| Study |
Bias |
| Randomisation process |
Deviations from intended interventions |
Missing outcome data |
Measurement of the outcome |
Selection of the reported results |
Overall |
| Authors' judgement |
Support for judgement |
Authors' judgement |
Support for judgement |
Authors' judgement |
Support for judgement |
Authors' judgement |
Support for judgement |
Authors' judgement |
Support for judgement |
Authors' judgement |
Support for judgement |
| Chalmers 2020 |
Low risk of bias |
Quote: "The randomisation schedule was created by the CTU using computer‐generated pseudo‐random code with permuted blocks of randomly varying size. The sequence was known only to the programmer until database lock." Baseline variables by treatment group do not suggest a problem with randomisation, characteristics well balanced (IPD available) |
Low risk of bias |
Not possible to blind participants’ carers, but there is no evidence that deviations arised because of the trial context. The control group rates of skin care application were consistent with other trials and observational studies (e.g. up to 75 % in Rendell et al. 2011). Quote: “Of families in the emollient group with complete questionnaire data on adherence at each time point, 466 (88%) of 532 had satisfactory adherence at 3 months, 427 (82%) of 519 at 6 months, and 375 (74%) of 506 at 12 months.” “No emollient was supplied to the control group, but self‐directed use of emollients at least three times per week to most of the body (contamination) occurred in 18% (82 of 457) at 3 months, 17% (62 of 372) at 6 months, and 15% (49 of 324) at 12 months.” |
Low risk of bias |
1236/1394 =89% have outcome. Sensitivity analysis using the IPD data shows conclusions do not change if all missing are assumed to have eczema, HR=0.94, 95% 0.79 to 1.12 or when all missing are assumed to have eczema, HR=0.92, 95% CI 0.79 to 1.11. |
Some concerns |
Outcome was parent report of doctor diagnosis of eczema and parents unblinded. |
Low risk of bias |
Full trial dataset provided by investigators and IPD meta‐analysis SAP followed. |
Some concerns |
Outcome assessors were reporters who were aware of the allocation. Low risk of bias for all other domains. |
| Dissanayake 2019 |
Low risk of bias |
Randomisation performed using minimization at a central location. Baseline variables by treatment group do not suggest a problem with randomisation, characteristics well balanced (IPD available). |
Low risk of bias |
Not possible to blind participants’ carers, but there is no evidence that deviations arised because of the trial context. Quote: “we did not prohibit the application of moisturizers in the no‐intervention group. As this could mask the effect of the moisturizer used as an intervention, we looked at the effect of any emollient (irrespective of the formal intervention assigned) on the development of AD. We did not observe any difference in incidence between children who received any type of emollient vs. those who did not (data not shown).” “Approximately 80% of the parents/guardians in the emollient groups reported that they applied the emollient at least twice a day. Adjusting for emollient application rate did not lead to a significant difference in the rate of AD development among the groups. The effect of emollient application was also investigated in all babies who received emollient (groups 1 and 3) versus those who did not (groups 2 and 4), but this too did not show any difference in the incidence of AD.” |
Low risk of bias |
Outcome missing for 21.5% (not all people with eczema outcome have time of eczema onset). Sensitivity analysis using the IPD shows conclusions do not change significantly if all individuals missing eczema outcome are assumed to not have eczema by 12months HR=1.27 [0.83 to 1.95], or to have eczema onsets by 12 months: HR= 1.16 [0.76 to 1.78]. |
Low risk of bias |
UK working party criteria and blinded assessor used. Quote: "AD diagnosis was further confirmed blindly" |
Low risk of bias |
Full trial dataset provided by investigators and IPD meta‐analysis SAP followed. |
Low risk of bias |
Low risk of bias in all domains. |
| Horimukai 2014 |
Low risk of bias |
Randomised trial, no information provided on allocation concealment. Baseline variables by treatment group do not suggest a problem with randomisation, characteristics well balanced (IPD available). |
Low risk of bias |
Not possible to blind participants’ carers, but there is no evidence that deviations arised because of the trial context. Control rates of skin care in line with those observed in other trials and observational studies (e.g. up to ~75 % in Rendell at al. 2011). Quote: “Because the IRB recommended permitting application of petroleum jelly when the parents thought it necessary, we calculated the amount of these 2 types of moisturizers used by each group based on their diaries. The mean daily amount of emulsion‐type moisturizer used by the intervention group was 7.86 6 4.34 g (0 g for the control group, excluding the 2 infants placed in the wrong group). The mean daily amount of petroleum jelly applied to the control group was 0.101 6 0.286 g (mean frequency of use, 0.235 d/wk). Petroleum jelly (20 g per bottle) was prescribed to all neonates born at the NCCHD, but we had no information about how much was used by the intervention group. Nevertheless, only a few of the parents occasionally used a small, almost ignorable, amount of the jelly on their infants.” |
