Tar level of cigarettes smoked and risk of smoking-related diseases

Abstract

Background: Opinions differ on the relationship between tar level and risk of smoking-related disease. However, except for lung cancer, few reviews have evaluated the epidemiological evidence. Here the relationship of tar level to risk of the four main smoking-related diseases is considered.

Methods: Papers comparing risk of lung cancer, COPD, heart disease or stroke in smokers of lower and higher tar yield cigarettes were identified from reviews and searches, relative risk estimates being extracted comparing the lowest and highest tar groups. Meta-analyses investigated heterogeneity by various study characteristics.

Results: Twenty-six studies were identified, nine of prospective design and 17 case–control. Two studies grouped cigarettes by nicotine rather than tar. Seventeen studies gave results for lung cancer, 16 for heart disease, five for stroke and four for COPD. Preferring relative risks adjusted for daily amount smoked, where adjusted and unadjusted estimates were available, combined estimates for lowest versus highest tar (or nicotine) groups were 0.78 (95% confidence interval 0.70–0.88) for lung cancer, 0.86 (0.81–0.91) for heart disease, 0.77 (0.62–0.95) for stroke and 0.81 (0.65–1.02) for COPD. Lower risks were generally evident in subgroups by publication period, gender, study design, location and extent of confounder adjustment. Estimates were similar preferring data unadjusted for amount smoked or excluding nicotine-based estimates.

Conclusions: Despite evidence that smokers substantially compensate for reduced cigarette yields, the results clearly show lower risks in lower tar smokers. Limitations of the evidence are discussed, but seem unlikely to affect this conclusion.

Keywords:

• Cigarettes
• tar level
• lung cancer
• COPD
• heart disease
• stroke

Introduction

Over the last 50 years or so, tar levels of cigarettes smoked have reduced substantially (Forey et al., 2006–2016; Lee, 2001; US Surgeon General, 2014). However, few reviews have considered whether the health effects of smoking are lower in smokers of lower yield cigarettes. Most attention has been given to lung cancer where opinions vary on whether smoking lower, compared to higher tar, cigarettes is associated with a reduction in risk (Kabat, 2003; Lee & Sanders, 2004), little or no change in risk (Alberg & Samet, 2003; Stratton et al., 2001; Thun & Burns, 2001) or even an adverse effect on specific histological types (Burns et al., 2011; US Surgeon General, 2014). The effect of tar level on other major smoking-related diseases has rarely been discussed, and a recent major report discussing 50 years of progress in studying the health consequences of smoking (US Surgeon General, 2014) did not even consider whether tar reduction had affected the risk of either respiratory or cardiovascular disease.

In this review, I summarize the evidence relating tar level of cigarettes smoked to the four major smoking-related diseases – lung cancer, heart disease, stroke and chronic obstructive pulmonary disease (COPD). As, by now, virtually all cigarettes smoked have filters, I do not consider comparison of risk in smokers of filtered and non-filtered (plain) cigarettes. For reviews of the evidence on filter/plain differences for lung cancer, the reader is referred to some earlier reviews (Kabat, 2003; Lee, 2001; Lee & Sanders, 2004). I restrict attention to findings from epidemiological case–control and prospective studies, deriving relative risks (RRs) and 95% confidence intervals (CIs) associated with the comparison of smokers of low and high tar cigarettes. As “deliveries of nicotine and tar are generally correlated” (Benowitz, 1989) and as there has been a “concomitant shift towards lowered levels of tar and nicotine” (Alberg & Samet, 2003) I also consider evidence from studies classifying smokers by nicotine rather than tar level. It should be noted that this review does not compare risk in smokers of low and high tar (or nicotine) cigarettes with that in those who have never smoked.

Methods

Papers comparing the risk of lung cancer, COPD, heart disease or stroke in smokers of lower and higher tar (or nicotine) yield cigarettes were identified from earlier reviews (Kabat, 2003; Lee et al., 2012; Lee & Sanders, 2004; National Cancer Institute, 2001; US Surgeon General, 2014), from Medline searches, and from reference lists in identified publications. As in previous reviews, the definitions of the outcomes studied were as follows: lung cancer – overall, including at least squamous cell carcinoma and adenocarcinoma (Lee et al., 2012), COPD – overall but not chronic bronchitis or emphysema specifically (Forey et al., 2011), heart disease – including ischemic heart disease, coronary heart disease or acute myocardial infarction, but not overall cardiovascular disease (Lee et al., 2017a), and stroke – overall, but not specific types (Lee et al., 2017a). Overlaps between studies were noted.

For each study, estimates of the RR and 95% CI were obtained for the lowest versus highest tar level for which results were reported. Where results by tar level were not available, comparisons of risk for the lowest versus highest nicotine yield or extremes of a combined tar and nicotine grouping were used instead. In a few studies RRs estimated for a given decrease in tar yield were also included. Studies relating risk only to a cumulative tar index (Meyers et al., 2017) were not included as they do not allow separation of effects of tar level and amount smoked. Where possible, estimates obtained were gender specific, as is usual in studies of smoking and disease.

For lung cancer and heart disease, I attempted to obtain, if possible, two different RR estimates together with their 95%CI. One of these estimates was adjusted for the daily number of cigarettes smoked, and one was unadjusted. Both estimates were adjusted for the maximum number of other potential confounding variables for which results were reported. For COPD and stroke, only estimates that were adjusted for number of cigarettes smoked were obtained.

Many estimates were taken directly from the source paper. Others were derived by inverting high versus low estimates to become low versus high estimates, or by various other procedures, similar to those used elsewhere (Lee et al., 2012) and described more fully in the Tables A1–A5. Excel spreadsheets are also available on request giving details of the calculations made.

Table 1. Studies of tar yield and major smoking-related diseases.

