Effects of lifestyle interventions and long-term weight loss on lipid outcomes – a systematic review


  • This work was conducted as part of the PROGRESS (PRevent Obesity GRowing Economic Synthesis Study) funded by the National Preventative Research Initiative and the Universities of Aberdeen and Melbourne.

  • The PROGRESS group consists of the following applicants: Lorna Aucott1, Alison Avenell2, Flora Douglas1, Alison Goode3, Kostas Mavromaras3, Mandy Ryan5, Matt Sutton4, Edwin van Teijlingen6 and Luke Vale2,5.

  • 1University of Aberdeen Section of Population Health; 2University of Aberdeen Health Services Research Unit (HSRU); 3University of Melbourne; 4University of Manchester; 5University of Aberdeen Health Economics Research Unit (HERU); 6University of Bournmouth.

  • HERU and HSRU are core funded by the Chief Scientist Office of the Scottish Government Health Directorates. AA was funded by a Career Scientist award from the Scottish Government Health Directorates.

Dr L Aucott, Medical Statistician, Section of Population Health, Polwarth Building, University of Aberdeen, Foresterhill, Aberdeen AB25 2ZD, UK. E-mail: l.aucott@abdn.ac.uk


Weight and lipids are critical components of the metabolic syndrome, diabetes and cardiovascular disease. Past reviews considering weight loss on lipid profiles have been for ≤1 year follow-up and/or were for very overweight, obese or morbidly obese participants.

This systematic review includes lifestyle interventions for adults (18–65 years), with a mean baseline BMI < 35 kg/m2, with weight and lipid differences over 2 years. Between 1990 and 2010, 14 studies were identified.

Mean differences for weight and lipids were modest. However, weight loss at 2–3 years follow-up, produced significant beneficial lipid profile changes. These were similar to previous reviews conducted on heavier target groups and/or over shorter follow-up periods; cholesterol (1.3% decrease per kg lost) and triglycerides (1.6% fall per kg). Weight loss sustained longer than 3 years was not associated with beneficial lipid changes, suggesting that other lifestyle changes not just weight loss needs maintaining.

Evidence linking lifestyle induced sustained weight loss with lipid profile changes in the long-term for this group is limited. Probable within-group differences (treatment vs prevention), would make further group separation prudent. Individual patient data analysis would facilitate this, uncover baseline, medication and confounding effects, and may identify successful program components enabling more effective obesity prevention and treatment strategies.


Obesity is a leading component of the Metabolic Syndrome (MetSyn) which in turn is an important risk factor for type 2 diabetes, cardiovascular morbidity and mortality (1–4). Key components of the MetSyn are elevated waist circumference, triglycerides, blood pressure, fasting glucose and reduced high-density lipoprotein (HDL). A patient with three or more of these components is currently defined as having the MetSyn (5,6). Consequently, weight reduction and blood-lipid regulation play important roles in treatment and prevention of these conditions.

In adults, body weight increases often incur detrimental changes to the blood-lipid profile particularly visceral obesity (7). A leading aim of contemporary public health policy is to reduce levels of obesity (8,9) in an attempt to reduce these other risk factors. Weight loss intervention studies and reviews have reported blood-lipid improvements associated with weight loss, particularly in the short term (10–13). Our previous systematic review on the long-term effects of weight loss on lipid outcomes (14) confirmed a significant positive linear relationship between cholesterol and weight change (r = 0.89), but focused predominantly on very overweight/obese populations [body mass index (BMI) ≥ 28], including the morbidly obese. Evidence for long-term effects of weight loss on serum lipids in more generalizable populations is lacking. The aim of this research was to systematically review evidence linking long-term weight and lifestyle changes to serum lipid changes for those with a BMI of <35 kg m−2.


This review considered longitudinal data rather than ‘between-treatment’ data linking weight differences with lipid measure differences. Consequently, recorded changes in weight and lipids [including cholesterol, HDL, low-density lipoproteins (LDL) and triglycerides], from clinical trials (CTs), including randomized controlled trials, controlled before and after studies (CBAs) and cohort studies (including interrupted time series) published between 1990 and 2008, were considered. Prior to 1990, obesity was not a research priority and study quality not as regulated.

Two systematic literature searches (part of a National Prevention Research Initiative-funded economic evaluation of obesity prevention for UK adults) were conducted. The first identified CTs and CBAs (collectively termed as trials from now on) with lifestyle interventions to prevent weight gain. The second identified exposure to lifestyle programmes within the cohort studies and also those studies without formal programmes where weight loss/prevention was intentional and lifestyle-based.

The searches used MeSH terms and text words for ‘trials’, ‘obesity’, ‘overweight’, ‘weight differences’ appropriately combined. The first search, for the trials was from 1990 onwards, was based on key reports, systematic reviews and primary studies indexed in Medline, Embase, PsycINFO, CINAHL, The Evidence Based Medicine Reviews Collection, CAB Nutrition Abstracts and Reviews, along with hand searching of International Journal of Obesity and Obesity Research. This search was up to October 2007 but updated in April 2008 (full details available from the authors). Studies relevant to longitudinal measures in both weight and health outcomes were considered in this review. These included lifestyle intervention arms (even if other arms were surgical or drug-based) provided the inclusion criteria were met. Thus, for our analysis, all the trial data were treated non-comparativally. Medline, Embase and CINAHL were also searched from 1990 to 2008 for cohort studies with the same search terms. In addition, reference lists from relevant primary and review papers were investigated.