Low risk of bias |
Data on time to eczema available for 116/118=98%, assuming censoring at random for 37/118=31% participants who dropped out of the trial prior to 9 months with no eczema outcome recorded prior to drop out. Sensitivity analysis using the IPD reveals conclusions do not change significantly if everyone censored is assumed to not have eczema by 9 months, HR=0.59 [0.33, 1.07], or if everyone missing is assumed to have eczema by 9 months 0.61, [0.39, 0.97]. |
Low risk of bias |
Modification of the UKWP criteria used to record eczema up to 32 weeks and blinded assessment. |
Low risk of bias |
Full trial dataset provided by investigators and IPD meta‐analysis SAP followed. |
Low risk of bias |
Low risk of bias in all domains. |
| Lowe 2018a |
Low risk of bias |
Quote from trial protocol: "A computer generated random allocation list in blocks of variable length (4‐12) will be used. This list will be held by The RCH Pharmacy Department, which will be independent from the participant recruitment or testing. At all times, the allocation list will be concealed from the study coordinator and the other study investigators, who will manage participant recruitment." Baseline variables by treatment group do not suggest a problem with randomisation, characteristics well balanced (IPD available). |
Low risk of bias |
Not possible to blind participants’ carers, but there is no evidence that deviations arised because of the trial context. The control group rates of skin care application were consistent with other trials and observational studies e.g. up to 75 % in Rendell et al. 2011. The IPD shows regular use of emollient (≥ 3 days a week) by 11/36 (31%) control participants and 30/38 (79%) with eczema outcome recorded). |
Low risk of bias |
Outcome available for 79/80=99%. |
Low risk of bias |
UK working party criteria used by blinded assessor. |
Low risk of bias |
Full trial dataset provided by investigators and IPD meta‐analysis SAP followed. |
Low risk of bias |
Low risk of bias in all domains. |
| McClanahan 2019 |
Low risk of bias |
Quote: "Randomization was completed via a computer‐generated randomization list issued by a statistician with block sizes of 10 and concealed from the recruitment coordinator until randomization." Baseline variables by treatment group do not suggest a problem with randomisation, characteristics well balanced (IPD available). |
Low risk of bias |
Not possible to blind participants’ carers, but there is no evidence that deviations arised because of the trial context.: "At 2, 6 and 12 months, the intervention group had 87%, 72.4% and 66.7% high adherence, respectively, vs. 45.2%, 45.8% and 45.5% in the control group, respectively." The control group rates are high but similar to those reported in another study by the same group (e.g. up to ~75 % in Rendell et al. 2011). |
Some concerns |
Data on time to eczema available for 100% (assuming censoring at random for 40/100 participants who dropped out of the trial prior to 2 years with no eczema outcome recorded prior to drop out). Loss to follow‐up was 40% for eczema at 2 years and incident eczema was captured through voluntary contacting of investigators. Sensitivity analysis of IPD reveals conclusions do not change if all censored before two years are assumed to not have eczema, 0.49 [0.17 to 1.40]. But conclusions do change if all censored before two years are assumed to have eczema, 0.91 [0.52 to 1.57] by the end of trial follow‐up (point estimate of RR HR >20% of the complete case estimate). |
Low risk of bias |
U.K. Working Party Criteria adapted to identify incident cases, some as young as 2 weeks. Some uncertainty about validity of this tool at this age. Incident cases not followed further after first diagnosis. Low risk of bias as this is our pre‐defined preferred tool. |
Low risk of bias |
Full trial dataset provided by investigators and IPD meta‐analysis SAP followed. |
Some concerns |
Some concerns due to missing data (40% censored before 2 years and results varied in sensitivity analysis) but low risk of bias in all other domains. |
| NCT03376243 |
Low risk of bias |
Baseline variables by treatment group do not suggest a problem with randomisation, characteristics well balanced (IPD available). |
Low risk of bias |
Not possible to blind participants’ carers, but there is no evidence that deviations arised because of the trial context. The control group rates of skin care application were consistent with other trials and observational studies e.g. e.g. up to ~75 % in Rendell et al. 2011. The IPD shows daily use of emollient over intervention period by 7/27 (26%) control and 19/22 (86%) intervention participants with eczema outcomes recorded. |
Some concerns |
5/54 (9%) participants are censored prior to 12 months due to loss to follow‐up/eczema status unknown. Sensitivity analysis of IPD reveals conclusions change if all missing are assumed to have eczema, HR=1.27, 0.46 to 3.53 (point estimate of HR >20% of the complete case estimate). Missingness could have depended on outcome, but this is not likely. |
Low risk of bias |
Quote: "assessor‐blind." UK WP criteria used. Patients were followed up until they had eczema or 12 months or were lost to follow‐up. |
Low risk of bias |
Full trial dataset provided by investigators and IPD meta‐analysis SAP followed. |