	References^a	Location, Study type^b	Lung Cancer	Heart disease	Stroke	COPD
1^c^,^d	Hammond et al. (1976); Lee & Garfinkel (1981)	USA, P	✓	✓	✓	✓
2	Higenbottam et al. (1982)	UK, P	✓	✓	–	–
3^e	Vutuc & Kunze (1982, 1983)	Austria, CC	✓	–	–	–
4^f	Kaufman et al. (1983)	USA, CC	–	✓	–	–
5	Lubin et al. (1984)	Western Europe, CC	✓	–	–	–
6	Alderson et al. (1985, 1986)	UK, CC	✓	✓	✓	✓^g
7^h	Petitti & Friedman (1985a,b); Sidney et al. (1993)	USA, P	✓	✓	✓	✓
8^i	Garfinkel & Stellman (1988); Harris et al. (2004)	USA, P	✓	–	–	–
9	Gillis et al. (1988)	UK, CC	✓	–	–	–
10	Wilcox et al. (1988)	USA, CC	✓	–	–	–
11	Wynder & Kabat (1988)	USA, CC	✓	–	–	–
12	Kaufman et al. (1989)	USA and Canada, CC	✓	–	–	–
13^f	Palmer et al. (1989)	USA, CC	–	✓	–	–
14^j	Kuller et al. (1991); Ockene et al. (1990)	USA, P	✓	✓	–	–
15	Negri et al. (1993)	Italy, CC	–	✓	–	–
16	Benhamou et al. (1994)	France, CC	✓	–	–	–
17	Parish et al. (1995)	UK, CC	–	✓	–	–
18^k	Tang et al. (1995)	UK, P	✓	✓	✓	✓
19	Bosetti et al. (1999)	Italy, CC	–	✓	–	–
20	Speizer et al. (1999)	USA, P	✓	–	–	–
21^l	Woodward (2001); Woodward et al. (1999)	Scotland, P	✓	✓	✓	–
22	Tavani et al. (2001)	Italy, CC	–	✓	–	–
23	Sauer et al. (2002)	USA, CC	–	✓	–	–
24^m	Gallus et al. (2007)	Italy, CC	–	✓	–	–
25	Zhang et al. (2011)	Germany, P	–	✓	–	–
26	Lee et al. (2017b)	5 European countries, CC	✓	–	–	–

Or surrogate measure as indicated by footnotes d and f.

^a Multiple references to the same study are shown on the same line.

^b Study type: CC: case–control; P: prospective.

^c The earlier reference gives results for lung cancer and heart disease, the later reference for all four diseases.

^d Comparisons based on a combined tar and nicotine index, see Table A1 for details.

^e The earlier reference gives results for women, the later reference for men.

^f Comparisons based on nicotine.

^g Originally described as “chronic bronchitis” and omitted from the earlier review (Forey et al., 2011), the outcome definition was subsequently corrected to COPD (Lee et al., 2014).

^h The earlier reference gives results for all diseases for 4 years follow-up, the later reference for lung cancer for 6 years follow-up.

ⁱ The earlier reference gives results for 4 years follow-up, the later reference for 6 years follow-up.

^j Both references give results for 10.5 years follow-up but give results with different confounder adjustment.

^k Combined evidence from four studies, including studies 2 and 9.

^l The earlier reference gives results for heart disease only for 8 years follow-up, the later reference for all three diseases listed for 13 years follow-up.

^m Combined evidence from females in study 19, all subjects in study 15 and subjects collected over a longer period than those considered in study 22.

Fixed and random effects meta-analyses were conducted using standard methods (Fleiss & Gross, 1991) though, as fixed effect estimates were quite similar, only random effects results are reported here. For lung cancer and heart disease, initial analyses included results from all the studies, giving summary results based separately on the estimates unadjusted and adjusted for amount smoked. The remaining analyses omit certain studies where overlaps had been identified in order to avoid double counting. For the reduced set of studies, results are shown based firstly on estimates that are unadjusted for amount smoked, secondly on estimates adjusted for amount smoked, thirdly preferring estimates unadjusted for amount smoked where there was a choice, and finally preferring estimates adjusted for amount smoked. For the final set of estimates heterogeneity was investigated by comparing the estimated relative risk by levels of specific factors (period of publication, gender, study design, number of confounding variables considered and continent). Results were also presented excluding estimates not specifically based on tar groupings.

Results of some meta-analyses are given in Figures as forest plots. Each result is presented as a square centered at the RR value and a horizontal line representing the width of the CI. This width is shown on a logarithmic scale. Within each Figure the area of each square is approximately proportional to the meta-analysis weight of the result (the inverse of its variance). The overall meta-analysis result is represented as a white square with a black cross, scaled as for the individual study results.

Unless stated otherwise, p < .05 is taken as statistically significant.

Results

Studies identified

Twenty-six relevant studies were identified, which are subsequently referred to by study number. Table 1 shows the relevant references, the location and which of the four diseases they provide data for. The final study, recently published, is of particular interest in that it is the only one studying risk in ultra-low tar level cigarettes. Fuller details of the studies are given in Table A1.

Of the 26 studies, 15 were conducted in Europe and 11 in North America (all in the USA and one additionally in Canada). Nine were prospective studies and 17 case–control studies. Half of the studies were first reported in or before 1990. Seventeen studies reported results for lung cancer, 16 for heart disease, five for stroke and four for COPD. Three studies used surrogate measures for tar; nicotine in two studies of heart disease, and a combined tar and nicotine index in a study of all four diseases.

Meta-analysis results

Individual effect estimates for lung cancer are presented in Table 2, followed in Table 3 by the detailed meta-analysis results. Similarly, Table 4 gives individual estimates and Table 5 meta-analysis results for heart disease. Given the more limited data for these diseases, Table 6 (stroke) and Table 7 (COPD) present both the individual and combined effect estimates. Further details, shown in Tables A2–A5, give, for each disease separately, fuller details of the comparisons made, the factors adjusted for, and the derivation of estimates. The results for the four diseases are considered separately in the paragraphs that follow.

Table 2. Relative risks of lung cancer in lower compared to higher tar cigarette smokers.

Study	References	Location, Study type^a	Comparison	Gender^b	RR (95% CI) Unadjusted for cigs/day^c	RR (95% CI) Adjusted for cigs/day^d
1	Lee & Garfinkel (1981)	USA, P	Low versus High T/N	M^e		0.82 (0.63–1.07)
				M		0.80 (0.58–1.10)
				F^e		0.57 (0.36–0.90)
				F		0.63 (0.42–0.95)
2	Higenbottam et al. (1982)	UK, P	18–23 versus 33+ mg tar	M	0.68 (0.46–1.00)	0.53 (0.34–0.83)
3	Vutuc & Kunze (1982)	Austria, CC	<15 versus 24 mg tar	F	0.32 (0.09–1.17)	0.29 (0.09–0.95)
3	Vutuc & Kunze (1983)	Austria, CC	<15 versus 24 mg tar	M	0.32 (0.09–1.08)	0.30 (0.11–0.81)
5	Lubin & Blot (1984)	West Europe, CC	Low versus high tar	M	0.98 (0.75–1.28)	0.71 (0.55–0.93)
				F	0.67 (0.38–1.20)	0.67(0.38–1.18)
6	Alderson et al. (1986)	England, CC	17–22 versus 29+ mg tar	M		0.83 (0.57–1.20)
				F		1.12 (0.77–1.64)
9	Gillis et al. (1988)	Scotland, CC	<23 versus 23+ mg tar	M	0.73 (0.52–1.01)	0.74 (0.53–1.03)
10	Wilcox et al. (1988)	USA, CC	≤14 versus 21 mg tar	M	0.53 (0.29–0.97)	0.61 (0.32–1.13)
11	Wynder & Kabat (1988)	USA, CC	<10 versus 15+ mg tar	M	1.32 (0.89–1.96)
				F	0.97 (0.63–1.49)
12	Kaufman et al. (1989)	USA, Canada, CC	<22 versus 29+ mg tar	M	0.11 (0.04–0.30)	0.25 (0.08–0.83)
				F	0.17 (0.07–0.43)	0.21 (0.05–0.91)
14	Kuller et al. (1991)	USA, P	<16 versus 20+ mg tar	M	0.73 (0.38–1.41)	0.88 (0.52–1.49)
7	Sidney et al. (1993)	USA, P	<11 versus >18 mg tar	M	0.92 (0.63–1.32)	0.79 (0.41–1.49)
				F	1.49 (0.78–2.87)	1.49 (0.76–2.94)
16	Benhamou et al. (1994)	France, CC	“Light imported” versus 30+ mg tar	M	0.26 (0.14–0.48)	0.30 (0.10–0.91)
18	Tang et al. (1995)	UK, P	Per 15 mg tar decrease	M		0.75 (0.52–1.09)
20	Speizer et al. (1999)	USA, P	low versus high tar tertile	F	0.50 (0.36–0.67)	1.00 (0.71–1.43)
21	Woodward (2001)	Scotland, P	<10 versus 15+ mg tar	B	0.31 (0.11–0.86)	0.43 (0.15–1.23)
8	Harris et al. (2004)	USA, P	0–7 versus 22+ mg tar	M	0.81 (0.63–1.05)	0.78 (0.60–1.00)
				F	0.60 (0.44–0.81)	0.56 (0.41–0.76)
26	Lee et al. (2017b)	Europe, CC	≤3 versus 10+ mg tar	B	0.86 (0.61–1.20)	0.73 (0.50–1.06)

