Via JAMA, via Petra Sander: Caloric Sweetener Consumption and Dyslipidemia Among US Adults, by Jean Welsh et al. (institutional subscription possibly required). The headline of this post is the last few words of the article. Despite my scare quotes, I don’t disagree, but I think the evidence is much more eloquent than the authors allow.
Welsh et al. collect a lot of data, most of which is tabulated, then thrown into regression models as covariates without serious analysis or discussion. On this, more later. For now, let’s look at the results they bother visualizing:
From top to bottom and then center, that’s a nice negative relationship between added sugars and HDL (“good”) cholesterol, a nice positive relationship between added sugars and triglycerides (bad at least inasmuch as triglyceride levels are negatively correlated with HDL levels), and an interesting interaction between sex, added sugars, and LDL (“bad”) cholesterol. From the abstract: “LDL-C levels modified by sex were 116, 115, 118, 121, and 123 mg/dL among women (P = .047 for linear trend). There were no significant trends in LDL-C levels among men.” What does that sound like? Added sugars increase LDL levels in women but not men. OK, but “increase” from what?
It’s not that women’s and men’s LDL both start low and only women’s increases with added sugar — men’s starts high, and women’s only gets that high if they’re consuming ~17.5% of their total calories from added sugar. So is the men’s LDL level healthy or unhealthy? If healthy, then maybe the effect in women doesn’t represent a problem (unless men need more LDL in their blood, which is possible). If the men’s level is unhealthy across the board, then added sugar isn’t the only problem. So what’s the healthy level for LDL cholesterol? 100 to 129 mg/dL is “near optimal.” That is to say, the mean values for both men and women were in a pretty healthy range at all levels of added sugar consumption.
What about the HDL effect? That’s a strong one — but even the lowest HDL levels don’t go under or even close to 40mg/dL. Still, though, the size of the effect is big in proportion to the range of the data, so someone who is unhealthily low would probably be well advised to cut sugar.
But let’s step back a bit: The problem is not really that added sugars raise the levels of lipids in our blood. The problem is that they lower them — specifically, they lower HDL cholesterol. (From the title: “dyslipidemia.” Someone really wanted to say “hyperlipidemia,” but they couldn’t.)
At this point, we’d like to start thinking about macronutrient consumption overall — how much fat and protein do people eat at different levels of added sugar? But here we’re hobbled by a constraint of measurement. Consumption of carbohydrate, fat, protein, and added sugar are all measured as a percentage of total food consumption, so they have to add up to 100 (sort of — obviously added sugar is a subset of carbohydrate); and total energy intake increases with consumption of added sugar. So the fact that the percentage of protein and fat falls as the percentage of added sugar rises is something resembling a mathematical necessity. It’s prima facie perfectly possible that people at all levels of added sugar consumption are eating exactly the same diet except for the added sugar — or, alternatively, that absolute consumption of fat or protein is going down as absolute consumption of added sugar is going up. Somebody knows, but not me. But one thing they do measure in absolute terms (grams) is cholesterol intake:
Check it out: The lowest-sugar group, which has the highest HDL levels and (in women) the lowest LDL levels, eats 31% more cholesterol than the highest-sugar group.
I haven’t addressed the most obvious shortcoming of the paper, which is that it’s observational and cross-sectional: Nothing’s manipulated, so no causality can be inferred. Maybe people with poor blood lipid profiles are more drawn to sugar; maybe a third variable explains both. This is obviously a problem worth surmounting, but research is hard; as long as the results constrain hypotheses, they’re interesting enough for me. The problem is that the paper doesn’t make the most of the evidence it has. It wants to say one thing: Eating added sugar raises your risk of heart disease. All the results are interpreted with that in mind. There’s no mention of the elements of the data that contravene medical convention about cholesterol, there’s no acknowledgement that the linear relationships only appear to obtain over healthy levels of blood lipids.
Predictably, I have some baggage regarding this issue. My family has a history of high cholesterol but not a history of heart disease; I’m currently trying to get my LDL levels down and my HDL levels up. My doctor has told me to restrict cholesterol — but, from this study, cholesterol intake seems to be positively correlated with good blood lipid profiles. I realize it’s hasty thinking to change one’s behavior based on one obviously flawed paper… which is why I wish they’d talked more about this aspect of the data. Whether it was authorial tunnel vision or reviewers’ dogma, I feel a bit personally deprived of some really useful discussion.
Anyway, I need to get lunch, so I’ll close as I began: I do think the paper’s conclusions are basically, if not strongly, supported by the evidence. But I don’t think that’s the whole story, or even the most striking finding.