A facet of the sleep cluster (nucleus Sleep Duration and Mortality), on a different outcome set
— incident type 2 diabetes and adiposity, rather than mortality. Two 2015 meta-analyses answer two
halves of one question and only together: is short sleep causally metabolically harmful, or does the
association merely mark illness? Shan supplies the observational shape (on T2D); Capers supplies the
experimental (RCT) arm — which direction-fixes the energy-intake link and so rebuts the
reverse-causation objection, rather than corroborating the T2D magnitude itself (Capers’ net-weight and
appetite-hormone endpoints were null). The composite is stronger than either alone.
(Capers et al., 2015; inferred from Shan et al., 2015)
The observational shape — a second sleep U-curve, on T2D [Shan 2015]
Shan pooled 10 prospective cohorts (11 reports, 18,443 incident T2D cases among 482,502 people, follow-up 2.5-16 y):
«A U-shaped dose-response relationship was observed between sleep duration and risk of type 2 diabetes, with the lowest risk observed at a sleep duration category of 7-8 h per day. Compared with 7-h sleep duration per day, the pooled relative risks for type 2 diabetes were 1.09 (95% CI 1.04-1.15) for each 1-h shorter sleep duration among individuals who slept <7 h per day and 1.14 (1.03-1.26) for each 1-h increment of sleep duration among individuals with longer sleep duration.» (Shan et al., 2015)
Category extremes: shortest (<=5 h) RR 1.37 (1.18-1.59); longest (>=9 h) RR 1.40 (1.08-1.80). Absolute scale (US incidence 7.6/1000/yr): ~68 extra T2D cases/100,000/yr per 1-h reduction; ~106 per 1-h increment. (Shan et al., 2015)
The two arms are NOT the same finding — apply the arm-level rule
This is The U-Shaped Association Artifact on a second exposure-outcome. Adjudicate the arm, not the curve:
- Short arm — causal footing. Shan reports no nonlinearity (P=0.22) and cites experimental corroboration: lab sleep-restriction studies show reduced glucose tolerance / insulin sensitivity, increased hepatic glucose production, sympathetic activation. A named, human-corroborated mechanism.
- Long arm — artifact-suspect, and left unadjudicated. The long-arm mechanism is weaker on Shan’s own account: «The potential mechanisms underlying the association between long sleep duration and increased diabetes risk are currently considered more speculative.» (Shan et al., 2015) The candidate explanations are the concept’s recipe — depression, low SES, undiagnosed disease, poor physical health, low activity, sleep apnoea — and Shan concedes reverse causation directly: «it is also possible that long sleep is a consequence of the sleep-inducing effects of the inflammatory state». (Shan et al., 2015) The tells match: the long arm carries higher heterogeneity (dose-response I2 79% vs 63% short) and is not robust to one study — omitting Tuomilehto renders it borderline (1.09, 0.99-1.12). (Shan et al., 2015)
Shan does NOT run the strong check (no referent-correction, no Mendelian randomization). Multivariable adjustment (which the long-T2D association survives) is the weak check — the same the alcohol artifact survived. So the long-sleep -> T2D arm is unadjudicated, not established; only the short arm is believed. This is the third sleep-arm instance confirming adjudicate-the-arm (Cappuccio mortality, Shan T2D). Not independent of Cappuccio — overlapping cohorts + shared Hu/Jackson lineage, so type-F (new outcome, same reverse-causation mechanism), not a fresh independent route.
The experimental arm removes pure reverse causation [Capers 2015]
The decisive weakness of any observational U-curve is that illness lowers sleep. Capers meta-analyzed RCTs that manipulated sleep, so its direction of causation is fixed by design:
«The available experimental literature suggests that sleep restriction increases food intake and total energy expenditure with inconsistent effects on integrated energy balance as operationalized by weight change.» (Capers et al., 2015)
- Food intake — consistent up. Across the food-intake studies (narratively pooled — measurement too varied to meta-analyze), sleep restriction reliably increased ad libitum intake, portion size and food purchasing. This is the causal link the observational data cannot prove.
- Extending sleep — modest weight benefit. Sleep-extension trials showed «overall significant effects on body weight, p = 0.02, with low heterogeneity I2 = 35%». (Capers et al., 2015)
- Net weight from restriction — inconsistent. Restriction gave weight gain / less loss in 3 of 4 trials but «no significant effect overall (p = 0.08)», I2=81%, and total energy expenditure also rose (NS) — intake up and expenditure up, so the net is unresolved. (Capers et al., 2015)
The mechanism refinement — leptin/ghrelin is NOT the operative pathway [type-F]
Sleep Duration and Mortality and Shan both invoke the leptin-down / ghrelin-up appetite axis as the short-sleep mechanism. The RCT evidence does not support it. Capers found no significant pooled effect of sleep restriction on leptin (p=0.84, I2=75%) or ghrelin (p=0.83, I2=77%), and cites Chaput & St-Onge:
«hormones such as ghrelin and leptin are not the main mechanism contributing to increased food intake during sleep restriction because hormone levels depend on sleep timing and the nutritional state of the participants. They assert that hedonic factors are likely stronger.» (Capers et al., 2015)
So the composite beats either source alone: the intake increase is real (Capers), but the popular hormonal explanation for it is downgraded — likely hedonic drive plus more waking hours to eat, not a leptin/ghrelin surrogate. Route the appetite-hormone story as mechanism only, discounted. (inferred from Capers et al., 2015)
Decision relevance
- Short sleep is a modifiable metabolic lever with causal footing. Chronic <7 h (especially <5 h) raises T2D risk (~9%/h observationally) with both a mechanism and RCT corroboration that restriction increases energy intake. Moving a chronically short sleeper toward ~7-8 h is the actionable move.
- The magnitude on weight is uncertain, not established. The causal RCT arm is short-term and net weight change is inconsistent — sleep is not a demonstrated weight-loss intervention; its firmer metabolic signal is on glucose/T2D, not the scale.
- Long sleep is a marker, not a target. Shortening a long sleeper’s sleep optimizes a symptom of occult illness, not a cause — the long arm is unadjudicated for T2D as for mortality.
- Ranking (layer 1): a moderate-effect, moderate-certainty lever — below the big rocks, above most refinements; relevant chiefly for the habitually sleep-deprived stratum.
Limits
- Shan: all exposure self-reported, single-timepoint questionnaires — actigraphy and polysomnography (the gold standard) were not used, so exposure misclassification is likely -> Measurement Error in Dietary Assessment; a meta-analysis of observational data cannot directly control for residual or unmeasured confounding. (Shan et al., 2015)
- Capers: few, small, short RCTs (mostly crossover, n<=20, most <4 weeks, only 2 >=4 weeks); sleep cannot be blinded (>60% did not blind assessors), and no long-term trial exists — the causal arm is direction-fixed but under-powered on magnitude.
- Coherence, not validity (R1): the U-shape is what the cohorts report; the short-arm causal reading rests on the experimental arm + mechanism, not the pooled association alone. (inferred from Capers et al., 2015)