Tables A1 and A2 give more details of the tar groupings used (and of the T/N [combined tar and nicotine] grouping used in study 1), as well as giving details on the factors adjusted for and on the derivation of estimates. Studies are shown in chronological order of the reference used.

Studies 2 and 9 are included in the analysis of study 18.

Studies 3 and 16 are included in the analysis of study 5.

^a Study type: P: prospective; C: case–control.

^b Gender: M: males; F: females; B: both.

^c Most-adjusted estimate not including cigarettes/day.

^d Most-adjusted estimate including cigarettes/day.

^e First estimate is for 1960–1966 follow-up, and second is for 1966–1972 follow-up.

Table 3. Meta-analyses for lung cancer.

Estimates selected^a	N	RR (95% CI)^b	Het chisq, df^c	Het p
Including studies 2, 3, 9 and 16
Unadjusted estimates only	20	0.65 (0.53–0.79)	71.4, 19	<.001
Adjusted estimates only	25	0.72 (0.65–0.81)	36.4, 24	.05
Excluding studies 2, 3, 9 and 16
Unadjusted estimates only	15	0.70 (0.55–0.87)	56.5, 14	<.001
Adjusted estimates only	20	0.76 (0.68–0.84)	25.9, 19	.13
Prefer unadjusted estimates	22	0.74 (0.63–0.86)	63.3, 21	<.001
Prefer adjusted estimates	22	0.78 (0.70–0.88)	34.0, 21	.04
– all
– by period of publication
1981–1985	6	0.73 (0.63–0.84)	2.8, 5	.73
1986–1995	11	0.88 (0.71–1.09)	18.3, 10	.05
1996+	5	0.73 (0.59–0.91)	7.2, 4	.13
Between periods			5.7, 2	.06
– by gender
Male	11	0.80 (0.71–0.91)	11.8, 9	.30
Female	9	0.78 (0.61–1.00)	20.7, 8	.008
Combined	2	0.69 (0.48–0.96)	0.9, 2	.35
Between genders			0.7, 2	.71
– by study design
Prospective	12	0.76 (0.67–0.86)	14.2, 10	.22
Case–control	10	0.80 (0.65–1.00)	19.0, 9	.03
Between designs			0.9, 1	.34
– by number of confounders
<5	14	0.84 (0.73–0.97)	21.8, 12	.06
5+	8	0.68 (0.56–0.82)	9.5, 7	.22
Between groups			2.7, 1	.10
– by continent
North America	15	0.78 (0.67–0.92)	28.2, 14	.01
Europe	7	0.78 (0.67–0.90)	5.8, 6	.45
Between continents			0.03, 1	.87
– excluding estimated based on tar and nicotine	18	0.80 (0.73–0.88)	30.8, 17	.02

^a Unadjusted and adjusted relate to cigs/day.

^b Estimates are random effects.

^c Het chisq: heterogeneity chi-squared. For single analyses, this relates to variable between estimates within analysis. For comparison of subgroups (e.g. by period of publication or gender), this relates to variation between subgroups.

Table 4. Relative risk of heart disease in lower compared to higher tar cigarette smokers.

Study	References	Location, Study type^a	Comparison	Gender^b	RR (95% CI) Unadjusted for cigs/day^c	RR (95% CI) Adjusted for cigs/day^d
1	Lee & Garfinkel (1981)	USA, P	Low versus High T/N	M^e		0.94 (0.84–1.05)
				M		0.81 (0.69–0.95)
				F^e		0.81 (0.69–0.96)
				F		0.81 (0.68–0.97)
2	Higenbottam et al. (1982)	UK, P	18–23 versus 33+ mg tar	M	0.82 (0.63–1.06)	0.75 (0.57–1.00)
4	Kaufman et al. (1983)	USA, CC	<0.8 versus 1.5+ mg nicotine	M	1.19 (0.67–2.10)
7	Petitti & Friedman (1985a)	USA, P	Per 5 mg tar decrease	B		0.80 (0.63–1.01)^f
				B		0.85 (0.68–1.06)^g
6	Alderson et al. (1986)	England, CC	17–22 versus 29+ mg tar	M		0.93 (0.56–1.57)^h
				M		1.16 (0.60–2.26)^i
				F		0.90 (0.51–1.59)^h
				F		1.33 (0.79–2.24)^i
13	Palmer et al. (1989)	USA, CC	<0.4 versus 1.3+ mg nicotine	F	1.12 (0.55–2.28)
14	Kuller et al. (1991)	USA, P	≤15 versus 20+ mg tar	M	0.94 (0.65–1.35)	0.84 (0.61–1.16)
15	Negri et al. (1993)	Italy, CC	<10 versus >20 mg tar	B	1.03 (0.56–1.87)	1.00 (0.56–2.00)
17	Parish et al. (1995)	UK, CC	<10 versus 10+ mg tar	B	0.83 (0.73–0.94)^j	0.86 (0.75–0.98)^j
				B	0.96 (0.81–1.12)^k	0.99 (0.84–1.17)^k
18	Tang et al. (1995)	UK, P	Per 15 mg tar disease	M		0.77 (0.61–0.97)
19	Bosetti et al. (1999)	Italy, CC	<10 versus 15+ mg tar	M	1.15 (0.72–1.82)
				F	0.71 (0.34–1.49)
22	Tavani et al. (2001)	Italy, CC	Low versus High tar	B	1.15 (0.65–2.04)
21	Woodward (2001)	Scotland, P	<10 versus 15+ mg tar	B	0.79 (0.46–1.37)	0.96 (0.46–2.04)
23	Sauer et al. (2002)	USA, CC	≤6 versus 12+ mg tar	B	0.45 (0.30–0.68)	0.53 (0.38–0.75)
24	Gallus et al. (2007)	Italy, CC	<10 versus 20+ mg tar	B	0.80 (0.51–1.25)	0.78 (0.50–1.23)
25	Zhang et al. (2011)	Germany, P	1–12 versus 13+ mg tar	M		0.75 (0.49–1.15)
				F		0.58 (0.20–1.70)

