It has long been known [1, 2] that current smokers have a lower concentration of high density lipoprotein cholesterol (HDL-C) than do non-smokers. Evidence that ex-smokers have higher HDL-C concentrations than do current smokers (e.g. [3]) also exists, though partly based on between-group comparisons, with current smokers and quitters differing on other relevant variables. A method of investigating how quitting affects HDL-C, which better avoids confounding, uses within-subject estimates of change. In 2003, Maeda et al. [4] reviewed evidence of effects of quitting on lipid/ lipoprotein profiles, using data on within-subject changes. Their meta-analysis for HDL-C, based on 29 studies with follow-up periods from 30 days to 2 years or more, showed an increase following quitting of 0.100 (95% CI 0.074 to 0.127) mmol/l. The estimates were highly heterogeneous (p<0.001), though significant (p<0.05) variations were only seen by gender (greater increases in females) and by baseline HDL-C (greater increases for higher pre-quitting levels). No significant variation was seen by study design, publication year or intra-measurement period (increases similar for periods less or greater than 8 weeks).

Although more recent reviews have considered effects of quitting on HDL-C (e.g. [57]), none involve an updated meta-analysis. This review presents an updated meta-analysis using data on changes, more fully investigating how the increase following quitting varies by other factors. Of particular interest is quantifying how the increase varies with time, some authors claiming “a complete and very fast reversibility of changes after smoking cessation” [8] and others an initial quite rapid “movement towards normalization of HDL-C” which “will continue to progress toward normal (non-smoking) levels as long as cessation continues” [6]. Some studies reporting changes in quitters also report changes in continuing and never smokers. We summarize these data.



Selection of studies and literature searches

For selected publications we examined abstracts, and where necessary full text, to find studies satisfying these inclusion criteria: (1) relevant to HDL-C and quitting; (2) clinical trial, human experimental, or epidemiological study; (3) prospective or longitudinal study; (4) at least five subjects quitting; (5) not restricted to those with coronary artery disease or taking antihyperlipidaemic drugs; (6) smoking habits recorded and blood sample taken concurrently on two or more occasions; (7) results reported separately for quitters during the study; (8) results available relevant to the change in HDL-C following quitting; (9) quitting for at least a day; and (10) results relate to conventional cigarette smoking.

Papers cited by Maeda et al. [4] were considered, the Medline searches they conducted being updated to July 2012. These used the search terms “(smoking OR tobacco OR cigar) AND (tobacco use cessation OR tobacco use cessation products OR stop OR quit OR cessation OR abstinence OR exsmokers) AND (cholesterol OR triglycerides OR lipoprotein OR HDL)”. New papers were then sought from Collaborative Trials within the Cochrane Library using the same keywords, from reference lists in accepted papers, and from an ongoing project on white blood cell changes (submitted for publication). The selected papers were then separated into studies, further study details being obtained from additional references if required.

Data entry

Relevant data were entered onto a study and a change database, each study being identified by a 6-letter reference (REF). The study database contains one record per study describing study attributes. The change database contains one or more records per study, describing estimates of HDL-C change from baseline.

Study attributes recorded include relevant publications, sexes considered, age range, location, years of start, finish and publication, length of follow-up, study design, nature of population studied (including smoking and medical criteria), study size, HDL-C measurement method, fasting or smoking abstinence requirements before measurement, diet or exercise modification during follow-up, and confounding and stratifying variables considered.

Details on the change database include smoking status (quitter, continuing smoker, never smoker), smoking habits at baseline and follow-up (products, cigarette types, amount smoked), biochemical validation methods, intra-measurement period, original measurement unit, data source, and population the data applies to (sex, age, intervention groups). Also recorded is information on the HDL-C change itself (mean change, or mean level at both baseline and follow-up), its variability (confidence limits, SD, SE, N, significance of change) and whether the change estimate was direct, or relative to never smokers or to continuing smokers. Where the mean change was unavailable, the median change, change in median, or an estimate based on the mean % change was accepted. Change data were entered for never smokers, continuing smokers and quitters, but not for smokers who quit during the follow-up period but resumed before the second HDL-C measurement. Change data were converted to mmol/l if necessary. Sex-specific data were preferred to combined-sex data. Data stratified on other variables were entered in addition to overall data, including stratification by later resumption of smoking, where available from studies with multiple follow-ups. Available information on baseline weight, body mass index (BMI) and other commonly reported physiological parameters, and on change in weight and BMI, was also recorded.

