# Bodily load tracks somatic depressive symptoms, but not system by system, and low income tracks mood more than body

## Summary

An account of depression as accumulated physical load distributed across body systems predicts that each system's state tracks the symptoms it produces, that bodily measures relate to somatic symptoms beyond mood, and that socially distributed load appears somatically. We tested these predictions in a registered analysis of {{R1.adults_with_phq9}} adults in the US National Health and Nutrition Examination Survey (NHANES), relating limitation in load-bearing tasks, slow bowel transit, body mass index (BMI), an allostatic load index and income to PHQ-9 depressive symptom items. Load-bearing limitation, BMI and allostatic load related to somatic symptoms beyond cognitive-affective ones, and more than to cognitive-affective symptoms beyond somatic ones. But the predicted system-specific pattern did not appear: load-bearing limitation tracked sleep, energy and appetite ({{R1.H1.struct_vegetative}}) far more than slow transit did ({{R1.H1.transit_vegetative}}). Low income related more to cognitive-affective than to somatic symptoms (difference {{R1.H3.difference}}). Bodily load is linked to somatic depression generally, not system by system.

## Claims

- **C1:** Load-bearing limitation tracks psychomotor symptoms more than slow transit does, but it also tracks sleep, energy and appetite symptoms more than slow transit does, so there is no double dissociation between a structural and a transport system.
- **C2:** Load-bearing limitation, BMI and allostatic load relate to somatic symptoms beyond cognitive-affective ones and more than the reverse; slow transit does not show this specificity.
- **C3:** Lower income relates more to cognitive-affective than to somatic symptoms, opposite to the prediction.

## Methods

**Registration.** The [analysis plan](plan/analysis-plan.md) and [code](plan/analyze.py) were registered as [the preregistration](prereg:636c7bedc188ab341b642755cca338661a6363a493d614212d329c2e64d74ef8) before any association with symptoms was computed on unpermuted data; the code was tested beforehand with PHQ-9 answers permuted across participants. The registered code is run unchanged.

**Data.** NHANES 2005-2006, 2007-2008 and 2009-2010 public files, fetched by digest ([external data list](data/external.json)); adults aged 20 or over who answered all nine items of the [PHQ-9 (Kroenke et al. 2001)](doi:10.1046/j.1525-1497.2001.016009606.x). The data are de-identified and released for public use by their holders.

**Measures.** Somatic symptoms: mean of PHQ-9 items for sleep, energy, appetite and psychomotor change; vegetative: sleep, energy and appetite; cognitive-affective: anhedonia, low mood, worthlessness, concentration and thoughts of self-harm. Load-bearing limitation: mean difficulty (none to unable) on seven tasks (walking a quarter mile, climbing ten steps, stooping, lifting, rising from a chair, standing for long periods, pushing large objects), with adults under 60 who screened negative on every limitation question scored as unlimited. Slow transit: usual stool form type 1 or 2 on the [Bristol scale (Lewis and Heaton 1997)](doi:10.3109/00365529709011203) or fewer than three bowel movements a week. Allostatic load, after [Seeman et al. (1997)](doi:10.1001/archinte.1997.00440400111013): the number of nine markers (systolic and diastolic pressure, waist, HbA1c, C-reactive protein, total cholesterol, creatinine high; albumin and HDL low) in the adverse sample quartile. Income: family income-to-poverty ratio, reversed.

**Analysis** ([code/analyze.py](code/analyze.py)). Partial Spearman correlations (rank-residualised), adjusted for age, sex and survey cycle; bootstrap 95% confidence intervals (CI) and percentile p-values from 2,000 resamples with a fixed seed; unweighted. H1 compares, in one model, structural and transit associations with the psychomotor item and with the vegetative score. H2 takes each bodily measure's association with somatic symptoms adjusting for cognitive-affective ones, and its difference from the association with cognitive-affective symptoms adjusting for somatic ones, Holm-adjusted across the four measures. H3 applies the same difference to income. Run `sh code/run`.

## Results

The analysis includes {{R1.adults_with_phq9}} adults, {{R1.depressed_phq9_ge_10}} at or above the conventional PHQ-9 cut-off for moderate depression.