Tables A1 and A3 give more details of the tar and nicotine groupings used (including the T/N [combined tar and nicotine] index used in study 1), as well as giving details on the factors adjusted for and on the derivation of estimates. Studies are shown in chronological order of the reference used.

Study 2 is included in the analysis of study 18.

Study 15 is included in the analysis of study 19 which is included in the analysis of study 24.

Study 22 is included in the analysis of study 24.

^a Study type: P: prospective; C: case–control.

^b Gender: M: males; F: females; B: both.

^c Most-adjusted estimate not including cigarettes/day.

^d Most-adjusted estimate including cigarettes/day.

^e First estimate is for 1960–1966 follow-up, and second is for 1966–1972 follow-up.

^f RR for acute myocardial infarction.

^g RR for other ischemic heart disease.

^h Estimates for age 35–54 years.

ⁱ Estimates for age 55–74 years.

^j Estimates for age 30–59 years.

^k Estimates for age 60–79 years.

Table 5. Meta-analyses for heart disease.

Estimates selected^a	N	RR (95% CI)^b	Het chisq, df^c	Het p
Including studies 2, 15, 19 and 22
Unadjusted estimates only	13	0.87 (0.77–0.98)	17.1, 12	.15
Adjusted estimates only	21	0.85 (0.81–0.90)	21.6, 20	.36
Excluding studies 2, 15, 19 and 22
Unadjusted estimates only	8	0.84 (0.72–0.99)	13.9, 7	.05
Adjusted estimates only	19	0.85 (0.80–0.90)	20.5, 18	.31
Prefer unadjusted estimates	21	0.85 (0.80–0.90)	23.5, 20	.26
Prefer adjusted estimates	21	0.86 (0.81–0.91)	22.3, 20	.33
– all
– by period of publication
1981–1989	12	0.87 (0.82–0.93)	9.5, 11	.58
1990–1999	4	0.88 (0.80–0.97)	3.4, 3	.33
2000+	5	0.67 (0.54–0.83)	3.5, 4	.48
Between periods			5.91, 2	.05
– by gender
Male	8	0.88 (0.81–0.95)	5.4, 7	.61
Female	6	0.85 (0.75–0.96)	5.3, 5	.38
Combined	7	0.83 (0.73–0.94)	11.2, 6	.08
Between genders			0.4, 2	.82
– by study design
Prospective	11	0.85 (0.80–0.90)	6.0, 10	.82
Case–control	10	0.90 (0.77–1.04)	15.4, 9	.08
Between designs			0.9, 1	.34
– by number of confounders
<5	8	0.90 (0.82–0.98)	6.4, 7	.50
5+	13	0.84 (0.79–0.90)	14.7, 12	.26
Between groups			1.2, 1	.26
– by continent
North America	10	0.83 (0.76–0.91)	13.7, 9	.13
Europe	11	0.89 (0.82–0.97)	7.9, 10	.64
Between continents			0.8, 1	.38
– excluding estimates not based specifically on tar	15	0.85 (0.78–0.92)	16.5, 14	.28
– excluding estimates based on nicotine	19	0.85 (0.80–0.90)	20.5, 18	.31

^a Unadjusted and adjusted relate to cigarettes/day.

^b Estimates are random effects.

^c Het chisq: heterogeneity chi-squared. For single analyses, this relates to variable between estimates within analysis. For comparison of subgroups (e.g. by period of publication or gender), this relates to variation between subgroups.

Table 6. Relative risks of stroke in lower compared to higher tar cigarette smokers.

Study	References	Location, study type^a	Comparison	Gender^b	RR (95% CI) Adjusted for cigs/day
1	Lee & Garfinkel (1981)	USA, P	Low versus High T/N	M	0.71 (0.57–0.88)
				F	0.98 (0.81–1.19)
7	Petitti & Friedman (1985b)	USA, P	Per 5 mg tar decrease	B	0.57 (0.37–0.89)
6	Alderson et al. (1986)	UK, P	17–22 versus 29+ mg tar	M	0.53 (0.31–0.91)
				F	1.00 (0.62–1.62)
18	Tang et al. (1995)	UK, P	Per 15 mg tar decrease	M	0.86 (0.50–1.50)
21	Woodward (2001)	Scotland, P	<10 versus 15+ mg tar	B	0.29 (0.04–2.38)
			Number of estimates		7
			Combined RR (95% CI)		0.77 (0.62–0.95)
			Heterogeneity chi-squared		11.79
			df		6
			p		.07

Tables A1 and A4 give more details of the tar groupings used (including the T/N [combined tar and nicotine] index used in study 1), as well as details on the factors adjusted for and on derivation of estimates. Studies are shown in chronological order of the reference used.

^a P: prospective study.

^b M: male; F: female; B: both sexes.

Table 7. Relative risks of COPD in lower compared to higher tar cigarette smokers.

Study	References	Location, study type^a	Comparison	Gender^b	RR (95% CI) Adjusted for cigs/day
1	Lee & Garfinkel (1981)	USA, P	Low versus high T/N	M	0.78 (0.53–1.15)
				F	0.59 (0.32–1.09)
7	Petitti & Friedman (1985b)	USA, P	Per 5 mg tar decrease	B	1.00 (0.90–1.10)
6	Alderson et al. (1986)	UK, P	17–22 versus 29+ mg tar	M	0.55 (0.35–0.86)
				F	1.03 (0.65–1.64)
18	Tang et al. (1995)	UK, P	Per 15 mg tar decrease	M	0.78 (0.41–1.48)
			Number of estimates		6
			Combined RR (95% CI)		0.81 (0.65–1.02)
			Heterogeneity chi-squared		10.48
			df		5
			p		.06

Tables A1 and A5 give more details of the tar groupings used (including the T/N [combined tar and nicotine] index used in study 1), as well as details on the factors adjusted for and on derivation of estimates. Studies are shown in chronological order of the reference used.

