197 - The science of obesity & how to improve nutritional epidemiology | David Allison, Ph.D

KEYPOINTS

• 01

  Obesity rose slowly for centuries before the field woke up in the early 1990s

  David Allison, a statistician, obesity researcher and dean at Indiana University, notes that obesity rates in Western industrialized countries have climbed fairly steadily since at least the 1700s. Robert Fogel, who won a Nobel Prize for this work, traced it in historical records such as British naval recruits. The 'wake-up moment' came in the early 1990s with NHANES III. Before that, obesity research was seen as niche, and treating extreme cases was often disdained even by doctors.

• 02

  Body weight is strongly heritable, as was clear long ago

  Allison says the genetic component could have been seen in Charles Davenport's 1923 family data for the Carnegie Institution, though Davenport is now viewed as a racist. Twin and adoption studies from the 1980s, notably by Stunkard, Sørensen and colleagues using Nordic registries, made it undeniable. The BMI similarity of identical twins is about the same whether they are raised together or apart.

• 03

  Bariatric surgery probably extends life

  Bjorn Torp's mentor laughed at the idea that it needed proving, saying everybody knew surgery helped. But Allison insists you have to do the experiment. Lars Sjöström's Swedish Obese Subjects study was controlled though not randomized. Across studies, all-cause mortality reductions vary widely but are usually on the order of 50%, sometimes more. Peter and Allison both suspect the benefit is larger in people who start with type 2 diabetes.

• 04

  The 'obesity paradox' depends on age, era and confounding

  The BMI with the lowest mortality (the 'nadir') rises with age: very roughly near 20 at age 20, and in the low 30s by 70–80. The nadir also shifts with calendar time. In both Denmark and the US, obesity looked less deadly in 1990s data than in 1970s data, a finding Allison's NIH grant is investigating. Much of the paradox stems from illness causing weight loss and smoking keeping people thin. A famous 1987 JAMA paper by JoAnn Manson and colleagues argued that analyses should include only never-smokers. A 1995 New England Journal paper from the same group added more exclusions, including people with variable weight, but even these didn't fully flatten the left side of the curve.

• 05

  BMI is crude but useful at population scale

  The index is old; Ancel Keys rediscovered it and named it in the 1960s–70s. For a sense of scale: Kate Moss at her peak was about 16–17, the underweight cutoff is 18.5, obesity starts at 30, and a top sumo wrestler is around 43–44. Allison compares BMI to an ordinary car, fine for daily driving but not precise enough for the Indy 500. For individuals, Peter prefers DEXA, which also shows visceral fat and bone density. The usual cutoffs fit adult white women well but fit other groups less reliably.

• 06

  Losing weight is easy compared with keeping it off

  Allison sees no single explanation for why maintenance is so hard. He rejects the popular belief that there is a diet that, eaten correctly, would let anyone maintain a healthy weight without counting calories or feeling hungry or deprived. Advocates disagree about which diet that is, whether low-carb, low-fat or not eating at night. He knows of no evidence that such a diet exists.

• 07

  Observational data can point in exactly the wrong direction

  In one of Allison's mouse studies, genetically identical C57BL/6J mice in identical housing were randomly assigned to eat less, a medium amount or as much as they wanted. Those forced to eat more lived shorter lives, consistent with decades of calorie-restriction research. But within the free-feeding group, the mice that chose to eat more lived longer. The observational association was the exact opposite of the causal effect, even with every usual confounder controlled. Allison calls the defence that 'it's just confounding' an admission of the problem.

• 08

  Childhood-obesity science is often analysed badly

  Cluster-randomized trials assign whole schools, classrooms or families to a treatment. That's fine with enough clusters, but the analysis must account for clustering and nesting. Allison says many studies don't, and he calls that 'science gone wrong in the worst sense'. He believes some researchers are obfuscating null results while the field repeats hackneyed advice.

• 09

  Separate honest mistakes from bad science

  Allison distinguishes normative errors, where good methods give a wrong answer at the edge of knowledge, from non-normative errors, which are avoidable. A normative example is Galileo's lantern experiment, which wrongly concluded light travels instantaneously. A non-normative one: early-20th-century norms for the size of children's thymus glands came from cadavers of poor, undernourished children. Normal thymuses in better-fed children then looked enlarged and were irradiated to 'prevent' sudden infant death, causing many thyroid cancers.