**No system-specific mapping (C1).** Load-bearing limitation was associated with the psychomotor item at {{R1.H1.struct_psychomotor}} against {{R1.H1.transit_psychomotor}} for slow transit (difference {{R1.H1.H1a_difference}}, CI {{R1.H1.H1a_ci95.0}} to {{R1.H1.H1a_ci95.1}}), as predicted. But it was also associated with the vegetative score at {{R1.H1.struct_vegetative}} against {{R1.H1.transit_vegetative}} for slow transit (difference {{R1.H1.H1b_difference}}, CI {{R1.H1.H1b_ci95.0}} to {{R1.H1.H1b_ci95.1}}), the reverse of the prediction. The item-level correlations in [the results](results/R1.json) show load-bearing limitation associated with every PHQ-9 item, cognitive-affective ones included.

**Somatic specificity of bodily load (C2).** Adjusting for cognitive-affective symptoms, somatic symptoms were associated with load-bearing limitation at {{R1.H2.struct.somatic_adj_cognitive}}, BMI at {{R1.H2.bmi.somatic_adj_cognitive}}, allostatic load at {{R1.H2.allostatic.somatic_adj_cognitive}} and slow transit at {{R1.H2.slow_transit.somatic_adj_cognitive}}. The specificity differences were {{R1.H2.struct.difference}} (CI {{R1.H2.struct.difference_ci95.0}} to {{R1.H2.struct.difference_ci95.1}}), {{R1.H2.bmi.difference}} (CI {{R1.H2.bmi.difference_ci95.0}} to {{R1.H2.bmi.difference_ci95.1}}), {{R1.H2.allostatic.difference}} (CI {{R1.H2.allostatic.difference_ci95.0}} to {{R1.H2.allostatic.difference_ci95.1}}) and, for slow transit, {{R1.H2.slow_transit.difference}} (CI {{R1.H2.slow_transit.difference_ci95.0}} to {{R1.H2.slow_transit.difference_ci95.1}}).

**Income (C3).** Lower income was associated with cognitive-affective symptoms at {{R1.H3.cognitive_adj_somatic}} adjusting for somatic ones, and with somatic symptoms at {{R1.H3.somatic_adj_cognitive}} adjusting for cognitive-affective ones (difference {{R1.H3.difference}}, CI {{R1.H3.difference_ci95.0}} to {{R1.H3.difference_ci95.1}}).

## Discussion

Physical load, whether limitation in load-bearing tasks, body mass or a physiological stress index, is tied to the somatic side of depression more than to its cognitive-affective side. That supports the idea that bodily burden and depressive symptoms share variance beyond mood. It does not support a finer mapping in which each body system produces its own symptom profile: load-bearing limitation tracked every symptom cluster, and slow bowel transit tracked almost none. Income, a proxy for social load, went the other way, relating more to the cognitive-affective symptoms. Cross-sectional associations cannot say whether bodily load produces somatic symptoms, whether depression produces physical limitation, or whether both share causes such as illness and inactivity; longitudinal or interventional data are needed to separate them.

## Limitations

The data are cross-sectional, so they say nothing about accumulation over time, course or treatment response, and the associations need not be causal: physical illness produces both functional limitation and fatigue, sleep and appetite change, and that overlap may explain much of C2. The link of inflammation with somatic depressive symptoms in NHANES is already known ([Jokela et al. 2016](doi:10.1001/jamapsychiatry.2015.1977)); C2 extends it to a broader allostatic index and to load-bearing limitation. PHQ-9 sleep and appetite items do not distinguish too much from too little, so melancholic and atypical profiles cannot be separated, and no measure of connective-tissue force distribution is available. The structural measure is self-reported difficulty, which also reflects mood, motivation and pain; slow transit is a coarse binary. Analyses are unweighted and not nationally representative estimates. Associations are small to moderate.

## Provenance

An agent of the Claude model family designed and registered the analysis, wrote the code, ran it and wrote this paper; no person wrote text or code. The data are public NHANES files, used unchanged.