^a P: prospective study.

^b M: male; F: female; B: both sexes.

Lung cancer

Of the 45 estimates shown in Table 2, 40 show a lower risk in the lower tar group (16 statistically significant), one shows the same risk in the two groups, and four show a higher risk (none significant). Not surprisingly, therefore, the meta-analysis results shown in Table 3 generally show a significantly lower risk in the low tar groups. Adjustment for amount smoked somewhat reduced the combined relative risk estimate. Using estimates adjusted for amount smoked where available, and avoiding double-counting by excluding results from studies 2, 3, 9 and 16, the overall relative risk for low versus high tar is 0.78 (95% CI 0.70–0.88), with modest heterogeneity (p = .04), the results being illustrated further in Figure 1. The lower risk in lower tar smokers is evident regardless of period of publication, gender, study design, number of confounders or continent, with no clear evidence of heterogeneity by level of these factors. Omitting results from study 1, which classified individuals based on a combined index of tar and nicotine, little affected the combined relative risk estimate.

Figure 1. Forest plot of relative risks of lung cancer in lower compared to higher tar cigarette smokers (preferring adjusted estimates and excluding studies 2, 3, 9 and 16 to avoid overlaps).

Heart disease

Of the 34 estimates shown in Table 4, 26 show a lower risk in the lower tar group (eight significant), one shows the same risk in the two groups, and seven show a higher risk (none significant). In line with this, the meta-analysis results shown in Table 5 generally show a significantly lower risk in the lower tar group. Estimates are similar, whether or not the estimates are adjusted for amount smoked, and whether the studies based on nicotine are included or not. A combined estimate using adjusted estimates where available, and omitting results from studies 15, 19 and 22 to avoid double counting, is 0.86 (95% CI 0.81–0.91) with no significant heterogeneity (see also Figure 2). Again there is evidence of a lower risk in lower tar smokers regardless of time, gender, study design, extent of adjustment for confounding variables or continent. There is some indication that the differential in risk is more marked in studies published this century. The estimate was little changed if estimates based on nicotine or a combined tar and nicotine index are excluded, or if only estimates based specifically on nicotine are excluded.

Figure 2. Forest plot of relative risks of heart disease in lower compared to higher tar cigarette smokers (preferring adjusted estimates and excluding studies 2, 15, 19 and 22 to avoid overlaps).

Stroke

The seven available estimates given in Table 6 show, overall, a modest reduction in risk for lower tar smoking, with a combined estimate of 0.77 (95% CI 0.62–0.95) and no substantial heterogeneity (p = .07).

COPD

As shown in Table 7, only six estimates are available. Here the relative risk is 0.81, equivalent to a risk reduction of 19%, which is similar in magnitude to that seen for lung cancer (22%), heart disease (14%) and stroke (23%). However, unlike for the other diseases, the reduction is not statistically significant, the 95% CI of the relative risk being 0.65–1.02.

Discussion

For the four smoking-related diseases considered the data summarized are consistent with smokers of lower tar yield cigarettes having a lower risk than smokers of higher tar yield cigarettes. Using estimates adjusted for amount smoked where possible, and avoiding overlaps, the risk in the lowest tar group is estimated to be 22% lower for lung cancer and 14% for heart disease than that in the highest tar group. There is little evidence of heterogeneity between the individual estimates. Estimates indicating a lower risk considerably outnumbered those indicating a higher increase. The number of studies providing estimates was much smaller for stroke and for COPD, though again the combined estimates indicated a similarly lower risk for low tar smokers; by 23% for stroke and by 19% for COPD. With the exception of COPD all the overall relative risk estimates were statistically significant, those for lung cancer and heart disease very highly so.

Although risks in smokers of low tar cigarettes are reduced compared to those in smokers of high tar cigarettes, it is clear that they are much higher than in those who have never smoked, the estimated reductions in risk for low versus high tar smokers being much less than the increase in risk associated with cigarette smoking (US Surgeon General, 2014).

The conclusions reached are not dissimilar from those in a review by Kabat (2003) who suggested that “smokers of lower tar cigarettes have a lower risk (by 20–30%) compared to smokers of higher tar cigarettes” and pointed to a tendency for risks also to be lower for heart disease and COPD. The estimate reported here of a 22% reduction for lung cancer is also not dissimilar to the estimates of 27% and 30% that I have reported earlier based on fewer studies (Lee et al., 2012; Lee & Sanders, 2004). While some authors (Alberg & Samet, 2003; Thun & Burns, 2001; US Surgeon General, 2014) have concluded that lower tar cigarettes provide little or no health advantage, or may even have an adverse effect, these opinions are not based on meta-analyses. Instead, they derive from a combination of concerns about the propensity for bias in the epidemiology, evidence that smokers of low tar/nicotine cigarettes may “compensate” in order to obtain their required nicotine dose, and the fact that, over a period when tar levels were decreasing, overall lung cancer risk rose in the US and the relative frequency of lung adenocarcinoma to squamous cell carcinoma rose in many countries. Sources of bias and compensation are discussed separately below. Detailed consideration of the data on lung cancer rate trends is beyond the scope of this paper, although the interested reader might wish to refer to publications of mine suggesting alternative explanations for the trends (Lee, 2016; Lee & Gosney, 2016; Lee & Sanders, 2004).

A limitation of the evidence is that subdivision into low and high tar level groups is based on the brands on the market at a particular time. Thus, while early studies included high tar groups with average tar yields of around 30 mg, they included low tar groups with average yields that were higher than for any cigarettes currently on the market. No studies are available which, say, directly compared smokers who, in their whole life, smoked, either 30 + mg tar cigarettes or <10 mg tar cigarettes, and indeed such a study would be impossible to design as tar reduction has been ongoing for more than 50 years.

The magnitude of the reduction seems relatively small, given that reductions over time in the average tar levels of cigarettes smoked have been substantial. There are two main reasons for this.

One is that, as noted above, comparisons of tar level within any given study do not reflect the full range of tar yields which have been available over the years. Indeed study 26 was specifically set up to investigate risk associated with smoking cigarettes with yields of ≤3 mg/cig, as risk of these so called “ultra-low” tar cigarettes, which have become increasingly popular in the last 30 years, had never previously been investigated.

The other reason lies in what is termed “compensation”, with smokers of products which have reduced yields (as measured by machines under standard smoking conditions) modifying how they smoke their cigarettes so that the reduction in yields is not reflected in the actual dose received by the smoker. While some authorities (National Cancer Institute, 2001) believe that compensation is complete enough for smokers of reduced tar cigarettes not to benefit at all from the nominal reduction in yield, a recent detailed review (Scherer & Lee, 2014) which I collaborated in revealed that though compensation is substantial, it is clearly not complete. Based on 19 brand-switching studies, in which nicotine uptake had been measured, a compensation index of 0.744 (95% CI 0.682–0.806) was estimated, where 1 indicates complete and 0 no compensation. A value of 0.744 is consistent with a 50% reduction in nominal yield being associated with a 17% reduction in uptake. Thus, even if tar yield is markedly reduced, one would not expect a substantial reduction in dose to the smoker, and consequently in risk of smoking-related disease.

