in principal studies, 67.8% relate to cross-sectional, 19.8% to prospective, and 12.4% to case-control studies. 81.2% of RRs are sex-specific. About half the RRs (52.0%) are adjusted for one or more variables. Of 1,488 adjusted RRs, age is adjusted for in 1,382 (92.9%) but only 490 (32.9%) are adjusted for variables other than age, sex or other smoking aspects. 34.0% of the RRs are given directly or calculated from a 2 × 2 or 2 × 2 × ℓ table, the rest being derived.

Of the 3,538 RRs, 1,439 are for major smoking indices, and 2,099 for dose-related indices (including 236 and 439 respectively in subsidiary studies). Of the 1,203 RRs in principal studies for major indices, 34.6% are for ever smoking, 37.8% current smoking and 27.6% ex smoking. 53.6% are for cigarette smoking ignoring other products, 33.8% any product smoking, and 12.6% cigarettes only. The unexposed group is typically never any product (55.8%) or never cigarettes (43.1%).

The distribution of smoking status for the 1,660 RRs in principal studies for dose-related indices differs considerably, with 22.8% for ever smoking, 59.6% current smoking and 17.6% ex smoking. Again, most (59.8%) RRs relate to cigarette smoking ignoring other products. The unexposed group is never smoking (any product or cigarettes) for 50.4% of these RRs, low smoking for 39.2%, and current smoking for 3.9%. 52.7% of RRs are for amount smoked, 8.1% age of starting, 19.8% pack-years, 4.4% years duration, and 15.1% years quit (about half compared to never smokers or long-term quitters, the rest compared to current smokers or short-term quitters). Based on RRs with an unexposed base of never smoking, there are 174 sets of categorical data for amount smoked, 18 for age of starting, 52 for pack-years, 11 for duration of smoking, and 26 for duration of quitting. For emphysema, there are few dose-related data other than for amount smoked

None of the RRs included in the meta-analyses and meta-regressions show more than minor failures of the validation tests used, attributable to rounding errors or small imprecisions or uncertainties in estimating the RRs and CIs. Additional File 3 provides further detail.

The main findings are summarized in the following sections, with tables and forest plots. Fuller results of the meta-analyses for the major smoking variables are given in Additional file 4 for COPD, Additional file 5 for CB and Additional file 6 for emphysema. Similar results for the dose-related smoking variables are given in Additional file 7 for COPD, Additional file 8 for CB and Additional file 9 for emphysema. An Excel file, available as Additional file 10, allows the user readily to view selected results from all these meta-analyses. Detailed meta-regression outputs are given in Additional file 11. For dose-related indices, Additional file 12 presents within-study plots of the dose-response relationships, while Additional file 13 gives results that were originally presented in a form unsuitable for meta-analysis. The interested reader should first refer to Additional file 1, which describes the content and structure of all these Additional files.