Some comments on the latest interaction in @IntJObesity between @KevinH_PhD Juen Guo and myself and @davidludwigmd concerning the diet intervention study published late last year in @bmj_latest 1/n
@IntJObesity @KevinH_PhD @davidludwigmd @bmj_latest A previous dispute focused on the choice of the reference point pre randomisation. Ludwig et al’s most recent IJO paper reiterates this debate showing that there is better correlation between their second chosen reference, rather than that originally registered.
@IntJObesity @KevinH_PhD @davidludwigmd @bmj_latest To an extent our IJO commentary makes this decision redundant. We showed that performing the analysis excluding any reference, using just the post randomised data confirms there is a difference between groups when TEE is analysed by ANCOVA. 3/n
@IntJObesity @KevinH_PhD @davidludwigmd @bmj_latest We repeated this ANCOVA analysis using the CO2 production data and found that the effect disappeared. What this analysis shows is that the claimed effect in the paper depends only on the assumed RQ which is used to convert the CO2 into TEE. 4/n
@IntJObesity @KevinH_PhD @davidludwigmd @bmj_latest Note: the DLW method directly measures the CO2 production not the TEE. 5/n
@IntJObesity @KevinH_PhD @davidludwigmd @bmj_latest Ludwig et al suggest our analysis of CO2 production deviates from an established protocol for analysis of DLW data. But that is not correct. There is no reason why CO2 values cannot be compared, and no established protocol I am aware of that does not permit such a comparison 6/n
@IntJObesity @KevinH_PhD @davidludwigmd @bmj_latest The problem is the RQ isn’t known and needs to be estimated from the FQ from the dietary intake. However, the dietary intakes are also unknown. This is because in common with almost all dietary interventions in free-living subjects individuals under-reported their intake. 7/n
@IntJObesity @KevinH_PhD @davidludwigmd @bmj_latest Under-reporting in this study is because the energy content of the supplied food (EI) in the different diet arms was always lower than the TEE. Hence there must have been off protocol food intake that was not reported, because on average they were not losing weight. 8/n
@IntJObesity @KevinH_PhD @davidludwigmd @bmj_latest We identified the largest dietary under-reporters by taking the difference between the TEE and the estimated energy intake (adjusted for weight change dES/dt) ie EI-TEE-dES/dt. If we eliminate these individuals then the diet effect is substantially attenuated. 9/n
@IntJObesity @KevinH_PhD @davidludwigmd @bmj_latest Ludwig et al suggest that this procedure is invalid because these individuals may not be dietary under-reporters but have a large difference because of an erroneously high TEE. They show the individuals we eliminated do have higher than average TEE. 10/n
@IntJObesity @KevinH_PhD @davidludwigmd @bmj_latest However, having an erroneously high TEE does not seem a great reason to keep them in an analysis of differences in TEE! They also suggest that our approach can be shown to be wrong by doing a different calculation. 11/n
@IntJObesity @KevinH_PhD @davidludwigmd @bmj_latest That is instead of using EI-TEE-dES/dt they use TEE/EI. When this is done the ones with the lowest ratio are disproportionately those with the lowest TEE and eliminating these makes the diet effect on TEE larger. 12/n
@IntJObesity @KevinH_PhD @davidludwigmd @bmj_latest What they don't reveal is that these individuals not only have low TEE but also the lowest discrepancy between TEE and EI. ie this analysis eliminates individuals that show the best agreement between methods, while our analysis eliminates those with the worst agreement. 13/n
@IntJObesity @KevinH_PhD @davidludwigmd @bmj_latest They also replicate our sensitivity analysis concerning the potential non-adherence of individuals. This shows that as non-adherence increases the absolute difference between diets declines. Interestingly the effect of a 10% difference in CHO intake remains almost constant. 14/n
@IntJObesity @KevinH_PhD @davidludwigmd @bmj_latest 15.This is expected because the absolute smaller difference in TEE is attributed to a smaller difference in CHO intake so the two effects cancel out. This number however depends on the initial 240 kcal/day difference which remains under dispute. 15/n
@IntJObesity @KevinH_PhD @davidludwigmd @bmj_latest The estimate is also contingent on a linear response across CHO levels. This is not supported by their own data (ie mid and high CHO are not significantly different in TEE) 16/n
@IntJObesity @KevinH_PhD @davidludwigmd @bmj_latest Nothing is really resolved by this reply. Reanalysis of these data continues to lead different people to different conclusions. I continue to think that if there is an effect of diet on TEE, the impact of lowered carbohydrate is likely much smaller than the CIM predicts. 17/n
@IntJObesity @KevinH_PhD @davidludwigmd @bmj_latest The fact we remain in dispute really only reflects how incredibly hard it is to perform such studies, and I continue to commend Ludwig and colleagues for performing this work and making the data available for such detailed scrutiny. 18/n
@IntJObesity @KevinH_PhD @davidludwigmd @bmj_latest I further agree with Ludwig at al that the veracity of the carbohydrate insulin model will likely only be resolved by further experimentation. Sorry this was so long! 19/19