Inclusion criteria were ≥2-year follow-up for studies with lifestyle interventions/programmes for weight loss (dietary, exercise, behavioural or environmental) or intentional weight loss (and weight cycling) along with records of long-term lipid profile change(s) for adult participants (18–65 years at recruitment; after 65 people tend to have ageing-related weight loss) (15,16). Studies were excluded if participants had a mean BMI ≥ 35 kg m−2, had eating disorders, were pregnant, or mentally or physically handicapped. There were no language restrictions and studies with ethnic groups were included provided the setting was UK-compatible. Small studies were excluded (<50 participants per subgroup at recruitment and/or <20 at follow-up, to maintain statistical robustness and acceptable dropout rates).

Foreign language papers were assessed by team members competent in the relevant language, or translated by a third party prior to assessment. All titles/abstracts and full text papers were independently assessed against the inclusion criteria by two reviewers, with disagreements arbitrated by discussion or by a third reviewer.

Statistical analysis

Lipid measure changes related to weight loss in the long term were considered longitudinally for clinical and statistical significance. Ideally, differences between follow-up and baseline were required along with associated precision. However, actual differences were not always available, follow-up and baseline summaries being commonly reported. In these cases, suitable imputed estimates were made for mean differences using the change between follow-up and baseline means.

Precision of the mean differences included standard deviations (SDs) of differences, standard error of mean differences (SE) or associated confidence intervals (CIs). For consistency, all were converted into SDs. For imputed mean differences, SDs were estimated using inline image, where Dif, F and B represent difference, follow-up and baseline, respectively. Theoretically, the difference variance would be inline imagewhere σFB is the follow-up/baseline covariance (17). However, when only baseline and follow-up variances are known, the proposed estimate gives a conservative measure with no covariance assumptions.

In some cases, percentage changes were reported and so converted into mean differences and SDs. Studies providing only BMIs had weights (kg) calculated by substituting average heights according to relevant population references (indicated in the tables) in place of individual heights. All measures were converted to SI units.

Lipid measure changes (and percentage changes) were correlated with weight differences and other variables, namely mean age, follow-up time, gender mix and consistent baseline variables. Meta-regression models were constructed using weighted least squares regression predicting lipid measure differences. Using the SE for each lipid measure mean difference, model weights were defined as 1/SE2. The SEs for the generated regression coefficients require adjustment before determining coefficient significance (18). Initially, all subgroups were included in the meta-regression. However, some studies did not control for medication (most importantly for this review, lipid-lowering medication) or were inconsistent and so subgroup analysis was conducted without these to determine their effect.


Lifestyle interventions/programmes need to be individual and flexible. Included papers indicate that interventions/programmes were intended for the study duration albeit monitored/modified during contact visits. Presented are lifestyle arms of the trials followed by cohort studies with lifestyle programmes.

The searches identified 4977 abstracts, from which 405 full text documents were assessed resulting in 54 potentially suitable papers. Several papers required extra information from authors and were only included once sufficient information was obtained. Of those with extractable data, 16 papers recorded lipid measures: nine related to seven trial studies (19–27) and seven were cohort studies with either lifestyle programmes or intentional weight loss (28–34). Tables 1 and 2 give the basic characteristics and also identify confounding factors of these studies.

Table 1.  Basic characteristics of included trials
Author, year, countryStudy descriptionLifestyle interventionProportion of femalesFollow-up proportion at last visitBaselinePotential confounding factors mentioned
Mean (SD)
  • *

    Baseline values provided for completers only rather than for all who started the programme.

  • BMI, body mass index; PA, physical activity; SCRIP, Stanford Coronary Risk Intervention Project.

