NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

Harris J, Felix L, Miners A, et al. Adaptive E-Learning to Improve Dietary Behaviour: A Systematic Review and Cost-Effectiveness Analysis. Southampton (UK): NIHR Journals Library; 2011 Oct. (Health Technology Assessment, No. 15.37.)

Cover of Adaptive E-Learning to Improve Dietary Behaviour: A Systematic Review and Cost-Effectiveness Analysis

Adaptive E-Learning to Improve Dietary Behaviour: A Systematic Review and Cost-Effectiveness Analysis.

Show details

Appendix 7Additional analysis of effectiveness

This appendix includes additional analyses of outcomes, including assessment of evidence for publication bias and evidence for subgroup effects.

Table 29 presents an assessment of statistical heterogeneity in the results for outcomes reported in the included studies. The chi-squared test p-value quantifies the evidence against the null hypothesis of homogeneity (similarity) of effects estimated by the trials. I-squared (derived from the chi-squared statistic and degrees of freedom) quantifies the percentage of total variation in effects attributable to heterogeneity. The assessment of heterogeneity is shown in separate columns according to whether or not outcome measures were reported as means or as a mean change. Heterogeneity is not quantified for some outcomes for which the study reports did not provide SDs (and it was not possible to obtain these from authors), and for which an outcome was reported by a single study only.

TABLE 29. Heterogeneity in effects on outcomes.

TABLE 29

Heterogeneity in effects on outcomes.

There was substantial heterogeneity (i.e. I2 > 50%) in the estimates of effect of e-learning interventions on many reported outcomes. There was relatively little evidence for heterogeneity in the estimates of effect of e-learning on total fat intake per day (p = 0.18 and I2 = 28%), energy intake per day (p = 0.33 and I2 = 13%) and BMI (p = 0.92 and I2 = 0%).

Fruit and vegetables

Figures 1518 show analyses of estimates of effect on fruit and vegetables, fruit only and vegetables only.

FIGURE 15. Forest plot showing the effect of e-learning on the mean intake of fruit and vegetables (servings per day).

FIGURE 15

Forest plot showing the effect of e-learning on the mean intake of fruit and vegetables (servings per day).

FIGURE 16. Funnel plot showing estimates of the effect of e-learning on the mean intake of fruit and vegetables (servings per day).

FIGURE 16

Funnel plot showing estimates of the effect of e-learning on the mean intake of fruit and vegetables (servings per day).

FIGURE 17. Forest plot showing the effect of e-learning on the mean intake of fruit (servings per day).

FIGURE 17

Forest plot showing the effect of e-learning on the mean intake of fruit (servings per day).

FIGURE 18. Forest plot showing the effect of e-learning on the mean intake of vegetables (servings per day).

FIGURE 18

Forest plot showing the effect of e-learning on the mean intake of vegetables (servings per day).

Egger's test for small-study effects:

Number of studies = 12, root MSE = 1.781

Std_EffCoef.SEtp > |t|(95% CI)
Slope −.2111621 .0847431 −2.49 0.032 −.3999816 −.0223426
Bias 2.616964 .7948991 3.29 0.008 .8458188 4.38811

HO, null hypothesis; MSE, mean squared error.

Test of H0: no small-study effects p = 0.008.

Figure 16 shows the estimates of effect of e-learning on mean intake of fruit and vegetables (servings per day) plotted against their standard errors. Smaller studies are associated with a larger treatment effect (i.e. larger increase in servings of fruit and vegetables per day, p = 0.008).

Fats

Figures 1926 show analyses of estimates of effect on total fat, saturated fat and percentage of energy from fat.

FIGURE 19. Forest plot showing the effect of e-learning on the mean intake of fat (g/day).

FIGURE 19

Forest plot showing the effect of e-learning on the mean intake of fat (g/day).

FIGURE 20. Funnel plot showing estimates of the effect of e-learning on the mean intake of fat (g/day).

FIGURE 20

Funnel plot showing estimates of the effect of e-learning on the mean intake of fat (g/day).

FIGURE 21. Forest plot showing the effect of e-learning on the mean intake of saturated fat (g/day).

FIGURE 21

Forest plot showing the effect of e-learning on the mean intake of saturated fat (g/day).

FIGURE 22. Funnel plot showing estimates of the effect of e-learning on the mean intake of saturated fat (g/day).

FIGURE 22

Funnel plot showing estimates of the effect of e-learning on the mean intake of saturated fat (g/day).

FIGURE 23. Forest plot showing effect of e-learning on mean percentage of energy from fat (%).

FIGURE 23

Forest plot showing effect of e-learning on mean percentage of energy from fat (%).

FIGURE 24. Funnel plot showing estimates of the effect of e-learning on the mean percentage of energy from fat.

FIGURE 24

Funnel plot showing estimates of the effect of e-learning on the mean percentage of energy from fat.

FIGURE 25. Forest plot showing the effect of e-learning on the mean change in percentage of energy from fat (%).

FIGURE 25

Forest plot showing the effect of e-learning on the mean change in percentage of energy from fat (%).

FIGURE 26. Forest plot showing the effect of e-learning on the mean percentage of energy from saturated fat (%).

FIGURE 26

Forest plot showing the effect of e-learning on the mean percentage of energy from saturated fat (%).

Egger's test for small-study effects:

Regress standard normal deviate of intervention

Effect estimate against its SE

Number of studies = 12, root MSE = 1.202

Std_EffCoef.SEtp > |t|(95% CI)
Slope −1.485873 .6410574 −2.32 0.043 −2.914238 −.0575083
Bias .3176107 .4487315 0.71 0.495 −.6822255 1.317447

HO, null hypothesis; MSE, mean squared error.

Test of H0: no small-study effects p = 0.495.

Figure 20 shows the estimates of effect of e-learning on mean intake of fat (g/day) plotted against their standard errors. There was no evidence that smaller studies are associated with a larger or smaller treatment effect (p = 0.495).

Egger's test for small-study effects:

Regress standard normal deviate of intervention

Effect estimate against its SE

Number of studies = 5, root MSE = 2.449

Std_EffCoef.SEtp > |t|(95% CI)
Slope .1064684 .7148023 0.15 0.891 −2.168352 2.381288
Bias −.3716999 1.744761 −0.21 0.845 −5.924307 5.180907

HO, null hypothesis; MSE, mean squared error.