Statistical analysis

For trials giving results for subjects continuously abstinent since baseline, the quit time was taken as the intra-measurement period. For other trials, it was estimated from the study design details. For observational studies, where quitting could have occurred any time in an interval, the quit time was estimated based on the interval midpoint.

Non-stratified data were generally selected for analysis if available, with certain exceptions. Sex-specific data were preferred to combined-sex data. Where a study stratified the participants by time successfully quit, with data for more time points for persistent quitters than for those resuming smoking by the end of the study, the stratified data were used. Stratified data were also used where the levels were relevant to a factor considered in the heterogeneity analyses (see below). For many studies, the only estimates available were direct (not relative to changes in continuing or never smokers) and unadjusted for covariates. Therefore, to make estimates as comparable as possible between studies, direct estimates and those least adjusted for covariates were selected from studies providing a choice of estimates.

If missing for intermediate time points, changes in weight and BMI were estimated assuming linearity. Tertiles of baseline parameters and of change in weight and in BMI were derived from the combined data on quitters, continuing smokers and never smokers. The weight and BMI tertile variables were then merged, preferring the weight tertile when both were available.

For changes in HDL-C, unweighted and inverse-variance weighted means and SEs were estimated using repeated measures analysis of variance, accounting for serial correlations between changes at different times within the same group of subjects by the Kenward-Roger method [9, 10]. For weighted analyses, an SE estimate was required for each change. If not provided, it was calculated from the p-value, 95% confidence interval (CI), or N and SD combined. Where information on variability was unavailable, the SE was estimated using the mean SD for changes where the SD was available.

The analyses of variance were conducted separately for quitters, continuing smokers and never smokers. Heterogeneity was investigated by intra-measurement period, estimated time quit, biochemical validation of quitting, study characteristics (type, continent, timing, constraint on diet/exercise), population details (age, sex), baseline levels of physiological parameters (HDL-C, LDL-C, triglycerides, systolic and diastolic blood pressure), and aspects of smoking before quitting. Change in weight was also included as a factor because it was considered important by several study authors, although interpretation is difficult as, unlike the other factors, this relates to changes concurrent with the HDL-C changes.

Formal within-study analysis of variation in change by intra-measurement period or by time quit for studies providing multiple estimates was not conducted, but such results if given by study authors are summarized.

To further assess effects of confounding, we compare change estimates from the same study that are comparable apart from level of adjustment. For direct estimates in quitters, the only possible factors are those relating to changes following quitting, such as BMI. Estimates relative to continuing or never smokers (or derived from a model involving two groups) may involve other factors.


ROELEE Version 3.1 (P.N.Lee Statistics and Computing Ltd., Sutton, Surrey, UK) was used for data entry and some statistical analyses, and SAS Version 9.2 (SAS Institute Inc., Cary, North Carolina, USA) to conduct the repeated measures analysis of variance.


Literature searches

Table 1 summarizes the literature searches. 661 papers were considered, 27 cited by Maeda et al. [4], 615 identified from the July 2012 Medline search, 13 from the August 2012 Cochrane search, and six from other sources. 508 were rejected by examining abstracts, and 100 by examining full text. This left 53 papers, 47 providing primary results, and 6 reviews. The 47 primary papers described 45 distinct studies, one study (GEPNER) being described in three papers. Ten further papers provided background information.

Table 1 Literature searches a


Table 2 summarizes details of the 45 studies, with further detail given in Additional file 1. Countries contributing the most studies were USA (15) and Japan (7). Eleven studies were randomized clinical trials (RCTs) of various smoking cessation treatments, while MOFFA3 was a RCT with an additional community control group. Twenty were non-randomized smoking cessation studies. These 32 trials were often small, 18 involving less than 100 subjects, and the largest (ALLEN and GEPNER) some 900. They were often short, 14 presenting results for less than 10 weeks, and the longest a year. The remaining studies were observational, three based around community health education studies, and ten on routine health checks or prospective studies. These were larger, most involving some thousands of subjects, and longer, all presenting changes over at least a year, and some for variable time periods. In particular, study BURNET recruited pre-menopausal women who returned for follow-up examination 1 and 2 years after reaching menopause.

