Older adults whose diets had narrowed since about age 60 had the highest death risk of four groups. They did not clearly lose capacity any faster

Try to remember how often you ate fish when you were 60. Now imagine you are 87. That, roughly, was the question behind a study published in Frontiers in Nutrition on 2 November 2026. In 2008, interviewers with the Chinese Longitudinal Healthy Longevity Survey asked older adults how often they ate eight foods: fish, legumes, garlic, dairy, nuts, mushrooms or algae, eggs and tea. They asked about the present and about life around age 60. The typical gap between the two was 29 years, and 35% of interviews were answered entirely by someone else on the participant's behalf.

The researchers sorted people into four groups: high scores at both times, low at both, improved and worsened. Then they followed them to 2018, through 9,359 deaths among the 16,068 people in the main models. The raw numbers, from a slightly larger sample of 16,230, look dramatic: 50.5% of the stable-high group died, against 61% to 66% of the others. But the stable-high group was also younger and more often urban, educated, married and active. After adjusting for those differences, what remained was modest. The worsened group had a 13.5% higher hazard of death than the stable-high group, and the stable-low group a 5.6% higher one.

The worsened result was the sturdier of the two. It survived different assumptions about when deaths occurred, weighting to the national survey (hazard ratio 1.201) and extra adjustment for sleep and access to medical care (1.132). The stable-low result lost statistical significance under the weighting and the extra adjustment. Even so, the worsened group's score fell by an average of only 8.1 points out of 100, the groups were split at fixed cut-points, and the authors' own sensitivity figures suggest a moderately strong unmeasured factor could explain the link. They also note that part of the association may reflect the decline that comes before death. A person who is already failing may simply eat a narrower diet.

The picture changes when the outcome is capacity rather than survival. Intrinsic capacity is a World Health Organization idea covering thinking, movement, mood, vitality and the senses, scored here from 0 to 5. The stable-low group lost it faster than the stable-high group, at 0.0289 against 0.0185 points a year. The worsened group, the one with the sturdiest death signal, was not significantly different at 0.0235. Remove vision and hearing from the score and even the stable-low contrast slips out of significance (p = 0.075). Among 1,673 people who started fully intact, no diet group had a significantly higher risk of developing impairment.

Then there is the headline-friendly number: low-quality diets, identified by a clustering method, declined four to eight times faster than the most diverse diet. It is real, but it moves. In a model with 26 diet items, the 'poor' class lost 0.0546 points a year against 0.0070 for the diverse class. In the main eight-food model, the 'poor' class (0.0143) did not differ significantly from the diverse class (0.0121), and it was a garlic-and-egg class (0.0490) that fell fastest. The authors chose these class solutions for simplicity and interpretability, and call them descriptive. Meanwhile, a high meat-and-egg class did not differ significantly from the diverse class on survival or rate of decline.

So what should a reader take from it? Eating a range of foods in old age went with slightly better survival, in one country's cohort, with effects the authors themselves call modest. Each 10-point higher late-life score went with about 3% lower hazard, and the recalled midlife score showed no independent link, but the authors warn against concluding that diet late in life matters more, since recall errors and uneven recall gaps cloud the comparison. The improved group did not differ significantly from the stable-high group either, which is encouraging but does not prove that catching up undoes anything. The score counted how often foods were eaten, not how much, and an observational study cannot show that changing a diet would change these outcomes.