Test of H0: no small-study effects p = 0.845.

Figure 22 shows the estimates of effect of e-learning on the mean intake of saturated fat (g/day) plotted against their standard errors. There was no evidence that smaller studies are associated with a larger or smaller treatment effect (p = 0.84).

Egger's test for small-study effects:

Regress standard normal deviate of intervention

Effect estimate against its SE

Number of studies = 10, root MSE = 1.994

Std_EffCoef.SEtp > |t|(95% CI)
Slope −2.535416 .4271861 −5.94 0.000 −3.520509 −1.550323
Bias 1.207866 .8719168 1.39 0.203 −.8027776 3.21851

HO, null hypothesis; MSE, mean squared error.

Test of H0: no small-study effects p = 0.203.

Figure 24 shows the estimates of effect of e-learning on mean percentage of energy from fat plotted against their standard errors. There was no evidence that smaller studies are associated with a larger or smaller treatment effects (p = 0.203).

Fibre, proteins, sugars

Figures 2729 show analyses of the estimates of effect on dietary fibre, percentage of energy from protein, and percentage of energy from carbohydrate.

FIGURE 27. Forest plot showing the effect of e-learning on the mean intake of dietary fibre (g/day).

FIGURE 27

Forest plot showing the effect of e-learning on the mean intake of dietary fibre (g/day).

FIGURE 28. Forest plot showing the effect of e-learning on the mean percentage of energy from protein (%).

FIGURE 28

Forest plot showing the effect of e-learning on the mean percentage of energy from protein (%).

FIGURE 29. Forest plot showing the effect of e-learning on the mean percentage of energy from carbohydrate (%).

FIGURE 29

Forest plot showing the effect of e-learning on the mean percentage of energy from carbohydrate (%).

Energy

Figures 30 and 31 show analyses of estimates of effect on energy intake.

FIGURE 30. Forest plot showing the effect of e-learning on the mean energy intake (kcal/day).

FIGURE 30

Forest plot showing the effect of e-learning on the mean energy intake (kcal/day).

FIGURE 31. Forest plot showing the effect of e-learning on the mean change in energy intake (kcal/day).

FIGURE 31

Forest plot showing the effect of e-learning on the mean change in energy intake (kcal/day).

Body mass index/weight

Figures 3236 show analyses of estimates of effect on BMI and weight.

FIGURE 32. Forest plot showing the effect of e-learning on the mean BMI (kg/m2).

FIGURE 32

Forest plot showing the effect of e-learning on the mean BMI (kg/m2).

FIGURE 33. Funnel plot showing estimates of the effect of e-learning on the mean BMI (kg/m2).

FIGURE 33

Funnel plot showing estimates of the effect of e-learning on the mean BMI (kg/m2).

FIGURE 34. Forest plot showing the effect of e-learning on the mean change in BMI (kg/m2).

FIGURE 34

Forest plot showing the effect of e-learning on the mean change in BMI (kg/m2).

FIGURE 35. Forest plot showing the effect of e-learning on the mean weight (kg).

FIGURE 35

Forest plot showing the effect of e-learning on the mean weight (kg).

FIGURE 36. Forest plot showing the effect of e-learning on the mean change in weight (kg).

FIGURE 36

Forest plot showing the effect of e-learning on the mean change in weight (kg).

Egger's test for small-study effects:

Regress standard normal deviate of intervention

Effect estimate against its SE

Number of studies = 9, root MSE = .6681

Std_EffCoef.SEtp > |t|(95% CI)
Slope .1203499 .5400326 0.22 0.830 −1.156624 1.397324
Bias −.2923767 .6274486 −0.47 0.655 −1.776057 1.191303

HO, null hypothesis; MSE, mean squared error.

Test of H0: no small-study effects p = 0.655

Figure 33 shows estimates of effect of e-learning on mean BMI plotted against their standard errors. There was no evidence that smaller studies are associated with larger treatment effects (i.e. larger decrease in BMI, p = 0.66).

Blood pressure

Figures 37 and 38 show analyses of the estimates of effect on systolic and diastolic blood pressure.

FIGURE 37. Forest plot showing the effect of e-learning on the mean change in systolic blood pressure (mmHg).

FIGURE 37

Forest plot showing the effect of e-learning on the mean change in systolic blood pressure (mmHg).

FIGURE 38. Forest plot showing the effect of e-learning on the mean change in diastolic blood pressure (mmHg).

FIGURE 38

Forest plot showing the effect of e-learning on the mean change in diastolic blood pressure (mmHg).

Lipids and lipoproteins

Figures 3942 show analyses of estimates of effect on low-density lipids, high-density lipids, total cholesterol and triglycerides.

FIGURE 39. Forest plot showing the effect of e-learning on the mean change in low-density lipids (mmol/l).

FIGURE 39

Forest plot showing the effect of e-learning on the mean change in low-density lipids (mmol/l).

FIGURE 40. Forest plot showing the effect of e-learning on the mean change in high-density lipids (mmol/l).

FIGURE 40

Forest plot showing the effect of e-learning on the mean change in high-density lipids (mmol/l).

FIGURE 41. Forest plot showing the effect of e-learning on the mean change in total cholesterol (mmol/l).

FIGURE 41

Forest plot showing the effect of e-learning on the mean change in total cholesterol (mmol/l).

FIGURE 42. Forest plot showing the effect of e-learning on the mean change in triglycerides (mmol/l).

FIGURE 42

Forest plot showing the effect of e-learning on the mean change in triglycerides (mmol/l).

Subgroup analysis for selected outcomes

Our published study protocol stated that in order to investigate possible sources of heterogeneity we would conduct sensitivity analysis by study quality and by SES of participants. Subgroup analysis has been conducted where outcomes were reported by five or more studies. Subgroups were also investigated for our secondary outcomes of BMI and weight, according to whether or not studies aimed to maintain or reduce BMI (or weight), and whether or not studies included a physical activity component.

The Stata output is shown for each subgroup and an interpretation provided.