Table 2 Study details

Two studies involve persons with CHD risk factors and one diabetic patients. Seven are studies in people of low physical activity, and one in participants described as obese. One study excludes people with low BMI, while four exclude those with high BMI. Otherwise, populations seemed reasonably representative, though some studies were restricted to heavy or more addicted smokers, or excluded subjects with defined medical conditions (Additional file 1). Two studies started in the 1970s, 15 in the 1980s (including RAHILL which started recruitment in 1961 but only measured HDL-C from 1981 onwards), 19 in the 1990s and 9 in the 2000s. Descriptions of methods for blood sampling and HDL-C determination varied considerably, being summarized in Additional file 1, as are requirements for overnight fasting and/or smoking abstinence before measurement, though not all publications provided such details.

Changes in HDL-C following quitting

The available data on changes in HDL-C (mmol/l) following quitting are shown in Additional file 2. Two studies provided no change estimate, merely stating non-significance at p < 0.05. Study YAMAMO analysed HDL-C as a dichotomous variable, and reported that significantly more quitters than continuing smokers increased their HDL-C level from below to above 40 mg/dl. The 94 change estimates for the remaining 42 studies provide the main data set for analysis. They relate to quit times from 1 day to 157 weeks. For 23 estimates, the SE of the change was either given or could be derived, while for 71 it was estimated from N and an SD estimate (0.263) derived from other studies.

The data concern 53 independent sets of subjects, including three from TAMURA, an observational study presenting results stratified by time between quitting and follow-up. Four studies provide sets stratified by sex, three by persistent quitting vs. later resumption, and one each by exercise training vs. control and by weight gain. The remaining 32 studies provide one set each. In addition to TAMURA, thirteen studies provide data on changes in non-independent multiple periods. The most intensive study was MOFFA2 presenting results for 11 periods up to 30 days, while four periods were presented by HAUSTE (up to 26 weeks) and by STUBBE and VANDEN (up to 52 weeks). Some additional data (not included in the analysis) are shown in footnotes.

The data are also shown in Figure 1, the x-axis showing time since quitting on a non-linear scale. The independent points from TAMURA are joined by a dotted line, while points for studies contributing multiple non-independent data are joined by solid lines. For these studies the available information is insufficient to determine whether variations between successive follow-ups are statistically significant, despite about half using repeated measures methods to analyze their data.

Figure 1
figure 1

Change in HDL-C by time since quitting. The data are also shown in Additional file 2 (Table A2-1). Independent estimates for different times from study TAMURA are joined by a dashed line. Non-independent estimates for different times from other studies are joined by solid lines and distinguished by symbol and colour, as shown in the figure legend. Where there is a single estimate for a study/stratum, this is shown as a blue diamond: Week 2 FEHER-B, QUENSE-M, PULS-B. Week 4 FERRAR-F, KONDO-M, MOFFA1-F-R, NORREG-B, PRIEME-B-R. Week 5 RICHAR-B. Week 6 ALLEN-B, MASARE-M, MASARE-F. Week 7 STAMFO-F-R. Week 8 ELIAS2-B. Week 9 BASLER-B, LEE-M, NIAURA-F-E, NIAURA-F-C. Week 10 RABKIN-M, RABKIN-F. Week 12, KORHON-F, TONSTA-B. Week 13 LUDVIK-B. Week 19 ZHANG-M. Week 23 TERRES-B. Week 44 YOON-M-L, YOON-M-H. Week 52 GEPNER-B. Week 65 GREEN-M. Week 78, FORTMA-B, RAHILL-M, YEH-B. Week 119 SHENNA-M. Week 130 KUSHIM-M. Week 156 GERACE-M. Note that each individual independent point and the start of each group of non-independent points effectively joins back to the (0,0) point, but this is not shown in the Figure for clarity. Note also that the scale on the x-axis is non-linear (with distances between 13 and 52 weeks reduced to a quarter, and over 52 weeks to a tenth). Abbreviations: B = both sexes, F = females, M = males; L=low weight gain (<1.3 kg), H=high weight gain (≥1.3 kg); P = persistent quitter, R = resumed smoking before end of study; E = exercise training intervention, C = control group.