Haskell (1994) (17) USASCRIP study investigating effect of intensive multiple risk factor reduction programme in men and women with coronary artery diseaseDiet, physical activity and medications (lipid and hypertension). Contact: risk reduction group baseline, every 2–3 months0.140.3930056 (7.4)27 (4)For each group: occupation %, medical history %, previous heart %, smoking % plus Baseline diet and PA%
Ditschuneit (1999) (18) Germany and Flechtner-Mors (2000) (19) GermanyProspective dietary, two-arm, parallel intervention for 12 weeks followed by a prospective, single-arm, 4-year trial on Slim Fast replacements for meals and snacksTwo subgroups, randomized to one of two dietary treatments for first 3 months, then all participants prescribed same diet using Slim Fast products. Contact: monthly, every 3 months bloods taken0.790.63 and 0.7510044 (10)34 (4)For each group: self-reported energy intake base and follow-up. Queationnaire about usual PA mentioned – no results
Kuller (2001) (20) USATo prevent rise of low-density lipoprotein, cholesterol and weight gain in premenopausal womenDiet (reduction of saturated fats, to reduce cholesterol and calories) and physical exercise. Some on hormone replacement therapy and/or diuretics. Contact: 6,18, 30, 42 and 50 months1.000.4653547 (2)25 (3)PA mean at Baseline for each group
Heshka (2003) (23) USAMulti-centre trial comparing self-help weight loss vs. a structured commercial weight loss programmeCounselling, self-help, food and activity and behaviour modification plan. Those taking glucose, hypertension or lipids medication were excluded from analysis. Contact: clinic 0, 1, 12, 26, 52, 78, 104 weeks. Self-help 2 × 20 min counselling weeks 0 and 12. Commercial group weekly session with diet and PA plan0.850.7342344.5 (10)34 (4)Smoking% in each group
Lindstrom (2003) (24)(25) FinlandFinnish Diabetes Prevention Study clinical trial investigating the effects of an intervention programme designed to prevent or delay the onset of type 2 diabetes in impaired glucose tolerance patientsDiet and exercise (29% and 5% on blood pressure and lipid-lowering medication respectively). Contact: nutritionist 0, 1–2, 5–6, weeks, 3, 4, 6 and 9 months then every 3 months thereafter0.670.4952255 (7)31 (5)Occupation%, education level%, alcohol%, energy intake% at baseline for each group
Mensink (2003) (26) Maastricht, the NetherlandsStudy of Lifestyle Interventions for those with impaired glucose tolerance, aimed to evaluate impact of 2-year combined diet and physical intervention programmeRegular dietary advice and encouraged to increase physical activity. Contact: 0, 4–6 weeks every 3 months thereafter0.500.7711457 (14)26 (7)Energy and diet intake means and standard errors for each group
Niebauer (2007) (27) Germany*Aim to assess long-term effects of physical exercise and low-fat diet on progression of coronary artery diseaseInstruction on lowering fat and cholesterol content of regular diet based on American heart association recommendations and daily exercise on cycle ergometer for 30 min and two group sessions 60 min per week. Contact: intervention group 3 weeks on hospital ward then every 3 months for the first year0.00.80113Mean 53.5 range (35–69)26 (2)Clinical history is alluded to
Table 2.  Basic characteristics of included cohort studies
Author, year, countryStudy descriptionLifestyle programmeProportion of femalesFollow-up percentageBaselinePotential confounding factors
Mean (SD)
  • *

    BMI estimated from average weight and average height of sample. Unknown SD.

  • Refers to the 980 individuals in the final analysis rather than at baseline.

  • Raw data for total sample of 2493 available from author (Sjostrom 1999 reports data from a subsample of 100) (32).

  • Average weight in kg converted to BMI assuming average European person height to be 1.681 m (40). SD for weight unknown.

  • BMI, body mass index; PA, physical activity.

Sedgwick (1990) (28) AustraliaAdelaide 1000 study. Two-year follow-up of 4-year health and fitness programme examining relationship between weight change and lipidsCommunity fitness programme – exercise and diet. Contact: baseline and 2 years.0.430.6698741.7 (10.5)25*Quantified for males and females: energy intake, smoking%, PA% alcohol
Eriksson (1991) (29) SwedenMalmö Prospective Study feasibility of long-term intervention to reduce risk factors for newly diagnosed type 2 diabetes mellitusDietary and lifestyle behavioural advice and increased physical activity. Contact: every 6 months for 2 years then yearly for 5 years0.00.8918148.1 (0.7)27 (3)Although not specified smoking is mentioned
Kaufmann (1992) (30) ChileProspective workplace study. Two-year follow-up of a cardiovascular risk control programmeDiet and lifestyle behavioural advice. Contact: every 2 years0.151.0083645 (8.67)25 (0)smoking is mentioned
Martinez-Gonzalez (1998) (31) SpainProspective workplace study. Three-year follow-up of a multifactorial workplace programme aimed at preventing cardiovascular diseaseDiet, exercise and lifestyle behavioural therapy and advice. Contact: baseline and 3 years0.230.63155541.45 (11.4)27Demographics, nutritional status, occupation and non-occupation PA, smoking habits mentioned with follow-up % for smoking
Sjostrom (1999) (32) Sweden (Raw data was made available)Prospective clinical study designed to test the short and long-term effectiveness of a 4-week residential programme to control obesity and related cardiovascular disease risk factorsDiet and exercise and behavioural therapy. Four-week residential programme. Contact: every day for 4 weeks, 6 months, 1 year, 5 years with in-house reinforcement0.560.33249350.30 (9.42)31 (5)Have individual data for smoking, education, occupation, some diet, PA . . . full dataset
Pawlowski (2003) (33) PolandSouthern Poland Epidemiological Survey. Two-year follow-up programme aimed at reducing hypercholesterolaemia and other coronary artery disease risk factorsLarge primary prevention programme. Diet and medication. Contact: unclear0.721.0025354.626For each group the education status numbers, occupation and smoking%
Welty (2007) (34) USAExamine effect of diet and exercise on weight loss and lipid levelsDiet and exercise and behavioural therapy. Contact: baseline, 3–6 months for diet, 1 year (or more) with physicianNot available0.75–1.008055 (12)30 (6)Mention that cardiovascular risk factors were assessed

Follow-up was between 24 and 72 months with single or mixed gender groups. Longitudinal differences in weight and lipid measures were tested for significance (Tables 3 and 4) using imputed mean differences and SDs where differences were not available. One study (22) provided difference data without precision. This was estimated using an average SD based on the included studies that had differences and associated SDs.