1. Study quality (EPHPP assessment)

Fruit and vegetables

Meta-regression number of studies = 12

Fit of model without heterogeneity (τ2 = 0): Q (10 d.f.) = 61.4446

p > Q = 0.000

Proportion of variation due to heterogeneity I2 = 0.837

Moment-based estimate of between-study variance: τ2 = 0.0936

WMDCoef.SEtp > |t|(95% CI)
_Iephpp_2 −.0587248 .2545131 −0.23 0.822 −.6258154 .5083657
_cons .2898408 .2220584 1.31 0.221 −.2049361 .7846178

There was no evidence to suggest that estimates of effect on servings of fruit and vegetables were associated with EPHPP global rating.

Total fat

. xi: metareg wmd i.ephpp, wsse(se) mm

i.ephpp _Iephpp_1–2 (_Iephpp_1 for ephpp = = MODERATE omitted)

Meta-regression number of studies = 12

Fit of model without heterogeneity (τ2 = 0): Q (10 d.f.) = 10.9395

p > Q = 0.362

Proportion of variation due to heterogeneity I2 = 0.086

Moment-based estimate of between-study variance: τ2 = 0.4840

WMDCoef.SEtp > |t|(95% CI)
_Iephpp_2 −4.414625 2.273575 −1.94 0.081 −9.480465 .6512156
_cons 3.059523 2.172567 1.41 0.189 −1.781259 7.900304

There was weak evidence to suggest that studies with ‘weak’ EPHPP global rating on average found larger reductions on fat intake (p = 0.081).

Percentage energy from fat

. xi: metareg wmd i.ephpp, wsse(se) mm

i.ephpp _Iephpp_1–2 (_Iephpp_1 for ephpp = = MODERATE omitted)

Meta-regression number of studies = 10

Fit of model without heterogeneity (τ2 = 0): Q (8 d.f.) = 28.5597

p > Q = 0.000

Proportion of variation due to heterogeneity I2 = 0.720

Moment-based estimate of between-study variance: τ2 = 1.6222

WMDCoef.SEtp > |t|(95% CI)
_Iephpp_2 −1.075684 1.170951 −0.92 0.385 −3.775902 1.624534
_cons −.6570533 .9659334 −0.68 0.516 −2.8845 1.570393

There was no evidence to suggest that estimates of effect on percentage of energy fat were associated with the EPHPP global rating.

Energy

. xi: metareg wmd i.ephpp, wsse(se) mm

i.ephpp _Iephpp_1–2 (_Iephpp_1 for ephpp = = MODERATE omitted)

Meta-regression number of studies = 5

Fit of model without heterogeneity (τ2 = 0): Q (3 d.f.) = 3.51876

p > Q = 0.318

Proportion of variation due to heterogeneity I2 = 0.147

Moment-based estimate of between-study variance: τ2 = 1.9e+03

WMDCoef.SEtp > |t|(95% CI)
_Iephpp_2 |
−91.91838
94.49451 −0.97 0.402 −392.6421 208.8053
_cons 53.8557 69.5182 0.77 0.495 −167.3822 275.0936

There was no evidence to suggest that estimates of effect on total energy intake were associated with EPHPP global rating.

Body mass index

. xi: metareg wmd i.ephpp, wsse(se) mm

i.ephpp _Iephpp_1–2 (_Iephpp_1 for ephpp = = MODERATE omitted)

Meta-regression number of studies = 9

Fit of model without heterogeneity (τ2 = 0): Q (7 d.f.) = 2.83317

p > Q = 0.900

Proportion of variation due to heterogeneity I2 = 0.000

Moment-based estimate of between-study variance: τ2 = 0.0000

WMDCoef.SEtp > |t|(95% CI)
_Iephpp_2 −.4149028 .6658704 −0.62 0.553 −1.989436 1.15963
_cons .1978034 .5780817 0.34 0.742 −1.169143 1.564749

There was no evidence to suggest that estimates of effect on BMI were associated with the EPHPP global rating.

Weight

. xi: metareg wmd i.ephpp, wsse(se) mm

i.ephpp _Iephpp_1–2 (_Iephpp_1 for ephpp = = MODERATE omitted)

Meta-regression number of studies = 4

Fit of model without heterogeneity (τ2 = 0): Q (2 d.f.) = 6.48674

p > Q = 0.039

Proportion of variation due to heterogeneity I2 = 0.692

Moment-based estimate of between-study variance: τ2 = 9.4399

WMDCoef.SEtp > |t|(95% CI)
_Iephpp_2 −3.087439 3.944292 −0.78 0.516 −20.05836 13.88348
_cons 2.48286 3.149766 0.79 0.513 −11.06949 16.03521

There was no evidence to suggest that estimates of effect on weight were associated with the EPHPP global rating.

2. Low attrition (0–20%) versus high attrition (> 20%)

Fruit and vegetables

. xi: metareg wmd i.CompleteEPHPP, wsse(se) mm

i.CompleteEPHPP _ICompleteE_1–2 (_ICompleteE_1 for Com∼P = = 60–79% omitted)

Meta-regression number of studies = 11

Fit of model without heterogeneity (τ2 = 0): Q (9 d.f.) = 25.0174

p > Q = 0.003

Proportion of variation due to heterogeneity I2 = 0.640

Moment-based estimate of between-study variance: τ2 = 0.0577

WMDCoef.SEtp > |t|(95% CI)
_IComplete∼2 .1077183 .2054918 0.52 0.613 −.3571365 .572573
_cons .2602589 .1213245 2.15 0.061 −.0141961 .5347139

There was no evidence to suggest that estimates of effect on daily servings of fruit and vegetables were associated with the level of attrition in studies.

Total fat

. xi: metareg wmd i.CompleteEPHPP, wsse(se) mm

i.CompleteEPHPP _ICompleteE_1–2 (_ICompleteE_1 for Com∼P = = 60–79% omitted)

Meta-regression number of studies = 11

Fit of model without heterogeneity (τ2 = 0): Q (9 d.f.) = 13.4312

p > Q = 0.144

Proportion of variation due to heterogeneity I2 = 0.330

Moment-based estimate of between-study variance: τ2 = 5.4695

WMDCoef.SEtp > |t|(95% CI)
_IComplete∼2 2.718195 3.868409 0.70 0.500 −6.032753 11.46914
_cons −.8227567 1.432562 −0.57 0.580 −4.063436 2.417923

There was no evidence to suggest that estimates of effect on total fat intake were associated with the level of attrition in studies.