The question of whether or not one should adjust RR estimates for daily amount smoked is an interesting one. If smokers switching to a reduced yield product increase their daily cigarette consumption as a form of compensation, then adjustment for amount smoked in analysis can be regarded as a form of “over-adjustment”, and would bias the RR estimate. However, if smokers choosing a lower yield product tend to be less “addicted” smokers and smoke fewer cigarettes a day than those choosing a higher yield product, then failure to adjust for amount smoked could lead to confounding. In principle, the approach to solve this dilemma is to compare smokers switching or not switching to a reduced yield product, and adjust for average daily cigarette consumption before the switch. However, such data are generally not available, and in the absence of this, I presented results for lung cancer and heart disease, both unadjusted and adjusted for daily amount smoked. In practice, the evidence for a modest reduction in risk was clear whether or not such adjustment was made. The estimated risk reductions by 22% for lung cancer and 14% for heart disease referred to above were derived preferring adjusted estimates where there was a choice. Had unadjusted estimates been preferred these reductions would be little changed, becoming 26% for lung cancer and 15% for heart disease. For stroke and COPD, I only presented adjusted results, mainly because, for four of the studies (1, 6, 7 and 18) which contributed almost all the estimates, only adjusted results were available.

Arguments about the appropriateness or not of adjusting for amount smoked also apply to other aspects of smoking which might be affected by the tar level of the brand smoked. This would particularly apply to adjustment for inhalation, given the evidence discussed above relating to “compensation”. However, only two studies (2 and 16) have presented results adjusted for inhalation as self-reported, and in both cases a reduction in risk in smokers of lower tar cigarettes was evident, whether or not adjustment for inhalation was made. Adjustment for age of starting to smoke is less problematic, the decision to start smoking being unlikely to depend materially on the tar level of the brand smoked. In general, adjustment for aspects of smoking, whether this is amount smoked, inhalation, age of starting to smoke or duration of smoking does not affect the conclusion of a lower risk of the diseases studied in smokers of lower tar cigarettes.

A limitation of the meta-analysis lies in the variation in the comparisons made in the different studies. The fact that two studies of heart disease compared smokers of cigarettes with a low and high yield of nicotine, and a further study considered yields of both tar and nicotine does not seem important as exclusion of results from these studies hardly affected the overall estimates. In any case, as noted in the introduction, tar and nicotine levels are correlated. More important is the variation in average yield levels between low and high groups, which one might expect to be reflected in differing RRs, if tar level is relevant to risk. Ideally one would wish to express all the comparisons in terms of a “per x mg difference”, where x is constant over study, but the data do not allow this, very few studies giving average yields of cigarettes smoked in the groups being compared. That said, the fact that the great majority of the RRs were below 1, many being significant, and that no significant increases were seen, gives confidence in the main conclusion of this review, namely, that there is a moderate but clear reduction in risk associated with smoking lower tar cigarettes. While the conclusion of a risk reduction could have been reached simply from the preponderance of RRs below 1, the meta-analyses, despite the limitations noted, give an indication of the magnitude of the risk reduction, and allow one to test for consistency of the reduction over subgroups.

A further limitation of the evidence is that subdivision into low and high tar level groups in the studies considered is often based on the brand smoked at a single time point, or over a limited period of the smokers’ lifetime. Even where lifetime histories are recorded, errors in recall may limit the ability to detect changes in risk associated with tar reduction.

Those who choose to use lower tar cigarettes (from the cigarettes on the market at any given time) may have other characteristics indicative of lower risk than those who choose to use higher tar cigarettes. For example, they may be more educated, or less employed in “risky” occupations. The extent to which risk factors other than smoking was taken into account in the studies considered was quite limited for lung cancer where, of the 17 studies considered (Table A2), only seven took into account education/occupation/social class, only two adjusted for diet and only one adjusted for a family history of lung cancer. However, the failure to consider well known risk factors may not be serious, since adjustment generally has little effect on relative risks comparing current and never smokers (Lee et al., 2012). For heart disease, adjustment for other risk factors was more extensive, with Table A3, between seven and 11 of the 16 studies considered adjusting for education/occupation/social class, history of heart disease, blood pressure, exercise/physical activity, obesity/BMI, alcohol, diet and cholesterol. Given that the advantage to lower tar smokers was no less for estimates adjusted for five or more potential confounding variables than for those adjusted for less than five, it seems unlikely that the advantage can be explained as being due to incomplete confounder adjustment.

Our analyses were limited, as explained, to the four main smoking-related diseases, without any consideration of other diseases, or sub-categories of the diseases that were considered. It has been argued (US Surgeon General, 2014) that tar reduction has led to an increase in the relative frequency of adenocarcinoma to squamous carcinoma of the lung, due to increased inhalation and deposit of particulate matter in the peripheral areas of the lung. While there is an unfortunate lack of data relating tar level (rather than filter/plain use) to histological type of lung cancer it is clear that this claim has little foundation for a number of reasons. Apart from the lack of evidence of an increased risk of adenocarcinoma in filter compared to plain cigarette smokers (Brooks et al., 2005; Lee et al., 2012; Lee & Sanders, 2004; Marugame et al., 2004; Papadopoulos et al., 2011), it is also clear that there has been a rise in relative frequency of adenocarcinoma to squamous carcinoma in non-smokers (Lee et al., 2016) and persuasive evidence that changes in methods for classification of histological type have contributed to the observed increase in relative frequency (Lee & Gosney, 2016).

Conclusion

Smokers of lower tar cigarettes have a clearly demonstrable lower risk of the four main smoking-related diseases than do smokers of higher tar cigarettes.

Acknowledgements

I thank Barbara Forey for her help in checking the data extraction and meta-analyses, and commenting on the paper and Yvonne Cooper and Diana Morris for assistance in literature searching and in preparing the manuscript.

Disclosure statement

Peter Lee, director of P.N. Lee Statistics and Computing Ltd. is a long-term consultant to various tobacco companies and organizations.

Additional information

Funding

This work was supported by Philip Morris International [project number 16].

Appendix

Table A1. Further details of the studies.