Some concerns |
Some concerns due to missing data (9% and result varied in sensitivity analysis) but low risk of bias for all other domains. |
| Simpson 2014 |
Low risk of bias |
Quote: "Infants were randomized at a 1:1 ratio using random block sizes to either the intervention or control group with a central, Web‐based, computer generated, Internet randomization service." Baseline variables by treatment group do not suggest a problem with randomisation, characteristics well balanced (IPD available). |
Low risk of bias |
Not possible to blind participants’ carers, but there is no evidence that deviations arised because of the trial context. Quote “Eight (13.3%) parents in the control group reported using emollients in a way that mirrored the intervention (ie, regular generalized application of emollient for reasons other than the treatment of cradle cap, nappy rash, or eczema).” The IPD shows regular use of emollient (≥ 3 days a week) by 50/55 intervention individuals and only 3/53 control participants with eczema outcome recorded). |
Some concerns |
Outcome is missing for 13% of randomised participants. Sensitivity analysis using the IPD reveals findings do not change significantly if everyone missing is assumed to not have eczema HR 0.48 [0.23 to 0.98] but if all are assumed to have eczema at 3 or 6 months HR = 0.65 [0.36 to 1.15] (point estimate of HR >20% of the complete case estimate). Missingness could have depended on outcome, but this is not likely. |
Low risk of bias |
Eczema was determined by an investigator. Outcome assessors were blind to treatment allocation. This only failed in 4% of cases. |
Low risk of bias |
Full trial dataset provided by investigators and IPD meta‐analysis SAP followed. |
Some concerns |
Some concerns due to missing data (13% and results varied in sensitivity analysis), but low risk of bias for all other domains. |
| Skjerven 2020 |
Low risk of bias |
Quote: “we used computer‐generated cluster randomisation based on 92 geographical living area blocks as well as eight 3‐month time blocks. All infants born in the same 3‐month period and belonging to the same postal code or city area were allocated to the same intervention group”. Baseline variables by treatment group do not suggest a problem with randomisation, characteristics well balanced (IPD available). |
Low risk of bias |
Not possible to blind participants’ carers, but there is no evidence that deviations arised because of the trial context. The IPD shows regular use of emollient (ceridal cream) (≥ 3 days a week averaged over intervention period) by only 1 control participant and 262/599 = 44% intervention arm. |
Some concerns |
Outcome available for 89.8%. Sensitivity analysis using the IPD reveals conclusions do not change significantly if everyone missing is assumed to not have eczema, HR=1.46 [1.00 to 2.12] but conclusions change if all missing are assumed to have eczema, HR=2.20 [1.66 to 2.93] (point estimate of HR >20% of the complete case estimate). Missingness could have depended on outcome, but this is not likely. |
Low risk of bias |
Used UK WP +Hanifin and Rajka criteria and blinded assessor. |
Low risk of bias |
Full trial dataset provided by investigators and IPD meta‐analysis SAP followed. |
Some concerns |
Some concerns due to missing data (10% and results varied in sensitivity analysis) but low risk of bias in all other domains. |
| Yonezawa 2018 |
Low risk of bias |
Quote: "Randomization was performed individually according to central registration after measuring the skin barrier function at baseline.” Baseline variables by treatment group do not suggest a problem with randomisation, characteristics well balanced (IPD available). |
Low risk of bias |
Not possible to blind participant carers, but there is no evidence that deviations arised because of the trial context. The control group rates of skin care application were consistent with other trials and observational studies e.g. up to ~75 % in Rendell et al. 2011. The IPD shows regular use of emollient (≥ 3 days a week) by 63/69 (91%) intervention participants 52/87 (60%) control participants with outcomes recorded). Quote: “The frequency of bathing (mean +/‐ standard deviation [SD], 0.81 +/‐ 0.22 vs 0.99 +/‐ 0.06 times/day, P < 0.001) and soap use was lower (mean +/‐ SD, 0.50 +/‐ 0.09 vs 0.98 +/‐ 0.05 times/ day, P < 0.001), and that of lotion use was higher (1.30 +/‐ 0.67 vs 0.48 +/‐ 0.39 times/day, P < 0.001) in the intervention group than in the control group during the study period.” |
Some concerns |
Eczema outcome available for 156/227 randomised = 69% participants. Sensitivity analysis of IPD reveals conclusions change if all missing do not have eczema: 0.63 [0.23, 1.73] or if all missing are assumed to have eczema at 12 or 24 months: 1.60 [1.03 to 2.47] (point estimate of HR >20% of the complete case estimate). Missingness could have depended on the outcome, but not likely. |
Some concerns |
Eczema was collected from parents using a self‐reported questionnaire. Outcome assessors were reporters who were aware of the allocation. |
Low risk of bias |
Full trial dataset provided by investigators and IPD meta‐analysis SAP followed. |
Some concerns |
Some concerns due to missing data (31% and results varied in sensitivity analysis) and how outcome reported by unblinded assessors (parents) but low risk of bias for other domains. |