Percentage energy from fat

. xi: metareg wmd i.CompleteEPHPP, wsse(se) mm

i.CompleteEPHPP _ICompleteE_1–2 (_ICompleteE_1 for Com∼P = = 60–79% omitted)

Meta-regression number of studies = 10

Fit of model without heterogeneity (τ2 = 0): Q (8 d.f.) = 38.8865

p > Q = 0.000

Proportion of variation due to heterogeneity I2 = 0.794

Moment-based estimate of between-study variance: τ2 = 2.3678

WMDCoef.SEtp > |t|(95% CI)
_IComplete∼2 .8740884 1.206316 0.72 0.489 −1.907681 3.655858
_cons −1.890358 .8921099 −2.12 0.067 −3.947567 .1668514

There was no evidence to suggest that estimates of effect on percentage energy from fat were associated with the level of attrition in studies.

Energy

. xi: metareg wmd i.CompleteEPHPP, wsse(se) mm

i.CompleteEPHPP _ICompleteE_1–2 (_ICompleteE_1 for Com∼P = = 60–79% omitted)

Meta-regression number of studies = 5

Fit of model without heterogeneity (τ2 = 0): Q (3 d.f.) = 3.51876

p > Q = 0.318

Proportion of variation due to heterogeneity I2 = 0.147

Moment-based estimate of between-study variance: τ2 = 1.9e+03

WMDCoef.SEtp > |t|(95% CI)
_IComplete∼2 91.91838 94.49451 0.97 0.402 −208.8053 392.6421
_cons −38.06268 64.00337 −0.59 0.594 −241.75 165.6246

There was no evidence to suggest that estimates of effect on energy intake were associated with the level of attrition in studies.

Body mass index

. xi: metareg wmd i.CompleteEPHPP, wsse(se) mm

i.CompleteEPHPP _ICompleteE_1–2 (_ICompleteE_1 for Com∼P = = 60–79% omitted)

Meta-regression number of studies = 8

Fit of model without heterogeneity (τ2 = 0): Q (6 d.f.) = 1.76955

p > Q = 0.940

Proportion of variation due to heterogeneity I2 = 0.000

Moment-based estimate of between-study variance: τ2 = 0.0000

WMDCoef.SEtp > |t|(95% CI)
_IComplete∼2 .7038533 .6068578 1.16 0.290 −.7810744 2.188781
_cons −.4702745 .4494989 −1.05 0.336 −1.570159 .6296096

There was no evidence to suggest that estimates of effect on BMI were associated with the level of attrition in studies.

Weight

. xi: metareg wmd i.CompleteEPHPP, wsse(se) mm

i.CompleteEPHPP _ICompleteE_1–2 (_ICompleteE_1 for Com∼P = = 60–79% omitted)

Meta-regression number of studies = 4

Fit of model without heterogeneity (τ2 = 0): Q (2 d.f.) = 21.7976

p > Q = 0.000

Proportion of variation due to heterogeneity I2 = 0.908

Moment-based estimate of between-study variance: τ2 = 12.1802

WMDCoef.SEtp > |t|(95% CI)
_IComplete∼2 −2.791344 10.5395 −0.26 0.816 −48.13913 42.55644
_cons 3.266 10.31643 0.32 0.782 −41.122 47.654

There was no evidence to suggest that estimates of effect on weight were associated with the level of attrition in studies.

3. Allocation concealment (Cochrane ‘low risk of bias’ vs other)

Fruit and vegetables

. xi: metareg wmd i.allocationconcealcoch, wsse(se) mm

i.allocationc∼h _Iallocatio_1–2 (_Iallocatio_1 for all∼h = = Unclear omitted)

Meta-regression number of studies = 12

Fit of model without heterogeneity (τ2 = 0): Q (10 d.f.) = 52.3053

p > Q = 0.000

Proportion of variation due to heterogeneity I2 = 0.809

Moment-based estimate of between-study variance: τ2 = 0.0819

WMDCoef.SEtp > |t|(95% CI)
_Iallocati∼2 −.0546828 .206455 −0.26 0.796 −.5146931 .4053276
_cons .2688521 .1447394 1.86 0.093 −.0536475 .5913517

There was no evidence to suggest that estimates of effect on daily servings of fruit and vegetables were associated with whether or not allocation was adequately concealed from investigators or was unclear.

Total fat

. xi: metareg wmd i.allocationconcealcoch, wsse(se) mm

i.allocationc∼h _Iallocatio_1–2 (_Iallocatio_1 for all∼h = = Unclear omitted)

Meta-regression number of studies = 12

Fit of model without heterogeneity (τ2 = 0): Q (10 d.f.) = 15.1604

p > Q = 0.126

Proportion of variation due to heterogeneity I2 = 0.340

Moment-based estimate of between-study variance: τ2 = 2.8665

WMDCoef.SEtp > |t|(95% CI)
_Iallocati∼2 −.3916092 2.275972 −0.17 0.867 −5.46279 4.679572
_cons −.6184049 1.251933 −0.49 0.632 −3.407885 2.171075

There was no evidence to suggest that estimates of effect on total fat intake were associated with whether or not allocation was adequately concealed from investigators or was unclear.

Percentage energy from fat

. xi: metareg wmd i.allocationconcealcoch, wsse(se) mm

i.allocationc∼h _Iallocatio_1–2 (_Iallocatio_1 for all∼h = = Unclear omitted)

Meta-regression number of studies = 10

Fit of model without heterogeneity (τ2 = 0): Q (8 d.f.) = 32.524

p > Q = 0.000

Proportion of variation due to heterogeneity I2 = 0.754

Moment-based estimate of between-study variance: τ2 = 1.9619

WMDCoef.SEtp > |t|(95% CI)
_Iallocati∼2 .669468 1.224406 0.55 0.599 −2.154018 3.492954
_cons −1.609689 .6854733 −2.35 0.047 −3.190393 −.0289842

There was no evidence to suggest that estimates of effect on percentage of energy from fat were associated with whether or not allocation was adequately concealed from investigators or was unclear.