Table 3.  Weight differences with lipids differences – trials
StudySettingSubgroup description (follow-up in months)Follow-up sampleWeight differenceCholesterol differenceHDL differenceLDL differenceTrig difference
nkg (SD)mmol L−1 (SD)mmol L−1 (SD)mmol L−1 (SD)mmol L−1 (SD)
  • a

    Assumes values from original text are SDs.

  • b

    P < 0.05.

  • c

    Mean estimated from (follow-up – baseline) values and SD estimated as √{var[baseline + var(follow-up)]}.

  • d

    n = 245 given for 54-month follow-up only. Used for 30 and 42 months as a conservative estimate.

  • e

    Pounds converted into kg.

  • f

    SD estimated from mean sd value of subgroups with actual differences.

  • g

    Sample size at follow-up ascertained from author may vary by 2 for extreme values.

  • h

    Converted from mg dL−1 to mmol L−1.

  • i

    BMI converted to kg assuming average European person height to be 1.681 m (40).

  • j

    SD converted from SE.

  • k

    Borderline P = 0.053.

  • Estimated values are in bold.

  • BMI, body mass index; HDL, high-density lipoprotein; low-density lipoprotein; Trig, triglycerides.

Haskell (1994) (19)aHospitalRisk reduction (48)118−3.00 (4.00)b−0.99 (0.83)b0.14 (0.23)b−0.95 (0.81)b−0.34 (0.87)b
Ditschuneit (1999) (20)c and Flechtner-Mors (2000) (21)cUniversity hospital clinicGroup A (27)31−7.70 (16.00)b−0.32 (1.12)−0.09 (0.37)0.36 (1.48)
Group B (27)3210.4 (19.16)b−0.48 (1.39)0.08 (0.87)−0.83 (1.33)b
Group A (51)38−4.10 (15.42)−0.43 (1.07)b−0.06 (0.43)−0.69 (1.48)b
Group B (51)37−9.5 (19.09)b−0.46 (1.11)b−0.05 (0.44)−0.94 (1.28)b
Kuller (2001) (22)d,e,fClinicIntervention group (30)245−2.13 (3.81)b0.14 (0.29)b−0.07 (0.80)0.08 (0.79)
Intervention group (42)245−1.0 (3.81)b0.05 (0.29)b−0.01 (0.80)0.13 (0.79)b
Intervention group (54)245−0.08 (3.81)0.06 (0.29)b0.09 (0.80)0.21 (0.79)b
Heshka (2003) (23)g,h,kMulti-centre academicSelf-help (24)108−0.10 (6.24)−0.29 (0.51)b0.02 (0.25)−0.00 (0.46)
Commercial (mainly diet) (24)94−3.00 (5.82)b−0.27 (0.50)b0.01 (0.20)−0.00 (0.44)
Lindstrom (2003) (24)(25)iMulti-centre studyMen and women with IGT (24)256−3.50 (5.50)b−0.10 (0.8)b0.10 (0.18)b−0.20 (0.58)b
Men and women with IGT (36)231−3.50 (5.10)b−0.10 (0.9)0.14 (0.20)b−0.10 (0.60)b
Mensink (2003) (26)jCommunity health intervention programmeIntervention group (24)40−2.40 (4.30)b+0.30 (0.63)b0.06 (0.19)k+0.32 (0.70)b−0.30 (0.76)b
Niebauer (2007) (27)Clinical intervention physical exercise and low-fat dietIntervention group (72)400.00 (5)−0.39 (1.03)b0.14 (0.28)b−0.24 (0.80)−0.33 (0.67)b
Table 4.  Weight differences with lipids differences – cohort studies
StudySettingSubgroup description (follow-up in months)Follow-up sampleWeight differenceCholesterol differenceHDL differenceLDL differenceTrig difference
nkg (SD)mmol L−1 (SD)mmol L−1 (SD)mmol L−1 (SD)mmol L−1 (SD)
  • *

    Follow-up-baseline lipids taken from graph.

  • Significant at P < 0.05.

  • % BMI changes converted to BMI changes and then converted into kg assuming average European male height to be 1.745 m (40).

  • §

    Mean estimated from (follow-up – baseline) values and SD estimated as √(var[baseline+var{follow-up}]).

  • Follow-up sample numbers differ for each health outcome difference.

  • ABMI converted to kg assuming average European person height to be 1.681 m (40).

  • **

    Raw data provided by authors (32).

  • ††

    Converted from mg dL−1 to mmol L−1.

  • ‡‡

    n = 60 for HDL, LDL and Trig follow-up differences.

  • BMI, body mass index; CHD, coronary heart disease; CVD, cardiovascular disease; HDL, high-density lipoproteins; LDL, low-density lipoproteins; Trig, triglycerides.

  • Italics indicate that sample sizes change depending on response variable at follow-up.

  • Estimated values are in bold.