	Study population^a	Controls^b	Age^c	Details of groupings used (only lowest and highest levels used in Tables 2–7)
1	Member of household with person age >45, excluding migrant workers		30+/30+	For 1960–1966 high T/N was defined as >2.0 mg nicotine and 2.5 mg tar, low T/N as less than 1.2 mg nicotine and less than 17.6 mg tar, and medium T/N as intermediate. For 1966–1972, the definition of low T/N was the same but that of high T/N was lowered “somewhat”
2	Civil servants		40–69/89	18–23, 24–32, 33+ mg tar
3	General	Diseased	Unrestricted	<15, 15–24, 24+ mg tar
4	General	Diseased	40–69	<0.8, 0.8–0.9, 1.0–1.1, 1.2–1.4, 1.5+ mg nicotine
5	General	Diseased	Unrestricted	Categories were the within-country 10, 25, 50, 75 and 90th percentiles. Mean tar values within categories were 15.6, 18.5, 20.6, 23.6, 25.2 and 29.8 mg.
6	General	Diseased	35–74	17–22, 23–28, 29+ mg tar
7	Attending screening		30+/30+	For lung cancer <11, 11–18, >18 mg tar For other diseases per 5.0 mg tar decrease
8	General		30+/30+	0–7, 8–14, 15–21, 22+ mg tar
9	General	Diseased	Unrestricted	<17, 17–22, 23–28, 29+ mg tar (as the extreme groupings had very little data the comparison used was <23 versus 23+ mg tar).
10	General	Healthy + decedents	Unrestricted	≤14, 14.1–17.5, 17.6–21.0, 21.1–28.0 mg tar
11	General	Diseased	Unrestricted	<10, 10–14, 15+ mg tar
12	General	Diseased	40–69	<22, 22–28, 29+ mg tar
13	General	Diseased	25–64	<0.40, 0.40–0.63, 0.64–0.75, 0.76–1.00, 1.01–1.06, 1.07–1.29, 1.30+ mg nicotine
14	Attending screening, with high coronary risk		35–57/69	≤15, 16–19, 20+ mg tar
15	General	Diseased	Unrestricted	<10, 10–15, >15–20, >20 mg tar
16	General	Diseased	Unrestricted	Light imported (unknown tar levels) for life, and use of 30+ mg tar cigs for ≤50%, 51–75% or >75% of years smoked
17	General	Healthy	30–79	<10, 10+ mg tar
18	4 cohorts: professional, civil service, factory, residents		35–78/98	Per 15 mg tar decrease
19	General	Diseased	<75	<10, 10–15, >15 mg tar
20	Nurses		30–55/71	Tertiles of tar (values not stated)
21	General		40–59/74	<10, 10–14, 15+ mg tar
22	General	Diseased	25–79	Low, high tar not further defined
23	General (with telephone and English speaking)	Healthy	30–65	≤6, 7–12, >12 mg tar
24	General	Diseased	Unrestricted	<10, 10–19, 20+ mg tar
25	General		25–74/90	1–12, 13+ mg tar
26	General	Diseased	Unrestricted	≤3 mg, 10+ mg tar

T/N: combined tar and nicotine index.

^a “general” indicates that population was of a general nature with no major inclusion or exclusion criteria.

^b For CC studies, the control group is described as diseased (includes hospital controls), healthy (includes community controls) or decedent.

^c For case–control studies the age range is shown, with “unrestricted” indicating no restriction on the age of cases. For prospective studies the age range at baseline and at final follow-up are shown separated by a “/”.

Table A2. Further details associated with Table 2 (lung cancer).

	RRs unadjusted for cigs/day^a		RRs adjusted for cigs/day^b
Study	Derivation of estimates from source tables^c	Covariates considered	Derivation of estimates from source tables^c	Covariates considered^d
1	–	–	Table 5 gives the RRs. The 95% CIs were estimated using method 6, with the additional assumption that the ratio of cases for low to high tar is 40:60 (based on additional data in Hammond et al. (1976))	Age, race, residence, education, occupational exposures, history of lung cancer, history of heart disease, age start, cigs/day
2	Estimated from Table 4, first estimating the rates combined over inhalation × cigs/day groups given, then method 5	Age, employment grade	Estimated from Table 4, first estimating the RR/CI within each inhalation × cigs/day group given, then method 3	+ inhalation, cigs/day
3	Estimated from Table 1 using method 1	Unadjusted	Estimated from Table 1 using method 4	Years smoked, cigs/day
5	Estimated from Table X using method 1	Unadjusted	Estimated from Table X using method 6	Age, study center, years smoked, years quit, cigs/day
6	–	–	Estimated from Table 13F using methods 6 and 8, with the additional assumption that the loss due to missing tar data was the same in controls as in cases	Age, cigs/day
9	Estimated from Table 5 using method 1	Unadjusted	Estimated from Table 5 using method 1	Cigs/day
10	Given in Table 2	Unadjusted	Given in Table 2	Cigs/day
11	Estimated from Figures 1 and 2, summing over LC types then method 1	Unadjusted	–	–
12	Estimated from Table 5 using method 1	Unadjusted	Estimated from Table 5 using method 8	Age, race, region, education, years of interview, age start, cigs/day
14	Estimated from Table 8 using method 1	Unadjusted	Estimated from Table 8 using method 8	Cigs/day
7	Estimated from Table 3 using methods 5 and 6	Age	Estimated from Table 3 using method 8	+ race, education, years smoked, cigs/day
16	Estimated from Table 3 using method 8	Age	Estimated from Table 3 using method 8	+ inhalation, tobacco type, use of filter cigarettes, years smoked, cigs/day
18	–	–	Given in Table 5	Age, study, cigs/day
20	Estimated from text on p477 using method 8	Age start	Estimated from text on p477 using method 8	+ cigs/day
21	Estimated from Table 3 using method 8	Age, gender	Estimated from Table 3 using method 8	+ social class, Bortner score, β-carotene, α-tocopherol and vitamin C consumption, BMI, urinary potassium, years smoked, cigs/day
8	Estimated from Table 5 using method 4	Education, race, marital status, occupation, vegetables and citrus fruit consumption, use of vitamin A, C or E, occupation, age start	Estimated from Table 5 using method 4	+ cigs/day
26	Given in Table 2	Unadjusted	Given in Table 2	Gender, age, country, education, age start, tar level^e, cigs/day^e

BMI: body mass index.

Comparisons used: See Table 2 and Table A1 for details of the groupings used and the comparisons made.

^a Most-adjusted estimate not including cigs/day.

^b Most-adjusted estimate including cigs/day.

^c The method numbers refer to the methods described fully in Lee et al. (2012) – see Additional File S1, section “Derivation of RR”. In brief, the methods used here are (1) calculated from 2 × 2 table; (3) combining independent estimates; (4) combining non-independent estimates by the method of Hamling et al. (2008); (5) ratio of rates; (6) CI for the adjusted RR estimated from the crude numbers; (8) low/high estimated from high/low by inversion.

^d + implies that the covariates listed are additional to those used for the unadjusted RRs.

^e Mean of brands smoked 21–50 years before interview.

Table A3. Further details associated with Table 4 (heart disease).