Energy

. xi: metareg wmd i.allocationconcealcoch, wsse(se) mm

i.allocationc∼h _Iallocatio_1–2 (_Iallocatio_1 for all∼h = = Unclear omitted)

Meta-regression number of studies = 5

Fit of model without heterogeneity (τ2 = 0): Q (3 d.f.) = 4.14494

p > Q = 0.246

Proportion of variation due to heterogeneity I2 = 0.276

Moment-based estimate of between-study variance: τ2 = 4.2e+03

WMDCoef.SEtp > |t|(95% CI)
_Iallocati∼2 50.12752 110.4682 0.45 0.681 −301.4315 401.6866
_cons −17.01646 72.97764 −0.23 0.831 −249.2639 215.231

There was no evidence to suggest that estimates of effect on total energy intake were associated with whether or not allocation was adequately concealed from investigators or was unclear.

Body mass index

. xi: metareg wmd i.allocationconcealcoch, wsse(se) mm

i.allocationc∼h _Iallocatio_1–2 (_Iallocatio_1 for all∼h = = Unclear omitted)

Meta-regression number of studies = 9

Fit of model without heterogeneity (τ2 = 0): Q (7 d.f.) = 2.98302

p > Q = 0.887

Proportion of variation due to heterogeneity I2 = 0.000

Moment-based estimate of between-study variance: τ2 = 0.0000

WMDCoef.SEtp > |t|(95% CI)
_Iallocati∼2 .2828604 .5793218 0.49 0.640 −1.087018 1.652739
_cons −.2368358 .3803496 −0.62 0.553 −1.13622 .662548

There was no evidence to suggest that estimates of effect on BMI were associated with whether or not allocation was adequately concealed from investigators or was unclear.

Weight

. xi: metareg wmd i.allocationconcealcoch, wsse(se) mm

i.allocationc∼h _Iallocatio_1–2 (_Iallocatio_1 for all∼h = = Unclear omitted)

Meta-regression number of studies = 4

Fit of model without heterogeneity (τ2 = 0): Q (2 d.f.) = .010586

p > Q = 0.995

Proportion of variation due to heterogeneity I2 = 0.000

Moment-based estimate of between-study variance: τ2 = 0.0000

WMDCoef.SEtp > |t|(95% CI)
_Iallocati∼2 5.246741 1.115829 4.70 0.042 .4457161 10.04777
_cons −2.8 .3627551 −7.72 0.016 −4.360809 −1.239191

There was some evidence to suggest that estimates of effect on weight were larger in studies in which allocation was adequately concealed from investigators (p = 0.042). (Note: this analysis does not account for whether or not studies aimed to reduce or maintain weight.)

4. Low income group versus other

Fruit and vegetables

. xi: metareg wmd i.Low_Inc, wsse(se) mm

i.Low_Inc _ILow_Inc_0–1 (naturally coded; _ILow_Inc_0 omitted)

Meta-regression number of studies = 12

Fit of model without heterogeneity (τ2 = 0): Q (10 d.f.) = 38.2686

p > Q = 0.000

Proportion of variation due to heterogeneity I2 = 0.739

Moment-based estimate of between-study variance: τ2 = 0.0453

WMDCoef.SEtp > |t|(95% CI)
_ILow_Inc_1 .4975182 .2271436 2.19 0.053 −.0085892 1.003626
_cons .1452362 .0912218 1.59 0.142 −.0580186 .348491

There was some evidence to suggest that estimates of effect on daily servings of fruit and vegetables were larger in studies with participants predominantly from low-income groups (p = 0.053).

Total fat

. xi: metareg wmd i.Low_Inc, wsse(se) mm

i.Low_Inc _ILow_Inc_0–1 (naturally coded; _ILow_Inc_0 omitted)

Meta-regression number of studies = 12

Fit of model without heterogeneity (τ2 = 0): Q (10 d.f.) = 13.8384

p > Q = 0.180

Proportion of variation due to heterogeneity I2 = 0.277

Moment-based estimate of between-study variance: τ2 = 1.9142

WMDCoef.SEtp > |t|(95% CI)
_ILow_Inc_1 3.042827 3.369411 0.90 0.388 −4.464689 10.55034
_cons −1.041848 .9845122 −1.06 0.315 −3.235478 1.151781

There was no evidence to suggest that estimates of effect on total fat intake were associated with studies including participants predominantly from low-income groups.

Percentage energy from fat

None of the studies measuring percentage energy from fat included participants predominantly from low-income groups.

Energy

None of the studies measuring energy intake included participants predominantly from low-income groups.

Body mass index

None of the studies measuring BMI included participants predominantly from low-income groups.

Weight

None of the studies measuring weight included participants predominantly from low-income groups.

5. Early outcome (< 3 months) versus later outcome (≥ 3 months)

Fruit and vegetables

. xi: metareg wmd i.FU, wsse(se) mm

i.FU _IFU_0–1 (naturally coded; _IFU_0 omitted)

Meta-regression number of studies = 12

Fit of model without heterogeneity (τ2 = 0): Q (10 d.f.) = 47.6783

p > Q = 0.000

Proportion of variation due to heterogeneity I2 = 0.790

Moment-based estimate of between-study variance: τ2 = 0.0732

WMDCoef.SEtp > |t|(95% CI)
_IFU_1 .1187708 .1980919 0.60 0.562 −.3226053 .560147
_cons .1813987 .1381617 1.31 0.219 −.1264447 .4892421

There was no evidence that estimates of effect on servings of fruit and vegetables differed according to time to follow-up.

Total fat

i.FU _IFU_0–1 (naturally coded; _IFU_0 omitted)

Meta-regression number of studies = 12

Fit of model without heterogeneity (τ2 = 0): Q (10 d.f.) = 15.159

p > Q = 0.126

Proportion of variation due to heterogeneity I2 = 0.340

Moment-based estimate of between-study variance: τ2 = 2.7992

WMDCoef.SEtp > |t|(95% CI)
_IFU_1 −.2754214 2.40662 −0.11 0.911 −5.637705 5.086862
_cons −.6704323 1.204002 −0.56 0.590 −3.353117 2.012252

There was no evidence that estimates of effect on total fat intake differed according to time to follow-up.

