| 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 |
| Dyson 2007 |
Low risk of bias |
"Randomization was undertaken by means of sealed envelopes equivalent to the number of subjects and filled fifty‐fifty with an indicator of either a low‐carbohydrate diet or healthy‐eating advice. Two separate sets of envelopes were prepared for both diabetic and non‐diabetic subjects. An independent observer witnessed randomization." Author reported in correspondence that no formal test of hypothesis was done for difference at baseline for non‐diabetic participants. Baseline characteristics table showed differences which were consistent with differences resulting by chance with very small sample sizes. |
Low risk of bias |
No information, however, blinding of participants and people delivering these dietary interventions is highly unlikely. n=1 of randomised participants in balanced group refused follow‐up as not allocated to low carbohydrate diet. Modified intention to treat, n=1 from control group refused follow‐up, thus data missing and not included. |
Low risk of bias |
In low carbohydrate diet group, 0% and in balanced diet group, 14% (n=1) of randomised participants were not included in the analyses. Based on reason provided, it is unlikely that missingness in data for HbA1c was influenced by its true value. |
Some concerns |
No information about method used for HbA1c measurements. No details about blinding of outcome assessors. Laboratory determined outcomes are not observer‐reported involving judgement. |
Some concerns |
No mention of a pre‐specified analysis plan. No information about blinding of outcome data. Analysis intentions are not reported in sufficient detail to enable an assessment. |
Some concerns |
No information regarding the analytical method used to measure the outcome, or whether assessors were blinded. No pre‐registered document available to assess analysis intentions. |
| Larsen 2011 |
Low risk of bias |
" Randomisation was carried out by a third party using computer‐generated random numbers (using block randomisation and random block sizes) and stratified according to diabetes treatment..." "Dietary assignment was performed by a third party on the day of the initial dietary counselling visit." "The baseline characteristics for the intention‐to‐treat population were well‐matched at baseline...although non‐significant differences were apparent in calcium excretion rate, systolic blood pressure and the proportion of men." |
Low risk of bias |
"...we were not able to blind participants or the study personnel involved in dietary counselling." Deviations reported are expected to arise in usual care, e.g. lost interest, dissatisfied with weight loss. Modified ITT conducted: "...study staff encouraged all participants to return for follow‐up assessments of primary and secondary outcomes, regardless of dietary adherence." |
Low risk of bias |
9/57 (16%) participants in the intervention and 6/51 (12%) participants in the control arm did not have data available. This includes 4 participants in intervention and 5 in control arm that changed their mind after randomisation but before starting the diets. Data imputed for ITT for 5/57 (intervention) and 1/51 (control) participants with missing data. "This study used the single imputation method of last measurement carried forward for missing data for primary and secondary outcomes." Based on reasons provided, it is unlikely that missingness in data for HbA1c was influenced by its true value. |
Low risk of bias |
"HbA1c was determined using immunoturbidimetric spectrophotometry with the Roche Integra 800 analyser (Roche Diagnostics, Castle Hill, NSW, Australia)." No evidence that a different method was used across intervention groups to measure HbA1c. "However, key study endpoints (anthropometrics and laboratory variables) were measured by personnel who were blinded to group allocation." |
Some concerns |
No mention of pre‐specified analysis plan, but trial registry at date prior to study publication mentions the outcome and timepoint "Study endpoints were assessed blinded to the diet group, but the statistical analysis was performed unblinded." The time point mentioned in trial registry at date prior to publication is reported on in publication. Analysis intentions are not reported in sufficient detail to enable an assessment. |
Some concerns |
Lack of information in the pre‐registered methods. |
| Sato 2017 |
Low risk of bias |
"Randomization was achieved by the minimization and biased coin method." "Allocation sequence conducted by third party organization (Soiken, Inc., Osaka, Japan) was concealed until interventions were assigned." No problematic differences in baseline characteristics of participants |
Low risk of bias |
"....and they could not be blinded due to the characteristics of dietary therapy." "Open ‐ no one is blinded" (trial registry). Deviations are expected to arise in usual care, e.g. dissapointed with assigned diet, lacked motivation, moved to a different hospital. "All statistical analyses were conducted per‐protocol analysis..." However, all participants randomised received their assigned interventions and all were analysed except for participants with missing outcome data. Thus, a modified ITT was done." |
High risk of bias |
8/33 (24%) intervention participants and 6/33 (17%) control participants did not have data. Modified ITT was done. High attrition in the intervention group. Reasons for lost to follow up and withdrawal are clearly stated for all participants and included transferring to other hospitals and admission for different diseases, including diabetes. Missingness could be and is likely to be dependent on the true value. |
Low risk of bias |
"...HbA1c ... were measured with standard techniques" No evidence that different methods were used across intervention groups to measure HbA1c. "Open ‐ no one is blinded" (trial registry). HbA1c is not an observer‐reported outcome involving judgement. |
Some concerns |
No mention of a pre‐specified analysis plan. Could not obtain trial registry record. Analysis intentions not reported in sufficient detail to enable judgement. |
High risk of bias |
High attrition in the intervention group, and reasons for withdrawal indicate that missingness could be and is likely dependent on the true value of the outcome. No pre‐registered document available to assess analysis intentions. |