As seen in Figure 1, several studies show a rapid increase in the first three weeks or so after quitting (FEHER, MOFFA2-females, STUBBE, SWANK), but little or no increase was seen in that time span from other studies (QUENSE, PULS), the exceptional pattern for MOFFA2-males showing an increase at day 2 subsequently dropping back to only a small increase from baseline. For quit times longer than 3 weeks, most data points show an increase in HDL-C, with only five decreases, of which only the largest from study KUSHIM (change −0.138) was significant (p<0.05). Of the longer-term studies with multiple quit times, some (HAUSTE, NILSSO, SUWAZO) show quite similar increases at the different times studied, one (MOFFA1) suggests an increasing change with time, another (MOFFA3, both sexes) suggests a change only starting after week 5, while others (ELIAS1, PRIEME, VANDEN) suggest a return towards baseline after an initial increase.

Table 3 summarizes the repeated measures analyses. The mean change was 0.107 (95% CI 0.085 to 0.128) unweighted and 0.060 (95% CI 0.044 to 0.075) weighted, the difference due to the absence of larger increases (>0.2) in the longer-term large observational studies (notably GERACE).

Table 3 Estimates of change in total HDL-C (mmol/l) following quitting a

In both unweighted and weighted analysis, there was significant (p<0.05) variation by intra-measurement period. For time quit, the variation was significant only in the weighted analysis, although the pattern of a lower, but still significant, increase in the last interval (52+ weeks) remained evident. When tested in univariate analyses (Additional file 3), several other factors showed significant variation, particularly in the weighted analysis (age, sex, continent, timing, constraint on diet/exercise, baseline HDL-C, change in weight/BMI). However many of these factors are inter-correlated, and correlated with time quit (e.g. observational studies had longer times and less validation of quitting or constraint on diet/exercise). In the further analyses in Table 3, we included time quit in the model (chosen as perhaps more biologically relevant than intra-measurement period). Except where based on few change estimates, increases following quitting were significant at each level of each factor. Inclusion of time quit markedly reduced the number of other factors showing significant variation between levels, with none significant in the unweighted analysis. In the weighted analysis, HDL-C increases were greater (p=0.003) in studies requiring subjects to maintain their pre-quitting diet and exercise habits than in studies encouraging healthier diets or more exercise, or imposing no constraint. Variation by baseline HDL-C was significant (p=0.03), but due to smaller changes in the studies where baseline HDL-C was unavailable, rather than to variation between the tertiles. There was also some evidence (p=0.06), that increases were greater in those with least weight gain following quitting.

Although no study presented data by amount smoked or any other aspect of pre-quitting smoking habits, a few provided data (see footnotes in Additional file 2) on HDL-C change by other factors, such as age and dose of nicotine patch. No significant (p<0.05) variation in change by any such factor was seen.

Several studies provided information on the effect of confounder adjustment (Table 4). In some, comparison is only possible between an unadjusted or age-adjusted estimate and a multiply-adjusted estimate, so inferences cannot be drawn concerning effects of adjustment for specific variables. However in two studies (KUSHIM, SUWAZO) estimates are somewhat higher when adjusted for BMI change, in one (STUBBE), the estimate is lower with adjustment for dietary fat change, while in study PRIEME adjustment for baseline HDL-C had little effect on change relative to continuing smokers. The significance of these differences cannot be assessed.

Table 4 Effects of adjustment on change in HDL-C (mmol/l) following quitting a

Changes in continuing smokers and never smokers

Change data for continuing smokers and never smokers from 20 studies are also shown in Additional file 2. For both groups the data are reasonably consistent with no change, and contrast markedly with the quitter data. In most studies, changes are minor and not significant, although two large studies (KUSHIM, YEH), which also surprisingly reported decreases in quitters, reported large decreases for both never and continuing smokers. The overall estimates from the repeated measures analyses (Additional file 3) were not significant except for the weighted analysis of never smokers, which showed an overall decrease (change = −0.048, p=0.004), due to the contributions of KUSHIM, and particularly YEH.


Despite considerable between-study variation in design, methods and populations, and many studies involving few subjects, the data, with few exceptions, show that HDL-C increases following quitting. From repeated measures analysis, mean increases (mmol/l) were 0.107 (95% CI 0.085 to 0.128) unweighted and 0.060 (95% CI 0.044 to 0.075) inverse-variance weighted.