Percentage energy from fat

. xi: metareg wmd i.FU, wsse(se) mm

i.FU _IFU_0–1 (naturally coded; _IFU_0 omitted)

Meta-regression number of studies = 10

Fit of model without heterogeneity (τ2 = 0): Q (8 d.f.) = 39.4114

p > Q = 0.000

Proportion of variation due to heterogeneity I2 = 0.797

Moment-based estimate of between-study variance: τ2 = 2.0669

WMDCoef.SEtp > |t|(95% CI)
_IFU_1 .6480017 2.137268 0.30 0.769 −4.280546 5.576549
_cons −2 2.05116 −0.98 0.358 −6.729982 2.729982

There was no evidence that estimates of effect percentage of energy from fat differed according to time to follow-up.

Energy

. xi: metareg wmd i.FU, wsse(se) mm

i.FU _IFU_0–1 (naturally coded; _IFU_0 omitted)

Meta-regression number of studies = 5

Fit of model without heterogeneity (τ2 = 0): Q (3 d.f.) = 1.46516

p > Q = 0.690

Proportion of variation due to heterogeneity I2 = 0.000

Moment-based estimate of between-study variance: τ2 = 0.0000

WMDCoef.SEtp > |t|(95% CI)
_IFU_1 176.0411 99.62481 1.77 0.175 −141.0095 493.0917
_cons −132.7 87.86581 −1.51 0.228 −412.3282 146.9282

There was no evidence that estimates of effect on energy intake differed according to time to follow-up.

Body mass index

. xi: metareg wmd i.FU, wsse(se) mm

i.FU _IFU_0–1 (naturally coded; _IFU_0 omitted)

Meta-regression number of studies = 9

Fit of model without heterogeneity (τ2 = 0): Q (7 d.f.) = 2.90451

p > Q = 0.894

Proportion of variation due to heterogeneity I2 = 0.000

Moment-based estimate of between-study variance: τ2 = 0.0000

WMDCoef.SEtp > |t|(95% CI)
_IFU_1 −.324209 .5759164 −0.56 0.591 −1.686035 1.037617
_cons .0332676 .3893472 0.09 0.934 −.8873923 .9539275

There was no evidence that estimates of effect on BMI differed according to time to follow-up.

Weight

. xi: metareg wmd i.FU, wsse(se) mm

i.FU _IFU_0–1 (naturally coded; _IFU_0 omitted)

Meta-regression number of studies = 4

Fit of model without heterogeneity (τ2 = 0): Q (2 d.f.) = 16.2713

p > Q = 0.000

Proportion of variation due to heterogeneity I2 = 0.877

Moment-based estimate of between-study variance: τ2 = 11.6652

WMDCoef.SEtp > |t|(95% CI)
_IFU_1 −2.9805 4.481034 −0.67 0.574 −22.26083 16.29983
_cons 2.634944 3.721698 0.71 0.552 −13.37823 18.64812

There was no evidence that estimates of effect on weight differed according to time to follow-up.

6. Overweight (BMI > 25 kg/m2) versus not overweight

Fruit and vegetables

. xi: metareg wmd i.Overweight, wsse(se) mm

i.Overweight _IOverweigh_0–1 (naturally coded; _IOverweigh_0 omitted)

Meta-regression number of studies = 12

Fit of model without heterogeneity (τ2 = 0): Q (10 d.f.) = 66.0638

p > Q = 0.000

Proportion of variation due to heterogeneity I2 = 0.849

Moment-based estimate of between-study variance: τ2 = 0.0835

WMDCoef.SEtp > |t|(95% CI)
_IOverweig∼1 −.2510207 .5715148 −0.44 0.670 −1.524435 1.022394
_cons .2510207 .1057554 2.37 0.039 .015383 .4866585

There was no evidence that estimates of effect on servings of fruit and vegetables differed according to whether or not included participants were overweight.

Total fat

. xi: metareg wmd i.Overweight, wsse(se) mm

i.Overweight _IOverweigh_0–1 (naturally coded; _IOverweigh_0 omitted)

Meta-regression number of studies = 12

Fit of model without heterogeneity (τ2 = 0): Q (10 d.f.) = 15.1632

p > Q = 0.126

Proportion of variation due to heterogeneity I2 = 0.341

Moment-based estimate of between-study variance: τ2 = 2.6260

WMDCoef.SEtp > |t|(95% CI)
_IOverweig∼1 −.9068454 3.274809 −0.28 0.787 −8.203574 6.389883
_cons −.6454614 1.091058 −0.59 0.567 −3.076491 1.785568

There was no evidence that estimates of effect on total fat intake differed according to whether or not included participants were overweight.

Percentage energy from fat

. xi: metareg wmd i.Overweight, wsse(se) mm

i.Overweight _IOverweigh_0–1 (naturally coded; _IOverweigh_0 omitted)

Meta-regression number of studies = 10

Fit of model without heterogeneity (τ2 = 0): Q (8 d.f.) = 27.5027

p > Q = 0.001

Proportion of variation due to heterogeneity I2 = 0.709

Moment-based estimate of between-study variance: τ2 = 1.4112

WMDCoef.SEtp > |t|(95% CI)
_IOverweig∼1 1.918306 1.140995 1.68 0.131 −.7128328 4.549444
_cons −1.891834 .5879203 −3.22 0.012 −3.247581 −.5360873

There was no evidence that estimates of effect on percentage of energy from fat differed according to whether or not included participants were overweight.

Energy

. xi: metareg wmd i.Overweight, wsse(se) mm

i.Overweight _IOverweigh_0–1 (naturally coded; _IOverweigh_0 omitted)

Meta-regression number of studies = 5

Fit of model without heterogeneity (τ2 = 0): Q (3 d.f.) = 3.39924

p > Q = 0.334

Proportion of variation due to heterogeneity I2 = 0.117

Moment-based estimate of between-study variance: τ2 = 1.5e+03

WMDCoef.SEtp > |t|(95% CI)
_IOverweig∼1 −102.272 98.09153 −1.04 0.374 −414.443 209.899
_cons 72.89176 80.47466 0.91 0.432 −183.2145 328.9981

There was no evidence that estimates of effect on total energy intake differed according to whether or not included participants were overweight.