Sedgwick (1990) (28)*CommunityFemale weight gain (24)804.80 (2.60)0.30 (0.11)−0.05 (0.30)0.20 (0.89)0.10 (0.50)
Female weight stable (24)1430.15 (1.20)0.09 (0.90)−0.05 (0.40)0.01 (0.75)0.01 (0.65)
Female weight loss (24)50−4.70 (2.80)−0.32 (0.89)−0.01 (0.25)−0.25 (0.80)−0.15 (0.65)
Male weight gain (24)1034.10 (1.90)0.30 (0.95)−0.10 (0.35)0.19 (0.85)0.30 (0.90)
Male weight stable (24)1870.07 (1.01)0.03 (0.70)0.03 (0.55)0.09 (0.70)−0.05 (0.70)
Male weight loss (24)92−4.08 (2.30)−0.20 (0.90)0.08 (0.35)−0.15 (0.90)−0.20 (0.90)
Eriksson (1991) (29)‡,§Diabetic clinicIGT (72)161−1.86 (4.35)0.05 (0.86)−0.18 (0.94)
Kauffmann (1992) (30)Workplace obesity programmeObesity programme adherence (24)80−2.20r = 0.24; P = 0.01
Martinez-Gonzalez (1998) (31)A,§WorkplaceAdequate intervention (36)4790.14 (8.2)−0.25 (1.12)
Inadequate intervention (36)5011.30 (5.16)0.16 (1.14)
Sjostrom M (1999) (32)**Residential education centreCVD risk women (60)380−0.04 (2.03)(n = 386)−0.02 (1.13)(n = 31) 0.19 (0.23)(n = 376)−0.02 (1.01)
CVD risk men (60)318−0.29 (2.08)(n = 318) 0.26 (1.15)(n = 23) 0.10 (0.26)(n = 309)−0.16 (1.99)
Pawlowski (2003) (33)††Hospital clinicGroup B (24)113−0.60 (5.1)−0.62 (1.32)
Group C (24)89−0.60 (4.7)−1.29 (1.34)
Welty (2007) (34)††,‡‡,§Medical centreParticipants with ≥ cardio risk factor or CHD (31)79−7.80 (66.13)0.12 (0.55)−0.34 (1.79)−1.14 (2.68)

While ethically, medication would not be withheld, only two studies (19,33) had large proportions of participants on lipid-lowering medication. These studies are initially included at the descriptive stage of this review and accounted for in sensitivity analysis for the meta-regression.

Descriptive analysis


Six of the seven intervention trials (Table 1) had diet and physical activity as their main components. The remaining study SlimFast was entirely diet-based with results being reported in two papers, after 27 months (20) and 51 months (21). Most had clinical or academic settings. The intervention delivery and design varied considerably but generally visits were initially frequent then annual after the first year; Table 1 shows the participant contact visits with the intervention organisers. The follow-up periods were 2–6 years.

The weight and lipid measure differences are given in Table 3. The Stanford Coronary Risk Intervention Project (SCRIP) trial, men and women with angiographically defined coronary atherosclerosis reported by Haskel et al. (19), actively put their participants on medication to favourably alter their lipoprotein profile. While the lipid results for this trial show significant lipid measurement improvements, these are likely to be because of the levels of medication rather than weight loss.

For the meal replacement study (20,21), the average weight loss ranged from 4 to 10 kg, all clinically significant, although the smallest maintained weight loss at just 4.1 kg for Group A at 51 months was not statistically significant because of the large estimated SD. The largest weight loss (10.4 kg) was for the ‘2 replacement meal’ arm (Group B) at 27 months with a maintained reduction of 9.5 kg at 51 months. Other significant average weight losses were approximately 3 kg (19,23–25). The largest cholesterol improvement of 0.99 mmol L−1 was on the SCRIP trial (19), but this is confounded with the previously mentioned lipid-lowering medication. Significant cholesterol reductions of between 0.3 and 0.4 mmol L−1 were reported by the meal replacement trial at both 27 (21) (not statistically significant) and 51 (21) months and the German CAD study (27). With respect to HDL, the largest improvements both clinically and statistically (+0.14 mmol L−1) were seen for studies by Kuller et al. (22) at just 30 months, the Finnish Diabetes Prevention Study (DPS) reported by Lindstrom et al. (25) at 36 months and the study reported by Niebauer et al. (27) at 72 months. The SCRIP study (19) also mirrored this improvement although this was probably confounded with medication. Not all studies reported LDL. Those that did (19,22,26,27) were inconclusive with only one, the SCRIP (19) study with its limitations, indicating significant reductions (−0.95 mmol L−1).

The largest and significant average drop in triglycerides was 0.94 mmol L−1 reported at 51 months for the ‘2 meal replacement’ arm of the meal replacement study (21), even larger than the reduction of 0.83 mmol L−1 at 27 months (20). Other significant drops included the ‘1 meal replacement’ arm (Group A) at 51 months (20) of 0.7 mmol L−1 and of those around 0.3 mmol L−1 reported by the ‘1 meal replacement’ arm (Group A) at 27 months (20), the SCRIP study (19), the Netherlands IGT intervention study evaluation (26) and the German CAD study (27). The Netherlands IGT intervention study (26) despite having significant weight and triglyceride reductions had significantly raised levels of both cholesterol and LDL levels.