We present some analyses by the period between measurements taken when still smoking and when quit. However, quitting may not start at baseline, and we also present analyses by estimated time quit. To derive this, studies were divided into three types. For cessation trials with quitters required to be continuously abstinent, intra-measurement period and time quit are identical. For other trials, where quitting may not have started immediately, time quit can be derived from the study design details. For observational studies, where quitting could have started at any time in an interval, we estimate time quit from the interval midpoint. This may be questionable, but should not affect our conclusions.

An increase in HDL-C was evident in subgroups by sex, age, location, timing, study type, baseline HDL-C, baseline weight, increase in weight, or whether quitting was validated biochemically. Our observation that quitting is associated with increased HDL-C, based on within-subject changes, is consistent with considerable evidence of lower HDL-C levels in smokers than quitters [1, 2]. A claim [6] that HDL-C in quitters increases with increasing time quit is not clearly confirmed by our analyses. Within the first year following quitting, the increase shows little overall variation by time quit, with different studies showing different patterns, and although estimates were lower for longer quit times, this difference is open to doubt. Notably, two large observational studies (KUSHIM, YEH) reported a decline in HDL-C following quitting, and an even larger decline in continuing smokers and never smokers, results inconsistent with other studies. Could some systematic difference between baseline and follow-up have biased the estimates of change from these studies, studies which contributed substantially to the lower estimates associated with longer-term quitting, and the lower weighted than unweighted overall estimate.

Previously, Maeda et al. [4], based on far fewer change estimates, found significantly (p<0.05) higher increases in HDL-C for females than males. Though our estimate was also higher for females, we did not find the difference to be significant. Nor could we repeat their finding that increases following quitting were significantly higher in those with higher levels before quitting, though we did find some trend in that direction. We studied some factors not considered in the earlier review [4]. The only significant result we found was that increases were substantially higher in trials where subjects were required to maintain pre-quitting diet and exercise habits (0.191 mmol/l) than in those that encouraged healthier diets or more exercise (0.064 mmol/l), or studies that did not impose any constraint (0.082 mmol/l). Taken together with the non-significant tendency for higher increases where subjects had less weight gain, and higher increases when adjusted for weight gain in two studies, the results suggest a role of diet and weight gain in mitigating the magnitude of the increase. We found no clear evidence that increases in HDL-C following quitting differed by mean age of subjects in the trial, study location, product smoked, validation of quitting, or baseline weight/BMI, all factors not considered earlier [4]. We also found no clear evidence that change estimates varied systematically by level of confounder adjustment.

Compared to the previously published meta-analysis [4] we consider many more studies, and studied effects of quitting by a much longer list of factors. However, the evidence considered must still be regarded as limited in a number of respects. Thus, many studies involved few subjects, no study provided data relating the change following quitting to initial amount smoked, no clinical trial investigated changes over more than a year, and few studies provided evidence on the effect of adjustment for confounders. Also, the variability of changes was often unavailable, and had to be estimated from other studies. Finally, many studies considered provided no evidence on changes in continuing smokers and in never smokers, though the available data are reasonably consistent with a lack of change, and contrast markedly with the results for quitters.

The mechanisms by which smoking might decrease or quitting might increase HDL-C are not fully understood. One popular explanation [6, 69] is that smoking alters catecholamine release, and hence free fatty acid release, which in turn affects VLDL and LDL concentrations to favour their accumulation in blood, contributing to a lower HDL concentration. However other hypotheses have been proposed [4, 70], involving smoking increasing cholesteryl ester transfer protein, reducing lecithin cholesterol acyltransferase activity, affecting apo-1 synthesis, or increasing triglycerides. It has also been suggested that some of the change in HDL-C following quitting is due to associated changes in diet [43, 51, 56, 57], but as mentioned above our analyses indicated that the increase in HDL-C tended to be greater in those with least weight gain or dietary change.


Quitting smoking is clearly associated with an increase in HDL-C concentrations. Generally the increase occurs rapidly, in less than three weeks, with no clear pattern of change thereafter. This emphasises that some, at least, of the adverse effects of smoking appear to be rapidly reversible on quitting, strengthening the argument for encouraging smokers to quit.