Body mass index

. xi: metareg wmd i.Overweight, wsse(se) mm

i.Overweight _IOverweigh_0–1 (naturally coded; _IOverweigh_0 omitted)

Meta-regression number of studies = 9

Fit of model without heterogeneity (τ2 = 0): Q (7 d.f.) = 2.88748

p > Q = 0.895

Proportion of variation due to heterogeneity I2 = 0.000

Moment-based estimate of between-study variance: τ2 = 0.0000

WMDCoef.SEtp > |t|(95% CI)
_IOverweig∼1 −.3317409 .5740735 −0.58 0.581 −1.689209 1.025727
_cons .0562025 .4122949 0.14 0.895 −.9187199 1.031125

There was no evidence that estimates of effect on BMI differed according to whether or not included participants were overweight.

Weight

. xi: metareg wmd i.Overweight, wsse(se) mm

i.Overweight _IOverweigh_0–1 (naturally coded; _IOverweigh_0 omitted)

Meta-regression number of studies = 4

Fit of model without heterogeneity (τ2 = 0): Q (2 d.f.) = 6.48674

p > Q = 0.039

Proportion of variation due to heterogeneity I2 = 0.692

Moment-based estimate of between-study variance: τ2 = 9.4399

WMDCoef.SEtp > |t|(95% CI)
_IOverweig∼1 3.087439 3.944292 0.78 0.516 −13.88348 20.05836
_cons −.6045792 2.374113 −0.25 0.823 −10.81956 9.610406

There was no evidence that estimates of effect on weight differed according to whether or not included participants were overweight.

7. Primary prevention versus diagnosed illness

Fruit and vegetables

None of the studies measuring daily servings of fruit and vegetables included participants with a diagnosed illness.

Total fat

None of the studies measuring total fat intake included participants with a diagnosed illness.

Percentage energy from fat

. xi: metareg wmd i.Secondary, wsse(se) mm

i.Secondary _ISecondary_0–1 (naturally coded; _ISecondary_0 omitted)

Meta-regression number of studies = 10

Fit of model without heterogeneity (τ2 = 0): Q (8 d.f.) = 18.9136

p > Q = 0.015

Proportion of variation due to heterogeneity I2 = 0.577

Moment-based estimate of between-study variance: τ2 = 1.4700

WMDCoef.SEtp > |t|(95% CI)
_ISecondar∼1 −1.350213 1.431508 −0.94 0.373 −4.651275 1.95085
_cons −1.149787 .5965762 −1.93 0.090 −2.525494 .2259199

There was no evidence that estimates of effect on percentage of energy from fat differed according to whether or not participants had a diagnosed illness.

Energy

None of the studies measuring energy intake included participants with a diagnosed illness.

BMI

None of the studies measuring BMI included participants with a diagnosed illness.

Weight

. xi: metareg wmd i.Secondary, wsse(se) mm

i.Secondary _ISecondary_0–1 (naturally coded; _ISecondary_0 omitted)

Meta-regression number of studies = 4

Fit of model without heterogeneity (τ2 = 0): Q (2 d.f.) = .010586

p > Q = 0.995

Proportion of variation due to heterogeneity I2 = 0.000

Moment-based estimate of between-study variance: τ2 = 0.0000

WMDCoef.SEtp > |t|(95% CI)
_ISecondar∼1 −5.246741 1.115829 −4.70 0.042 −10.04777 −.4457161
_cons 2.446741 1.055217 2.32 0.146 −2.093492 6.986975

There was some evidence to suggest that estimates of effect on weight were larger in studies with participants with a diagnosed illness (p = 0.042). (Note: this analysis does not account for whether or not studies aimed to reduce or maintain weight.)

8. Aimed to reduce versus maintain body mass index/weight

Fruit and vegetables

. xi: metareg wmd i.BMI_down, wsse(se) mm

i.BMI_down _IBMI_down_0–1 (naturally coded; _IBMI_down_0 omitted)

Meta-regression number of studies = 12

Fit of model without heterogeneity (τ2 = 0): Q (10 d.f.) = 66.0638

p > Q = 0.000

Proportion of variation due to heterogeneity I2 = 0.849

Moment-based estimate of between-study variance: τ2 = 0.0835

WMDCoef.SEtp > |t|(95% CI)
_IBMI_down_1 −.2510207 .5715148 −0.44 0.670 −1.524435 1.022394
_cons .2510207 .1057554 2.37 0.039 .015383 .4866585

There was no evidence that estimates of effect on servings of fruit and vegetables differed according whether or not studies aimed to maintain or reduce BMI.

Total fat

. xi: metareg wmd i.BMI_down, wsse(se) mm

i.BMI_down _IBMI_down_0–1 (naturally coded; _IBMI_down_0 omitted)

Meta-regression number of studies = 12

Fit of model without heterogeneity (τ2 = 0): Q (10 d.f.) = 15.0849

p > Q = 0.129

Proportion of variation due to heterogeneity I2 = 0.337

Moment-based estimate of between-study variance: τ2 = 2.5378

WMDCoef.SEtp > |t|(95% CI)
_IBMI_down_1 −1.535807 3.487471 −0.44 0.669 −9.306377 6.234762
_cons −.6056245 1.06856 −0.57 0.583 −2.986523 1.775275

There was no evidence that estimates of effect on total fat intake differed according whether or not studies aimed to maintain or reduce BMI.

Percentage energy from fat

. xi: metareg wmd i.BMI_down, wsse(se) mm

i.BMI_down _IBMI_down_0–1 (naturally coded; _IBMI_down_0 omitted)

Meta-regression number of studies = 10

Fit of model without heterogeneity (τ2 = 0): Q (8 d.f.) = 36.9256

p > Q = 0.000

Proportion of variation due to heterogeneity I2 = 0.783

Moment-based estimate of between-study variance: τ2 = 2.7521

WMDCoef.SEtp > |t|(95% CI)
_IBMI_down_1 1.642182 1.275113 1.29 0.234 −1.298234 4.582598
_cons −2.32978 .9474181 −2.46 0.039 −4.51453 −.1450303

There was no evidence that estimates of effect on percentage of energy from fat differed according whether or not studies aimed to maintain or reduce BMI.