Cohort studies

All of the seven included cohorts (Table 2) had diet components and most included exercise. Some specified elements including lifestyle behavioural advice (29,30,32) or behavioural therapy programmes (31,34). Programmes were delivered in different settings, ranging from residential clinics (32) to free-living work places (30,31); follow-up periods varied from 2 to 6 years. In addition, the frequency and duration of contact for each intervention differed; the residential intervention (32) was an intense 4-week period with further checks of varying monthly gaps then yearly, while others had an initial delivery with further visits at 3–6 months (29,34). The contact is summarized in Table 2. Some of the cohort subgroups were retrospectively decided depending on the adequacy of the programme implementation and/or of the success of the individual patients' weight loss.

A Polish study reported by Pawlowski (33) was another that did not control for lipid-lowering medication, indeed communication with the authors, and revealed that medication was part of the programme. While cholesterol was seen to decrease for relevant groups in this study, there was no major weight loss, and so lipid improvement was probably medication-induced.

Welty et al. (34) reported the largest average weight loss (7.8 kg after 31 months) of the cohorts along with HDL (+0.12 mmol L−1) and LDL (−0.34 mmol L−1) improvements, although none of these were significant because of large estimated SDs. The mean triglyceride reported here showed the largest decrease of the whole review (1.14 mmol L−1) which despite the large estimated SD was statistically significant. Cholesterol was not reported for this cohort.

Sedgwick (28) and Kauffmann (30) had subgroups with weight loss of 2–5 kg, with all subgroups showing significant benefits for cholesterol. Other cohort studies had inconsistent weight lipid−1 change relationships.

Meta-mean analysis

Figure 1 shows mean differences, with 95% CIs, for weight and lipid measures. The Polish (33) and SCRIP (19) studies both admitted to lipid-lowering medication that probably confounded any weight loss effects on lipid measurement differences. To illustrate the effects of these studies, Fig. 1 includes data with and without these studies.

Figure 1.

Forest plot of mean differences in weight (x0.1), triglycerides, cholesterol, LDL (x.5) and HDL at follow-up of at least 2 years. inline image Cohort study; inline image Clinical Trial; ▵ Meta analysis mean; ▿ Reduced group meta regression; HDL, high-density lipoprotein; LDL, low-density lipoprotein. † Random effects model.

Combining the results from all papers (excluding the SCRIP (19) and Polish (33) studies) required random effects models. These demonstrate average significant differences of −1.20 kg (95% CI, −1.90, −0.50) for weight, −0.10 mmol L−1 for triglycerides (95% CI, −0.17, −0.03) and 0.05 mmol L−1 for HDL (95% CI, 0.01, 0.08). However, the differences of −0.08 mmol L−1 for cholesterol (95% CI, −0.20, 0.05) and −0.07 mmol L−1 for LDL (95% CI, −0.23, 0.08) were not significant. Formal associations are further examined by meta-regression.

Meta-regression analysis

In total, 28 subgroups had information on mean weight and lipid measure changes. Correlations of mean weight differences and per cent weight changes between mean lipid measure differences were similar. Hence, for easier interpretation, only mean weight differences were used to develop regression models. While the focus of this analysis is to investigate the effect of weight loss on lipid measures, demographic and baseline lipid measures have been considered in the models. It was not possible to adequately and consistently adjust for other factors such as physical activity as their reporting was limited (Tables 1 and 2). A sensitivity analysis was conducted to assess the impact of the Spanish workplace study (31), given the suggestion of gender confounding by the authors (35). This indicated little difference and has not been further reported. Similarly, a sensitivity analysis was performed to investigate the impact of the two studies (19,33) that specifically allowed lipid-lowering medication. This confirmed that these studies had contrary results and should be excluded from the following meta-analyses, reducing the total number of subgroups to 25. The reported models reflect all remaining data and then are split into follow-up times of 2–3 years and 3 + years (Table 5).

Table 5.  Meta-regression to predict lipid measure differences with weight differences using weighted least squares
Cholesterol (baseline 5.78 mmol L−1*)LDL (baseline 3.21 mmol L−1*)
ModelAdj R2gVariablesBetaAdj tModelAdj R2gVariablesBetaAdj t
C1 all FU0.73121Constant−0.015−1.212L1 all FU0.41612Constant0.0452.126
        Weight difference (kg)0.0384.142
  Weight difference (kg)0.05818.749§L2 all FU0.69611Constant0.0351.680
C2 ≤ 36 months0.76515Constant0.0010.074   Weight difference (kg)0.0434.588
  Weight difference (kg)0.06018.238§L3 ≤ 36 months0.4909Constant0.0491.872
        Weight difference (kg)0.0384.133
C3 > 36 months0.1236n/a for weight  L4 ≤ 36 months0.8928Constant0.0311.175
        Weight difference (kg)0.0434.348
     L5 > 36 months<03n/a for weight  
HDL (baseline 1.3 mmol L−1*)Triglycerides (baseline 1.94 mmol L−1*)
ModelAdj R2gVariablesBetaAdj tModelAdj R2gVariablesBetaAdj t
  • *

    Baseline lipid levels for this population were estimated from the median baseline values of the included studies adjusted for the number of participants.