	RRs unadjusted for cigs/day^a		RRs adjusted for cigs/day^b
Study	Derivation of estimates from source tables^c	Covariates considered	Derivation of estimates from source tables^c	Covariates considered^d
1	–	–	Table 7 gives the RRs. The 95% CIs were estimated using method 6, with the additional assumption that the ratio of cases for low to high tar is 40:60 (based on additional data in Hammond et al. (1976)).	Age, race, residence, education, occupational exposures, history of lung cancer, history of heart disease, history of stroke, history of diabetes, history of blood pressure, exercise, obesity, coffee/tea, alcohol, aspirin, occupation, age start, cigs/day.
2	Estimated from Table 6, first estimating the rates combined over inhalation × cigs/day groups given, then method 5	Age, employment grade	Estimated from Table 6, first estimating the RR/CI within each inhalation × cigs/day group given, and using method 2 for a group with zero cases, then method 3	+ inhalation, cigs/day
4	Estimated from Table 3 using method 4	Age, region, treatment for hypertension, cholesterol, treatment for diabetes, family history of MI or stroke, personality, alcohol, religion, marital status	–	–
7	Estimated from Table 3 model 3 (for AMI and other IHD) using method 8	–	Estimated from Table 3 (for AMI and other IHD) using method 8	Age, gender, race, Quetelet index, cholesterol, history of hypertension, alcohol, cigs/day
6	–	–	Estimated from Tables 15F and 16F using methods 6 and 8, with the additional assumption that the loss due to missing tar data was the same in controls as in cases	Age, cigs/day
13	Estimated from Table 3 using method 4	Age, history of hypertension, history of angina, history of diabetes, history of raised cholesterol, family history of MI before age 60, menopause, BMI, type A score, exercise, education, residence, use of OCs, use of estrogen, coffee, alcohol	–	–
14	Estimated from Table 8 using method 1	Unadjusted	Estimated from Table 8 using method 8	Age, cholesterol, blood pressure, cigs/day
15	Estimated from Tables II and III using method 4	Age, gender, education, cholesterol, history of diabetes and hypertension, family history of AMI, BMI, coffee	Estimated from Table III using method 8	+ cigs/day
17	Estimated from Table III using method 8	Age, gender	Estimated from Table III using method 8	+ cigs/day
18	–	–	Given in Table 5	Age, study, cigs/day
19	Estimated from Table 2 using method 4	Age, education, study center, cholesterol, history of diabetes, hypertension and hyperlipidemia, family history of AMI, coffee, alcohol, BMI	–	–
22	Estimated from Table 3 using method 4	Age, gender, education, BMI, cholesterol, coffee, alcohol, physical activity, hyperlipidemia, diabetes, hypertension, family history of AMI	–	–
21	Estimated from Table 3 using method 8	Age, gender	Estimated from Table 3 using method 8	+ social class, Bortner score, β-carotene, α-tocopherol and vitamin C, potassium, years smoked, cigs/day
23	Estimated from Table 2 using method 1	Unadjusted	Estimated from Table 8 using method 8	Age, gender, race, BMI, history of CHD, hypertension, diabetes or hypercholesterolemia, family history of CHD, exercise, vitamin use, education, years smoked, cigs/day
24	Estimated from Table 4 using method 8	Age, gender, study, education, BMI, alcohol, coffee, cholesterol, family history of AMI, history of diabetes and hypertension	Estimated from Table 4 using method 8	+ years smoked, cigs/day
25	–	–	Estimated from Table 3 using method 4	Age, survey, alcohol, hypertension, cholesterol/HDL-C ratio, physical inactivity, diabetes, cigs/day

AMI: acute myocardial infarction; BMI: body mass index; CHD: coronary heart disease; HDL-C: high density lipoprotein cholesterol; IHD: ischemic heart disease; OC: oral contraceptives.

Comparisons used: See Table 4 and Table A1 for details of the groupings used and the comparisons made.

^a Most-adjusted estimate not including cigs/day.

^b Most-adjusted estimate including cigs/day.

^c The method numbers refer to the methods described fully in Lee et al. (2012) – see Additional File S1, section “Derivation of RR”. In brief, the methods used here are (1) calculated from 2 × 2 table; (4) combining non-independent estimates (Hamling et al., 2008); (6) CI for the adjusted RR estimated from the crude numbers; (8) low/high estimated from high/low by inversion.

^d + implies that the covariates listed are additional to those used for the unadjusted RRs.

Table A4. Further details associated with Table 6 (stroke).

Study	Derivation of estimates from source tables^a	Covariates considered
1	Table 6 gives the RRs. The 95% CIs were estimated using method 6, with the additional assumption that the ratio of cases for low to high tar is 40:60 (based on additional data in Hammond et al. (1976)).	Age, race, residence, education, occupational exposures, history of lung cancer, history of heart disease, age start, cigs/day
7	Estimated from Table 3 model 3, using method 8	Age, sex, race, Quetelet index, serum cholesterol, history of hypertension, alcohol, cigs/day.
6	Estimated from Table 17F using methods 6 and 8, with the additional assumption that the loss due to missing tar data was the same in controls as in cases	Age, cigs/day
18	Given in Table 5	Age/study, cigs/day
21	Estimated from Table 3 using method 8	+ social class, Bortner score, β-carotene, α-tocopherol and vitamin C consumption, BMI, urinary potassium, years smoked, cigs/day

BMI: body mass index.

Comparisons used: See Table 6 and Table A1 for details of the groupings used and the comparisons made.

^a The method numbers refer to the methods described fully in Lee et al. (2012) – see Additional File S1, section “Derivation of RR”. In brief, the methods used here are (6) CI for the adjusted RR estimated from the crude numbers; (8) low/high estimated from high/low by inversion.

Table A5. Further details associated with Table 7 (COPD).

Study	Derivation of estimates from source tables^a	Covariates considered
1	Table 7 gives the RRs. The 95% CIs were estimated using method 6, with the additional assumption that the ratio of cases for low to high tar is 40:60 (based on additional data in Hammond et al. (1976)).	Age, race, residence, education, occupational exposures, history of lung cancer, history of heart disease, age start, cigs/day
7	Given in Table 3	Age, gender, race, cigs/day
6	Estimated from Table 14F using methods 6 and 8, with the additional assumption that the loss due to missing tar data was the same in controls as in cases	Age, cigs/day
18	Given in Table 5	Age/study, cigs/day

COPD: chronic obstructive pulmonary disease.

Comparisons used: See Table 7 and Table A1 for details of the groupings used and the comparisons made.

^a The method numbers refer to the methods described fully in Lee et al. (2012) – see Additional File S1, section “Derivation of RR”. In brief, the methods used here are (6) CI for the adjusted RR estimated from the crude numbers; (8) low/high estimated from high/low by inversion.