Energy

. xi: metareg wmd i.BMI_down, wsse(se) mm

i.BMI_down _IBMI_down_0–1 (naturally coded; _IBMI_down_0 omitted)

Meta-regression number of studies = 5

Fit of model without heterogeneity (τ2 = 0): Q (3 d.f.) = 2.88746

p > Q = 0.409

Proportion of variation due to heterogeneity I2 = 0.000

Moment-based estimate of between-study variance: τ2 = 0.0000

WMDCoef.SEtp > |t|(95% CI)
_IBMI_down_1 −380.5853 291.8849 −1.30 0.283 −1309.493 548.3226
_cons 377 288.8699 1.31 0.283 −542.313 1296.313

There was no evidence that estimates of effect on energy intake differed according whether or not studies aimed to maintain or reduce BMI.

Body mass index

. xi: metareg wmd i.BMI_down, wsse(se) mm

i.BMI_down _IBMI_down_0–1 (naturally coded; _IBMI_down_0 omitted)

Meta-regression number of studies = 9

Fit of model without heterogeneity (τ2 = 0): Q (7 d.f.) = 3.20983

p > Q = 0.865

Proportion of variation due to heterogeneity I2 = 0.000

Moment-based estimate of between-study variance: τ2 = 0.0000

WMDCoef.SEtp > |t|(95% CI)
_IBMI_down_1 −.0619419 .5753033 −0.11 0.917 −1.422318 1.298434
_cons −.0861856 .391764 −0.22 0.832 −1.01256 .840189

There was no evidence that estimates of effect on BMI differed according whether or not studies aimed to maintain or reduce BMI.

Weight

All studies that estimated mean weight as an outcome aimed to reduce weight.

9. Physical activity component in intervention versus none

Fruit and vegetables

. xi: metareg wmd i.PA, wsse(se) mm

i.PA _IPA_0–1 (naturally coded; _IPA_0 omitted)

Meta-regression number of studies = 12

Fit of model without heterogeneity (τ2 = 0): Q (10 d.f.) = 57.6919

p > Q = 0.000

Proportion of variation due to heterogeneity I2 = 0.827

Moment-based estimate of between-study variance: τ2 = 0.0800

WMDCoef.SEtp > |t|(95% CI)
_IPA_1 .1512111 .2807434 0.54 0.602 −.4743242 .7767464
_cons .2175574 .1114721 1.95 0.080 −.0308179 .4659328

There was no evidence that estimates of effect on servings of fruit and vegetables differed according to whether or not interventions included a physical activity component.

Total fat

. xi: metareg wmd i.PA, wsse(se) mm

i.PA _IPA_0–1 (naturally coded; _IPA_0 omitted)

Meta-regression number of studies = 12

Fit of model without heterogeneity (τ2 = 0): Q (10 d.f.) = 10.0671

p > Q = 0.435

Proportion of variation due to heterogeneity I2 = 0.007

Moment-based estimate of between-study variance: τ2 = 0.0337

WMDCoef.SEtp > |t|(95% CI)
_IPA_1 −7.125047 3.147323 −2.26 0.047 −14.13772 −.1123749
_cons −1.033167 .4412864 −2.34 0.041 −2.016414 −.0499195

There was some evidence to suggest that estimates of effect on total fat intake were larger in studies in which interventions also included a physical activity component (p = 0.047).

Percentage energy from fat

. xi: metareg wmd i.PA, wsse(se) mm

i.PA _IPA_0–1 (naturally coded; _IPA_0 omitted)

Meta-regression number of studies = 10

Fit of model without heterogeneity (τ2 = 0): Q (8 d.f.) = 17.1085

p > Q = 0.029

Proportion of variation due to heterogeneity I2 = 0.532

Moment-based estimate of between-study variance: τ2 = 1.0507

WMDCoef.SEtp > |t|(95% CI)
_IPA_1 −1.43327 1.032209 −1.39 0.202 −3.813548 .9470078
_cons −.8038327 .6508597 −1.24 0.252 −2.304718 .6970525

There was no evidence that estimates of effect on percentage of energy from fat differed according to whether or not interventions included a physical activity component.

Energy

. xi: metareg wmd i.PA, wsse(se) mm

i.PA _IPA_0–1 (naturally coded; _IPA_0 omitted)

Meta-regression number of studies = 5

Fit of model without heterogeneity (τ2 = 0): Q (3 d.f.) = 4.29626

p > Q = 0.231

Proportion of variation due to heterogeneity I2 = 0.302

Moment-based estimate of between-study variance: τ2 = 3.7e+03

WMDCoef.SEtp > |t|(95% CI)
_IPA_1 78.65628 140.2495 0.56 0.614 −367.6804 524.9929
_cons −9.159267 58.73057 −0.16 0.886 −196.0662 177.7476

There was no evidence that estimates of effect on total energy intake differed according to whether or not interventions included a physical activity component.

Body mass index

. xi: metareg wmd i.PA, wsse(se) mm

i.PA _IPA_0–1 (naturally coded; _IPA_0 omitted)

Meta-regression number of studies = 9

Fit of model without heterogeneity (τ2 = 0): Q (7 d.f.) = 3.20835

p > Q = 0.865

Proportion of variation due to heterogeneity I2 = 0.000

Moment-based estimate of between-study variance: τ2 = 0.0000

WMDCoef.SEtp > |t|(95% CI)
_IPA_1 .06573 .5749834 0.11 0.912 −1.29389 1.42535
_cons −.145655 .3932479 −0.37 0.722 −1.075538 .7842284

There was no evidence that estimates of effect on BMI differed according to whether or not interventions included a physical activity component.

Weight

All studies that estimated mean weight as an outcome included a physical activity component.

© 2011, Crown Copyright.

Included under terms of UK Non-commercial Government License.

Bookshelf ID: NBK98322

Views

Other titles in this collection

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...