  • Model using weighted least squares, weighted in each case by 1 (mean systolic difference variance), significance adjusted for the meta-regression (18).

  • P < 0.01.

  • §

    P < 0.001.

  • Excludes Mensink et al.(26) (see Fig. 2c).

  • Chol, cholesterol difference; FU: follow-up; g, number of subgroups; HDL, high-density lipoprotein difference; LDL, low-density lipoprotein difference; n/a, adj R2 relates to non-significant relationship with weight difference – also not significant with other baseline variables – unless shown; Trig, triglyceride difference.

  • Italics represent a variable name.

H1 all FU0.18622Constant−0.047−6.627§T1 all FU0.38623Constant0.0221.477
  Weight difference (kg)−0.014−5.593§   Weight difference (kg)0.0438.137§
H2 all FU0.49422Constant−0.369−7.263§T2 all FU0.56623Constant0.2635.615§
  Age (years)0.0098.502§   Weight difference (kg)0.0325.602§
H3 ≤ 36 months0.35115Constant0.0323.713   Base Trig (mmol L−1)−0.170−5.315§
  Weight difference (kg)−0.019−7.716§T3 ≤ 36 months0.49515Constant0.0060.342
H4 ≤ 36 months0.50715Constant−0.363−6.821§   Weight difference (kg)0.0376.329§
  Age (years)0.0099.555§T4 > 36 months0.3128n/a for weight  
H5 > 36 months0.2377n/a for weight  T5 > 36 months0.6828Constant1.7757.128§
        Base weight mean (kg)−0.024−7.132§

Mean weight loss has a strong relationship with mean cholesterol reduction. As a result of lifestyle modifications, mean weight change accounts for 73% of the overall variation in the mean cholesterol change (Model C1, Table 5). This result is dominated by the 2- to 3-year studies (Model C2, Table 5) with the longer-term subgroups (Model C3, Table 5) having no linear relationship with any of the variables considered.

While there is a significant relationship between mean HDL changes and mean weight changes after lifestyle interventions, mean weight change only accounts for 19% and 35% of the variation depending on whether all the subgroups are considered or just the 2- to 3-year studies (Models H1 and H3, Table 5). Both show the benefit of increasing HDL levels with weight loss, as indicated by the negative coefficient. The best fit model for HDL is determined by the average baseline age of participants (45–53 years for this review) accounting for around 50% of the variation (Models H2, Table 5). Model H4, Table 5, suggests that HDL levels after lifestyle changes would benefit for as long as 2–3 years, with estimated increases of 0.04 mmol L−1 to 0.11 mmol L−1 over the age range, with the older participants benefiting the most. The longer-term follow-ups were again non-significant with respect to HDL benefits for any considered variables (Model H5, Table 5).

While the relationship between mean LDL changes and sustained weight loss as a result of lifestyle changes is reasonable (Table 5: Models L1 adj R2 = 42%; and L3 adj R2 = 49%), one study (26) stands out by having an average increase in LDL despite reasonable average weight losses (Fig. 2c). The values have been verified by the authors Mensink et al. However, if removed (sensitivity analysis), the fit improves to 70% (Model L2, Table 5) which again increases to 89% (Models L4, Table 5) in the first 2–3 years of follow-up. The relationship for even longer-term maintained weight loss is unpredictable (Model L5, Table 5).

Figure 2.

Raw scatter plots of mean weight changes (kg) with lipid measure changes (mmol/L) identified by study names (see tables 1 & 2). HDL, high-density lipoprotein; LDL, low-density lipoprotein.

While a model based only on mean weight differences (Model T1, Table 5) accounts for 39% of the overall variation, the best predictive model for mean triglyceride difference includes the mean baseline triglyceride (Model T2, Table 5). This model accounts for 57% of the variation in mean triglyceride changes and suggests that a patient with a baseline triglyceride level of 1.94 mmol L−1 (the middle of the range classed as borderline high) (5) may reduce this level by 0.10 mmol L−1 after a weight loss of 1 kg with lifestyle changes; this is a reduction of more than 5%. The best fit for the follow-up studies of 2–3 years (Model T3, Table 5) is a less complex model using only weight differences. The longer-term follow-up studies (Model T4, Table 5) were not significant but there was a strong relationship with baseline weight (Model T5, Table 5) whereby an initially heavier participant may expect greater benefits, in terms of triglycerides, after lifestyle changes regardless of maintained long-term weight loss.

In this review, baseline lipid levels (taken as the median of the baseline values from all the included studies adjusted for sample size, Table 5) show that overall, participants were borderline high or at least above optimal levels as categorized by the National Cholesterol Education Program ATP III guidelines (5). From these, a general percentage change in each lipid measure for such a population was derived using the estimated changes from the above meta-regression models. Consequently, a 1 kg maintained weight loss in the long term (2–3 years) could be expected to result in reductions of 1.3% in cholesterol, 1.6% for triglycerides and 0.34% for LDL (0.4% if the study by Mensink et al. (26) is excluded) and a 4% increase of HDL.


This review investigates normal, overweight and obese individuals but not the morbidly obese. It indicates significant benefits to blood serum lipids associated with sustained weight loss for as long as 2–3 years as a result of modified lifestyle. While 5–10% (approximately 5–10 kg) is the target weight loss recommended for risk reduction for those in the BMI 25–35 kg m−2 category (36), maintaining this is often difficult. From this review, the meta-mean analysis of sustained weight loss during follow-up was 1.2 kg ranging from a loss of 10 kg (20) to a gain of 4.8 kg (28).

The associations found here for a maintained weight loss of 1 kg are similar to those suggested elsewhere. Dattilo and Kris-Etherton (10) reviewed the effects of weight reduction through dietary management, using correlation meta-analysis and suggested that cholesterol and triglycerides decreased by 1% and 1.9%, respectively (for studies mostly less than 1 year), compared with this review's-values of 1.3% and 1.6% for more than 2-year follow-up. The expected decrease in LDL by Dattilo and Kris-Etherton (10) of 0.68% with every 1 kg weight loss seems ambitious as the current meta-regression indicates 0.34%. A review by Anderson and Konz (11) predicted an increase of 2% in HDL from a 1 kg weight loss whereas the current review far exceeds this at 4%. Our previous review on the long-term effect (at least 2 years) of weight loss for very overweight/obese and morbidly obese participants (14) only considered the effect of cholesterol in detail as the primary diagnostic measure for general practitioners. This suggested that for 10 kg weight loss a 5% drop in cholesterol may be expected which is directly comparable with the current model results, if 10 kg were sustainable. The benefits of cholesterol reductions in conjunction with improved HDL levels, while age-dependent, have been seen to reduce the risk of ischemic heart disease mortality (6) as well as to reduce the impact of the MetSyn (7).

The current results show that in the longer term (3 or more years) the relationship between maintained weight loss and lipid levels is less determined perhaps suggesting that more than weight loss is required in the longer term. This split of follow-up time has not previously been investigated although Volek and Sharman (37) have suggested that blood-lipid response is not always consistent.

One study, Kauffmann et al. (30), although included descriptively, was not combined analytically because of lack of detailed results. Nonetheless, this study suggested that by 2 years of follow-up weight loss was sustained and associated with improved cholesterol but only for a group that adhered to the programme.

The last search for this review was conducted in April 2008. It was extended to March 2010 in Medline only. Two potentially relevant papers were identified, one based in New Zealand (38) and the other in Boston, USA (39). Both were 2 × 2 factorial designs comparing weight loss for different diet compositions. The New Zealand study also compared a relatively expensive dietician/exercise specialist-led programme (like the Finnish DPS) with a simple nurse-coordinated programme. Both of these new papers reported very similar results to the Finnish DPS (24,25) and as such do not change the conclusions of this current review.

It was not possible to adequately account for confounders such as medication, in this review; only those subgroups that appeared to exclude participants on lipid-lowering medication were included in the meta-regression of this review. It was possible to investigate some confounders like follow-up time and baseline weight category, although only at study level rather than for individuals. Other important confounders, particularly physical activity, smoking and alcohol, were not consistently reported let alone quantified even at study level. While these confounders (especially physical activity) may contribute to lipid changes, this review was unable to compare and validate such effects.

There is concern about the LDL result seen from the Netherlands study (26) which otherwise reflects the primary objective of this review. Despite modest but representative average weight losses, the direction of change in LDL was contrary to other included studies. Consequently, analysis with and without this study was conducted. This resulted in little difference in the meta-regression model coefficients, although the overall fit of the model was greatly reduced by its inclusion.

This review suggests that lifestyle change interventions/programmes in this target group, of normal to obese participants, are effective in improving blood-lipid serum levels even with modest weight losses and are beneficially associated with sustained weight loss seen here to range up to as much as 10 kg. However, while the effect is seen for long-term sustained weight loss (2–3 years), it is not consistent over even longer periods despite maintained weight loss. This may suggest that other factors should be considered in the much longer term in order to maintain the benefits of lifestyle interventions/programme on lipid profiles.

The target group considered in this review is important for not only obesity treatment but also obesity prevention. A better understanding of treatment and prevention may be possible using individual patient data analysis particularly on studies whose primary objective is weight reduction. This would also allow for covariate and confounding factor adjustments, more flexible subgroup analysis, assessment of variation between individuals [modest group average weight losses (with benefits) may include individuals with little or no weight loss and others with more impressive losses], and finally IPD would facilitate comparisons between interventions/programmes and hence aid evaluation.


This review was funded by the National Prevention Research Initiative. We appreciate the help of Linda McIntyre for initial searches, María Angélica de la Torre and Rodolfo Hernández (HERU, University of Aberdeen) for translation and Dr Amudha Poobalan for support with the cohort search strategy. The initial search for the CT papers was created by reviewers from the University of Teesside run by Carolyn Summerbell (now at the School of Medicine and Health, Durham University): Laurel D. Edmunds, Tamara Brown, Helen Moore, Vicki Whittaker, Leah Avery under co-supervision by Alison Avenell, Health Services Research Unit, University of Aberdeen.

Conflict of Interest Statement

No conflict of interest was declared.